Datasets:

TomTBT

/

pmc_open_access_section

like 1

PMC10786460

38223395

Introduction “Aging with Disability” is a term that refers to people living with the long-term effects of disabilities acquired early in life who are now surviving into mid- and later life [ 1 ]. Increased life expectancy among those with disabilities acquired at birth (e.g., cerebral palsy, spina bifida) is attributed to advancements in medicine, technology, and public health [ 2 , 3 ]. From a life course perspective, individuals aging with disability experience the dynamic processes of aging superimposed on their disability, putting them at risk of worse health outcomes compared to those who develop a disability in later life [ 1 ]. Evidence suggests that adults aging with disability exhibit signs of accelerated or “premature” aging and are likely to enter mid- to late-life in worse health than the general population [ 1 , 4 ]. Despite sparse data at the population level in the United States, clinical and survey research indicate that chronic health conditions in people aging with disability typically occur about 20–25 years sooner than those without disability [ 5 ]. Compared to those without disability, individuals aging with disability experience higher rates of chronic disease, worse self-rated health, and premature mortality [ 6 - 9 ]. Additionally, there are risks of developing secondary health conditions, such as increased spasticity, osteoporosis and osteoarthritis that originate directly or indirectly from the primary disability [ 10 - 12 ]. Thus, individuals aging with a physical disability have complex healthcare needs, including appropriate care for their primary disability, routine preventive services (e.g., screenings), and care for chronic and secondary conditions, that require ongoing medical monitoring and coordinated care management over time [ 13 , 14 ]. Continuity of care (COC) across settings and over time is considered a key metric of many care delivery models [ 15 - 17 ], especially for older adults with multiple chronic conditions that require comprehensive medical management [ 18 ]. For adults aging with disability, complex care needs must be met by professionals with a range of skills and specialties [ 13 ], and continuous care reflects the degree to which these services are delivered in coordinated and uninterrupted succession over time [ 19 ]. There is ample evidence that people with disabilities receive poor standard healthcare, including disparities in screening and preventive services, cancer diagnosis and treatment, reproductive and pregnancy care, communication with health care professionals, and satisfaction with care [ 14 , 20 - 25 ]. People with disabilities also encounter a multitude of environmental barriers to accessing care, including lack of transportation and distance to treatment centers, which create challenges for continuous care delivery [ 21 , 26 - 28 ]. Thus, proximity to healthcare facilities, availability of primary care and specialist providers, and access to transportation are critical for supporting continuous care for people aging with disability. Yet, COC in adults aging with disability and the community factors that contribute to care continuity in this population are largely unknown. Person-level factors associated with high COC in the general population have been well characterized, including older age, female sex, white race and fewer comorbid health conditions [ 29 , 30 ]. There is also emerging evidence on the role of the community environment in care continuity amongst individuals with psychiatric disabilities, where a greater density of mental health centers and practicing psychologists were associated with more continuous care [ 31 ]. However, for individuals aging with a physical disability, there may be additional community features that are important for accessing care, including public transit and broadband internet access (for telehealth), and access to tertiary care specialists [ 22 , 32 - 34 ]. In order to address the limitations in the current state of knowledge, the current work leverages data from a large nationwide medical claims database to examine COC in a cohort of adults aging with cerebral palsy and spina bifida in the United States. Cerebral palsy (CP) is the most common pediatric onset disability with increased survival in recent decades [ 35 ], and spina bifida (SB) is a congenital birth defect that often results in severe life-long disability and morbidity [ 9 ]. Individuals with CP and SB have significant and progressive motor impairment, excessive sedentary behavior, inadequate muscle and bone development, and are at risk of secondary chronic disease as they age [ 36 , 37 ]. In this study we provide one of the first estimates of COC in adults aging with CP or SB. Using residential ZIP codes in medical claims data linked with geographic data sources, we examined how proximity to health care facilities, availability of care providers, and accessible environments were associated with more continuous care. The overall aim of the study was to examine the association between community factors and care continuity in adults aging with CP/SB. Understanding the factors associated with greater COC is important for identifying individuals aging with disability who are at risk of fragmented care, thereby informing appropriate clinical and population-level interventions.

Materials and Methods Data Source and Analytical Cohort Data for this study were obtained from Optum’s de-identified Clinformatics ® Data Mart (CDM) Database (2007–2018). This is a nationwide, single private payer, administrative health claims database for over 80 million beneficiaries with commercial and Medicare Advantage health plans in the United States. Data on patient demographic characteristics, inpatient/outpatient records, diagnoses, procedures, and filled prescriptions are available. We used the Clinical Modification (ICD-9-CM) codes from the International Classification of Diseases 9th Edition to identify adults (age 18+ years) with a CP or SB diagnosis (see Supplementary Table S1 ). Inclusion criteria required that individuals had at least four years of continuous enrolment on the insurance plan to ensure stable membership ( N = 15,456). Individuals were excluded if they had both a CP and SB diagnosis ( N = 256) given the lack of clinical plausibility and different disease etiologies. We also excluded 6604 individuals with less than 4 outpatient visits in the year following the date they enrolled on the plan, in order to compute estimates of COC [ 30 ]. The final analytic sample consisted of 8596 individuals. To make linkages with data on community characteristics, residential ZIP codes were obtained from CDM. However, when CDM provides ZIP codes to researchers, information on individual-level income, education and race is removed to protect patient privacy. Since secondary data analyses of de-identified datasets cannot be tracked to a human subject, this study was reviewed and categorized as exempt human subjects research by the University of Michigan Institutional Review Board. Measuring Continuity of Care Continuity of care was measured using outpatient evaluations and management visits (with unique provider identification numbers) in the one-year period following their enrollment. The Bice-Boxerman COC Index, which captures the degree to which a patient’s outpatient/office visits are concentrated among providers [ 38 ], was calculated using the following formula: ((Σ i =1 n i 2 ) − N/(N(N − 1)) (where N is the total number of visits, and ni is the number of visits with the provider i ). COC scores range from 0 to 1 with higher scores indicative of a greater share of total visits concentrated within a few unique providers. The Bice-Boxerman COC score has no inherent meaning and needs to be converted from a continuous to binary or categorical variable for interpretation [ 30 ]. Since there is no widely accepted cut-off value, we operationalized COC as a binary variable, consistent with previous studies [ 39 ], where individuals with scores above the median (>0.25) were considered to have high COC (high continuity; concentrated care) and those at or below the median were considered to have low COC. To describe the types of providers visited and the frequency of those visits, we also calculated the proportion of all visits in the one-year period to different types of health care providers using the physician-reported specialty in the CDM provider data file (e.g., internal medicine, family/general medicine, obstetrics/gynecology). When clinicians reported more than one specialty, the first reported specialty was used. Measures of Community Characteristics Measures of the community environment were obtained from the National Neighborhood Data Archive (NaNDA). NaNDA is a publicly available data archive containing contextual variables derived from a variety of data sources and available at various spatial scales. The measures were linked to the study cohort using residential ZIP codes converted to ZIP Code Tabulation Areas (ZCTA), which are spatial representations of ZIP codes generated by the U.S. Census Bureau with an average population of about 9000 people. NaNDA variables were selected a priori based on associations with COC noted in the literature, or features of the environment that may impact availability and/or accessibility of healthcare services, providers or facilities and affect the ability to maintain a continuous relationship with a set of providers [ 31 ]. Data on healthcare establishments included counts of ambulatory care services, hospitals, and residential/skilled nursing facilities in each ZCTA [ 40 ]. The availability of broadband internet was based on the number of households with any broadband internet connections per ZCTA [ 41 ]. Because public transit may be an important means for accessing health services among people with disability [ 34 ], we also included data on the number of public transit stops [ 42 ]. For all neighborhood characteristics, we used a measure of per capita density (count divided by total ZCTA population). Spatial accessibility to healthcare providers, including family medicine doctors (FM), nurse practitioners (NP), medical specialists, and chiropractors, were created using the Variable-distance Enhanced 2 Step Floating Catchment Area (VE2SFCA) method, which includes a distance decay weight accounting for travel time, and a metric of provider to population ratio in each ZCTA [ 43 ]. NaNDA measures of neighborhood socioeconomic affluence and disadvantage were used to capture a broader indicator of neighborhood investment and disinvestment [ 44 ]. Values range from 0 to 1 with higher scores indicating higher levels of disadvantage or affluence. In order to account for non-linearity in the relationships between neighborhood resources and health [ 45 , 46 ], all neighborhood variables were operationalized as tertiles (T1 = low, T2 = medium, T3 = high). Due to high collinearity between FM and NP availability, we created a composite measure to capture combinations of these providers as follows: low spatial accessibility (low FM and NP, or low/medium FM/NP), medium spatial accessibility (medium FM and NP, or low/high FM/NP), and high spatial accessibility (high FM and NP, or high/medium FM/NP) (See Supplementary Table S2 ). Statistical Analyses Generalized estimating equations logistic regression models were used to examine the relationship between community factors and odds of high COC, adjusting for individual factors that could increase the risk of worse COC in poor resource neighborhoods. Covariates included age (categorized as age 18–40, 41–64 and 65+ for analysis), sex (male or female), and comorbid conditions (Elixhauser Comorbidity count (range 0–31)) [ 47 ]. Year was included to account for both structural improvements in neighborhoods and changes in healthcare policy over the 7-year period in which individuals entered the study cohort. Tests were 2-sided and significance was assessed at p < 0.05. All analyses were conducted in SAS version 9.4 (SAS Institute, Cary, NC, USA).

Results Descriptive statistics for the study cohort are presented in Table 1 by levels of care continuity. Individuals aging with CP/SB were around 50 years of age on average and 59% were female. Individuals had almost 3 comorbid health conditions on average. The mean COC score was 0.30 overall. Amongst individuals with high COC (categorized as above the median >0.25) the mean Bice-Boxerman score was 0.52, compared to 0.14 for those with low COC (at or below the median). Individuals who had more concentrated care were older than those with low COC (mean age 49.9 years vs. 47.3 years, respectively). Compared to males, females were over-represented in the group with low COC (59% of those with high continuity were female compared to 64% in the low COC group). Co-morbidity burden was similar in those with high or low COC. Individuals with low and high COC also varied in terms of the characteristics of the communities in which they lived ( Table 1 ). Individuals with high COC were more likely to live in areas with a higher density of residential care/skilled nursing facilities (34.7% vs. 32.0% for high vs. low COC respectively). Compared to those with more continuous care, individuals with low COC were more likely to live in areas with more broadband internet connections (36.6% vs. 30.0%), a greater proximity to a variety of healthcare providers (chiropractors (35.1% vs. 31.5%), medical specialists (35.0% vs. 31.6%), and FM/NP (40.9% vs. 37%, respectively), and in areas characterized as more socioeconomically advantaged (e.g., 37.8% of those with low COC resided in highly affluent communities vs. 28.7% of those with high COC). Figure 1 depicts the distribution of healthcare visits to different specialty types across those with high and low care continuity. Individuals aging with CP/SB, irrespective of COC score, saw more than 14 different provider specialties including orthopedic, neurology, and psychiatry specialties. The most common specialties seen for both groups were family/general medicine (FM/GM) and internal medicine (IM) physicians. However, amongst those with high COC, a greater proportion of total visits were concentrated in these primary care providers. For example, 55% of total visits in those with high COC were to IM and FM/GM specialties compared to 45% for their counterparts with low COC ( Figure 1 ). Compared to those with high COC, a greater share of the health care visits of individuals with low COC were spread between a variety of different types of specialties, including obstetrics/gynecologists (OBGYN) (4% vs. 2.8%), orthopedics (6.2% vs. 5%) and dermatologists (3.9% vs. 2.8%). Table 2 presents the odds ratios (OR) and 95% confidence intervals (CI) from the multivariable logistic regression model examining the association between community characteristics and the odds of receiving high COC (vs. low COC). The model adjusts for individual age, sex, comorbidity count, and year. After adjusting for individual factors, a greater density of hospitals and residential care facilities was significantly associated with receiving high COC. Residing in areas with a lower density of hospitals was associated with 16% lower odds of high COC (OR for medium vs. high density: 0.84, 95% CI: 0.72–0.98). Compared to areas with a high density of residential care/skilled nursing facilities, areas with low density were associated with 28% lower odds of concentrated care (OR 0.72, 95% CI: 0.59–0.88). Low accessibility of primary care providers (FM and NP) was significantly associated with more concentrated care, net of individual and other community characteristics (low vs. high: OR 1.26, 95% CI: 1.09–1.46). No significant findings were observed for spatial accessibility to medical specialists or chiropractors. Adjusting for health care resources, residence in less affluent areas was associated with higher odds of receiving concentrated care (low vs. high affluence: OR 1.55, 95% CI: 1.29–1.86), with a dose-response relationship observed. No significant associations were found for density of transit stops or broadband internet access and COC ( Table 2 ).

Discussion In this large nation-wide cohort of members from large commercial and Medicare Advantage health plans in the United States, we provide one of the first characterizations of COC in adults aging with CP/SB, and highlight how proximity to health care facilities, availability of care providers, and community socioeconomic context are associated with more continuous care. Continuity of Care in Adults Aging with CP/SB CP and SB are congenital conditions that result in lifelong limitations in movement. As they age, individuals with CP and SB have progressive motor impairment, excess sedentary behavior, inadequate muscle and bone development, and increased risk for obesity and other high-burden medical conditions (e.g., chronic pain, osteoporosis, and cerebrovascular disease) [ 48 ]. These complex care needs create challenges in communicating across different healthcare providers, with increased risk for potentially preventable psychological, cardiometabolic, and musculoskeletal morbidities in adulthood [ 49 ]. It is thus not surprising that we found that individuals aging with CP/SB saw a variety of different specialty types and generally had discontinuous care. Extensive work has been done to characterize COC in older adults or those with chronic health conditions, but there is a dearth of research on care patterns in the growing number of adults aging with CP/SB. In comparison to findings in the general population, our results suggest that individuals aging with congenital disabilities have lower mean COC scores. For example, Medicare/Medicaid patients >65 years with a diagnosis of Diabetes, Congestive Heart Failure, and Chronic Obstructive Pulmonary Disease, had mean Bice-Boxerman COC scores of 0.50, 0.55 and 0.60, respectively [ 50 ]. These scores are notably higher than that which we observed in our study, where the mean COC score was 0.30. Amongst those with high continuity, visits were more concentrated in primary care providers (PCP) (i.e., FM, NP and IM). In a previous study of Medicare beneficiaries with multi-morbidity, those reporting a specialist (compared to PCP) as their primary care provider had worse COC score [ 51 , 52 ]. PCP as the central provider may promote better coordination by comprehensively managing conditions and referring to specialists only when necessary [ 51 ]. Whereas previous work has largely used Medicare data (>65 years of age), our use of private claims data meant we included a younger cohort of women. For these women, visits to obstetrics/gynecology specialists for reproductive health needs may represent an important source of care and increase the number of specialists seen. Research has highlighted the lack of adequate training to address reproductive health needs of women with disabilities, which might challenge one’s ability to maintain consistent care [ 53 ]. Community Resources and Care Continuity We found that the community context was related to the level of care continuity received. Individuals who lived in areas with more hospitals and residential care facilities received more continuous care than those with limited availability of these resources. Continuous care is more likely when complex medical care can be received in the same location [ 19 ]. For those seeing multiple providers, integration of services is more likely when care is confined to a single setting [ 19 ]. In hospital settings, there is organization of care, and a care manager, allowing for greater coordination across different types of providers and greater consistency in providers seen. Hospitals also have care coordinators available to manage inpatient and outpatient care, which have been noted in studies to have positive effects on the patient-provider relationship and to reduce coordination problems among patients with complex health care needs [ 54 , 55 ]. Hospital staff often play a role in care plans and arrange for follow-up care, which may not be available in outpatient settings in the community [ 56 ]. Similarly, in residential care settings, where the majority of older adults with disabilities reside [ 57 ], care coordination can be provided within a single institution. Adjusting for health care resources, residence in more affluent areas was associated with receiving more fragmented care. Affluent neighborhoods are not just indicative of low disadvantage, but also specific norms such as higher levels of social control and leverage over local institutions that foster environments supportive for health [ 58 ]. Healthcare providers often prefer to practice in more affluent areas where residents have more discretionary income, thereby affording individuals with greater choice and shorter travel times to different types of healthcare providers [ 51 , 59 ]. But beyond provider density, COC may be more fragmented in affluent communities where residents tend to have higher levels of health literacy, which may facilitate provider “shopping around” for multiple care providers [ 60 , 61 ]. In many managed care organization health plans, individuals must be proactive and advocate for their healthcare needs from a variety of healthcare providers. Similarly, lower spatial availability of FM/NP was associated with receiving more continuous care. Out of network providers are generally not covered by insurance in managed care organizations making it less likely for individuals who reside in areas with less availability of FM/NP to seek care from multiple different providers outside their communities and insurance networks [ 22 ]. Despite expectations, the availability of public transit and broadband internet were not found to be associated with COC after adjusting for other community and individual factors. Strengths and Limitations Claims-based measures of care continuity provide a comprehensive record of billed services for beneficiaries aging with congenital disabilities [ 38 ]. Claims data provide a more accurate measure of health care encounters across unique physicians and specialty types, mitigating misclassification of the outcome attributed to recall bias in self-reports of care continuity. However, medical diagnoses do not necessarily reflect disability, and we were unable to investigate potential variation in care continuity across the severity of disability in those with CP and SB. Because these data are intended for administrative purposes, there are inherent limitations due to errors in coding. In addition, claims-based measures of COC are not necessarily reflective of health care quality or patient-reported continuity measures [ 38 , 62 ]. The COC index that we used in this study does not capture other important dimensions of coordinated care, including direct communication or co-management between clinicians, and does not consider a patient’s perception of an integrated relationship with their providers. Moreover, calculation of the Bice-Boxerman COC Index requires at least 4 outpatient visits, which meant that we excluded adults with less frequent engagement in the healthcare system who may be in better health. We categorized high and low COC based on the median value of the Bice-Boxerman COC Index, but future research should consider alternate thresholds for high and low COC in this population, perhaps in conjunction with qualitative research that informs the meaning of care continuity in adults aging with CP and SB. The use of private health insurance data in this study limits generalizability and future studies should conduct research using data from publicly insured individuals. By using a measure of spatial accessibility of health care providers that accounted for boundary effects and distance decay functions, we addressed commonplace limitations of density-based measures of spatial accessibility and more accurately reflected the availability of these service providers. However, we were unable to account for the quality of the infrastructure in the built environment. It is plausible that despite a high density of community healthcare facilities, these may be inaccessible for individuals with a physical disability. If this was the case, we likely underestimated the true association between healthcare availability and COC. Our study did not seek to identify causal relationships between community factors and COC, and we were unable to measure the actual use of community resources. Mixed methods research should aim to better understand the relationship between availability, access and use amongst individuals aging with disability. Finally, due to patient confidentiality, information on individual socioeconomic status (SES) was not available when requesting the geographic identifiers in CDM. Although we adjusted for area-level SES, which is a key factor for understanding the socioeconomic experience of populations, neighborhood SES is not a strong proxy for individual SES [ 63 ]. Thus, residual confounding by individual SES may result in an over-estimate of the effect estimates in our results.

Conclusions Collectively, these findings suggest that individuals aging with CP/SB represent a population particularly prone to fragmented care, perhaps more so than other complex-care populations. Adults living with pediatric onset disabilities in the United States report challenges accessing appropriate care once they transition out of the pediatric care setting [ 49 ]. Policy solutions to address care fragmentation have largely focused on shifting payment structures, use of electronic medical records, and decreasing specialty care [ 64 ]. Our work contributes to existing calls to promote systematic, coordinated care for people with CP and SB throughout the life course [ 49 ] by drawing attention to the spatial distribution of health care resources at the community level, which must also be incorporated when considering ways to address health care disparities in this medically underserved and high-risk population.

Author Contributions: Conceptualization, A.M.K., E.M. and P.C.; Formal analysis, A.M.K. and P.L.; Methodology, A.M.K., P.L., N.K. and E.M.; Resources, N.K.; Supervision, P.C.; Writing—original draft, A.M.K.; Writing—review & editing, A.M.K., P.L., N.K., E.M. and P.C. All authors have read and agreed to the published version of the manuscript. Continuity of care is considered a key metric of quality healthcare. Yet, continuity of care in adults aging with congenital disability and the factors that contribute to care continuity are largely unknown. Using data from a national private administrative health claims database in the United States (2007–2018). we examined continuity of care in 8596 adults (mean age 48.6 years) with cerebral palsy or spina bifida. Logistic regression models analyzed how proximity to health care facilities, availability of care providers, and community socioeconomic context were associated with more continuous care. We found that adults aging with cerebral palsy or spina bifida saw a variety of different physician specialty types and generally had discontinuous care. Individuals who lived in areas with more hospitals and residential care facilities received more continuous care than those with limited access to these resources. Residence in more affluent areas was associated with receiving more fragmented care. Findings suggest that over and above individual factors, community healthcare resources and socioeconomic context serve as important factors to consider in understanding continuity of care patterns in adults aging with cerebral palsy or spina bifida.

Supplementary Material

Funding: The contents of this manuscript were developed under a grant from the National Institute on Disability, Independent Living, and Rehabilitation Research (NIDILRR grant number 90RTHF0001). NIDILRR is a Center within the Administration for Community Living (ADL), Department of Health and Human Services (HHS). The contents of this manuscript do not necessarily represent the policy of NIDILRR, ACL, or HHS and you should not assume endorsement by the Federal Government. Data Availability Statement: As part of a Data Use Agreement, authors are not allowed to share the data. Upon reasonable request, the first author will provide statistical programming code used to generate results.

CC BY

no

2024-01-16 23:35:04

Disabilities (Basel). 2023 Jun 12; 3(2):295-306

oa_package/97/ba/PMC10786460.tar.gz

PMC10786622

38223535

Background The widespread adoption of electronic health records (EHRs) affords an unprecedented opportunity to leverage clinical data for research. EHRs contain a large amount of clinical information that, when combined with other data sources such as ‘omics, enhances the potential for research discovery. The secondary use of EHR data does not involve subject recruitment or prospective data collection. It often qualifies for expedited rather than full institutional review board (IRB) review and may qualify for exemption if the data is deidentified. Vast reservoirs of such patient data derived from clinical encounters promises to accelerate patient-centered research and advance medical discoveries at decreased cost [ 1 ]. Data warehouses can store enormous quantities of structured, semi-structured, and unstructured data extracted from EHRs and other sources. When combined with other modalities such as images, claims data, public health outcome data, prescription data, radiology, lab databases, and wearables [ 2 ], these data are called a Health Data Warehouse, or Clinical Data Warehouse. The cloud-based clinical data warehouse at the University of Colorado is called Health Data Compass (HDC), and has been described elsewhere [ 3 ]. Health data warehouses increasingly support the storage and re-use of large quantities of clinically-derived electronic data for clinical and translational research, clinical operations, and quality improvement. These data are longitudinal and amenable to descriptive analytics or trend analysis to evaluate patient outcomes over time in a longitudinal patient record. Secondary use of well curated EHR data can be relatively inexpensive and efficient, and includes a vast amount of clinical detail not available from administrative claims data [ 4 , 5 ]. EHR data can be linked to additional data sources and provide a detailed and longitudinal approach to understanding health across the lifecycle. There are some caveats to these approaches, however. For example, to protect patient privacy these data necessarily have tightly restricted access controls. In addition, healthcare data can be difficult to analyze until carefully validated and transformed into analytic datasets. Healthcare data are also vast and heterogenous, requiring large data storage and delivery platforms to manage which come with their own complexities and expenses. Finally, healthcare data are famously non-interoperable and harmonizing with data from outside the original health system can pose significant challenges. Despite the many advantages of EHR data for research, knowing how to access and use these resources is often not intuitive and can lead to frustrations as researchers may require repeated attempts to get the data necessary to answer the desired research questions. The objective of this paper is to describe the opportunities, challenges, and mechanics of using a health data warehouse for clinical research. We explain the types of data available, limitations of EHR data, and steps necessary to obtain the data desired in a timely fashion.

Conclusions In summary, this review considers the advantages, potential drawbacks, and approaches to utilize EHR data for secondary research. We discuss methods to optimize efficiency in obtaining large amounts of data while protecting patient privacy, methodological and statistical considerations to maintain data integrity, integration with additional data sources, and ongoing privacy and security concerns. Health data warehouses have tremendous potential to support research and discovery. Repurposing EHR data for research is an iterative process that requires familiarity with institutional resources and process requirements. Advances in EHR data abstraction, integration with biobanks, and linkage across institutions are essential to fully realize the potential of health data warehouses for clinical research. Furthermore, inter-institutional collaboration and harmonization are requisite to support the translation of EHR-based research to inform clinical care.

Authors’ contributions CA wrote the manuscript; BS provided a preliminary draft of the manuscript; FM, AV, RP, EZ, and ME provided a critical review of the manuscript; IB critically reviewed the manuscript and provided extensive edits; AM conceived of the idea for the manuscript and provided extensive edits. The author(s) read and approved the final manuscript. Electronic health records (EHRs) and linked biobanks have tremendous potential to advance biomedical research and ultimately improve the health of future generations. Repurposing EHR data for research is not without challenges, however. In this paper, we describe the processes and considerations necessary to successfully access and utilize a data warehouse for research. Although imperfect, data warehouses are a powerful tool for harnessing a large amount of data to phenotype disease. They will have increasing relevance and applications in clinical research with growing sophistication in processes for EHR data abstraction, biobank integration, and cross-institutional linkage.

Main text Overview of using EHR data for research Data types and steps for data request Data abstracted from the EHR for secondary use research falls broadly into three categories, each with specific compliance expectations. Datasets are classified as identified, limited, or de-identified. Identified data contains personal health information (PHI), limited data has partial removal of PHI, and de-identified data does not include PHI. The investigator should request the minimum amount of PHI necessary to accomplish the goals of the research. Datasets containing PHI are necessarily managed with the greatest precautions. Identified data requires full IRB review while de-identified data often qualifies for exemption. Investigators using only limited data may obtain a waiver of consent. In addition, institutions may have specific requirements for data use not addressed by IRB regulations [ 6 ]. Prior to submitting a data request, we would strongly recommend that the investigator or their team has engaged with a statistician or analyst to accurately determine which data are likely to be needed, and to have a plan for post-abstraction cleaning and curation (e.g., one-hot-encoding of categorical variables, imputation of missing data). Knowing in advance what challenges the data may pose will often result in a faster data delivery and is excellent prep-to-research training. Once one knows which specific variables (and thus PHI) are needed, one might obtain IRB approval [ 7 , 8 ]. Furthermore, fostering a team-science approach from the outset allows subject matter experts to weigh in when appropriate. For example, certain data may not be in your clinical data warehouse and thus having a clinician on hand might facilitate locating these data. Although data analysts have access to the entire EHR, the EHR format can differ from what clinicians and investigators see when referencing the same information. An initial chart review that identifies the variables of interest and where they reside in the medical record will streamline the analyst request, subsequent data validation, and delivery process. The data analyst responsible for coding will match these specific variables with those appearing within the “back end” view of the EHR data warehouse to ensure that the investigator receives the correct data. Timelines and considerations for data acquisition The timeline for data deliveries often depends upon multiple factors. Administrative factors might include other IRB approvals, the availability of research information technology (IT) services to provision a suitable environment for data storage, or availability of specialist statistical software. Informatics factors will include the speed with which the analyst can identify and retrieve the data from the warehouse, the amount and complexity of data to be pulled, the availability of the data analyst team, and verifying that the data delivery is consistent with the request. Initial planning meetings with data analysts should aim to determine the scope of the project, including whether the project is a feasibility study, clinical research study, or dissemination and implementation study. As a first step, the investigator and analyst should use pertinent inclusion and exclusion criteria to define a specific cohort for investigation within the bounds of their approvals. Alternatively, the investigator may specify a cohort in advance and provide personal identifiers (e.g., medical record number or encounter number) as a guide for data extraction. The investigator will also specify the clinical variables requested. Being specific in this step facilitates accurate and efficient data retrieval. Although tempting for some investigators to ask for the entirety of a subject’s medical record, this approach results in unwieldy amounts of data that take longer to obtain and renders the data cleaning process excessively time consuming. Therefore, it is prudent for researchers to have spent time working with a statistician to identify what specific variables are essential to their study and only request these data. After the data delivery In many cases, and especially if receiving raw EHR data, the investigator should assess for accuracy and arrange for the data to be cleaned, and if necessary, mapped to a common data model. One method of ensuring accuracy involves verifying a subset of the data delivered with medical records by chart review. There is risk for miscommunication during the data request and retrieval process and there are multiple locations within the EHR where the desired variables might reside. Comparing the data delivered with manual chart review on a subset of the study cohort is critical to ensure delivery of the correct variables. Data cleaning is a crucial step in using EHR data for research. EHR data are originally intended for clinical care rather than research, and there is variation in specific data points that must be recoded for statistical analyses. For example, many factors influence how a drug is displayed within the medical record and how it is subsequently delivered in the final report for medication data. Factors that can make the same medication look different within the data report include: generic versus brand name, long acting versus immediate release formulation, which department ordered the medication, the form used to record medication history, the pharmacy that filled the prescription, the unit of measure, or the route of delivery. Laboratory values may also require review because reference ranges can vary depending on laboratory location and equipment used; each source of common lab tests will result in a separate data entry. For instance, sodium from a point of care machine, blood, and urine will all have their own separate discrete entries in the EHR. In fact, there are 146 different sodium lab values with distinct providence and structure in our local EHR. Although arduous, data cleaning is essential to ensure the integrity of the data prior to performing any analyses. Finally, the cleaned dataset can be linked with additional data sources, such as metabolomic or genomic data. The amount of server space required to store all files for ‘omics-based precision medicine studies can be very large. For example, a whole genome survey (WGS) file for one subject is at least 30 GB and a raw metabolomics dataset can be over 60 GB. These server space requirements should be anticipated in preparation for this stage of the study. The processes for integrating data from our data warehouse with the Biobank at the Colorado Center for Personalized Medicine (CCPM Biobank) have been detailed elsewhere [ 9 ]. Figure 1 provides an overview. Utilization of genomic data The logistics for genomic data delivery from CCPM are specific to the data available from Translational Informatics Services (TIS). These data are mapped to HDC by either medical record number (MRN), which matches to the person ID, or contact serial number (CSN), which matches to the encounter ID. As expected, the initial data acquisition steps remain the same with regard to meeting with the HDC analyst, defining variables of interest, and working through the data request process. More than 180,000 people have consented to participate in the CCPM biobank and more than 60,000 participants have been genotyped on the Illumina Infinium MEGA-EX chip. The investigators and faculty at the University of Colorado provided input regarding which single nucleotide polymorphisms (SNPs) are of greatest research interest with a significant focus on SNPs with known or suspected pharmacogenomic associations. The version of the Illumina Infinium MEGA-EX chip used to genotype CCPM biobank participants includes many of the SNPs they selected. This chip genotypes ~ 2.2 million SNPs with the ability to impute ~ 50 million SNPs. The Access to Biobank Committee (ABC) reviews applications for use of the CCPM biobank. The investigator must complete a study proposal form specific to the ABC to receive consideration for study approval. It is worth noting that the IRB protocol may need to be revised if using genotyping data, which is considered PHI. As with the IRB submission, the ABC may require revisions to the study protocol or additional clarifications prior to granting access to biological specimens, genotyping data, PHI, and recontact of participants. After obtaining approval, the investigator must decide how to receive genomic data. This will depend primarily on available storage space and analytical capabilities. TIS delivers genotyping data in one of four ways depending on the investigator’s preference and experience. Each of these delivery methods has advantages and potential drawbacks. In the first, CCPM delivers non-imputed genotype data from the Illumina chip. This method requires significant time and effort on the part of the investigator, who must either possess or have access to the requisite expertise, but also offers flexibility in approach to statistical analyses. In the second, TIS delivers genotype data that is already imputed and has undergone quality control. This method may require significant storage space but the format is more accessible. TIS provides imputation rates with the data delivery such that the investigator can interrogate the data as appropriate, but the investigator does not have access to the original raw genotype dataset. A third option is that TIS delivers genotype data for a prespecified region of a gene or list of SNPs. These data are typically derived from prior WGS or genome wide association studies (GWAS). This data delivery method is relatively expeditious and most appropriate for candidate gene studies or validation studies. TIS staff are available to confirm which genes and SNPs have sufficient coverage on the MEGA-EX chip. This resource helps ensure that previously identified genomic associations are not excluded in the analyses. Finally, as a fourth option, TIS completes all imputation, quality control, regression analyses, and basic manuscript figure development. As a prerequisite, the investigator must provide the arbitrary identification numbers assigned by HDC, the outcome variables, and the study covariates. This data delivery method demands the fewest resources and the least amount of expertise from the investigator. Building the genotype and imputation pipeline takes 1–3 months as outcome variables are well defined and ancestry principle component analyses are validated. For established genomic pipelines, principle component analyses are relatively static, saving time for developing the genomic analysis pipeline. Following pipeline refinement, TIS provides the delivery via the server OneDrive, and the turnaround time after submission of the requisite variables is approximately 1 to 2 weeks. Challenges of using data warehouses for research EHRs are designed for clinical care, and there are challenges to consider when repurposing EHR data for research. These challenges stem from imperfect and fragmented data as well as patient privacy and security protections. Variability in EHR data format and quality EHR data include both structured and unstructured data. Structured data consist of discrete variables that capture controlled vocabulary. Common examples include laboratory values, medications, allergies, immunizations, vital signs, and encounter diagnoses. Unstructured data comprise free-text, narrative notes in the medical record. EHR data are complex and structured and unstructured data may overlap. For example, flowsheets used to record vital signs may include free-text descriptors. Both data types have advantages and drawbacks. Structured data reduce ambiguity but limit expressivity. They are relatively easy to retrieve, can reduce the time required for coding, and can streamline data analysis. Unstructured data provide complex and detailed clinical information, but with a spectrum of quality and completeness in documentation [ 10 ]. Natural language processing (NLP) is a form of machine learning that scans narrative data to extract computable results for analysis. Many NLP programs are currently available, although limited in ability to provide sufficiently detailed analysis to appreciate linguistic nuances and clinical concepts [ 2 ]. Defining clinical Phenotypes from EHR data Several methodological approaches are available to define clinical phenotypes in EHRs. These approaches include manual chart review, rule-based phenotyping, machine learning, and phenome-wide approaches (PheWAS) [ 11 , 12 ]. Each of these strategies requires a range of clinical and informatics expertise, and none comprehensively captures the information available in EHRs [ 11 , 12 ]. Since accurate phenotyping is crucial for correct identification of study cohorts, further refinement of phenotyping algorithms will inform rigorous EHR-based research. Clinical phenotyping algorithms facilitate the extraction of clinical data from diverse EHRs in a consistent manner. The Electronic Medical Records and Genomics (eMERGE) Network is a multisite network of health systems that combines DNA biorepositories with EHRs. eMERGE applies electronic algorithms to investigate phenotypic associations with genetic variants [ 13 ]. The network highlights the importance of accurate phenotyping and the potential for these methods to accelerate discovery research. Accuracy of EHR data Although EHRs contain vast amounts of data for clinical phenotyping, the utility of EHR data for research purposes is limited by EHR inaccuracy [ 8 ]. Common sources of EHR inaccuracy include diagnostic uncertainty, billing errors, and incomplete documentation. Diagnoses evolve over time, and a clinician may initially bill for a suspected diagnosis that is later deemed incorrect. A clinician may enter the wrong billing code or upcode diagnoses for higher reimbursement. Billing limited to a certain number of codes per visit often biases toward codes with the highest reimbursement such that “cheaper” codes are underrepresented in the medical record. Furthermore, clinical data are often fragmented since EHRs are not centralized. Patients may utilize multiple healthcare systems depending on insurance status and geography, and this can result in missed cases [ 2 ]. Missing data Missing data are a common problem in all areas of medical research, including EHR-based studies. The investigator should anticipate that data will be missing and have a method to address this as part of the study’s analytic plan. Missing values often do not occur at random, such that the amount of missing data and subject characteristics associated with missingness should be identified. Although subjects with missing data could be excluded from the analysis, this approach results in bias and loss of power. A better approach is to perform multiple imputation of missing values. Multiple imputation is a rigorous statistical technique that replaces each missing value with a plausible value based upon known characteristics of the dataset [ 14 – 17 ]. This allows for the inclusion of all subjects in the analysis, including those that would have been excluded due to missing values, and a less biased and more precise estimate of the outcomes under study [ 15 , 17 , 18 ]. Lack of discrete validated clinical outcome elements Many EHRs, including the EHR associated with the University of Colorado Health System, do not routinely include discrete tools for common outcomes. Validated clinical tools (e.g., HEART score for cardiac events, PHQ-9 for depression severity) would be helpful if incorporated into the medical record. Investigators could use scores from these tools, if routinely captured, to more efficiently collect and assess retrospective data. Privacy and security EHR-linked biobanks pose privacy and security concerns. They have historically operated under various consent models, with the traditional gold standard of informed consent considered problematic for four reasons: (1) Informed consent considers only the individual and does not account for close relatives; (2) Consent cannot be “informed” at the time it is obtained since future research on those samples is unknown; (3) Biobanks are a resource for a vast number of research projects such that obtaining informed consent for each project is impossible; and (4) It is unlikely that an individual’s right to later withdraw consent can be fully respected [ 19 ]. The NIH’s Genomic Data Sharing policy, which went into effect January 25, 2015, requires patients to consent to sharing of their DNA and made many existing opt-out consent models untenable [ 2 , 20 ]. EHR-linked biobanks are associated with various institutions and currently fragmented. Linking this data is desirable for large-scale research studies, although data become more difficult to deidentify with increasing quantities of linked data [ 21 ].

Funding This work was funded by NIH grants, F32HL160123 (CA) R35GM124939 (AAM) and CTSI UL1 TR001082. Availability of data and materials Not applicable. Abbreviations Electronic health record Institutional review board Health Data Compass Personal health information Information technology

CC BY

no

2024-01-16 23:35:02

Transl Med Commun. 2023 Mar 2; 8:7

oa_package/cf/bf/PMC10786622.tar.gz

PMC10786681

0

Physician and geneticist Dr. Funmi Olopade is the founding director of the Center for Cancer Genetics and Global Health at the University of Chicago ( Figure 1 ). Olopade’s research is focused on gaining a better understanding of the root causes and genomic basis of cancer in diverse populations. She is internationally renowned for her work in inherited cancer syndromes and for her clinical expertise in early detection and prevention of breast cancer in high-risk women. To hear her talk about doing physics problems for fun, see the full video on the JCI website at www.jci.org/videos/cgms JCI: What were you like as a child? Olopade: I grew up in Nigeria. I started school in a very small town called Ijebu Igbo, although I was born in Abeokuta. I’m number five of six children and was very quiet as a child. My father was a pastor. In his days, there were only three professions you could consider: a teacher, a pastor, or a doctor. He had the opportunity to go to school to be a teacher, but then he got into the missionary society and became a pastor. Being in a missionary family meant we grew up in very small towns and the focus was to get people educated and thinking about the world larger than themselves. Most of my early days were spent reading books and magazines and newspapers. We would pour over Time magazine every week when it got delivered like it was the best thing. Like many who grew up in a small town in Nigeria, you have to leave home to get the best education, so I went to boarding school when I was 12. By the time I was in sixth form, there were only 18 of us in the advanced science class in my all-girls school. We had wonderful teachers, and we tracked in either arts or science early on. I was quite strong in mathematics and physics. My brother had already become an engineer, and my father really wanted a doctor in the family, so after my O-levels, which is really where you did more science, I went to Queen’s College in Lagos. That’s where for A-levels, I did physics and dropped math for biology. I did not like my biology classes as much as physics, but still excelled to get to medical school. JCI: Did you have any early clinical inclinations, like toward cardiology or anesthesiology or surgery? Olopade: I enjoyed medicine and pathology the best. And Robbins’ Textbook of Pathology was so fun to read, and I had this habit of reading every book in medical school from cover to cover, and people thought that was nerdy, but I enjoyed it. Physiology was great and biochemistry too. Anatomy felt like a lot of tedious memorizations, but once we finished the preclinical part of medicine and got to the clinical years, it was so wonderful to be a medical student and to learn that we didn’t have all the answers. My pediatric rotation was fascinating because we had an American-trained pediatrician, and in medicine, we had a diabetologist who practiced the American way, which was very different from the British way. I began really thinking about whether any of those specialties would interest me. I left Nigeria still eliminating things that I didn’t like. Pathology, I would’ve done, except I didn’t like morbid anatomy; I couldn’t imagine having to do autopsies. In coming to America, I wanted to do internal medicine and I had role models who were in cardiology. We had done one heart transplant during my rotating internship in Nigeria. He was a kid with rheumatic heart disease, and it was the first cardiac transplant in Nigeria. I was on that service; I worked hard but unfortunately the patient died and I kept thinking it was a futile exercise. If the child had had penicillin, he would never have damaged his heart — that got me thinking about prevention. Those were moments that were transformative for me in thinking about how I wanted to spend the rest of my life as a doctor. JCI: How did you happen to move to Chicago for your medical residency? Olopade: One of my habits in medical school was to wake up to the radio program Voice of America . During the Iran hostage crisis, I loved Jimmy Carter and I loved everything that was happening in America. Given the profits from finding oil in Nigeria, we all went to medical school for free, and the government gave us some money to invest in our education that allowed us to discover the world. My brother was already in graduate school at Stanford, and so during the summer I decided to use my bursary to travel and visit him. I loved it. I thought everything was big in America. JCI: Well, that is often true. How did you pick Chicago and the program at Cook County for your residency? Olopade: I tried to interview at different places. Since my brother was at Stanford, I tried to get interviews at San Francisco General and Stanford. But those programs didn’t take international graduates. I got advice to start at a public hospital. We had a friend in Chicago, and he was willing to welcome us. In parallel, at that time we didn’t know what caused AIDS, but people were dying, and we didn’t know why they were dying; the public hospitals were short staffed. The minute I walked into Cook County Hospital in Chicago, they offered me a job. It was like — any warm body standing just come on in. JCI: You moved on to do a clinical fellowship at the University of Chicago? Olopade: I was looking to do research. That was hard at Cook County, as we were so busy and so understaffed. I ended up doing another year of chief residency and during that year, I had time to reflect on what I wanted to do. There was so much cancer in the community, and the patients suffered so much. But those who came from no matter what part of the city got better. I didn’t see that in Nigeria; everyone in Nigeria died, and there was so much stigma around cancer. During my chief residency I used autopsies to learn. I took it upon myself to consent families of nearly every patient who died on our services to do autopsies. These were mostly Black patients in Chicago, and during that year, we got our autopsy rate to almost 70%. We learned a lot and got better at taking care of very sick patients. Everyone I asked said, “Sure.” I thought, if I could go to a place where I can do research, this will be phenomenal because then people are not going to die from cancer because we will figure it out one day. I applied to University of Chicago and embraced immersing myself in research. JCI: You joined the lab of the legendary leukemia geneticist Janet Rowley. Was this the dawn of your interest in genetics? Olopade: Nobody goes to University of Chicago without learning about Janet Rowley! But she was honestly very approachable. You only needed to be in a room with her once to want to be like her. I was lucky that she was welcoming and supported late bloomers like me. By this time, I had three children; my colleagues in the fellows’ room all were very good at babysitting my children while I finished an experiment in the lab. Dr. Rowley was very adamant I should do molecular biology. So I knew molecular biology, and she knew cytogenetics. We worked together to curate chromosome aberrations in solid tumors. I realized that we didn’t really have much on solid tumors. At the time, you couldn’t grow them in culture, there was no such thing as organoids, and we didn’t know nearly enough about them. The few cell lines that were available for solid tumors were only from melanoma, head and neck cancers, and lung cancers. I studied all of those cancers except leukemia because I knew I couldn’t compete with Dr. Rowley and all the other great hematologists at the University of Chicago. That turned out to be really a good strategy for me. JCI: Was your perspective that genetics played a larger role in breast cancer in the African diaspora (more than poverty or medical neglect) considered heretical or revolutionary at the time? How did you go about building the coalitions and data to support it? Olopade: I happened to have gone to a Gordon Conference and run into Francis Collins and Mary-Claire King, who were mapping genes on different chromosomes, but using human families as their model. This inspired me and gave me an idea. While I was looking for a tumor-suppressor gene on chromosome 9 — I was going to knock it out and have an animal model to do functional studies — Mary-Claire and I both got scooped. I didn’t get the p16 gene on chromosome 9 even though I had mapped all that region, and she didn’t get BRCA1 . And misery loves company, right? But it also allowed me to play to my strengths — in the clinic. My patients became the laboratory for me, as we were looking for melanoma families to map the gene on chromosome 9. But I didn’t have enough melanoma families; on the contrary, the same year BRCA1 was identified, and I realized I had a boatload of breast cancer families. As soon as the race was on to find tumor-suppressor genes, patients were calling and coming to my clinic from south, north, everywhere. When we started the Cancer Risk Clinic, I was still racing to find the gene on chromosome 9, but families with breast cancer were coming in large numbers. I had a startup package that included a genetic counselor and a nurse. The University of Chicago embraced the idea that I could create something unique and different from Janet so that my scholarship couldn’t be attributed to her. I was able to move the needle for BRCA1 research because the first few families that helped nail BRCA1 were Black families, including families with five generations of slavery in their family history. I was single-minded about tracking this. I went to family reunions down south. The patients trusted me, while also wanting to be part of the solution. I was furthermore fortunate to have an opportunity to apply for the Department of Defense Idea Award, which asked for anyone who had ideas on what to do about breast cancer to write a proposal. We had published our first paper on breast cancer in extended African-American families. They didn’t have what at that time everyone was thinking were “Jewish mutations” — they just had heterogeneous mutations/variants of unknown significance. The minute you see variants of unknown significance, you have to continue to dig deeper. We were able to link up with those who were studying breast cancer, like Charles Perou, who had looked at basal-like breast cancer. We got BRCA1 in ‘94; P16 was 1994 also. My husband and I had gone back to our Nigerian medical school in 1996 as alumni speakers, and we were trying to tell our colleagues and the medical school students about the great discovery of BRCA1 . Everybody was like, what? This is highfalutin; it doesn’t apply to us. I had left in 1983 and they were still teaching students the common things: if you see a mass, it’s tuberculosis until proven otherwise. I started off thinking that it was my obligation to teach them about the future of medicine; remember, my father being a missionary had taught me that knowledge is power and that ignorance kills. I wanted to give back to Nigeria and show them the data and convince them to join the effort. In 2004, we had the first international workshop on breast and cervical cancer. That’s eight years from my encounter in 1996. By the end of that meeting in Lagos, we had a ten-point communique: my Nigerian colleagues said, “We want immunohistochemistry, we want to study breast cancer, we want to study cervical cancer, we want to do genetics research.” They put in terms of how we could continue the collaboration. That ten-point communique is what I’ve used to think about what I study and how I bring my colleagues in Nigeria along so that we keep pushing the science forward. JCI: If you could not have been a physician or a scientist, what other profession do you think would’ve kept you as engaged? Olopade: I love debates, and I also love acting. I wanted to be a lawyer because law as a career would really allow me to be able to debate. But then the science and the math could always get me focused. At one point, I thought I’d go to divinity school, after I’d finished with medicine. But then I realized I believed too much in science and technology to go to divinity school. I love being a doctor, and I’ve been fortunate that my patients trust me, and the fact that I don’t know nearly enough to help every patient just keeps me going.

01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:16

J Clin Invest.; 134(2):e178495

oa_package/01/04/PMC10786681.tar.gz

PMC10786682

0

Over several decades the JCI has published key advances in our understanding of glucagon-like peptide 1 (GLP-1) biology. The first incretin peptide characterized in the 1970s, glucose-dependent insulinotropic polypeptide (GIP), was isolated from porcine gastric extracts. Subsequently, the sequence of GLP-1 was identified following the cloning of the glucagon cDNAs and genes, soon followed by the demonstration that GLP-1 potentiated glucose-dependent insulin secretion in cells, animals, and humans (comprehensively reviewed in Drucker, et al.; ref. 1 ). Incretin action in islets and implications for diabetes The findings that the acute insulinotropic actions of GLP-1, but not GIP, were relatively preserved in people with type 2 diabetes (T2D) ( 2 ) focused greater attention on the therapeutic potential of GLP-1, ultimately supporting multiple clinical development programs for GLP-1 receptor (GLP-1R) agonists (GLP-1RA). Physiologically, the essential roles of incretin receptors for glucose homeostasis have been demonstrated in single and double incretin receptor knockout mice. Glp1r –/– mice, and, to a greater extent, Glp1r –/– : Gipr –/– mice, exhibit defective glucose-stimulated insulin secretion, subnormal upregulation of insulin gene expression in response to high-fat diet (HFD) feeding, and impaired glucose tolerance ( 3 , 4 ). In contrast, Gipr –/– mice exhibit greater resistance to diet-induced obesity, relative to Glp1r –/– mice ( 4 ). The physiological importance of GLP-1R signaling has also been revealed in humans treated with GLP-1 receptor agonists such as exendin(9-39). Schirra and colleagues infused exendin(9-39) into healthy male human subjects, under euglycemic or hyperglycemic conditions, with or without concomitant i.v. administration of GLP-1 or GIP ( 5 ). Exendin(9-39) blocked the stimulation of insulin and the inhibition of glucagon secretion in the presence of exogenous GLP-1 administration but had no effect on the insulinotropic actions of GIP. Importantly, infusion of exendin(9-39) alone increased levels of plasma glucagon under conditions of both euglycemia and hyperglycemia, and decreased levels of plasma insulin when the glucose was elevated. Collectively, these findings revealed the essential physiological actions of GLP-1R and GIPR signaling for islet hormone secretion in mice and humans ( 5 ). Among the holy grails of human islet research is the identification of methods to safely and effectively stimulate replication of human islet β cells. Dai and colleagues studied the uncoupling of GLP-1 responses linked to cell proliferation from those that potentiate glucose-dependent insulin secretion in juvenile versus adult human islets ( 6 ). Exendin-4 stimulated glucose-dependent insulin secretion in both juvenile and older adult human islets. However, examination of the proliferative response identified age-associated impairments in components of the calcineurin/NFAT signaling pathway that were responsive to exendin-4 in juvenile, but not in adult, human islets ( 6 ). As GIP and GLP-1 exert their actions through structurally similar G protein coupled receptors, the differential mechanisms underlying preserved GLP-1, but not GIP, insulin stimulatory responses in diabetic β cells have remained enigmatic. Oduori and colleagues probed this anomaly in studies of mice and both murine and human islets exposed to hyperglycemia, and determined that a Gs/Gq signaling switch in β cells arises following exposure to sustained hyperglycemia ( 7 ). Notably, GLP-1 but not GIP, is able to activate both Gq and Gs, while GIP seems only to activate Gs, suggesting a possible mechanism for the diminished insulinotropic response to GIP in diabetic β cells. GLP-1 and the reduction of food intake Following the demonstration that intracerebroventricular administration of GLP-1 inhibited food intake in mice and rats, treatment of animals with peripherally administered GLP-1RA was associated with reduction of food intake and weight loss ( 1 ). Flint and colleagues examined the effects of acute GLP-1 infusion on sensations of hunger and satiety in healthy human volunteers. GLP-1 infusion increased sensations of fullness and satiety and reduced solid food intake after breakfast and lunch ( 8 ). Observations in those treated with GLP-1RA subsequently confirmed weight loss in people with T2D, and later obesity ( 1 ). Understanding the mechanisms underlying the anorexic effects of GLP-1 is of great interest. The GLP-1R is widely expressed in multiple regions of the rodent and human brain, and activation of GLP-1R + neurons in the hypothalamus and brainstem reduces food intake and promotes weight loss. Chemogenetic activation of murine preproglucagon neurons in the hindbrain reduces food intake and metabolic rate and suppresses hepatic glucose production in normal mice ( 9 ). Activation of these GCG neurons in HFD-fed mice revealed a persistent reduction of food intake and body weight, without changes in glucose homeostasis or stress responses. Hence, this population of GCG neurons is likely important for fine tuning the control of food intake, but less essential for the control of whole-body glucose homeostasis. Furthermore, the relative importance of endogenous GLP-2 versus GLP-1 or glucagon as orchestrators of these chemogenetic responses was not determined and is clearly less important for weight control relative to pharmacological actions of the same peptides. Sisley and colleagues used mouse genetics to inactivate the Glp1r in the mouse brain, demonstrating that the acute anorectic and chronic weight loss–inducing pharmacological actions of GLP-1RA required GLP-1R expression in the central nervous system ( 10 ). In contrast, loss of GLP-1Rs in the central or autonomic nervous system did not impact the physiological control of food intake or body weight, even under HFD conditions ( 10 ). These findings emphasize the robust GLP-1R–dependent pharmacological induction of weight loss, yet a comparatively modest importance of basal GLP-1R signaling for food intake or long-term energy homeostasis ( 1 ). Secher and colleagues studied the importance of hypothalamic GLP-1R signaling for the anorectic actions of liraglutide in mice. Injection of fluorescent liraglutide labelled neurons in circumventricular organs, as well as the arcuate nucleus, showed brain uptake of labelled liraglutide was abolished in Glp1r –/– mice, showing that brain uptake of liraglutide is dependent on the canonical GLP-1R ( 11 ). GLP-1 directly stimulated populations of POMC/CART neurons and inhibited the activity of neuropeptide Y+ and agouti-related peptide (AgRP) neurons. It is now appreciated that multiple regions within the hypothalamus, brainstem, and beyond, transduce pharmacological GLP-1R-dependent signals in the brain to reduce food intake, enabling weight loss with chronic administration of GLP-1RA ( 1 ). Rupp et al. used single nucleus RNA-Seq to identify a population of GABAergic Glp1r -expressing LepRb neurons exhibiting robust expression of leptin-regulated genes in the mouse hypothalamus ( 12 ). Mice subjected to fasting followed by refeeding exhibited increased FOS immunoreactivity in dorsomedial hypothalamic Glp1r neurons with a distribution overlapping with that exhibited by LepRb + Glp1r + neurons. Activation or deletion of Lepr in these neurons revealed an essential role for this neuronal population in the basal control of food intake. Similarly, selective rescue of the GLP-1R in this hypothalamic neuronal population of Glp1r –/– mice restored an anorexigenic response to GLP-1R agonism, evident following acute liraglutide administration ( 12 ). GLP-1 actions beyond insulin secretion and body weight Clinically, GLP-1RAs are used to treat people with T2D and/or obesity ( Figure 1 ), based on mechanisms described above linked to control of insulin and glucagon secretion, as well as reduction of food intake. Initial reports in animals showed that GLP-1 acutely increases blood pressure (BP) and heart rate (HR) in rats and mice, actions mediated through activation of the autonomic nervous system, including medullary catecholamine neurons, providing input to sympathetic preganglionic neurons ( 13 ). In humans, GLP-1R agonism frequently reduces BP; however, increases in HR are common and may be sustained with prolonged GLP-1R agonism. Notably, GLP-1RAs were subsequently shown to produce cardioprotective actions in animals ( 1 ). Importantly, starting in 2016, the first in a series of cardiovascular outcome studies demonstrated that long-acting GLP-1RAs reduce the rates of myocardial infarction, stroke, cardiovascular death, and all-cause mortality in people with T2D ( 1 ). More recent studies have extended the cardiovascular benefits of GLP-1R agonism to people with obesity, and subjects with heart failure and preserved ejection fraction (HFpEF). Intriguingly, GLP-1RAs are neuroprotective in animals and several trials have examined the actions of exenatide in people with Parkinson’s disease ( 1 ). Aviles-Olmos and colleagues examined the effects of twice-daily exenatide over 12 months in a randomized controlled trial of people with Parkinson’s disease (PD) ( 14 ). Modest but detectable improvements were noted in PD activity scores and dementia rating scales; however, the small number of subjects studied (44 in total, 20 randomized to exenatide) and the limited duration of the trial limits definitive conclusions from being drawn. The future of GLP-1–based medicines There are currently multiple once-weekly GLP-1RAs used to treat T2D, and two, liraglutide and semaglutide, are approved for therapy of people with obesity. A GIP-GLP-1R coagonist is now used to treat people with T2D and produces substantial weight loss, resulting in its approval for treatment of obesity in 2023. Oral semaglutide is also available as a once daily option, and several small molecule GLP-1R agonists, exemplified by orforglipron are in late stage clinical development ( Figure 1 ). Newer GLP-1RAs and GLP-1–based coagonists also appear promising and are being studied in separate trials for T2D, diabetic kidney disease, peripheral artery disease, and metabolic liver disease ( Figure 1 ). GLP-1RAs such as semaglutide have proven efficacy in HFpEF and are being studied in people with PD as well as in trials for Alzheimer’s disease. Hence, the expanding role of GLP-1-based medicines, together with newer more powerful GLP-1-based medicines ( Figure 1 ), holds great promise for achieving improved health for substantial populations of individuals living with the complications of chronic cardiometabolic disorders.

DJD is supported by a Banting and Best Diabetes Centre-Novo Nordisk Chair in Incretin Biology, a Sinai Health Novo Nordisk Foundation Fund in Regulatory Peptides, and CIHR Foundation Grant 154321. 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:16

J Clin Invest.; 134(2):e175634

oa_package/03/0e/PMC10786682.tar.gz

PMC10786683

0

Conclusions and future prospective CAR-T cells have been transformative for B-cell acute lymphoblastic leukemias (ALL) and NHL. As we extend the benefits of CAR-T cells to other hematologic malignancies, new CARs and combinations are taking the stage, and additional challenges are emerging. More toxicities with new targets may occur when trying to differentiate between normal- versus tumor-specific antigens, since, as opposed to B-cell targeting, we cannot consider as acceptable risks the long-term myeloid aplasia or T-cell aplasia that could occur, for example, with CD33- or CD123-CARs for the treatment of acute myeloblastic leukemia, and with CD5- or CD7-CARs for T-ALL and NHL. Another evolving issue is tumor escape caused by antigen loss. Epitope loss has been reported with CD19 CAR-T cells, particularly in patients with leukemia ( 19 ), opening the need for multiple CAR-T cell targeting. However, lack of response durability after CAR-T cell treatment is most frequently caused by suboptimal CAR-T cells expansion and persistence, and by the presence of local inhibitory factors and/or cells ( 17 , 18 ). Understating the exact cause calls for different combinatorial approaches. We believe that in-patient comparisons can be particularly handy when testing new strategies to overcome T-cell dysfunction, like persistence using dual costimulations ( 20 ) or co-expressing CARs and chemokine receptors to correct suboptimal trafficking of CAR-T cells to the tumor, as we are currently exploring in patients with Hodgkin lymphoma (NCT03602157). As we reflect on the evolution of CAR-T cells over the past 30 years, with 6 CAR-T cell products approved by the FDA between 2017 and 2022, there is substantial anticipation that what we learned from CD19 CAR-T cells provides the foundations for expanding this approach to other hematologic malignancies and eventually to solid tumors. Setbacks or slow advances may pave the way in the solid tumor arena, but even if CAR-T cells do not become the holy-grail therapy, we still should not discount that this strategy has revolutionized how we treat B-cell malignancies and positively impacted the outcome of patients with hematologic malignancies.

An advanced treatment for hematological malignancies For decades, the success of allogeneic hematopoietic stem cell transplant has highlighted the role of T cells in eliminating hematological tumors. This proof of concept pioneered the practice of the ex vivo generation and adoptive transfer of tumor-specific T cells (ATC), which became an effective modality, for example, for virus-associated diseases, like Epstein-Barr-virus-associated post-transplant lymphoproliferative disorders ( 1 ). A remarkable advance to the ATC field was pioneered by Eshhar and colleagues in 1993, when, rather than replicating the T-cell receptor (TCR)-mediated recognition of tumor molecules, which requires antigen processing and presentation through major histocompatibility complex (MHC) molecules, they generated chimeric antigen receptors (CAR) that target unprocessed antigens expressed on the cell surface of tumor cells in an MHC unrestricted fashion ( 2 ). Providing T cells with the ability to engage targets via an antibody-derived single-chain-variable-fragment (scFv) linked to the intracellular T-cell signaling domains was game changing because it overcomes HLA barriers and opens a whole new antigen repertoire for tumor recognition. While the first clinical application of CAR-T cells was for treatment of patients with HIV ( 3 ), this approach gained the most traction in hematological malignancies and was driven by the availability of lineage restricted antigens homogeneously expressed by hematological tumors. Specifically, targeting B-cell malignancies offered a much-needed route to success ( 4 – 6 ). The CD19 antigen promptly took the spotlight as meeting the fundamental requirements for ideal CAR targeting, namely: its high expression on tumor cells, expression at all stages of B-cell development and thus on derived tumors, and absent expression on vital tissues, with the potential for B-cell aplasia considered a risk-benefit ratio worth taking ( 7 ). The other major advantage provided by hematological malignancies is their accessibility. CAR-T cells infused intravenously are much more likely to interact with tumor cells that are circulating in the blood or located in the bone marrow and lymph nodes (or lymphatic system), compared with the those that require penetrating through a solid tumor stroma. Moving this technology to cancer treatment also required several advances in the laboratory. First, genetic modification of human cells became more mainstream, thanks to the development of efficient viral vectors, like lenti- and retro-virus based vectors, combined with the development of better protocols for vector production, cell transduction, and newer technologies for their safety monitoring ( 3 , 8 ). Second, recognizing the need for a costimulatory signal (known as signal 2) to promote full T-cell activation was key. The provision of just the CD3ζ or CD3γ endodomains that trigger the proximal signaling of the TCR to the first-generation CAR was in fact unable to replicate the physiological sequence of activation events that occur upon TCR engagement and proved to be sub-optimal in supporting CAR-T cell expansion and persistence ( 9 ). In contrast, including the intracellular domains of molecules such as CD28 or 4-1BB in tandem with the CAR supported the transmission of signals capable of producing the required sustained activation, proliferation, and effector function ( 10 – 12 ). Informative first-in-human studies for advancing the field To this end, in 2011 we conclusively demonstrated the essential role of co-stimulation in helping CAR-T cell expansion in vivo in the context of the human immune system ( 13 ). Our studies employed a simultaneous-infusion strategy, in which two CAR-T cell products were provided in the same patient. Each CAR-T cell product carried the identical scFv targeting CD19, but one included only the CD3ζ endodomain and the other included the CD28 costimulatory endodomain. These products were generated in parallel, starting from the same blood collection to further minimize any potential difference outside of CD28 co-stimulation. Although requiring the generation of two products for each individual, this study showed that the inclusion of the CD28 signaling motif supported superior activation, proliferation, and effector function of CAR-T cells in patients with CD19 + Non-Hodgkin lymphoma (NHL) ( 13 ). The major strength of the simultaneous-infusion study is that it removed from the evaluation all confounding factors, such as the heterogeneity of B-cell lymphomas, prior treatments, patient’s comorbidities or other intrinsic aspects, and allowed us to reach meaningful conclusions even with the small sample size that characterizes phase-I studies by allowing for comparisons within an individual patient ( Figure 1 ). We suggest that the design of small phase I clinical trials is also well poised to elucidate specific biological questions that remain critical in hematological malignancies. For example, we demonstrated that grafting the CD19 antigen on antigen-specific CAR T cells, such as virus specific T cells, yields alternative costimulatory signals when the native TCR of antigen-specific T cells engages with the cognate antigen expressed by effective antigen presenting cells ( 14 ). These types of biological implications, in fact, cannot be easily modelled in immune-deficient mice, which are typically obtained by engrafting with human tumors and treated with human CAR-T cells. Immune-deficient mouse models are also limited when used to address the even more complex setting of human diseases, such as the effects of the tumor microenvironment on the functionality of CAR-T cells. In light of these restrictions, immune-competent mouse models are being used more frequently, but differences in the evolutionary pathways and molecules, and requirements for tumor implant, rather than spontaneous development, may remain insufficient in recapitulating the clinical scenario. From our perspective, well-designed studies in humans also facilitate the reverse engineering process, or help with translation, informing us of where to go next. This bench-to-bedside, back-to-the-bench approach has proven instrumental to correlate prolonged survival of CAR-T cells enriched in naïve-central cells and/or memory-like cells, and helped us and others in implementing simple changes in manufacturing, like using different cytokines or enrichment protocols and shorter cultures ( 15 , 16 ). One of these studies also underscored how responses can be better predicted if the infused product is further formulated with a defined ratio of CD4 + and CD8 + cells ( 16 ), a particularly important observation since the peripheral blood of patients with hematologic malignancies is usually characterized by variable proportions of these subsets due to prior therapies or underlying diseases. Early phase CAR-T cell clinical trials have also addressed the importance of the conditioning therapy before CAR-T cell administration, even in patients with hematologic malignancies who are often fairly cytopenic. These studies demonstrated not only that conditioning therapy is needed ( 17 , 18 ), but also that the inclusion of fludarabine in the lymphodepleting regimen optimize CAR-T cell expansion and persistence, likely because the addition of fludarabine promotes superior bioavailability of homeostatic cytokines, such like IL7 and IL15, to CAR-T cells and halts the cell-mediated elimination of CARs of murine origin ( 16 , 18 ). As CAR-T cell therapies are being moved earlier in the treatment schedule, we consider these correlative analyses paramount. For example, patients will likely have received less chemotherapy, which may positively impact the quality of the infusion-product. On the other hand, when infused in the context of a more competent immune environment we will learn whether current lymphodepletion regimens will continue to prevent CAR immune-mediated rejection. Technologies advance at the speed of light and we should apply them not just to improve CAR design, but also to study in depth what happens to these cells upon infusion in patients, so we can create fitted cells for improved outcomes.

01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:16

J Clin Invest.; 134(2):e177160

oa_package/0e/f5/PMC10786683.tar.gz

PMC10786684

37943622

To the Editor: Dilated cardiomyopathy (DCM) and peripartum cardiomyopathy frequently associate with heterozygous truncating variants in TTN ( TTN tvs) (1). TTN codes for titin, a spring-like protein that spans the sarcomere from the Z-disk to the M-line and is essential for sarcomeric assembly, homeostasis, and regulation of contractility ( 2 , 3 ). How TTNtvs cause DCM remains unclear. Recent work by us ( 4 ) and others ( 5 ) demonstrated that the truncated titin proteins (hereafter referred to as TTNtvs) encoded by TTNtvs are expressed and detectable in human DCM hearts, concomitant with reduced full-length titin. TTNtvs have been shown in human induced pluripotent stem cell–derived (iPSC-derived) cardiomyocytes to incorporate into nascent myofibril-like structures ( 6 ). Disease may thus be caused by titin haploinsufficiency, i.e., insufficient intact titin, or by a direct negative effect of the TTNtv. With respect to the latter possibility, it is important to determine whether TTNtvs incorporate into the sarcomere in TTNtv-bearing hearts and, if so, whether they bear force. Alternatively, TTNtvs could have negative effects such as extra-sarcomeric protein aggregation ( 5 ). To specifically detect TTNtvs in human myocardium, we developed a patient-specific TTNtv-specific antibody. One previously described DCM patient bearing a TTNtv (patient 1371) (4) had a frameshift introduced into the TTN exon 329, appending a proteome-unique sequence of 32 amino acids to the C-terminal end of the TTNtv ( Figure 1A ). A rabbit polyclonal antibody (FS-Ab) raised against this frameshift antigen (FS-Ag) detected the TTNtv, but not full-length titin ( Figure 1B ). Probing skinned cardiomyocyte fragments from patient 1371 with this FS-Ab revealed paired stripes 238 nm ± 24 nm from the sarcomeric M-line ( Figure 1C ), matching the predicted location of the TTNtv C-terminus along the thick filament, assuming integration matching that of full-length titin. We conclude that this TTNtv integrated into the sarcomere at the expected thick filament binding location and did so despite not binding to the M-line. We next investigated whether TTNtvs can bear the length-dependent forces generated by the entropic spring behavior of titin’s I-band. Upon stretch of cardiomyocyte fragments from very short to supraphysiologically long sarcomere lengths (SLs), TTNtv from patient 1371 remained attached to the relatively inelastic thick filament ( Figure 1D ). The distance between TTNtv C-termini (i.e., across the M-line) remained invariant, whereas the distance to the Z-disk (i.e., spanning the I-band) increased with SL ( Figure 1D quantification). Labeling the N-terminus of titin (titin-Z, in blue in Figure 1 ) revealed no signal peaks outside the Z-disk, demonstrating that the N-terminus of the TTNtv remained attached to the Z-disk. We conclude that TTNtvs remained attached to both the Z-disk and thick filament upon stretch, indicating that TTNtv can bear load across the sarcomere. To test in a different manner whether TTNtvs bind to the Z-disk, we stretched cardiomyocyte fragments from patient 1371 and added 400 mM KCl, a treatment known to disrupt the thick filament. We reasoned that release of TTNtv from the thick filament would enable entropic spring forces to pull its truncated C-terminus toward the Z-disk. Under high-salt treatment, the TTNtv C-terminus signal relocated from the thick filament to near the Z-disk ( Figure 1E , Supplemental Figure 1 , and Supplemental Methods ; supplemental material available online with this article; https://doi.org/10.1172/JCI170196DS1 ), although, interestingly, not precisely to the Z-disk, consistent with previous observations that titin binds the thin filament outside the Z-disk ( 3 ). In contrast, full-length titin remained attached to the M-line ( Supplemental Figure 2 ). These data confirm that TTNtvs transmitted force across the I-band region of the sarcomere, even at supraphysiological SLs, but detached more readily than full-length titin from the thick filament. The observed recoil of the truncated C-terminus to the Z-disk also demonstrates that the recognized protein was not encoded by the Cronos transcript, as the latter lacked the I-band spring and the Z-disk attachment regions. The data also demonstrate that attachment of titin to the M-line was dispensable for binding to the Z-disk and incorporation into the sarcomere. Together, these observations demonstrate that TTNtv protein was present in myocardium from a TTNtv-bearing DCM patient; that the TTNtv protein was incorporated into the sarcomere akin to full-length titin, binding both the Z-disk and the thick filament; and that this A-band overlapping truncated titin bore force across the I-band, even under supraphysiological strain, but detached more readily from the thick filament compared with full-length titin ( Figure 1F ). Caveats of our study include the restriction to a single patient; we were not able to generate antibodies against other TTNtv frameshift variants. The data also do not provide stoichiometric information about truncated versus full-length titin protein. While not ruling out contributions of haploinsufficiency or of aberrant protein aggregation, our study supports the notion that truncated titin proteins may have a direct effect on sarcomeric behavior, affecting sarcomeric structure, contractility, or signaling, and thus contributing to TTNtvs-associated DCM pathophysiology. Supplementary Material

This work was supported by the National Heart, Lung and Blood Institute (NHLBI) (HL152446, HL126797, HL149891); the National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) T-32 (AR-53461-12); the Children’s Hospital of Philadelphia Frontier Program; the Foundation Leducq (research grant 20CVD01); and the Center for Engineering Mechanobiology through a grant from the National Science Foundation’s Science and Technology program (15-48571). Parts of the figure were created with BioRender.com. 11/09/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:16

J Clin Invest.; 134(2):e170196

oa_package/c3/d0/PMC10786684.tar.gz

PMC10786685

0

Conclusions and implications By uncovering distinct mechanisms and essential roles for SEL1L and HRD1 variants, these exciting findings by Qi and colleagues raise intriguing new questions, answers to which should generate further insights into ERAD mechanisms that promote proteostasis. For example, impaired ERAD function is known to cause proteotoxic stress to cells due to an accumulation of misfolded proteins in the ER lumen, which activates the unfolded protein response (UPR) ( 17 ). Conceivably, chronic proteotoxic stress due to constitutive ERAD defect in developing neurons and nervous system tissues expressing SEL1L and HRD1 variants would cause UPR-induced apoptosis ( 17 ). However, despite ERAD defects and increased levels of misfolded proteins, there was no increase in UPR activity in cells expressing SEL1L or HRD1 variants. As this phenotype may be specific to fibroblasts or HEK293T cells in which the variants were analyzed, resolving exactly how the SEL1L and HRD1 variants cause ENDI-symptoms would require future studies modelling and examining the SEL1L-HRD1 crosstalk with other proteostasis pathways in relevant cell and tissue types. Additionally, SEL1L has been implicated as functioning outside of ERAD in regulating lipoprotein lipase secretion in adipocytes ( 18 ), suggesting that these SEL1L variants might regulate tissue development by potentially also modulating non–ERAD-related pathways. In light of the reported roles for ERAD in the generation and maintenance of multiple hematopoietic cell lineages in mice ( 5 – 7 ), another intriguing issue is how these variants are dispensable for hematopoiesis and appear to only impair development of B cells in humans. Does this difference simply reflect the hypomorphic activity of the SEL1L and HRD1 variants, or does it indicate a mouse-versus-human nuance wherein human immune cells (compared with neurons) express distinct partners that are still able to interact with the SEL1L-HRD1 variants? Notably, none of the patients with ENDI showed any overt defects in hematopoiesis, unlike mice, in which SEL1L-deficient HSCs failed to efficiently generate blood cells ( 6 , 9 ). In individuals with ENDI-A with the SEL1L p.C141Y variant, only B cells (and consequently antibodies) were deficient, and these patients also lacked detectable COVID-specific memory T cells after COVID infection ( 12 ). As these memory defects could be due to a lack of CD4 + T cell help, it would be worth investigating specific functions of T cells harboring these variants given reported roles for SEL1L and HRD1 in T cell development and function ( 7 , 19 ). Moreover, while T cell numbers were normal in patients with ENDI-A, their skewed CD4/CD8 ratios (normal: 1.5–2.5; ENDI-A: 0.77–0.83) suggest defects in T cell homeostasis. In conclusion, these reports from Qi and colleagues establish the importance of ERAD in humans and set the stage for precise dissection of ERAD and its crosstalk with other pathways that promote proteostasis. Their findings establish ENDI and ENDI-A as clinically relevant neurodegenerative disorders that may be corrected by reversal of the mutant genes. Importantly, by modeling these variants in cell lines and animal models, future studies can now uncouple SEL1L-HRD1 interactions from substrate degradation to better understand nuances in ERAD function.

The suppressor of lin-12-like–HMG-CoA reductase degradation 1 (SEL1L-HRD1) complex of the endoplasmic reticulum–associated degradation (ERAD) machinery is a key cellular proteostasis pathway. Although previous studies have shown ERAD as promoting the development and maintenance of many cell types in mice, its importance to human physiology remained undetermined. In two articles in this issue of the JCI , Qi and colleagues describe four biallelic hypomorphic SEL1L and HRD1 variants that were associated with neurodevelopment disorders, locomotor dysfunction, impaired immunity, and premature death in patients. These pathogenic SEL1L-HRD1 variants shine a light on the critical importance of ERAD in humans and pave the way for future studies dissecting ERAD mechanisms in specific cell types.

ERAD pathway proteins The endoplasmic reticulum (ER) is a major site of protein synthesis, folding, and maturation for membrane and secreted proteins. It is therefore imperative to have ER-intrinsic quality-control mechanisms that prevent accumulation of misfolded proteins in the ER lumen to prevent proteotoxic stress. The suppressor of lin-12-like–HMG-CoA reductase degradation 1 (SEL1L-HRD1) complex is the most conserved branch of the ER-associated degradation (ERAD) pathway, which senses and degrades misfolded proteins to maintain proteome homeostasis (termed “proteostasis”) ( 1 ). During ERAD, the molecular chaperones enhancing α-mannosidase like protein 1 (EDEM1), osteosarcoma amplified 9 (OS9), and ER lectin 1 (ERLEC1/XTP3B) bind to misfolded proteins in the ER lumen and bring them to SEL1L for degradation ( 1 ). Subsequent interaction of SEL1L with HRD1 facilitates the cytoplasmic export of substrates through its retrotranslocation channel and their ubiquitination by HRD1’s cytoplasmic tail ( 2 ). SEL1L and HRD1 also promote the stability of each other ( 3 ). As germline deletion of Sel1l or Hrd1 genes in mice results in embryonic or perinatal lethality ( 3 , 4 ), conditional cell type–specific knockout models have been used to characterize the importance of these proteins and ERAD in animals. To date, critical requirements for SEL1L-HRD1 have been established in a wide range of cell types, including adipocytes, pancreatic β cells, and gut epithelial cells as well as in immune cells, notably hematopoietic stem cells (HSCs) and B and T cells ( 5 – 10 ). That defects in many of these cells and processes are incompatible with fetal and postnatal life provides one potential reason why pathogenic SEL1L-HRD1 variants have not been identified in humans. In the absence of any such natural variants, it was unclear how SEL1L-HRD1 and ERAD contributed to human physiology. HRD1 and SEL1L pathogenic variants In two complementary articles in this issue of the JCI , Qi and colleagues ( 11 , 12 ) report the identification of one HRD1 (p.P398L) and three SEL1L (p.G585D, p.M528R, and p.C141Y) pathogenic variants ( Figure 1 ) arising from missense DNA mutations that altered amino acid residues within protein-protein interacting domains of SEL1L and HRD1. All four SEL1L-HRD1 variants were biallelic mutations, suggesting that the presence of WT alleles might have protected against the development of any detectable phenotype in monoallelic carrier parents. Notably, children and adolescents in which the variants were identified presented with multiple signs of impaired neurodevelopment, severe hypotonia, and recurring infections. These SEL1L-HRD1 variants independently disrupted ERAD function and were associated with a spectrum of phenotypes that the investigators term ERAD-associated neurodevelopmental disorder with onset in infancy (ENDI) syndrome. Disease due to one of these variants ( SEL1L p.C141Y), which was associated with a complete loss of B cells and antibodies, was termed ENDI-agammaglobulinemia (ENDI-A) syndrome. In Wang et al. ( 11 ), Qi and colleagues describe SEL1L-HRD1 variants in six patients from three unrelated consanguineous families from disparate geographical backgrounds (Italian, Saudi Arabian, and Moroccan). Whole-exome sequencing revealed that they all had biallelic mutations in the ERAD complex: SEL1L p.G585D, SEL1L p.M528R, and HRD1 p.P398L. While all patients presented with dysmorphisms, developmental delays, intellectual disability, and short statures, only four of the six patients presented with seizures and microcephaly. Similarly, in Weis et al. ( 12 ), the authors report a SEL1L p.C141Y variant in five siblings from two consanguineous Slovakian families exhibiting ENDI symptoms along with frequent infections and premature death. Immunological tests failed to detect CD19 + lymphocytes or immunoglobulins in the blood of these patients, consistent with a loss of ERAD function in B cells as reported in mice ( 5 ). The basis of this variability in phenotype, despite similar impacts of these variants on ERAD activity, is unclear but suggests nuanced functions and interactions by SEL1L and HRD1. Importantly, that these SEL1L and HRD1 variants were found in patients with seemingly different ancestry suggests a broad ethnic inheritance or acquisition of these mutations. It is notable that all the SEL1L and HRD1 variants were associated with neurodevelopmental disorders in patients given the critical importance of proteostasis and ERAD to normal brain and nervous system development ( 13 – 15 ). Mechanistically, ERAD dysfunction in the brain could result in the accumulation of misfolded proteins that cause proteotoxic stress, damaging neurons, and brain cells, which would then lead to impaired neuronal development and poor cognitive function. In line with this possibility, SEL1L silencing in neuronal stem cells caused the accumulation of autism spectrum disorder–related postsynaptic molecules CADM1 and Shank3 ( 14 ). Interestingly, this requirement for SEL1L-HRD1 in neurodevelopment appears to be shared across mammals, as the Qi group also recently described a SEL1L variant (p.S658P) in Finnish hounds that impaired SEL1L-HRD1 interaction and was associated with ENDI-like symptoms due to a loss of Purkinje neurons in the cerebellar cortex ( 16 ). In comprehensive structure-function assays, Qi and coworkers elegantly demonstrate that these variants likely altered ERAD by disrupting the structure, stability, and interacting partners of both proteins ( Figure 1 ). Specifically, the p.G585D variant, which is located in the substrate -binding domain of SEL1L, impaired SEL1L interaction with the chaperones OS9 and ERLEC1, thereby inhibiting substrate recruitment and degradation. The p.M528R, part of an α-helix in the SLR-M domain of SEL1L, resulted in reduced SEL1L stability, probably by impairing dimerization mediated by the SLR-M region or disrupting its interaction with HRD1. The C141Y mutation in SEL1L resulted in HRD1-mediated complex degradation due to a disruption in the disulfide bridge within the FNII domain. The only variant in HRD1, p.P398L, is located in the proline-rich region within the cytoplasmic tail that mediates its interaction with other ER membrane proteins ( 13 ) and resulted in a reduction in substrate ubiquitination and HRD1 autoubiquitination. Collectively, functional assays revealed that protein products encoded by the SEL1L and HRD1 variants are hypomorphic, as they only partially impaired ERAD activity and were nonlethal, unlike germline murine Sel1l or Hrd1 deficiency. Moreover, while all variants caused an increase in ERAD substrates, including IRE1α and CD147, they appeared to do so by different mechanisms. These observations suggest a multilayered role for SEL1L-HRD1 in the ERAD machinery in which distinct regions (represented by the variants) of either protein capture unique interacting partners to actuate different ERAD outputs. The hypomorphic activity of the variants may also explain the spectrum of phenotype and disease severity across patients with ENDI. Hypothetically, depending on cell type, the hypomorphic variants may detect and still provide some ERAD activity sufficient to mitigate low levels of proteotoxic stress.

Work in our laboratory is supported by the Intramural Research Program of the NIH, National Cancer Institute, Center for Cancer Research (ZIA BC 012135). 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:16

J Clin Invest.; 134(2):e175448

oa_package/f2/05/PMC10786685.tar.gz

PMC10786686

37962956

Introduction Targeted metagenomic sequencing is commonly used for the identification of disease-causing bacteria, archaea, and fungi. In addition, 16S rRNA gene surveys are increasingly being adopted as diagnostic tools to profile microbiome communities that contribute to clinical pathogenesis ( 1 ). However, to understand what constitutes a disease-associated or -causing microbiome, it is necessary to define the characteristics and functions of a healthy human microbiome community across diverse genetic and environmental confounders ( 2 ). Consortium-driven studies such as MetaHIT, the Human Microbiome Project, LifeLines, and the American Gut Project have made significant progress in compositional profiling of the human microbiome by establishing standard operating procedures, including DNA extraction protocols, 16S primer design, and bioinformatics pipelines ( 3 ). The next phase in identifying robust host-microbiome interactions that modulate human disease requires integrated and sufficiently powered multicenter trials to account for human genetic and environmental variation. However, optimal study designs are often cost-prohibitive and logistically difficult to manage. Reanalyzing large deposits of publicly available 16S sequencing data represents an attractive alternative approach to mine clinical microbiome associations, in order to facilitate precision diagnosis and microbiome-based therapy. Nevertheless, reanalysis of individual microbiome surveys remains a significant bioinformatics challenge owing to the lack of a gold standard analytical pipeline that provides accurate taxonomic profiling of sequences generated from distinct 16S variable regions across multiple technology platforms. Gastrointestinal disease is a prime example of where clinical microbiome surveys have provided promising insights into microbiome associations and mechanisms. However, systematic review of these largely single-site cohort studies has demonstrated inconsistent findings, largely due to variations in methods for data generation and analysis, which introduce significant bias for cross-comparisons ( 4 , 5 ). Chronic diarrhea is a significant cause of morbidity in developed countries, and overlapping disease symptoms often make diagnosis and management challenging. Thus, there is a pressing need for noninvasive approaches to differentiate the clinical spectra of common diarrheal symptoms, particularly in irritable bowel syndrome (IBS), inflammatory bowel diseases (IBDs) such as Crohn’s disease and ulcerative colitis, and Clostridioides difficile infection (CDI), which affect up to 20% of the population, and misdiagnosis is frequent. With few reliable disease-specific fecal biomarkers reported for IBD or IBS ( 6 , 7 ), endoscopy remains the gold standard for diagnosis, combined with laboratory testing and questionnaires. Clinical diagnosis and treatment are further complicated by antibiotic use and susceptibility to CDI, which is often a postinfectious trigger of IBS, while IBD patients are frequently asymptomatic carriers of toxigenic C . difficile ( 8 , 9 ). As the gut microbiome is compositionally different, yet implicated in the pathogenesis of these diarrheal diseases, we investigated whether 16S profiling could stratify patients with these commonly misdiagnosed diseases.

Methods Simulation of full-length and region-specific 16S amplicon data Two reference databases were used for data simulation: the NCBI 16S rRNA RefSeq database (downloaded in July 2019) and the Ribosomal Database Project (RDP) database (release 11.5) ( 33 ). To extract sequence fragments as full-length amplicons of targeted 16S variable regions (V1–V3, V3–V5, V4, and V6–V9), the cutadapt tool (version 2.4) ( 34 ) was used, based on the forward and reverse primers listed in Supplemental Table 5 . During sequence extraction, an error rate of 0.2 was permitted. For specific benchmarking purposes, further sequence length trimming, as well as random simulation of sequence abundance and quality score, was performed as indicated below. Benchmarking of sequence clustering and denoising using simulated amplicons with variable length To benchmark the accuracy of clustering or denoising for amplicon data with variable sequence lengths, a random count ranging from 1 to 50 was assigned to each parent full-length amplicon extracted from NCBI 16S rRNA RefSeq sequences. As traditional 454 data are typically generated from the reverse orientation, length trimming from either the forward or reverse orientation was applied to each type of amplicon data, resulting in sequence lengths of 100, 150, 170, 200, 250, 300, 350, 400, and 450 bases for V1–V3, V3–V5, and V6–V9 amplicon data and 100, 150, 170, 200, and 250 bases for V4 amplicon data. For sequencing denoising only, a random Phred quality score (ASCII_BASE=33) ranging from 30 to 42 was assigned to each base. Each simulated amplicon of a specific sequence length represented 1 sample. All samples with the same sequence orientation from the same 16S region were included for closed-reference or de novo clustering using UCLUST (v1.2.22) ( 35 ) or VSEARCH (v2.9) ( 36 ), or denoising using DADA2 (v1.8) ( 37 ). Sequence similarity thresholds of 0.97, 0.99, and 1.00 were evaluated for each clustering strategy. The comprehensive SILVA database (release 132) was used for closed-reference OTU picking. Because simulated amplicons of variable length originating from the same parent full-length amplicon had the same sequence counts, pairwise Spearman correlation analysis was performed for sequence counts of any 2 sequence lengths (as 2 independent samples) in the OTU count tables. Benchmarking of taxonomic overclassification Controlling false positives resulting from taxonomic overclassification of short amplicon data is an important consideration. To this end, the default parameters in the BLCA tool ( 11 ) and its default database NCBI 16S rRNA RefSeq were used to annotate random and repeat sequences previously generated for benchmarking IDTAXA and other annotation tools ( 10 ). Full-length 16S amplicons of unannotated sequences (at least down to the family rank; 868,902 sequences) extracted from the RDP database (release 11.5) were used for further testing of BLCA. BLASTN search ( 38 ) of unannotated sequences against the NCBI 16S rRNA RefSeq database confirmed that no best hits were identified at the 97% threshold applied to both sequence identity and coverage. Simulated amplicons of unannotated RDP sequences were tested using different thresholds of sequence coverage and identity ranging from 0.85 to 1.00 in BLCA. Ten iterations of random subsampling (1%) and BLCA annotation on those unannotated amplicons were performed to statistically determine the optimal sequence coverage and identity required for BLCA. Taxonomic overclassification rate was defined as the classifiable proportion of unannotated amplicons at the species level. The confidence score of taxonomic assignment was not considered at this stage. Benchmarking of taxonomic accuracy using simulated amplicons of variable length To evaluate the taxonomic accuracy of BLCA, a series of simulated amplicons were generated by trimming of full-length amplicons obtained from NCBI 16S RefSeq from either forward or reverse orientation, resulting in variable sequence lengths (100, 150, 170, 200, 250, 300, 350, 400, and 450 bases for V1–V3, V3–V5, and V6–V9 amplicon data, and 100, 150, 170, 200, and 250 bases for V4 amplicon data). The known taxonomic lineage of the parent 16S sequences of the simulated amplicons was present in the BLCA default reference database, allowing for the evaluation of taxonomic misclassification. The misclassification rate was defined as the proportion of incorrectly annotated simulated amplicons with known taxonomic lineage. To determine the optimal confidence threshold of BLCA for mitigating misclassification, simulated amplicons with selected sequence length ranges were combined to calculate the proportion of correct versus incorrect annotations using defined thresholds. The true-positive and false-negative hits were used to represent correct annotations, while true-negative and false-positive hits represented incorrect annotations, given the known taxonomic lineage of the data input. Design of the Taxa4Meta pipeline Based on our benchmarking results, we developed a computational pipeline named Taxa4Meta for analyzing 16S amplicon data. This pipeline incorporated various open-source programs such as VSEARCH ( 36 ) for stringent clustering at 99% identity optimized for 16S amplicon data with selected variable lengths, BLCA ( 11 ) with optimal region-specific confidence thresholds for stringent taxonomic annotation of OTUs, and IDTAXA ( 10 ) for annotating OTUs that could not be identified by BLCA using identity and coverage thresholds of 99% during sequence alignment against NCBI 16S RefSeq sequences. Since merging de novo OTU tables from different 16S variable regions can be challenging, we used collapsed taxonomic profiles from OTU tables for downstream analysis during 16S meta-analysis. To generate relative abundance of collapsed taxonomic profiles without rarefaction for OTU tables, we used total sum scaling. However, it is worth noting that controlling the batch effect, i.e., removing contaminated reads from different sequencing labs, is not feasible in this pipeline because data of negative controls and sample DNA yields were commonly missing in publicly available data sets. Benchmarking of taxonomic profiling accuracy using Taxa4Meta versus other 16S pipelines We evaluated the feasibility and accuracy of commonly used 16S pipelines for processing simulated and experimental data sets ( 12 , 28 ) to achieve precise sequence clustering and enhanced taxonomic accuracy. The simulated data sets were derived from the NCBI 16S RefSeq database and included full-length amplicons of V1–V3, V3–V5, V4, and V6–V9. Each full-length amplicon was randomly assigned a sequence count between 1 and 50, and a Phred quality score (ASCII_BASE=33) ranging from 30 to 42. Further trimming was performed for each amplicon from the forward and reverse orientations to generate different sequence lengths for each variable region: V1–V3 forward amplicons (200, 250, 300, 350, 400, and 450 bases), V1–V3 reverse amplicons (300, 350, 400, and 450 bases), V3–V5 forward amplicons (250, 300, 350, 400, and 450 bases), V3–V5 reverse amplicons (300, 350, 400, and 450 bases), both forward and reverse amplicons of V4 (200 and 250 bases), V6–V9 forward amplicons (300, 350, 400, and 450 bases), V6–V9 reverse amplicons (250, 300, 350, 400, and 450 bases). Trimmed amplicons from the same sequence orientation of each 16S variable region were combined into a single sample to enable benchmarking of various 16S pipelines. As the sequence abundance was known for the simulated data, the NCBI 16S taxonomic lineage was used as a reference annotation (ground truth) for the comparison of different taxonomic profilers. The Korean stool microbiome data set ( 12 ) was utilized as a real-world microbiome data set for the benchmarking of different 16S pipelines. Identical DNA extracts were sequenced using 454 V1–V4, Illumina V1–V3, Illumina V3–V4, Illumina V4, and Illumina shotgun metagenomic sequencing. To prepare the data set for benchmarking, primers retained in the sequence reads were removed by positional trimming, and Illumina paired-end reads were merged using USEARCH (v8.1.1831) with default parameters. 16S pipelines were tested, including EzBiome, DADA2-IDTAXA, DADA2-RDP, UCLUST-UCLUST, USEARCH-RDP, Taxa4Meta, Kraken2, and MetaPhlAn2. These were benchmarked using the aforementioned simulated amplicons and healthy human fecal microbiome data set. Collapsed taxonomic profiles from each pipeline were generated using the total sum scaling method without rarefaction procedure. The specific analysis procedure for each pipeline is described below. EzBiome pipeline. In EzBioCloud, the 16S microbiome taxonomic profiling (MTP) pipeline together with its pre-built database PKSSU4.0 was used for analysis of 16S data using default parameters. Since its output contains many genome accession IDs for species annotation, its species profile was not compared with the MetaPhlAn2 species profile. DADA2-IDTAXA pipeline. DADA2 (v1.8) was used to denoise amplicon data after quality filtering with a maximum expected error of 2 and a minimum length of 200 bases. IDTAXA together with its pre-built RDP training set (version 16; curated by program developer) was used for taxonomic annotation (down to genus rank) with confidence threshold of 70 using 100 bootstraps. DADA2-RDP pipeline. DADA2 (v1.8) was used to denoise amplicon data after quality filtering with a maximum expected error of 2 and a minimum length of 200 bases. RDP Naive Bayesian Classifier algorithm implemented in DADA2’s assignTaxonomy function together with its preformatted RDP training set (version 16) was used for taxonomic annotation (down to species rank) using minimum bootstrap confidence of 50. UCLUST-UCLUST pipeline. UCLUST (v1.2.22q) was used to cluster amplicon data with 97% sequence similarity after quality filtering with a minimum quality threshold of 20 and a minimum length of 140 bases. Representative sequences of OTUs were selected using pick_rep_set.py script default parameters. UCLUST implemented in assign_taxonomy.py script together with the SILVA database (release 123; choice of silva_132_97_16S.fna) was used for taxonomic annotation, down to species rank using minimum bootstrap confidence of 0.5. All procedures were completed in the QIIME platform (v1.9.1) ( 39 ). This pipeline is similar to the meta-analysis method used by Mancabelli et al. ( 21 ). USEARCH-RDP pipeline. USEARCH was used to cluster amplicon data with 100% sequence similarity after quality filtering with a maximum expected error of 2 and a minimum length of 200 bases. RDP classifier (v2.12) together with RDP training set (v16) was used for taxonomic annotation down to species rank using minimum bootstrap confidence of 0.5. This pipeline is similar to the meta-analysis method used by Duvallet et al. ( 20 ). Taxa4Meta pipeline. Taxa4Meta (v1.22) was used to cluster amplicon data after quality filtering with maximum expected error of 2 and selected range of variable lengths as described above. Taxonomic annotation was provided down to species rank. Benchmarking of Taxa4Meta confidence thresholds was performed using the previously described Korean human microbiome data set with (a) Tolerant setting: genus score of 0 and species score of 0; (b) Strict setting: genus score of 100 and species score of 100; (c) Default region-specific optimized thresholds in Taxa4Meta. Metagenomic pipelines. Paired-end sequences were trimmed and filtered to meet a maximum expected error of 2 with a minimum read length of 50. Kraken2 (v2.0.8) with its pre-built database (minikraken2_v2_8GB_201904_UPDATE) with default parameters was used for taxonomic profiling of shotgun metagenomic data. MetaPhlAn2 (v2.7.7) with its default database (mpa_v20_m200) and default parameters was used for taxonomic profiling of shotgun metagenomic data. Kraken2 family-level abundance results were used as the reference for comparisons across different 16S pipelines. Given the high precision of species identification, MetaPhlAn2 species-level abundance results were used as the reference for evaluating species calls by different 16S pipelines. A pseudo-sample was created by averaging of each species-level or family-level abundance of all 27 WGS samples; then the Spearman correlation or abundance-weighted Jaccard distance was calculated between the pseudo-sample and real-world samples analyzed by the different pipelines. Patient cohorts and clinical definitions In this study, a meta-analysis was conducted using 27 patient cohorts with available raw 16S sequencing data that had been previously published. An additional 13 patient cohorts were used for independent validation purposes ( Supplemental Table 3 ). Of the 40 data sets that were initially considered, it was found that 5 did not have any publications documenting specific clinical metadata. For the majority of cohorts, sample grouping information was available after NCBI BioSample registration during data deposition. In cases in which grouping information was missing, contact investigator follow-up was conducted for individual studies. For CDI cohorts, symptoms of diarrhea, PCR-based toxin gene detection, and enzyme immunoassay tests for C . difficile toxins were commonly reported for diagnosis per the 2017 Infectious Diseases Society of America/American Gastroenterological Association guidelines ( 40 ). However, it was noted that 2 Mayo cohorts (training data sets 24 and 25) ( 41 , 42 ) may not have adopted this practice. For IBD diagnosis, colonoscopy, Montreal classification, and disease activity index were used in the majority of cohorts ( 43 , 44 ). Disease severity in UC was measured using several quantitative methods, including the Mayo score and the Simple Clinical Colitis Activity Index, which have been found to correlate well with endoscopic disease activity ( 43 ). For CD diagnosis, the Crohn’s Disease Activity Index is commonly used ( 44 ). It should be noted, however, that the disease status was not clearly indicated for all patients and specimens. Therefore, data collected during active and remission stages were combined for meta-analysis and classification. Diagnosis of IBS and its disease status relied on a questionnaire and the Rome III criteria ( 45 ). Cases of IBD and IBS in 2 large-scale community data sets, namely the American Gut and LifeLines-Deep cohorts, contained self-reported metadata of prior clinical diagnosis provided by a physician and were only used for classifier validation. It should be noted that, since fecal samples from the American Gut cohort were transported at room temperature, microbial blooms (i.e., Gammaproteobacteria) were filtered out of the final taxonomic profiles, as previously described ( 15 ). Microbiome meta-analysis of diarrheal microbiome data sets Each diarrheal data set was processed through the Taxa4Meta pipeline using optimal taxonomic thresholds for each 16S variable region. The specific Taxa4Meta command used for each data set is indicated in Supplemental Table 3 . The relative abundance of collapsed species profiles generated from each Taxa4Meta OTU count table was used without rarefaction, but a minimum of 1,000 reads per sample was required. If a species was assigned by Taxa4Meta-BLCA, the taxonomic lineage from NCBI 16S RefSeq was adopted for that species to avoid any inconsistencies in the taxonomic lineage. The merging of Taxa4Meta collapsed profiles from all data sets was based on the taxonomic lineages. In calculating the percentage of classifiable sequences generated by Taxa4Meta, only clean reads that passed QC were used for proportional calculations, excluding reads assigned as human or PhiX sequencing controls. For benchmarking of the classification performance of input taxonomic profiles generated by Taxa4Meta, a side-by-side comparison with the DADA2-RDP pipeline was performed across all the meta-analysis training and validation cohorts. Diversity and pathobiome analyses Two α-diversity indices were calculated at OTU level: the Shannon index (alpha_diversity.py in QIIME v1.9.1) and the richness index (breakaway package v4.7.5). Unless otherwise stated, principal coordinate analysis (PCoA) with abundance-weighted Jaccard distance metric was applied for β-diversity analysis using combined collapsed species-rank profiles in QIIME v1.9.1. Analysis of similarities (ANOSIM) test for group comparison was performed using the β-diversity distance profile and 999 permutations. The taxonomic abundance of potential pathobionts (pathobiome) including Enterococcus , Streptococcus , Clostridioides , Escherichia / Shigella , Klebsiella , and Pseudomonas was calculated for each sample. Kullback-Leibler (KL) divergence analysis was performed between any 2 specific populations in the meta-analysis training cohorts: the relative abundance of pathobiome in each disease or case-control population was normalized using the total sum scaling method prior to analysis using the KL() function in the R package philentropy (v0.7.0). Hierarchical clustering and heatmap visualization At the OTU level, two α-diversity indices were computed: the Shannon index and the richness index. The Shannon index was calculated using the alpha_diversity.py script from QIIME v1.9.1, while the richness index was computed using the breakaway package v4.7.5. For β-diversity analysis, the PCoA with an abundance-weighted Jaccard distance metric was used, unless explicitly stated otherwise. The combined collapsed species-rank profiles were used for β-diversity analysis in QIIME v1.9.1. To highlight less abundant microbiome features, relative abundance profiles from each 16S pipeline and WGS pipeline were transformed using a –log 2 calculation, and the imputed value for microbiome features with a relative abundance of zero prior to clustering analysis was set as the maximum value of all transformed data. The R package pheatmap (v1.0.12) was used for heatmap generation and hierarchical clustering analysis. Clustering analysis was performed using the default parameters, with the Euclidean distance measure and complete method. Abundance-based Spearman correlation analysis was performed for species (L7 rank) and parent genera (L6 rank) for the entire training set. Fitting factors onto β-diversity ordination plot The process of placing fitting factors, or taxa, onto a 2-dimensional ordination plot based on the first 2 coordinates was carried out using the envfit function found in the vegan package (v2.5-7). To facilitate a side-by-side comparison, the taxonomic abundance profile at the family-rank level was used in this analysis. To establish the significance of the fitted factors, 999 permutations were implemented in the envfit run. Supervised classification and independent cohort validation All supervised classification procedures were executed using Orange software (v3.20) ( 46 ) on the reported cohorts with clinical definitions. To maintain consistency with previous studies, we adopted the original sample grouping information from each cohort, using gold standard diagnostic criteria for CDI, IBD, and IBS. In order to select the top 100 input taxa features for downstream supervised learning, we used random forest–based feature ranking as a first pass. All input samples were used for training, unless subsampling of samples was performed. Individual learning algorithms, including random forest (RF), support vector machine (SVM), naive Bayes (NB), and neural network (NN), were used for supervised classification. Furthermore, a stack model was evaluated as an aggregated meta-learner of RF, SVM, and NB. A 5-fold cross-validation method was applied for subsampling of training and test data during training, unless otherwise specified. The training results were subjected to receiver operating characteristic and precision-recall analyses using the R package precrec (v0.14.2), and values of area under the curve (AUC) and classification accuracy (CA) were calculated to evaluate the performance of each classification model. CA represents the proportion of correctly predicted samples from the classification model in comparison with the original clinical diagnosis. Independent validation of classification models was performed using data sets of recently published microbiome surveys of human diarrheal diseases that were not included in the training set. Taxonomic profiles were generated for validation of classification models using the Taxa4Meta pipeline and DADA2-RDP pipeline. CDI and IBD scores refer to the predicted scores of each sample from the binary classifier of the respective disease diagnosis. Statistics In the absence of any other specifications, we conducted comparisons between 2 groups using the nonparametric Mann-Whitney-Wilcoxon 2-tailed test. Likewise, comparisons involving more than 2 groups were made using the nonparametric Kruskal-Wallis 2-tailed test. To account for multiple comparisons and pairwise Spearman or Pearson correlations, we used the Benjamini-Hochberg FDR ( P < 0.05, considered statistically significant). Unless stated otherwise, box plots are presented with the interquartile range (IQR), median, and whiskers extended to values less than 1.5 × IQR from the first and third quartile, respectively. Study approval The study used deidentified sequencing and metadata available through publicly available databases with prior institutional IRB approval. Data availability Data accession numbers and reference to publicly available 16S data sets including 1 restricted data set (LifeLines-Deep cohort) are listed in Supplemental Table 3 . Values for all data points are available in the Supporting Data Values file. The source code for the Taxa4Meta pipeline is available at https://github.com/Savidge-lab/Taxa4Meta (Commit ID: 77dec14e0b41579ac02b724f9b957e640a833f06). Scripts for amplicon data simulation and benchmarking analyses can be accessed at https://github.com/Savidge-lab/Taxa4Meta-ParameterBenchmarking (Commit ID: 5a1b007359c689edfe6eda390a612f711ca2f748).

Results Optimal sequence length for accurate taxonomic profiling of 16S amplicons. Most pipelines that process 16S amplicon reads apply quality control (QC) procedures and trim the reads to short, equal lengths. This approach can introduce taxonomic and compositional bias ( Figure 1A ). An alternative method is to submit both short and long 16S reads for downstream processing, but this presents a bioinformatics challenge. Furthermore, the optimal amplicon length for sequence clustering/denoising and taxonomic resolution needs to be determined for each microbial species of interest. To assess the feasibility of this bioinformatics approach, we simulated 16S amplicon data with variable length and an identical allocated sequence count (randomly assigned from 1 to 50) to benchmark the accuracy of different sequence clustering/denoising tools. We found that for commonly used 16S variable regions (V1–V3, V3–V5, V4, and V6–V9), closed-reference analysis using UCLUST discarded a large proportion of amplicon reads, even when using a comprehensive reference database such as SILVA release 132. Moreover, the results were strongly biased toward higher sequence identity and longer reads ( Supplemental Table 1 ; supplemental material available online with this article; https://doi.org/10.1172/JCI170859DS1 ). While the DADA2 pipeline retained more reads in this simulated analysis, it still discarded more than 2% of sequences, with singleton reads being disproportionately excluded. By contrast, de novo clustering tools retained all sequence reads, setting a precedent for accurate compositional profiling ( Supplemental Table 1 ). To determine the optimal amplicon length thresholds and ranges for sequence clustering of variable length input data, we performed pairwise Spearman correlations between any 2 variable lengths (as 2 independent samples) in operational taxonomic unit (OTU)/amplicon sequence variant output tables. We found that applying 99% similarity for clustering amplicons in VSEARCH conferred the highest correlation coefficients across wider length ranges in all 16S variable regions tested ( Supplemental Table 2 ). Spearman coefficients increased progressively with longer reads, allowing us to establish minimum amplicon length thresholds (Spearman’s ρ > 0.75) and optimal amplicon length ranges for sequence clustering of variable length input data ( Figure 1B and Supplemental Figure 1 ). The selected amplicon sequence ranges were subsequently used to evaluate the accuracy of taxonomic annotation provided by qualified variable read lengths generated from various 16S regions. To achieve this, we used random and repeat sequences previously reported for benchmarking of taxonomic overclassification by Murali et al. ( 10 ). Our findings indicated that the default settings in the Bayesian-based Lowest Common Ancestor (BLCA) tool ( 11 ) did not annotate these sequences, while other commonly used taxonomic classifiers, including Ribosomal Database Project (RDP) classifier and SINTAX, produced high false-positive hits ( 10 ). It is important to note that random and repeat sequences do not accurately reflect uncharacterized or unidentified species that may contribute to taxonomical overclassification in a microbiome community. As a result, we used simulated amplicon data of unannotated 16S sequences (down to the family rank from the RDP database 11.5) to determine the optimal settings in BLCA. Our results suggested that taxonomic overclassification is heavily dependent on the 16S variable region, identity, and coverage of sequence alignment in BLCA ( Supplemental Figure 2 ). By increasing both identity and coverage thresholds of sequence alignment to 0.99, without applying bootstrap confidence thresholds for taxonomic selection, we were able to reduce overclassification rates to below 5% (for V1–V3, V3–V5, and V6–V9) and 10% (for V4) ( Supplemental Figure 2 ). Therefore, we used sequence identity and coverage thresholds of 0.99 in BLCA to conduct subsequent benchmarking. The aforementioned threshold settings were used in BLCA to annotate simulated amplicons of variable length, which were generated from known taxonomic lineages in the NCBI 16S RefSeq database. To determine taxonomic accuracy, we compared BLCA annotations with curated input lineage 16S data (ground truth). Optimal confidence scores and proportions of correctly assigned taxonomic annotations were calculated for each qualified sequence length, and our analysis revealed a significant increase in the proportion of correctly assigned amplicons with longer read length ( Figure 1, C and D , and Supplemental Figure 3 ). Interestingly, we also observed a significant increase in confidence scores for incorrect annotations with longer read length, and that misclassification rates were highly dependent on 16S sequence orientation and the variable region analyzed ( Supplemental Figure 3 ). This finding is related to the observation that increasing amplicon length generally improved taxonomic accuracy at the species rank compared with the genus rank, as the latter has a greater capacity for degeneracy. Based on our finding that universal confidence thresholds should not be applied to all types of 16S amplicon data, we determined optimal region-specific confidence thresholds to achieve accurate taxonomic annotation for all common types of amplicon data that could be used in a meta-analysis ( Supplemental Figure 4 ). This conceptual approach provided the foundation for our 16S meta-analysis using a new taxonomic binning strategy. Taxa4Meta: a “best practices” taxonomic profiler for 16S meta-analysis. Based on our benchmarking results of simulated 16S amplicon data, we developed the bioinformatics pipeline Taxa4Meta to enable accurate taxonomic profiling of 16S ribosomal DNA amplicon data generated from different sequencing strategies ( Figure 2A ). The pipeline was designed to maximize the utilization of clinically archived 16S data sets by employing a variable sequence length analysis strategy that can be applied to multiple amplicon regions. To achieve precise taxonomic profiles, we implemented 2 key workflow-specific settings. First, VSEARCH-based de novo sequence clustering with 99% similarity was used to cluster the 16S amplicon data, while keeping in mind the optimal sequence length range identified for each amplicon data type. Second, we used BLCA with stringent sequence alignment criteria (99% identity and 99% coverage) to obtain confident species calls, while applying region-specific confidence scores as determined above. Any OTUs not annotated by BLCA were processed by the IDTAXA program, which utilized its pre-built RDP training set (version 16; curated by the program developer) for classification purposes. Finally, we generated Taxa4Meta feature tables by collapsing the taxonomy of de novo OTUs down to the species rank without processing for random rarefaction, which could potentially result in a biased taxonomic profile. To evaluate the taxonomic profiling accuracy of Taxa4Meta, we generated complex mock communities comprising defined and cultivable bacteria as benchmarking input. Initially, we simulated variable length amplicons of diverse 16S sequences sourced from the NCBI 16S RefSeq database. This database consists of over 20,000 bacterial strains representing more than 14,000 species from over 2,900 genera. We selected amplicon length ranges for benchmarking Taxa4Meta that provided optimal sequence clustering and taxonomic annotation for each distinct 16S variable region, as illustrated in Supplemental Figures 1 and 3 . To assess Taxa4Meta’s performance, we critically compared it against state-of-the-art 16S pipelines and the curated input data (ground truth). As commonly used 16S pipelines rely on different reference databases for taxonomic annotation, we interpreted taxonomic profiles at the family rank, which is more consistently represented across databases compared with genus and species ranks. We used simulated data sets containing defined sequence abundances and taxonomic lineages (ground truth) to generate Spearman correlations and compared qualified input data with compositional profiles generated by individual 16S pipelines. In side-by-side comparisons of Taxa4Meta against EzBiome-, DADA2-, UCLUST-, and USEARCH-based 16S pipelines, we demonstrated that Taxa4Meta outperformed the other taxonomic profilers, generating significantly higher Spearman correlation coefficients across all 16S regions tested ( Figure 2B ). Using an independent method of hierarchical clustering, we further demonstrated that only Taxa4Meta profiles clustered with ground truth input profiles ( Supplemental Figure 5A ). In contrast, other taxonomic profilers failed to detect a significant number of families across the four 16S regions tested. Specifically, up to 30% of families were not detected, depending on the specific taxonomic profiler, compared with only 0.4% omitted by Taxa4Meta ( Supplemental Figure 5B ). These results underscore the utility of Taxa4Meta in generating accurate taxonomic profiles of complex microbiome communities, which is evident down to species rank, as demonstrated by the stringent detection of C . difficile , a pathogen required for the clinical diagnosis of CDI ( Supplemental Figure 5C ). To evaluate how Taxa4Meta performs with real-world microbiome data sets, we benchmarked different 16S pipelines using a cohort of healthy subjects ( 12 ). Here, individual fecal DNA extracts underwent comprehensive 16S profiling and shotgun metagenomic sequencing. Similar to our observations with complex simulated microbiome communities, Taxa4Meta family-rank profiles generated from Illumina sequencing platforms provided significantly more sequencing depth than 454 pyrosequencing and clustered together with Kraken2-generated annotations ( Supplemental Figure 6 , A and B). Kraken2 was regarded as a gold standard reference method given its high family-rank taxonomic accuracy using metagenomic data ( 13 ). Furthermore, adopting an independent method of pairwise abundance-weighted Jaccard distance calculations, Taxa4Meta profiles were found to have the best close distance to Kraken2 profiles ( Figure 2C ), which was consistently observed across all 16S data types investigated, regardless of sequencing depth ( Supplemental Figure 6 , B and C). We also evaluated the accuracy of Taxa4Meta species-rank profiles compared with MetaPhlAn2-generated taxonomy, which has higher precision in avoiding species misclassification ( 14 ). We found that Taxa4Meta stringently controlled for species misclassification ( Supplemental Figure 7A ), and its species abundance profiles showed significantly improved correlations with MetaPhlAn2 species profiles ( Supplemental Figure 7B ). Moreover, our benchmarking analysis demonstrated that the optimized (default) parameters within Taxa4Meta exhibited a higher degree of consistency in correlation results with the reference profile. This higher specificity is in contrast to the varied outcomes observed when stricter or more lenient parameter settings were employed in Taxa4Meta for taxonomic profiling across diverse platforms and regions ( Supplemental Figure 7C ). Collectively, our findings demonstrate that collapsed taxonomic profiles generated by Taxa4Meta are highly accurate and suitable for 16S meta-analysis of amplicon data generated from diverse sequencing strategies. Population-scale meta-analysis to define the healthy human gut microbiome. Defining the healthy human gut microbiome is a significant challenge because of the numerous individual factors that influence it, including age, genetics, diet, environment, lifestyle, and transmission ( 2 ). In addition to these factors, inconsistent analytical methods and small cohort sizes play a crucial role in determining the reliable characterization of the healthy human microbiome. To address these challenges, we used the Taxa4Meta pipeline to perform a meta-analysis of diverse 16S regions and sequencing platforms, identifying common microbiome features in over 900 subjects with no documented gastrointestinal (GI) disease across North America, Europe, Asia, and Australasia ( Supplemental Table 3 ). We further compared the taxonomic profiles of control subjects with those of over 13,000 participants in the American Gut Project ( 15 ) and LifeLines cohorts ( 16 ). Our findings using Bray-Curtis dissimilarity distance-based β-diversity analysis showed that control subjects sequenced across diverse technology platforms shared a similar sample distribution or microbiome variation pattern with the American Gut cohort at both genus- and family-rank abundance profiles ( Supplemental Figure 8 ). We also identified a significant enterotype bias when comparing the American Gut cohort or meta-analysis controls with the European LifeLines cohort. Therefore, we designed our meta-analysis to include study controls spanning all the major classical gut enterotypes to facilitate accurate downstream disease classification at a population-scale level. Our analysis revealed that the healthy gut microbiome in controls from our 16S meta-analysis cohorts was dominated by non- Prevotella enterotypes, which were largely composed of Bacteroidaceae, Lachnospiraceae, and Ruminococcaceae ( Supplemental Figure 8 ). Furthermore, we identified some outlier controls that were dominated by high abundance of pathobiome, which was defined as the presence of Enterococcus , Streptococcus , Clostridioides , Escherichia / Shigella , Salmonella , Klebsiella , and Pseudomonas ( Figure 3A ). Given that Prevotella and pathobiome-dominated gut microbiota are associated with chronic inflammatory conditions ( 5 , 17 , 18 ), our results emphasize the importance of population-scale analyses that consider enterotypes and pathobiome when defining the healthy human microbiome, particularly in the context of dysbiosis-associated GI disease. Dysbiosis in chronic human diarrheal disease. Using the Taxa4Meta pipeline, we conducted an analysis of fecal microbiome data obtained from multiple 16S regions sequenced on Illumina and 454 pyrosequencing platforms. Our study involved the examination of more than 5,500 matched controls and clinically confirmed diarrheal patients with various conditions, including CDI, IBD, IBS, and non-IBS functional gastrointestinal disorders (FGIDs) from diverse geographical regions including North America, Europe, Asia, and Australasia. Our inclusion criteria for clinical cohorts required adherence to internationally recognized diagnostic guidelines ( Supplemental Table 3 ) and the use of 16S amplicon data that met our QC standards. We calculated α-diversity indices (Shannon and richness) from Taxa4Meta feature tables, which revealed that CDI cases had significantly lower diversity compared with controls or other diarrheal diseases ( Supplemental Figure 9A ). However, there was inconsistency in α-diversity indices among clinical cohorts sequenced across different 16S regions ( Supplemental Figure 9B ). The Taxa4Meta pipeline classified 85% of total sequences that successfully passed QC across the various meta-analysis cohorts. This outcome indicates that this data set comprehensively represents the substantial proportion of mined sequence reads, thus underpinning the robustness and reliability of the data interpretation ( Supplemental Figure 10A ). The collapsed species profiles generated by Taxa4Meta included both classified and unclassified members, representing 54% and 46% of total abundance, respectively ( Supplemental Figure 10B ). This allowed for confident assignment of species calls and further data mining. An abundance-weighted Jaccard distance-based β-diversity analysis revealed a healthy-like microbiome community structure in patients with IBS and ulcerative colitis (UC), while significant dysbiosis was consistently detected in cases of CDI and Crohn’s disease (CD) ( Figure 3A ). This observation is consistent with prior case-control matched studies ( 19 ), which reported subtle microbiome differences in IBS and FGIDs compared with healthy controls ( Supplemental Figure 11 ). In contrast, we found that CDI and CD patients significantly differed from patients with other diarrheal diseases in terms of pathobiome abundance ( Figure 3A ), reflecting gut dysbiosis favoring engraftment and expansion of potential pathogens. Our findings corroborate previous research ( 5 , 18 ), which found that the abundance of Enterobacteriaceae and Enterococcaceae was significantly higher in CDI and CD patients compared with matched controls or IBS and UC patients ( Figure 3B and Supplemental Figure 11 ). Pathobiome-dominated microbiome communities in CD and CDI patients primarily consisted of Enterobacteriaceae and Enterococcaceae, as demonstrated independently of 16S region, sequencing platforms, age, or geography ( Supplemental Figure 11 ). To further explore pathobiome compositional differences in patients versus disease controls, we conducted a Kullback-Leibler divergence analysis ( Figure 3C ). Disparity was particularly pronounced among patients diagnosed with IBS and FGID, in whom a diminished abundance of pathobiome was observed. This subtlety could not be readily discerned through conventional statistical methodologies ( Figure 3, A and B ) and is noteworthy in that while pathobionts are typically characterized as minor constituents of the IBS microbiome, their implication in the pathogenesis of IBS is well established ( 4 ). Therefore, specific β-diversity distance metrics and pathobiome abundance calculations are useful tools for defining the core microbiome features of specific diarrheal disease types. We conducted an abundance-based Spearman correlation analysis to examine the relationship between identified species and their respective parent genera across the comprehensive cohorts within the meta-analysis. The results indicated that a significant correlation is observed for the majority of identified species with their corresponding parent genera. However, notable disparities in the Spearman ρ values were evident for some identified species ( Figure 3D ). This observation suggested that specific cohort and technical variations could potentially affect the accuracy of some feature detection. Notably, species exhibiting lower correlation values might not consistently demonstrate the identical pattern observed in the genus-based abundance differential analysis. Nevertheless, using hierarchical clustering analysis to examine family abundance profiles, we demonstrated that 4 of 8 UC cohorts were clustered together with control and IBS patients. Meanwhile, the remaining UC cohorts and the majority of CD cohorts formed a distinct IBD-specific cluster ( Supplemental Figure 11 ). This noteworthy finding was not replicated by the microbiome meta-analysis conducted by Duvallet et al. ( 20 ), as UC and CD patients were combined for microbiome comparisons against controls. In a recent systematic literature review, a significant reduction in Faecalibacterium prausnitzii , an antiinflammatory gut commensal, was reported in both UC and CD patients ( 5 ). Our meta-analysis of CD and UC cases corroborated this finding ( Supplemental Table 4 ). However, we were unable to demonstrate significant alterations in Eubacterium rectale and Escherichia coli abundance in UC patients, as previously reported ( 5 ). These discrepancies may reflect microbiome variations seen in UC cohorts, as demonstrated in our meta-analysis. We identified several previously unappreciated top-ranked disease-associated species, including Fusicatenibacter saccharivorans (control-specific) and Bacteroides xylanisolvens and Romboutsia timonensis (less prevalent in IBD), which have not been reported in prior studies. Our unique findings also included decreased relative abundances of Anaerostipes hadrus and Eubacterium rectale in CDI patients only ( Supplemental Table 4 ), features that we exploited to develop disease-specific classifiers. Collectively, the findings of our study revealed both consistent and inconsistent outcomes generated by Taxa4Meta across diverse platforms and regions. While the taxonomic profiles demonstrated remarkable consistency, certain discrepancies necessitate focused attention. Specifically, the following observations warrant consideration: (a) Taxonomic profiles derived from 16S V6–V9 data exhibited a distinct separation from profiles originating in other regions ( Supplemental Figure 5A ). (b) Platforms employing 454 pyrosequencing tended to yield a reduced number of amplicon reads in comparison with Illumina platforms, leading to potential oversight in detecting critical microbiome features ( Supplemental Figure 6 , A and B). (c) Within our meta-analysis cohorts, a conspicuous demarcation emerged not only between 454 and Illumina platforms within the control population but also between V4/V3–V4 regions and alternative regions ( Supplemental Figure 11 ). These identified deviations hold the potential to impact downstream applications. Therefore, accurate classification analysis must adequately address region-specific or platform-specific effects to ensure the robustness and reliability when disease-specific classifiers are selected. In part, we overcame these limitations using the pan-microbiome profiling strategy described below. Pan-microbiome profiling outperforms individual 16S region–specific or platform-specific analysis for disease classification. Disease classification represents a crucial emerging application of gut microbiome surveys for biomarker discovery. To investigate the potential benefits of pan-microbiome profiling in disease classification, we merged core microbiome communities that had been adjusted for demographic and technical bias. In this context, we defined pan-microbiome as representing core microbiome features identified across different sequencing strategies. To assess the efficacy of our approach, we conducted pilot studies using different sequencing modalities, focusing on our center’s Human Microbiome Project (HMP) cohort of pediatric FGID cases. We used 16S V1–V3 and V3–V5 amplicons generated on the 454 pyrosequencing platform to profile the cases. Our analysis of β-diversity from collapsed Taxa4Meta taxonomy profiles did not separate FGID cases from healthy controls ( Figure 4A ). As expected, we observed suboptimal classification accuracy (CA < 0.85) when profiling individual V1–V3 and V3–V5 data sets ( Figure 4B ). To identify core microbiome genera that discriminated between FGID and healthy controls, we used feature ranking generated by the random forest algorithm. We selected >85% of genera abundance features that were common to both 16S regions, which identified Roseburia , a previously underappreciated genus, as a top and consistent core microbiome feature ( Supplemental Figure 12 ). Our results indicated that supervised training of pan-microbiome profiles significantly improved classification accuracy (CA) when compared with individual microbiome surveys ( Figure 4B ). As another example, we conducted an analysis on amplicon data derived from multiple CDI cohorts, which were generated using various sequence deposits from different 16S regions and technology platforms ( Supplemental Table 3 ). In contrast to the subtle differences observed in FGID cases, CDI patients displayed a consistent and severe dysbiosis, which was evident across multiple geographic locations and sequencing methods ( Figure 4C ). Using an approach similar to the one mentioned above, classification models specific to the platform demonstrated good performance during the training phase in distinguishing CDI from controls. However, these models were unable to cross-validate subjects across different sequencing platforms, thereby posing a significant limitation for meta-analysis. This classification inaccuracy was overcome by use of merged pan-microbiome profiles for training ( Figure 4D ). By minimizing the impact of pattern variation and retaining common microbiome features, pan-microbiome patterns facilitated the discovery of biomarkers. Utility of pan-microbiome features for diarrheal disease classification. We employed 2 distinct strategies to generate comprehensive and binary disease classification models, using deposited 16S amplicon data based on pan-microbiome profiles ( Supplemental Figure 13A ). The primary objective of developing classification models was to effectively differentiate between CDI, IBD, and IBS patients using alternative microbiome-based classifiers. The diagnosis of diarrheal patients included in the study was based on internationally recognized clinical guidelines, as outlined in individual clinical cohorts ( Supplemental Table 3 ). The original sample grouping information provided in each published cohort was used to develop our classification models. As disease subgroup information was not consistently provided for all individual patients, we generated disease classifiers that combined the respective IBD or IBS subgroups. We also included IBS-constipated cases in our meta-analysis, given that these patients often exhibit alternating symptoms, and no significant differences in microbiome community structure were observed, as previously reported ( 19 ). Using Taxa4Meta taxonomic profiles, we identified several key features, including presence of C . difficile ( Supplemental Figure 13B ), as the top discriminating features that are pathophysiologically relevant in differentiating CDI from other diarrheal diseases. It is worth noting that this top-ranking feature was not identified as a classifier in 2 prior microbiome meta-analyses, highlighting the technical bias in previous studies ( 20 , 21 ). Taxa-based classification models were developed for the 5 clinical groups under investigation (control, CDI, IBS, UC, and CD) and demonstrated excellent AUC results, but moderate CA scores, indicating suboptimal disease classification across all cohort groups ( Supplemental Figure 13C ). We reasoned that this underperformance could be attributed to the similarity of microbiome features in control, UC, and IBS subjects, and as such represented a challenge for reliable cross-classification ( Figure 3A ). Nonetheless, in contrast to the multiple-group classification, binary models provided excellent disease classification, with improved AUC and CA scores, especially when differentiating CDI from other patients and healthy controls ( Supplemental Figure 13D ). Prototypical workflow for clinical diarrheal disease classification. With the urgent need to differentiate common symptoms in CDI, IBD, and IBS, we assembled a prototypical workflow to assist in stratifying these patients based on our Taxa4Meta-generated binary algorithms ( Figure 5A ). We prioritized the need to diagnose CDI based on the clinical necessity for rapid treatment and patient contact isolation. By applying a binary classifier that differentiated CDI from combined IBD or IBS subtypes, we demonstrated a CA of 0.95 ( Figure 5A ). Although we rationalized employing disease subtype–agnostic classifiers, this decision was underpinned by our substantiated demonstration of CA in the context of CDI cases. However, it is noteworthy that our CDI-centric model encountered limitations in effectively distinguishing between IBD and IBS disease subtypes ( Figure 5B ). In light of this, we pursued a secondary objective that entailed the development of an additional binary classifier. This classifier was designed to discriminate between combined instances of IBD (encompassing both UC and CD) and cases of IBS. Remarkably, the resultant classifier yielded an overall accuracy of 0.96 ( Figure 5A ), indicating that our pan-microbiome analytical approach has potential diagnostic utility beyond CDI. To independently validate our 2-step diagnostic workflow, we tested 16S data generated from (a) recently published clinical CDI, IBD, and IBS microbiome cohorts, and (b) real-world data obtained from self-reported IBD and IBS cases in the American Gut and LifeLines population cohorts. Our classifiers exhibited a robust performance, identifying CDI patients at a rate of 93.6%. Furthermore, the overall accuracy of 0.97 achieved in discriminating between clinically confirmed instances of IBD and IBS ( Figure 5B ) underscored the proficiency of our approach. The validation cohorts used in this study serve as robust substantiation of the viability of pan-microbiome–based classification in the advancement of companion diagnostics aimed at the stratification of diarrheal diseases. To compare the classification performance of Taxa4Meta with that of other state-of-the art 16S profilers, we conducted a benchmark analysis using the taxonomic profiles generated via both the Taxa4Meta and DADA2-RDP pipelines. This assessment was conducted across the same meta-analysis and validation cohorts. Notably, our findings revealed that Taxa4Meta exhibited a superior capacity, for example in detecting instances of C . difficile within CDI patients in comparison with the DADA2-RDP pipeline. Despite these disparities in performance, it is noteworthy that no significant differences were observed in the training statistics of the 2 classification models ( Supplemental Figure 14 , A and B). However, Taxa4Meta emerged as notably more accurate than the DADA2-RDP pipeline when validating the models through independent cohorts ( Supplemental Figure 14C ). This divergence was particularly conspicuous using the Taxa4Meta classification models, which significantly outperformed their DADA2-RDP counterparts across all 3 categories of diarrheal diseases ( Supplemental Figure 14D ). Finally, to ascertain whether antibiotic exposure represented the dominant determinant in categorizing a patient’s classification as CDI, we undertook a series of subanalyses of the data sets in the meta-analysis. Notably, in primary instances of CDI in which no prior antibiotic exposure was reported, a comparable classification score and β-diversity clustering pattern were evident with both primary and recurrent cases where antibiotics were administered ( Supplemental Figure 15A ). Given the acknowledged challenges in capturing accurate records of prior antibiotic exposure in CDI patients, we broadened our investigation to include healthy volunteers who underwent diverse antibiotic treatment regimens ( Supplemental Figure 15B ). Upon analysis of the longitudinal microbiome data of individuals subjected to single antibiotic treatments, CDI classification scores were generally not achieved, especially with administration of broad-spectrum β-lactam antibiotics, which induced subtle gut microbiome alterations ( 22 ). Although a transient CDI classification score became apparent in some individuals with clindamycin, this was not evident after ciprofloxacin exposure, both of which are recognized as high-risk antibiotics with regard to CDI development ( 23 ). Further, even though antibiotic use is common in IBD patients, our classification models still confidently differentiated IBD from CDI cases. These subanalyses unveil a distinct microbiome signature in CDI patients that is not exclusively associated with antibiotic exposure. These findings represent a significant advance in diarrheal disease classification, highlighting that compositionally distinct microbiome communities are discernible between infectious colitis (CDI), IBD, and IBS patients.

Discussion The field of microbiome science is a rapidly evolving area of research. Recent advancements in sequencing strategies and bioinformatics have significantly improved our understanding of host-microbiota interactions ( 1 – 3 ). Nonetheless, it is important not to disregard the value of retrospective microbiome data. The scientific community recognizes the significance of previous sequencing efforts that have investigated microbiome community dynamics in human pathogenesis ( 4 , 5 , 20 , 21 ), as these studies could collectively offer vital insights into disease-specific associations. However, individual clinical microbiome surveys often employ cohort-specific sequencing platforms, 16S primer regions, and bioinformatics pipelines, which we and others systematically demonstrate require a consolidated bioinformatics approach to mitigate technological and demographic bias, as well as taxonomic misclassification ( 24 , 25 ). Our research findings indicate that previous microbiome meta-analyses have not adequately addressed these limitations ( 20 , 21 , 26 ). To address this gap, we developed Taxa4Meta, a bioinformatics pipeline that ensures accurate taxonomic profiling by systematically benchmarking sequence orientation and length, so that data output can be reliably utilized from different 16S variable regions. Given the challenges of accurately merging OTU/amplicon sequence variant tables generated from different 16S variable regions, we implemented a new binning approach by collapsing taxonomic annotations of Taxa4Meta feature profiles, which facilitates meta-analysis of diverse 16S amplicon data. Supervised classification is a significant downstream application of clinical microbiome surveys, particularly for GI diseases, where altered community dynamics are commonly observed ( 4 , 5 , 18 , 19 , 27 – 30 ). The construction of large, curated databases is typically required for diagnostic workflows to facilitate cohort-specific classifier training and cross-validation of disease-specific biomarkers. Population-scale meta-analysis is an appealing approach for powering microbiome surveys for disease classification, as it enables control of large variations in human genetics and demographics ( 2 ), as well as the technology bias ( 12 ) that contributes to false discovery rates. It is worth noting that the application of the Taxa4Meta pipeline to identical DNA extracts sequenced using different strategies revealed several prominent limitations in disease classification due to this bias. To overcome these technological hurdles, we developed a pan-microbiome profiling concept that achieves superior disease classification accuracy. The initial step in enabling accurate case-controlled disease comparisons ( 2 ) is to establish a clear definition of the healthy human gut microbiome, as we have done in our study. We have also developed robust binary classifiers for CDI, IBD, and IBS using pan-microbiome profiles. These classifiers were independently validated using clinical and real-world population-scale cohorts that were excluded during the construction of our classification models. Our binary 16S-based classifiers exhibited superior classification accuracy compared with a previously reported shotgun metagenomics survey of IBD and IBS patients ( 27 ). While shotgun metagenomics profiling demonstrates high precision for species calls, the taxonomic abundance accuracy of the entire microbiome community is heavily reliant on sequencing depth and the reference genome databases used ( 31 ). These limitations are less pronounced using 16S profiling because comprehensive databases are already available and higher detection sensitivity can be readily achieved using this method. Our application of the pan-microbiome profiling strategy to population-scale 16S amplicon data also revealed prominent enterotypes that should be considered when developing clinical diagnostic pipelines. Our study successfully employed Taxa4Meta as a prototypical pan-microbiome–based profiling strategy for diarrheal disease classification. However, there are several potential limitations associated with this approach. First, we relied on publicly deposited cross-sectional metadata, which often lacked detailed clinical confounders such as disease activity, comorbidities, medication use, and dietary modulators including prebiotics and probiotics. Therefore, prospective clinical trials with detailed clinical metadata and dietary and pain symptom diaries are required to validate our diagnostic classifiers, with special attention paid to patients with overlapping comorbidities, for example, CDI cases with IBD symptoms. Second, the limited availability of clinical metadata prevented us from identifying all patients with potentially confounding antibiotic use. In addition, specific diarrheal subtype information was not available for all cases used in training IBD and IBS classification models. Further studies are required to generate disease subtype–specific classifiers, as well as more careful consideration of how to classify patients with no overt dysbiosis. Nevertheless, our meta-analysis was primarily focused on reliably differentiating CDI from IBD and IBS patients, and we show that many of these confounders are unlikely to adversely impact feasibility. Third, although our classifiers are based on high-confidence genus-rank features, short 16S amplicon reads resulted in many unannotated species. Whole-genome sequencing (WGS) and 16S long-read sequencing strategies could provide species- and strain-level annotation ( 32 ), but this requires a different profiling strategy. Additionally, retrospective data are currently not available to adequately power such a meta-analysis using deep or long-read sequencing. In this regard, we demonstrate that Taxa4Meta features can be linked to WGS data in parallel analysis to provide deeper taxonomic insight if needed. Finally, sequencing depths between 454 and Illumina platforms significantly impact the sensitivity of microbial detection. To avoid data rarefaction, which introduces bias to abundance profiles used for biomarker identification, we used a pan-microbiome approach. This approach minimizes technical variation for downstream classification, but it remains a common challenge for any microbiome meta-analysis that incorporates 454 data. In summary, our study addressed a significant bioinformatics challenge by utilizing a new workflow (Taxa4Meta) to accurately cluster sequences and annotate taxonomy across multiple 16S regions. Taxa4Meta was applied to comprehensively reanalyze diverse 16S data sets generated from multiple retrospective GI disease cohorts investigated across four continents. By combining collapsed species abundance for each 16S data set, we successfully interpreted the downstream microbiome and performed supervised classification of diarrheal patients who are difficult to diagnose because of overlapping symptoms. This “best practices” approach allowed us to develop a prototypical diagnostic workflow based on disease-specific pan-microbiome biomarkers.

Targeted metagenomic sequencing is an emerging strategy to survey disease-specific microbiome biomarkers for clinical diagnosis and prognosis. However, this approach often yields inconsistent or conflicting results owing to inadequate study power and sequencing bias. We introduce Taxa4Meta, a bioinformatics pipeline explicitly designed to compensate for technical and demographic bias. We designed and validated Taxa4Meta for accurate taxonomic profiling of 16S rRNA amplicon data acquired from different sequencing strategies. Taxa4Meta offers significant potential in identifying clinical dysbiotic features that can reliably predict human disease, validated comprehensively via reanalysis of individual patient 16S data sets. We leveraged the power of Taxa4Meta’s pan-microbiome profiling to generate 16S-based classifiers that exhibited excellent utility for stratification of diarrheal patients with Clostridioides difficile infection, irritable bowel syndrome, or inflammatory bowel diseases, which represent common misdiagnoses and pose significant challenges for clinical management. We believe that Taxa4Meta represents a new “best practices” approach to individual microbiome surveys that can be used to define gut dysbiosis at a population-scale level. A pan-microbiome-based profiling strategy facilitates meta-analysis of diverse 16S amplicon data for diarrheal disease classification

Author contributions QW and TCS led the conceptualization and data interpretation of the study. QW, SB, SYS, and TJT performed the data acquisition and analysis. QW and TCS drafted the manuscript. Supplementary Material

The authors thank the LifeLines data management team for providing access to the microbiome data of the LifeLines-Deep cohort, and EzBiome Inc. for granting access to the 16S MTP pipeline in EzBioCloud. This work was supported by NIH grants from the National Institute of Diabetes and Digestive and Kidney Diseases (P30-DK56338 and R01-DK130517), National Institute of Allergy and Infectious Diseases (R01-AI10091401, U01-AI24290, and P01-AI152999), and National Institute of Nursing Research (R01-NR013497). Address correspondence: Tor C. Savidge, TCH Department of Pathology, Baylor College of Medicine, 1102 Bates Avenue, Houston, Texas 77030, USA. Phone: 832.824.2839; Email: [email protected]. 11/14/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:16

J Clin Invest.; 134(2):e170859

oa_package/9e/dc/PMC10786686.tar.gz

PMC10786687

0

Introduction Cues from the innate immune system determine the priming, recruitment, and functionality of adaptive immunity. This crosstalk is generally mediated through antigen-independent pattern recognition receptor (PRR) signaling and results in costimulation, antigen presentation activity, and cytokine/chemokine signaling ( 1 ). Spontaneous populations of antitumor T cells have long been noted in the context of both murine and human cancers ( 2 – 5 ), despite the absence of apparent pathogen infection. Over the last decade, the stimulator of interferon (IFN) genes (STING) protein has emerged as a critical player in mediating endogenous PRR-induced inflammation that enables the priming of spontaneous antitumor T cell responses ( 6 ). STING senses cyclic dinucleotides (CDNs; e.g., 2′,3′-cGAMP) generated by cyclic GMP–AMP synthase (cGAS) and other viral dsDNA sensors, as well as those produced by intracellular bacteria, to induce TBK1/IRF3– and NF-κB–driven proinflammatory responses ( 7 ) ( Figure 1A ). DNA or CDNs from dying cancer cells can trigger STING signaling to induce antigen presentation activity, type I IFN signaling required for appropriate T cell priming by antigen-presenting cells (APCs), costimulatory ligand expression on APCs, and chemokines (e.g., CXCL9 and CXCL10) that enable T cell trafficking to the tumor site ( 8 ). Accordingly, STING agonists were proposed to mediate antitumor immunotherapy and are being developed for clinical use — primarily as intratumoral in situ vaccines — despite initial setbacks from early clinical trials. Moreover, numerous preclinical studies have demonstrated the role of STING signaling, particularly in APCs, on tumor antigen cross-presentation and antitumor T cell immunity after various treatments, e.g., chemo/radiation ( 9 ), CD47 blockade ( 10 ), telomerase-targeting agents ( 11 ), and DNA damage ( 12 ). However, the understanding of how STING signaling occurs within the native tumor microenvironment (TME) as well as the routes by which STING signaling can be leveraged for cancer immunotherapy lack mechanistic understanding. Moreover, the expression of cGAS and STING is notably low in tumor cells from several cancer types, particularly in central nervous system (CNS) tumors ( 13 , 14 ). Here we discuss compartment-specific characteristics of STING expression and signaling in the TME; recent evidence for STING pathway suppression in the CNS; efforts to target STING therapeutically in cancer clinical trials; and potential future implications of STING modulation for the efficacy of virotherapy, standard-of-care chemo/radiation, and other modalities being investigated for targeting CNS tumors ( Figure 1B ).

Since the discovery that cGAS/STING recognizes endogenous DNA released from dying cancer cells and induces type I interferon and antitumor T cell responses, efforts to understand and therapeutically target the STING pathway in cancer have ensued. Relative to other cancer types, the glioma immune microenvironment harbors few infiltrating T cells, but abundant tumor-associated myeloid cells, possibly explaining disappointing responses to immune checkpoint blockade therapies in cohorts of patients with glioblastoma. Notably, unlike most extracranial tumors, STING expression is absent in the malignant compartment of gliomas, likely due to methylation of the STING promoter. Nonetheless, several preclinical studies suggest that inducing cGAS/STING signaling in the glioma immune microenvironment could be therapeutically beneficial, and cGAS/STING signaling has been shown to mediate inflammatory and antitumor effects of other modalities either in use or being developed for glioblastoma therapy, including radiation, tumor-treating fields, and oncolytic virotherapy. In this Review, we discuss cGAS/STING signaling in gliomas, its implications for glioma immunobiology, compartment-specific roles for STING signaling in influencing immune surveillance, and efforts to target STING signaling — either directly or indirectly — for antiglioma therapy.

Compartment-specific roles for STING signaling in the TME While the importance of STING signaling in priming antitumor immunity is well established, less is known about the precise contributions of STING activation in the cells comprising the tumor and TME (i.e., neoplastic cells, immune cells, and stroma/vasculature) to antitumor immunity. When implanted with intracranial GL261 murine glioma cells, mice lacking functional STING had shorter survival than wild-type counterparts, showed increased immature myeloid suppressor cells and regulatory T cells, and decreased IFN-γ + CD8 + T cells ( 15 ). In melanoma, STING signaling is necessary in APCs for spontaneous T cell priming, wherein tumor-derived DNA stimulates host APC production of type I IFNs ( 8 ) ( Figure 2 ). In addition, selectively inducing STING signaling in dendritic cells (DCs) was also shown to engage antitumor T cells more effectively than nontargeted STING activation in mice ( 16 ). Collectively, these and other reports establish an important role for STING signaling in the TME. Recent work also indicates roles for cGAS/STING signaling in the neoplastic compartment in dictating both cancer cell fitness and immune surveillance. KRAS/LKB1–mutant lung cancer cells suppress STING expression, wherein exogenous expression of STING led to the detection of cytoplasmic mitochondrial DNA, induction of TBK1/IRF3 innate signaling, and decreased cell fitness ( 17 ). Collectively, these findings imply that loss of STING function may suppress immunogenic and cytotoxic effects of DNA damage in cancer. In melanoma, neoplastic cell STING activation enhanced antigenicity, induced MHC class I, and improved CD8 + T lymphocyte–mediated killing of melanoma cells in vitro ( 18 , 19 ). Similarly, STING expression in small cell lung cancer correlated with MHC class I expression and responsiveness to immune checkpoint blockade ( 20 ). Inhibition of topoisomerase resulted in tumor cell DNA damage and STING activation, which induced type I IFN responses that potentiated PD-1 blockade therapy ( 21 ). These results collectively suggest a role for neoplastic cell–intrinsic STING signaling in controlling the immunogenicity of tumor cells during the effector phase. In addition, cGAS activity in neoplastic cells may accentuate cancer cell sensitivity to DNA damage during radiation/chemotherapy ( 22 ). Release of CDNs from malignant cells can trigger STING signaling in APCs in a paracrine manner via the CDN transporter SLC19A1 ( 23 ) or by a recently described plasma membrane–localized STING isoform ( 24 ). STING activation in APCs potentially further accentuates their activation and capacity to prime T cells ( Figure 2 ). STING pathway suppression in glioblastoma STING signaling is suppressed in several cancers, including melanoma ( 25 ), colon cancer ( 26 ), and KRAS/LKB1–mutant lung cancer, perhaps as a consequence of STING’s role in antitumor immunity and sensing of DNA damage ( 17 ). STING suppression occurs through various mechanisms, including loss-of-function mutations in the genes encoding cGAS and STING ( STING1 ) and by DNA methylation of their promoters ( 14 , 17 ). We and others have found that in human glioma samples, STING1 is expressed in stromal and immune cells but is uniformly suppressed in neoplastic cells ( 13 , 27 ). In contrast, human vascular cells and glioma-associated myeloid cells respond to STING agonism, and STING activation in intracranial glioblastoma (GBM) murine models drives infiltration of innate immune cells, including macrophages, neutrophils, and NK cells ( 27 ). The relative importance of STING activation in specific types of T cells, NK cells, and myeloid cells remains unknown and may have important implications for designing optimal therapeutic approaches. Whether STING suppression in glioma cells contributes to the characteristically immunosuppressed nature of gliomas remains unclear. Mutations in the STING1 gene are rare in GBM, but methylation of a region of the STING promoter near the transcriptional start site is nearly universal. Interestingly, normal fetal and adult brains also exhibit STING promoter methylation, suggesting that epigenetic STING silencing may be characteristic of the GBM cell of origin ( 13 ). Indeed, treatment with the demethylation agent decitabine reversed STING1 promoter methylation and rescued STING expression in GBM cell lines ( 13 ). Whether such silencing can be reversed in vivo to potentiate antitumor immunity and/or sensitize GBMs to immunotherapy, DNA damaging agents, and therapeutically delivered CDNs also remains to be determined. STING as a therapeutic target in GBM Immunotherapy has yielded significant treatment success in several solid malignancies ( 28 ). However, despite favorable results in preclinical models, immunotherapeutic approaches have generally failed to improve survival over standard treatments in GBM ( 29 , 30 ). This failure may be due in part to the immunologically “cold” nature of GBM, whereby spontaneous antitumor T cell responses are either absent or suppressed ( 31 – 33 ). The reasons for such T cell silencing are incompletely understood and are likely multifactorial; GBM tumors recruit immunosuppressive regulatory T cells ( 34 ) and myeloid cell populations ( 35 ), secrete immunosuppressive cytokines, induce T cell apoptosis, sequester T cells in the bone marrow ( 36 ), and harbor low levels of tumor-infiltrating T cells in the TME ( 37 ). STING pathway activation may represent another approach for activating antitumor immunity given its role as a key upstream mediator of type I IFN signaling. GBM harbors extensive cytoplasmic extrachromosomal DNA ( 38 ) that could, in principle, induce the cGAS/STING pathway. However, as discussed above, STING signaling appears to be innately silenced epigenetically in the brain ( 13 ). Both direct and indirect routes to engage STING are currently being explored for glioma therapy, including STING agonists, alternating electric field therapy (e.g., tumor-treating fields [TTFs]), radiation therapy, and oncolytic viruses. STING agonism. STING agonism involves the exogenous introduction of synthetic agonists designed to enhance STING signaling and the resulting IFN response. A wide range of STING agonists of varying potency and specificity have been investigated preclinically and in early-stage clinical trials for several tumor types ( 39 , 40 ). Preclinical studies have demonstrated the ability of synthetic CDNs to induce tumor-specific CD8 + T cells ( 41 ) and reduce tumor growth when administered intratumorally in murine models in combination with immune checkpoint blockade ( 42 ). However, despite promising results in animal models, the first STING agonist to enter clinical trials, DMXAA (vadimezan), demonstrated poor efficacy against solid tumors either alone or in combination with chemotherapy ( 43 ). This disappointing outcome is potentially explained by the poor binding affinity of DMXAA for human STING despite strong binding to murine STING ( 44 , 45 ). Treatment of rat esophageal adenocarcinoma models with the CDN ADU-S100 (MIW815, Novartis) resulted in stimulation of CD8 + T cell–mediated antitumor responses ( 46 ) and phase I clinical trial results of this agent in solid tumors demonstrated systemic immune activation ( 47 ). However, interim results from ongoing clinical trials of ADU-S100 and MK-1454 (Merck) administered intratumorally demonstrated very poor overall responses in advanced solid tumors and lymphomas ( 47 , 48 ). Even when combined with pembrolizumab, MK-1454 yielded an overall response rate of only 24% (ClinicalTrials.gov NCT03172936). Multiple other intratumorally administered CDNs are currently in clinical trials. First-generation CDNs are inherently structurally unstable and generally administered intratumorally. Non-CDN STING agonists have been designed with better stability and affinity for STING to allow for systemic delivery (reviewed in ref. 39 ), including amidobenzimidazole-based compounds ( 49 ). Additionally, alternative approaches are in preclinical development, including bacterial vectors, antibody-drug conjugates ( 50 ), and nanoparticle vaccines ( 51 ). The reasons for the poor clinical efficacy of STING agonists observed thus far in human trials are still being elucidated. While STING activation has primarily been studied in APCs where it stimulates IFN signaling and primes T cell responses, STING signaling can also be activated in T cells themselves, where they may activate cell stress and apoptotic pathways in addition to IFN stimulation ( 52 ). STING agonism may additionally stimulate the production of regulatory cytokines ( 53 ) and immune checkpoints ( 54 ) that actively limit antitumor responses. For example, systemic STING agonist treatment stimulates immunosuppressive B cells that suppress NK cell–mediated antitumor responses ( 55 ). In GBM, tumor cells highly express CD47, an antiphagocytosis signal ( 56 ). Combination treatment of an anti-CD47 antibody and temozolomide induced ER stress, activated the STING pathway, and increased glioma cell phagocytosis by APCs, resulting in increased antigen cross-presentation and T cell priming ( 10 ). These results were not seen when anti-CD47 antibody was used alone. A second study demonstrated that nanoparticles encapsulating a STING agonist and coated with dual anti-CD47/anti–PD-L1 antibodies mediated robust antitumor efficacy in murine gliomas ( 56 ). These nanoparticles induced glioma-associated myeloid cell phagocytosis of tumor cells via CD47–PD-L1 ligation, and activation of T cell–supportive myeloid cell phenotypes due to STING agonist–mediated effects. Collectively, these studies suggest that STING activation in different cell populations may result in varying immunomodulatory phenotypes, and combination approaches that target specific STING regulatory programs might be required for optimal antitumor activity. STING agonists for patients with infiltrating gliomas have not yet entered human clinical trials, although there have been initial promising results in animal models. Injection of the STING agonist c-di-GMP into the tumors of glioma-bearing mice significantly improved survival, enhanced type I IFN signaling, and increased T cell migration into the brain ( 15 ). These effects were not observed in mice homozygous for the nonfunctional Goldenticket (Gt) STING variant, establishing the necessity of STING expression in the TME. Additionally, the combination of c-di-GMP and a peripheral vaccine significantly increased survival in glioma-bearing mice, as compared with monotherapy with either c-di-GMP or peripheral vaccine alone. Despite these promising results, the lack of spontaneous murine gliomas limits the translatability of murine results to human gliomas. This limitation is particularly important when studying the tumor immune microenvironment, which is markedly more proinflammatory in immunocompetent murine models and greatly abrogated in human-derived xenografts as compared with human gliomas. Some of these challenges may be resolved by studying spontaneous canine gliomas, whose molecular landscapes more closely resemble human gliomas ( 57 ). In a recent phase I trial, Boudreau et al. treated 5 dogs with spontaneously arising GBM with the small-molecule STING agonist IACS-8779 via intratumoral injection ( 58 ). In 3 of the 5 treated dogs, the treatment was well tolerated and reduction in the contrast-enhancing tumor volume was noted on follow-up magnetic resonance imaging (MRI). One dog, which had received the lowest dose of IACS-8779, showed tumor growth on serial MRIs following intratumor treatment. The final dog developed a fatal acute intracranial inflammatory response following intratumoral injection of IACS-8779, with postmortem evaluation showing perivascular and leptomeningeal inflammation and a mixed inflammatory polymorphonuclear leukocyte infiltrate. While small, this study provides promising support for intratumoral STING agonist treatment as a therapeutic approach for gliomas. To further advance these proof-of-concept results into the clinic, it will be important for future studies to investigate the duration of the clinical effect, optimal dose, scheduling, and potential inflammatory sequelae of STING agonism. Additionally, intratumoral administration poses technical limitations due to the need for surgery and limits the frequency of administration, while systemic administration must overcome the challenges of the blood-brain barrier and systemic inflammatory responses. Finally, the relative contribution of specific cell types of the glioma TME in mediating the clinical benefit of STING agonism is unknown and will need to be determined in future preclinical studies and larger scale trials. TTFs. Alternating electric field therapy (e.g., TTFs) combined with standard-of-care temozolomide is recommended as an option by the National Comprehensive Cancer Network for the treatment of newly diagnosed GBM and as monotherapy for recurrent GBM. For newly diagnosed GBM, the EF-14 clinical trial demonstrated a median survival of 20.9 months when TTFs were used together with temozolomide, as compared with 16 months with temozolomide alone ( 59 ). For recurrent GBM, TTF monotherapy resulted in similar survival as compared to physician’s choice of chemotherapy ( 60 ). The non-uniform alternating electric fields of TTF therapy are thought to alter the spatial orientation of polar amino acids and disrupt their proper alignment at the mitotic spindle ( 61 ). This disruption ultimately inhibits tumor cell division and forms the mechanistic basis by which TTFs target dividing cancer cells ( 62 , 63 ). Recent preclinical studies have shown that TTFs can disrupt cellular membranes ( 64 ) as well as promote autophagy and ER stress ( 65 ) in addition to their known effect of disrupting mitosis. Additionally, there have been several reports of increased contrast enhancement on MRI imaging after initiation of TTF therapy followed by durable clinical and radiographic responses ( 66 , 67 ). This pseudoprogression observed in some patients receiving TTF treatment has led to the hypothesis that TTFs may also induce an inflammatory response. Indeed, in patient-derived glioma stem cell lines and established human glioma lines, TTF treatment induced formation of cytosolic micronuclei clusters and activation of type I IFNs in an AIM2- and STING-dependent manner ( 68 ). In syngeneic KR158 and GL261 murine gliomas models, TTF treatment stimulated antitumor immune memory that resulted in a cure rate of 42%–66%. Paired transcriptomic analysis of PBMCs from patients with GBM before and after TTF treatment showed activation of adaptive immune signatures. These studies have motivated several clinical trials investigating the combination of TTFs with immune checkpoint blockade ( 69 ). Radiation therapy. Radiation therapy (RT), long integral to the standard of care for GBM, likely exerts profound effects on the immune microenvironment. RT was established as an effective therapy for GBM with the report in 1978 that whole-brain radiation more than doubled the median overall survival (35 versus 14 weeks) ( 70 ). Subsequent studies clarified the effective dose and treatment fields ( 71 , 72 ). Modern guidelines call for 60 Gy delivered in 30 daily fractions to the postsurgical resection cavity, suspected residual tumor, and areas of MRI T2–hyperintense nonenhancing tumor with a 2 to 3 cm anatomic expansion. The primary mediator of RT efficacy is thought to be production of double-strand DNA breaks, which can lead tumor cells to undergo apoptosis and/or mitotic catastrophe. However, an appreciation for RT’s effects on the immune microenvironment has recently emerged. RT can affect the GBM immune microenvironment in several ways. RT triggers type I IFN expression through STING or through STING-independent mechanisms in a variety of in vivo models of extracranial tumors (for a review, see ref. 73 ). RT may promote DC and other APC activation to facilitate cross-presentation of tumor-derived antigens ( 74 ). Thus, RT might stimulate the innate immune response and stimulate antigen presentation. However, RT may deplete tumor-infiltrating lymphocytes given these cells’ intrinsic radiosensitivity ( 75 ), which could thwart adaptive immune responses. Intriguingly, several studies have shown that RT synergizes with immune checkpoint blockade in mouse models of GBM ( 76 ), suggesting that RT may play a unique role in stimulating the immune system. However, irradiation of the normal brain was found to blunt the effects of checkpoint blockade and stimulate more aggressive tumor growth in another study ( 77 ). Oncolytic viruses. Oncolytic viruses (OVs) are an emerging class of immune-oncologic agents capable of promoting a robust antitumor immune response through selective tumor lysis and the induction of antitumor immunity ( 78 , 79 ). OVs offer a targeted approach for the treatment of brain tumors, and as such, a litany have been tested, albeit with varying results ( 78 , 79 ). Among the OVs trialed, engineered oncolytic herpes simplex virus type 1 (oHSV) has been extensively researched and several constructs have shown substantial promise in preclinical models/clinical trials in both pediatric and adult brain tumors ( 80 – 83 ). Of the oHSVs examined, G207 has been the most widely studied and has proven safe in the CNS of both children and adults ( 79 – 86 ). In addition, G207 treatment induced tumor-infiltrating lymphocytes and some prolonged responses in children with progressive high-grade glioma ( 83 ). These promising safety/efficacy data have led to a first-in-human phase I trial of intratumoral G207 in recurrent cerebellar brain tumors and the development of a multi-institutional phase II trial (ClinicalTrials.gov NCT04482933) in pediatric high-grade glioma at first relapse/progression slated to open in 2023 ( 87 ). STING is a critical determinant of both oHSV-mediated oncolysis and the development of innate/adaptive inflammation ( 88 ). Given that PRR pathways are central to host responses for most pathogens, this role is perhaps unsurprising ( 89 ). The cGAS/STING signaling pathway detects cytosolic DNA and triggers many downstream immune responses, and in response to HSV infection induces type I IFN gene expression ( 90 ). Not surprisingly, the STING pathway is the target of a wide range of strategies utilized by herpes viruses to evade the immune response ( 88 , 91 ). Thus, the STING pathway has emerged as a therapeutically relevant pathway, with a strong rationale for STING modulation in the context of oHSV-mediated therapy. Recent evidence indicates that STING signaling is required for durable antitumor effects related to oHSV ( 90 ). However, as discussed above, primary brain tumors lack STING expression and exhibit hypermethylation of a region of the STING promoter ( 13 ). Interestingly, STING expression and signaling can be reconstituted in glioma cell lines via exposure to decitabine, a DNA hypomethylating agent that has been shown to enhance immune recognition and killing of glioma-initiating cells ( 92 ). Given the established role of STING signaling in cancer immunity and the potential for its modulation/reconstitution in neoplastic cells, the rational modulation of this axis may lead to therapeutic benefits when combined with oHSVs and/or RT. However, while early and robust STING activity may antagonize oHSV infection by suppressing viral replication via IFN, STING reactivation downstream may enhance oHSV efficacy by facilitating nuclear import of HSV DNA during infection, augmenting/sustaining an antitumor immune response initiated by the virus ( 90 , 93 ). Future work will therefore be required to determine the optimal timing of any combinatorial treatment strategies related to STING and oHSVs in relevant preclinical models as a translational bridge to the clinic. Future perspectives in STING-directed therapies Despite the disappointing results from initial STING agonist trials, STING activation remains an enticing target for combinatorial therapeutic approaches due to its central role in priming innate antitumor immunity. The CNS environment presents additional challenges for STING activation due to its epigenetic silencing in the brain parenchyma and neoplastic cells. This epigenetic silencing could present a potential opportunity for the use of epigenetic modulation therapy to de-repress STING in neoplastic cells. Release of innate epigenetic silencing may permit the recognition of cytosolic DNA within neoplastic cells, presenting the tantalizing possibility of sensitizing gliomas to therapy-induced DNA damage, including RT, chemotherapy, and TTFs ( Figure 1B ). Additionally, STING de-repression may activate innate immunity and sensitize gliomas to immunotherapy, including checkpoint blockade and OVs. That STING is silenced in the normal brain parenchyma raises the question of whether de-repressing STING may lead to undesired neurotoxicity; reassuringly, however, the use of decitabine for hematologic malignancies has not resulted in significant neurotoxicity. STING expression is regulated by negative feedback to prevent its constitutive activation. Chronic activation in fact appears to be tumorigenic in certain contexts ( 94 – 96 ). Thus, the degree and timing of STING activation to inflame the TME will need to be determined. Optimal STING pathway activation may require the concurrent inhibition of regulatory programs that may attenuate the impact of STING signaling and antitumor immunity, such as the antiphagocytosis signal CD47, regulatory B and T cells, and immune checkpoints. The unique immunosuppressive CNS environment means that the results from systemic cancer studies must be applied with caution to CNS tumors. The toxicity profile of STING-activating strategies in the CNS may also differ substantially from those observed in systemic cancers; a particular concern is the relative intolerance of the CNS to immune activation. Additionally, sex-specific differences in response to STING activation therapy requires further exploration; a recent publication noted significantly reduced cGAS/STING activation in females as compared with males in murine models of traumatic brain injury ( 97 ). In this Review, we have focused on canonical STING/TBK1/IRF3–driven activation of type I IFNs. However, crosstalk between STING pathway components and other signaling nodes means that activation of STING pathway components is not necessarily proinflammatory or antitumor in all contexts. Despite its key antitumor innate immune role in mediating STING-dependent activation of type I IFNs, TBK1 contrastingly has distinct immune evasion roles that are independent of STING and IRF3. For example, TBK1 inhibition has been shown to improve the efficacy of immune checkpoint blockade in several tumor models ( 98 ). While STING is well known to activate NF-κB, it has recently been shown that NF-κB activation also activates STING via microtubule depolymerization, which prevents the trafficking of STING to lysosomes ( 99 ). Finally, STING activation may enhance the frequency of brain metastases from breast and lung cancer ( 100 , 101 ). The TME changes with disease progression and in response to treatment. An understanding of these changes could influence the optimal timing of STING activation and combinatorial regimens. However, our understanding of these dynamic processes is hampered by the paucity of patient tissue samples collected before and immediately after treatments, in particular RT, and the heterogeneity of samples analyzed at the time of recurrence. Unfortunately, murine models are unable to account for intra- and intertumoral heterogeneity, and they do not accurately model the extent of tumor evolution that occurs during standard-care therapy in GBM. Thus, phase 0/surgical window-of-opportunity studies may be best suited to answer questions regarding STING and innate immune activation in human tumor samples. In summary, while STING epigenetic silencing is characteristic of primary CNS tumors, it presents both a challenge to existing treatment approaches as well as a promising potential therapeutic target. Much work remains to determine how best to exploit this key innate immune pathway and design combination treatment regimens for optimal therapeutic effect.

GKF is supported by the US Food and Drug Administration (grant R01FD006368), Cannonball Kids Cancer Foundation, the Rally Foundation for Childhood Cancer Research, CureSearch for Children’s Cancer, The V Foundation for Cancer Research, Hyundai Hope on Wheels, Andrew McDonough B+ Foundation, the National Pediatric Cancer Foundation, and the Pediatric Cancer Research Foundation. ZJR received support from the National Cancer Institute (NCI) (grant K08256045, Mentored Clinician Scientist Development Award), the NIH (grant U19CA264385, Glioblastoma Therapeutics Network), developmental funds from the Duke Cancer Center Support Grant P30CA014236, ChadTough Defeat DIPG Foundation, the SoSo Strong Foundation, the Pediatric Brain Tumor Foundation, and the St. Baldrick’s Foundation. ZJR and JTL received developmental funds from NCI grant P50CA190991 (Duke SPORE in Brain Cancer). MCB received support from NCI grant R00CA263021. JMM received support from NIH R01CA222903. 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e163452

oa_package/80/93/PMC10786687.tar.gz

PMC10786688

38032738

Introduction Insulin resistance underlies the development of type 2 diabetes (T2D) and metabolic syndrome ( 1 – 3 ), which together affect 20%–30% of adults in Westernized populations ( 4 ). In addition, an almost equal fraction of nondiabetic people within the general population have been shown to have a level of insulin resistance similar to that of patients with T2D when assessed by levels of circulating insulin, homeostatic model assessment for insulin resistance (HOMA-IR), euglycemic clamp, or steady-state plasma glucose in response to a fixed insulin and glucose infusion ( 5 , 6 ). The molecular determinants underlying insulin resistance in these states include both cell-autonomous factors, such as genetics and epigenetics, and effects of extrinsic circulating factors, including lipids, cytokines, and miRNAs, that can modify insulin action ( 7 ). These factors likely play a role in insulin resistance in the nondiabetic population. While both T2D and metabolic syndrome affect men and women almost equally, different factors may play a role in development of insulin resistance and its metabolic complications in men and women, including differences in fat distribution and effects of circulating sex hormones ( 8 , 9 ). These may lead to differences in disease progression at different stages of prediabetes ( 10 , 11 ) and differences in insulin resistance–associated diseases such as atherosclerosis, fatty liver disease, and Alzheimer’s disease ( 12 – 14 ). To begin to identify the cell-autonomous factors driving insulin resistance, in previous studies, we have developed a unique model using induced pluripotent stem cell–derived (iPSC-derived) myoblasts (iMyos) taken from either type 2 diabetic patients and controls ( 15 ) or insulin-sensitive (I-Sen) and insulin-resistant (I-Res) nondiabetic individuals ( 16 ) and investigated basal and insulin-stimulated protein phosphorylation using quantitative global phosphoproteomics. This revealed a broad network of cellular signaling defects associated with insulin resistance in both patients with T2D and I-Res individuals without diabetes. We also found phosphorylation differences in the iMyos taken from the male versus female individuals, especially in the cells from the nondiabetic population, reflecting possible sex-dependent differences in cellular effects ( 16 ). The goal of the current study was to investigate the determinants of insulin resistance and sex-dependent differences within the nondiabetic population by analysis of gene expression. In the current study using RNA-Seq, we found a major effect of insulin resistance on gene expression with almost 600 up- or downregulated genes in iMyos from I-Res individuals, including many genes with SNPs linked to T2D. In addition, we observed over 1,500 differences in gene expression that were linked to the sex of the cell donor, most independent of the insulin sensitivity status, over 90% of which were on autosomal chromosomes. Furthermore, we found an increase in global DNA methylation in iMyos created from cells of female versus male individuals, and treatment with a DNA methyltransferase (DNMT) inhibitor, 5-azacytidine (5-Az), reversed some of the sex-related differences in gene expression, and a functional readout of sexual dimorphism, RhoA activation. By contrast, 5-Az did not impact the gene expression changes or differences in glucose uptake associated with insulin resistance. Thus, iPSC-derived human myoblasts exhibit differences in gene expression based on insulin resistance status and sex of the donor. The latter appear to be in large part the result of epigenetic DNA methylation changes, whereas the former appear to be mediated by a complex of genetic differences or epigenetic mediators other than DNA methylation.

Methods Study participants, SSPG, and iPSC reprogramming. The iPSC lines were generated from 20 human study participants who had been recruited and assessed for insulin sensitivity using steady-state plasma glucose (SSPG) obtained from the modified insulin suppression test at the Stanford Clinical and Translational Research Unit ( 69 ). iPSC lines were generated as described previously ( 18 ), and those used in the study were chosen from 8 in the upper quintile of insulin sensitivity and 8 in the lowest quintile of insulin sensitivity, matched for age, sex, and race/ethnicity based on the SSPG as previously described ( 16 ). iPSC culture, myogenic differentiation, and 5-Az treatment. The iPSCs were cultured on plates coated with hESC-qualified Matrigel (Corning) using the mTeSR1 medium containing the 5× complement (Stemcell Technologies) and passaged as aggregates using ReLeSR (Stemcell Technologies). For differentiation into myoblasts, a modified version of the 2-step differentiation protocol was used ( 70 ). For this, approximately 7 × 10 3 iPSCs/cm 2 were seeded onto collagen I–coated plates (Biocoat, Fisher) in skeletal muscle cell growth basal medium (Lonza) containing 5% horse serum, 50 μg/mL fetuin, 3 μM CHIR99021, 2 μM Alk5 inhibitor, 1 ng/mL bFGF, 10 ng/mL human recombinant epidermal growth factor (hrEGF), 10 μg/mL insulin, 0.4 μg/mL dexamethasone, 10 μM Y27632 (Rock inhibitor), and 200 μM ascorbic acid with change of medium every 2 days, which resulted in myogenic precursor/satellite-like (SC-like) cells within 10 days. The SC-like cells were then trypsinized and plated at approximately 7 × 10 3 iPSCs/cm 2 onto collagen I–coated plates (Biocoat, Fisher) in skeletal muscle cell growth basal medium (Lonza) containing 5% horse serum, 50 μg/mL fetuin, 10 μg/mL insulin, 0.4 μg/mL dexamethasone, 10 μM Y27632 (Rock inhibitor), 10 ng/mL hrEGF, 20 ng/mL human recombinant hepatocyte growth factor (HGF), 10 ng/mL human recombinant platelet-derived growth factor (PDGF-AB), 10 ng/mL oncostatin M, 20 ng/mL bFGF, 10 ng/mL insulin-like growth factor 1 (IGF-1), 2 μM SB431542, and 200 μM ascorbic acid with change of medium every 2 days. This resulted in well-differentiated myoblasts (iMyos) within another 10 days as characterized by high levels of MyoD1 ( 16 ). For the treatment with 5-Az, the differentiated iMyos were treated with 20 μM 5-Az for 24 hours followed by processing for RNA extraction as described below. RNA isolation, qPCR, and RNA-Seq. Total RNA from all the cell types was isolated using TRIzol (Thermo Fisher Scientific) following the chloroform/isopropanol/ethanol extraction method. Complementary DNA (cDNA) was synthesized from 400 ng of RNA using a High Capacity cDNA Reverse Transcription kit (Applied Biosystems), and the resulting cDNA was used for the qPCR reaction with iQ SybrGreen Supermix (Bio-Rad, catalog 1708884) performed on a C1000 Thermal Cycler (Bio-Rad, catalog CFX384). TATA box binding protein (Tbp) was used as a housekeeping gene to normalize gene expression unless stated otherwise. Primer sequences used were TBP (forward: CCACTCACAGACTCTCACAAC; reverse: CTGCGGTACAATCCCAGAACT), AR (forward: GACGACCAGATGGCTGTCATT; reverse: GGGCGAAGTAGAGCATCCT), ESR1 (forward: GAAAGGTGGGATACGAAAAGACC; reverse: GCTGTTCTTCTTAGAGCGTTTGA), DNMT1 (forward: AGGCGGCTCAAAGATTTGGAA; reverse: GCAGAAATTCGTGCAAGAGATTC), EZH1 (forward: GTCACTGAACACAGTTGCATTG; reverse: TGCACAAAACCGTCTCATCTTC), GLIPR1 (forward: TCCGATCAGAGGTGAAACCAA; reverse: GGCTTCAGCCGTGTATTATGTG), COL8A1 (forward: AAAGGGGAAATTGGGCCTATG; reverse: CTGGTTGCCCTGGTAACCC), USP11 (forward: CATTGAACGCAAGGTCATAGAGC; reverse: AACAGTGTGAGATTTGCCCAA), GALNT18 (forward: CCAGAGGTGAGCATCGTGTT; reverse: CTCCTTGAGCAGATGTGGGG), ARHGAP25 (forward: CTGAGAGACGCTTTTGATGCT; reverse: TCTCGGAGGTAGAGCTTTAACA), HSD17B14 (forward: TAGGGCCACAATCCGAGAGG; reverse: GAGCAGTTCAATGCCCGTG), and NNAT (forward: ACTGGGTAGGATTCGCTTTTCG; reverse: ACACCTCACTTCTCGCAATGG). For RNA-Seq, total RNA samples that passed the quality tests were submitted to the Biopolymers Facility at Harvard Medical School. The KAPA mRNA HyperPrep kit for Illumina sequencing was used. mRNA was pulled down using oligo-dT beads, and the resulting mRNA was converted into cDNA. The resulting cDNA then became a library through adapter ligation and post-PCR cleanup. RNA-Seq raw reads were 100-bp reverse-stranded paired-end reads with 50 million reads per sample. The reads were trimmed for adapters and filtered by sequencing of Phred quality (≥Q15) using fastp ( 71 ). The count table was generated by aligning of reads to the human transcriptome (Ensembl version 98) using kallisto ( 72 ), and conversion of transcript counts to gene counts using tximport ( 73 ). To filter out low-expressed genes, we kept genes that had counts per million of more than 0.5 in at least 4 samples. Counts were normalized by the weighted trimmed mean of M values ( 74 ). To detect differential genes, we used limma, an R package to investigate differential expression analyses ( 75 ). P values were corrected using the Benjamini-Hochberg false discovery rate (FDR). Hierarchical clustering analysis was used to determine differential gene clusters using a variable cut height approach ( 76 ). Glucose uptake, active Rho pull-down assays, and DNA methylation ELISA. For the glucose uptake assay, iMyos grown in a 96-well plate were serum-starved (DMEM/F12 plus 0.25% BSA) overnight, washed, and incubated with Krebs-Ringer bicarbonate HEPES (KRBH) buffer (120 mM NaCl, 10 mM NaHCO 3 , 4 mM KH 2 PO 4 , 1 mM MgSO 4 , 1 mM CaCl 2 , 30 mM HEPES) with 5.5 mmol/L glucose for 15 minutes at 37°C. The cells were then stimulated with 100 nM insulin for 30 minutes and then incubated with 100 μM of 2-( N -(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2-NBDG) dye in KRBH buffer for 1 hour at 37°C. Cells were washed 3 times with PBS, and the fluorescence was recorded using a plate reader. For the active RhoA pull-down assay, differentiated iMyos were processed and assayed according to the manufacturer’s protocol. The kit components, including the Rho rabbit antibody, was purchased from Cell Signaling (catalog 8820). For quantification of global DNA methylation, DNA was extracted from the differentiated iMyos, and DNA samples were processed according to the manufacturer’s protocol to measure levels of 5-methylcytosine through an ELISA reaction (P-1030-48, EpigenTek). Statistics. Data analysis was performed using appropriate unpaired or paired 2-tailed Student’s t test (version 8.4.3, GraphPad Prism Software), and P < 0.05 was considered to be significant. Study approval. The iPSC lines used in this study were generated from 20 human study participants who had been recruited and assessed for insulin sensitivity at the Stanford Clinical and Translational Research Unit at Stanford University (Stanford, CA) as part of the NIH-sponsored GENESiPS project, which had approval to conduct the study ( 69 ). Data availability. RNA-Seq raw data were deposited to the Gene Expression Omnibus (GEO) database (accession GSE244307). Raw values for all data points in graphs are reported in the Supporting Data Values file.

Results Transcriptional profiling reveals the cell-autonomous changes in gene expression associated with insulin resistance. iPSCs were created from 20 individuals without diabetes, half in the top quintile of insulin sensitivity (I-Sen) and half in the bottom quintile of insulin sensitivity, i.e., most insulin resistant (I-Res), previously identified by population screening using the steady-state plasma glucose (SSPG) approach ( 17 ); the iPSCs were derived from blood cells using nonintegrative Sendai virus ( 18 ). Both the I-Sen and I-Res cohorts were equally divided between male and female individuals and had an average age of approximately 60 years (clinical details in ref. 16 and Supplemental Figure 1 ; supplemental material available online with this article; https://doi.org/10.1172/JCI172333DS1 ). The iPSCs were converted to myoblasts (iMyos) using a 2-stage cocktail approach, and both groups of cells showed similar myogenic differentiation capacity ( 16 ). To identify the full spectrum of gene expression changes associated with the differences in insulin sensitivity, we performed RNA-Seq of iMyos from 8 I-Sen and 8 I-Res donors ( Figure 1A ). Principal component analysis (PCA) of these data demonstrated a clear separation based on 2 factors: the sex of the cell donor, which was the largest driver of variance (component 1), and insulin sensitivity status, which was the second largest driver (component 2) ( Figure 1B ). Interestingly, in male individuals, insulin resistance status shifted the relative PCA coordinates to the left (filled squares vs. open squares, Figure 1B ), whereas in female individuals, insulin resistance shifted the coordinates toward the right (filled circles vs. open circles, Figure 1B ) in the PCA plot, suggesting an interaction between insulin sensitivity and donor sex at the level of gene expression. Hierarchical clustering analysis of the expression data focusing on genes that were differentially abundant between I-Sen and I-Res iMyos in both male and female individuals revealed 271 genes that were significantly decreased and 306 genes that were significantly increased in cells from I-Res donors as compared with I-Sen donors ( P < 0.05; Figure 1C and Supplemental Table 1 ). The genes altered in expression in insulin resistance extended well beyond genes typically linked to insulin action but did include a number that have been previously linked to diabetes. Thus, among the decreased genes were WD repeat domain 46 ( WDR46 ), which has been associated with diabetic retinopathy ( 19 ), integrin subunit α2 ( ITGA2 ), which has been associated with T2D and its complications ( 20 ), and matrix metalloproteinase 11 ( MMP11 ), which protects against T2D in mice ( 21 ), all of which showed 50%–70% decreases in cells from the I-Res donors ( Figure 1D ). Some representative examples of genes increased in expression in I-Res cells included peripherin 2 ( PRPH2 ), which is associated with inherited retinal dystrophy ( 22 ), the secretin receptor ( SCTR ), which has a GWAS risk allele for development of T2D ( 23 ), and GATA5 , a transcription factor involved in multiple processes, including pancreatic development ( 24 ), all of which exhibited 2- to 3-fold increases in cells from I-Res donors. Database for Annotation, Visualization and Integrated Discovery (DAVID) Gene Ontology analysis of the genes with decreased expression in insulin resistance revealed that the most enriched biological processes were for genes involved in the regulation of transcription (30 genes), cell adhesion (10 genes), axon guidance (5 genes), calcium ion–dependent exocytosis (3 genes), and mitotic sister chromatid segregation (3 genes) ( Figure 1E , blue bars), whereas Gene Ontology analysis of genes whose expression was increased in insulin resistance identified enrichment of genes involved in negative regulation of transcription (14 genes), intracellular signal transduction (12 genes), protein localization (7 genes), Wnt signaling pathway (7 genes), and lipid catabolic process (6 genes) ( Figure 1E , red bars). Mapping of the specific genes associated with each of these biological processes revealed a unique network of genes associated with insulin sensitivity status that is maintained in vitro in these differentiated iMyos ( Supplemental Figure 2A ). Positional gene enrichment analysis of the genes that were increased or decreased in iMyos in relationship to insulin resistance ( Figure 1C ) revealed that these genes are spread throughout all autosomes and the X, but not Y, chromosome ( Supplemental Figure 2B ). Despite its small size, there were 29 genes with differential expression on chromosome 19. Whether this represents some enrichment or simply reflects the gene-rich nature of chromosome 19, which contains roughly 1,500 genes or 6% of all the genes, remains to be determined. Overlapping the genes that were significantly increased or decreased in I-Res iMyos ( Figure 1C ) with the genes associated with T2D via SNPs ( 19 ) revealed a set of 32 genes ( Supplemental Figure 3 ), 5 of which were also associated with the most changed biological processes in I-Res iMyos (indicated by asterisks in Supplemental Figure 2A ). Among the genes associated with T2D and increased expression in I-Res iMyos were the zinc finger homeobox gene TSHZ3 and tumor suppressor WT1 , both of which are negative regulators of transcription. By contrast, genes showing decreased expression in I-Res iMyos included positive regulators of transcription, such as PBX2 , ZNF213 , and IRF2BP1 . Interestingly, the lysophosphatidic acid hydrolase ACP6 , which is associated with T2D via SNPs ( Supplemental Figure 3 ), was increased in expression in iMyos of I-Res donors and has also been shown to be increased in expression in skeletal muscle of individuals with a family history of T2D ( 25 ). Likewise, TRIM63 (also known as MURF1 ), a muscle-specific E3 ubiquitin ligase, was increased in I-Res iMyos and has also been found to be increased in muscle of streptozotocin diabetic mice ( 26 ). Conversely, FBXW7 , an F-box protein that serves as the substrate recognition component of SCF E3 ubiquitin ligase, was decreased in expression in I-Res iMyos and has been found to be decreased in muscle of the Goto-Kakizaki rat model of T2D ( 27 ). Thus, iMyos exhibit a gene expression signature associated with insulin resistance even in the absence of the influence of extrinsic factors, many of which, are similar to gene expression differences in muscle of patients with T2D. Sex-specific gene expression changes associated with insulin resistance. Because the sex of the patient has a significant effect on gene expression, we also analyzed the RNA-Seq data from iMyos of the male and female individuals separately, and this revealed an even larger set of genes impacting insulin sensitivity ( Supplemental Table 2 ). Thus, expression of 718 genes was significantly decreased and of 926 significantly increased comparing the cells of I-Res with those of I-Sen male donors ( Figure 2A , left), whereas slightly smaller numbers (349 decreased and 356 increased) were observed in the I-Res cells from the female donors ( Figure 2A , right). Among the protein-coding genes that were differentially expressed, paired immunoglobulin-like type 2 receptor α ( PILRA ) showed a decrease, and collagen 6 α2 ( COL6A2 ) showed a significant approximately 2-fold increase of mRNA expression in I-Res as compared with I-Sen cells from male donors but showed no changes in cells from female donors ( Figure 2B ). Conversely, thrombospondin 1 ( THBS1 ) showed a significant 50% decrease, and solute carrier family 26 member 7 ( SLC26A7 ) showed a significant approximately 5-fold increase in I-Res as compared with I-Sen cells from female individuals, with no significant changes in the cells from male individuals ( Figure 2B ). Interestingly, PILRA has also been found to be decreased in skeletal muscle of patients with obesity and T2D ( 28 ), and thrombospondin 1 has been linked to β-cell lipotoxicity and diabetic retinopathy ( 29 ), suggesting an important role of these sex-specific changes in diabetes pathogenesis. DAVID Gene Ontology analysis revealed that the biological processes associated with increased expression in I-Res male individuals ( Figure 2C , left, red bars) included genes involved in protein transport (45 genes), the apoptotic process (35 genes), intracellular signal transduction (32 genes), endocytosis (21 genes), and extracellular matrix (ECM) organization (17 genes), while the most enriched biological processes in cells from females were involved in regulation of transcription (38 genes), skeletal system morphogenesis (8 genes), axon guidance (8 genes), pattern specification (7 genes), and cellular response to hypoxia (6 genes) ( Figure 2C , right panel, red bars). On the other hand, biological processes associated with lower gene expression in I-Res male individuals ( Figure 2C , left, blue bars) included genes involved in DNA repair (24 genes), cell division (21 genes), negative regulation of transcription (21 genes), cellular response to DNA damage stimulus (15 genes), and chromatin organization (14 genes), while downregulated genes in female cells were related to negative regulation of transcription (23 genes), cell adhesion (22 genes), positive regulation of apoptosis (13 genes), negative regulation of cell proliferation (12 genes), and actin cytoskeleton organization (11 genes) ( Figure 2C , right, blue bars). Thus, in addition to the 577 gene expression differences in insulin sensitivity of both male and female individuals ( Figure 1C ), iMyos derived from nondiabetic I-Res and I-Sen individuals showed over 2,000 changes in gene expression based on insulin sensitivity, which were distinct in male and female individuals. In addition to these protein-coding genes, bulk RNA-Seq also revealed differential expression of a few encoded long noncoding RNAs, such as AL158832.2 and AL512625.3, and a few miRNAs, including miR8075 and miR570, which were decreased in I-Res iMyos from the male donors. Further exploration of these using small RNA-Seq is warranted. Cell-autonomous sexual dimorphism in gene expression and their associated genomic distribution. Both the PCA analysis and the volcano plots in Figure 2A demonstrate that in addition to insulin resistance, the sex of the cell donor is a major modulator of differences in gene expression. Hierarchical clustering analysis of the expression data focused on sex of the cell donor rather than insulin sensitivity status revealed 1,552 genes that differed significantly in expression between male and female cells, with 766 genes being significantly higher in expression in cells from male individuals as compared with those from female individuals (i.e., male dominant) and 786 genes being significantly higher in cells from female individuals (i.e., female dominant) ( Figure 3A and Supplemental Table 3 ). DAVID Gene Ontology analysis revealed that the biological processes associated with the male-dominant cluster included genes involved in cell adhesion, ECM organization, response to hypoxia, cell migration, and axon guidance, while genes more highly expressed in female cells were involved in muscle tissue development, regulation of ion transport, muscle contraction, GPCR signaling, and exocytosis ( Figure 3B ). The magnitude of these differences ranged from 2- to 10-fold. For example, in the male-dominant cluster, ICAM1 showed 2-fold higher levels in cells of male versus female individuals, COL8A1 and KDR showed 2.8-fold differences, and IFI35 showed a 1.7-fold difference ( Figure 3C ). Representative genes associated with the biological processes identified in the female-dominant clusters included TFAP2B , which showed a 2-fold increase, CNTN1 , which had a 2.9-fold increase, and the glutamate receptor GRIA2 , and doublecortin DCX genes, which showed 8-fold higher levels in female as compared with male cells ( Figure 3C ). While DCX is an X chromosome–encoded gene, GRIA2 is encoded on chromosome 4 with sex differences in expression in cochlea ( 30 ) and the brain ( 31 ). These sex-specific differences were, in general, independent of the insulin sensitivity status (compare dark- vs. light-shaded squares and circles in Figure 3C ) and occurred in vitro in the absence of sex hormones, i.e., represented cell-autonomous sex-specific changes in gene expression. Since male and female cells differ in copy number of genes represented on the X and Y chromosomes, we performed positional gene enrichment analysis of the most differentially expressed male- and female-dominant genes to determine the chromosomal distribution of the sex-specific gene expression changes. Using criteria of fold change >1.5 and P < 0.05, we identified 243 male-dominant and 497 female-dominant genes. Analysis of these sex-biased genes mapped to their genomic coordinates is shown in Figure 3 D. Importantly, 93% of the male- and female-specific genes were distributed across the autosomal chromosomes (nos. 1–22), and only 7% were localized to the sex chromosomes (X, Y) ( Figure 3E ). As expected, all the genes that mapped to the Y chromosome were male-dominant genes, and most of the genes mapping to the X chromosome were female dominant ( Figure 3D ). Most sex-differential genes that were encoded on the autosomes were widely dispersed but showed a few potential “hot spots” of activity, including a female-dominant cluster of 20 genes on chromosome 3 (genomic coordinates: 6.8 × 10 6 to 52.8 × 10 6 ) and a second female-dominant cluster of 9 genes on chromosome 16 (genomics coordinates: 0.98 × 10 6 to 6.1 × 10 6 ) ( Figure 3D ). Autosomal sex-specific gene expression changes are independent of X chromosome dosage and androgen receptor action. Although only 7% of the sex-specific genes were localized on the X or Y chromosomes, there might be differences in X chromosome dosage in the female cells arising from the difference in the extent of X chromosome inactivation, i.e., the developmental process in which the one X chromosome in female cells is silenced by being packed into transcriptionally inactive heterochromatin ( 32 ). It is known that reprogramming of somatic cells from female donors into iPSCs results in reactivation of the silenced X chromosome, leading to 2 active X chromosomes (Xa Xa), and this is associated with a decrease in the DNA methylation of both the X chromosome and many autosomal genes ( 33 ). When iPSCs are induced to differentiate into myoblasts or other differentiated cell types, the cells undergo the process of renewed X chromosome inactivation resulting in 1 active and 1 inactive chromosome in the female cells (Xa Xi). However, this process may not be complete in all cells ( 34 ). The long noncoding RNA XIST is the major marker of X chromosome inactivation ( 35 , 36 ). As expected, XIST expression was undetectable in iMyos from all the male donors ( Figure 4A ). By contrast, differentiated iMyos from female individuals showed variable XIST levels, with 4 of 8 iMyos having high levels of XIST expression (XIST high) and the other 4 having very low or undetectable levels of XIST mRNA (XIST low). This distribution was true in cells from both I-Sen and I-Res donors ( Figure 4A ). To determine the effect of the sex chromosomes and different expression levels of XIST on the sex-specific gene expression data, we performed PCA excluding the genes encoded on the X or Y chromosome but annotated for female donors based on whether they had high or low XIST levels ( Figure 4B ). Even focusing only on autosomally encoded genes, the gene expression differences showed a clear separation based on sex of the donors (PC1 in Figure 4B ), although the female donors also showed a tendency to separate by level of XIST , with those having low XIST mapping more similarly to the male donors at high values along the PC2 axis. To further explore the role of active X dosage and XIST expression in the changes in autosomal gene expression, we performed comparisons of the gene expression data for the XIST high (likely Xa Xi) and XIST low (likely Xa Xa) female cells versus the male cells (all XIST low) ( Figure 4C ). Focusing on 1,840 male-dominant and 1,607 female-dominant autosomal genes ( P < 0.05) for these comparisons, we found that the majority of the sex-related changes in autosomal gene expression were independent of the XIST level and X chromosome dose ( Figure 4D ). Thus, the X chromosome dosage and the variation in XIST expression in female individuals do not account for most autosomal sex-specific gene expression changes. In addition to the X chromosome dose, we also investigated the potential effect of sex hormone receptor action on the differential gene expression. Notably, estrogen receptor ( ESR1 ) mRNA was not detected by either RNA-Seq or quantitative PCR (qPCR) in iMyos from either male or female donors ( Figure 4E and Supplemental Figure 4A ). By contrast, expression of the androgen receptor ( AR ), which, interestingly, is encoded on the X chromosome, was detected in cells of both sexes by RNA-Seq, with significantly higher levels of AR in the cells of I-Sen male individuals as compared with I-Res male and both I-Sen and I-Res female individuals ( Figure 4E ). This was confirmed by qPCR ( Supplemental Figure 4A ), suggesting that differences in AR levels in I-Sen and I-Res male individuals could potentially contribute to some of the insulin resistance changes in male individuals, as well as the sex-specific changes. Indeed, incubation of I-Sen and I-Res male iMyos with 10 μM dihydrotestosterone (DHT) for 4 days normalized the expression level of AR in I-Res male iMyos ( Figure 4F ). Interestingly, in addition, the impaired glucose uptake ability upon insulin stimulation in I-Res iMyos was also rescued upon incubation with DHT ( Figure 4G ). These results suggest an important role of AR action in regulating insulin resistance changes in male iMyos. On the other hand, overlapping the autosomal sex-specific gene expression changes in both male and female iMyos ( P < 0.05) with RNA-Seq data of muscle from an independent study of mice with or without DHT stimulation ( 37 ) revealed that only 7.2% of the male-dominant changes and only 3% of the female-dominant changes overlapped with the DHT-induced muscle gene expression changes ( Supplemental Figure 4B ), suggesting that varying AR levels in male and female individuals do not seem to contribute to the sex-specific gene expression changes. Given that no sex hormones were added to the media used for differentiation and growth of the iMyos, and that so few of the differences in expression correspond to androgen-responsive genes, the differences in level of androgen receptor in male cells as compared with female cells do not appear to have a major impact on the expression of autosomal sex-specific genes. DNA methylation contributes to sexual dimorphism, but not insulin resistance. The I-Sen and I-Res iPSCs were originally derived from circulating blood cells of adult men and women with an average age of 60 years, i.e., all donors were postpubertal, and all or most of the women postmenopausal. Thus, the donor had been exposed for many years to varying levels of sex hormones prior to cellular isolation for iPSC derivation. Although reprogramming of blood cells into iPSCs is known to erase most of the epigenetic marks exerted by hormonal action and other factors in vivo ( 38 ), it is possible that some residual epigenetic marks remain and contribute to the differences observed in the iMyos. To investigate this possibility, we assessed expression differences for some of the genes involved in epigenetic regulation in the I-Sen and I-Res iMyos from male and female donors. Interestingly, the expression of the major DNA methyltransferase DNMT1 encoded on chromosome 19, as well as of the histone-lysine N -methyltransferase ( EZH1 ) encoded on chromosome 17, was significantly higher in the cells from female donors as compared with those from male donors, independent of the insulin sensitivity status ( Supplemental Figure 5A ). These differences in gene expression were even magnified at the protein level, with 39% and 35% increases in protein expression of DNMT1 and DNMT3A ( P < 0.0001 and P < 0.007), respectively, in cells from female donors as compared with those from male donors, independent of insulin sensitivity status ( Figure 5A ). Consistent with the difference in the expression of the methylation enzymes in the postpubertal iMyos, we found significantly high levels (~15% increased, P < 0.05) of global DNA methylation in cells from the postpubertal female individuals as compared with the postpubertal male individuals ( Figure 5B ). To determine the potential role of postpubertal sex hormones in these epigenetic effects, we used an independent set of iPSCs derived from blood cells of normal prepubertal, i.e., less than 10 years old, male and female donors and differentiated these into iMyos. Interestingly, in the prepubertal iMyos, no differences were observed in the mRNA level of DNMT1 and EZH1 between the sexes ( Supplemental Figure 5A ). Likewise, no difference was observed in global DNA methylation in the iMyos from prepubertal donors ( Figure 5B ), suggesting that sex hormone exposure in vivo may result in persistent DNA methylation epigenetic marks and contribute to some of the gene expression differences observed in the cells derived from the postpubertal donors. These findings are supported by a recent study using human skeletal samples from 222 male and 147 female individuals, which revealed that, of the differentially methylated regions, 94% were hypomethylated in male participants as compared with female participants ( 39 ). Overlapping the male and female differences on autosomal genes in iMyos ( n = 3,447 genes, P < 0.05) with this analysis of autosomal sex-biased methylation in human muscle samples ( n = 15,724 genes, P < 0.05) revealed 1,356 genes in iMyos (39%) that overlapped with the genes showing sex-biased methylation in human muscle samples ( Figure 5C ). Thus, a significant proportion of the genes with sex-biased expression observed in iMyos contain differentially methylated positions. To further investigate whether these differences in DNA methylation play a role in the sex or insulin resistance differences in gene expression, we treated postpubertal male and female iMyos with 5-azacytidine (5-Az), an inhibitor of DNMT ( Supplemental Figure 5B ). This did not affect cell density as assessed by crystal violet staining or protein content ( Supplemental Figure 5B ). mRNA expression of 2 normally male-dominant genes, glioma pathogenesis–related protein 1 ( GLIPR1 ) and collagen type VIII α1 chain ( COL8A1 ), showed a significant rescue of the gene expression in the female iMyos following treatment with 5-Az ( Figure 5D ). Similarly, mRNA expression of two female-dominant genes, ubiquitin-specific peptidase 11 ( USP11 ) and N -acetylgalactosaminyltransferase 18 ( GALNT18 ), showed a significant reversal of the increased expression in the female iMyos following treatment with 5-Az ( Figure 5D ), suggesting that epigenetically mediated DNA methylation contributes to these sex-specific gene expression differences. On the other hand, mRNA expression of genes increased in I-Res male and female iMyos, including Rho GTPase activating protein 25 ( ARHGAP25 ), 17-β-hydroxysteroid dehydrogenase ( HSD17B14 ), and neuronatin ( NNAT ), remained unaffected following treatment with 5-Az ( Figure 5E ), suggesting that DNA methylation is not a major contributor to the gene expression changes associated with insulin resistance. Using phosphoproteomics, we previously showed that iMyos exhibit multiple sex-specific differences in a broad network of protein phosphorylations, several of which were related to the Rho GTPase pathway ( 16 ), leading to enhanced activation of RhoA in iMyos from males versus female donors, as measured in a pull-down assay ( 16 ). This increase in Rho GTPase activity correlated with significantly higher levels of RhoA mRNA in postpubertal male iMyos as compared with the female iMyos ( Supplemental Figure 5C ). We therefore used RhoA activation as a functional readout to study the impact of DNMT inhibition on sexual dimorphic functional changes. Again, we found that there were higher levels of RhoA activation in the male versus female cells from I-Sen iMyos, and this was abolished by treatment of the cells with 5-Az ( Figure 5F and Supplemental Figure 5D ), indicating that a methylation-dependent epigenetic modification was contributing to this sexually dimorphic functional difference. We also showed that the higher level of DNMT1 protein in the cells from female individuals as compared with those from male individuals observed by Western blotting was lost after blockade of DNMT1 upon 5-Az treatment ( Figure 5F ). Thus, treatment with 5-Az was sufficient to reverse the sex-biased changes in a functional readout of RhoA activation. This reversal of a sexually dimorphic phenotype was not observed for insulin resistance as measured by glucose uptake. Thus, when we assessed glucose uptake of the I-Sen and I-Res iMyos with and without treatment with 5-Az using 2-( N -(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2-NBDG) fluorescent glucose, we found that the ability of insulin to stimulate increased glucose uptake in I-Sen iMyos was markedly reduced in the I-Res iMyos, and this effect was not reversed by treatment with 5-Az ( Figure 5G ). Thus, while DNA methylation contributes to sexual dimorphism in gene expression and Rho activation, it does not appear to contribute to differences in gene expression related to insulin resistance or to the reduction in insulin-stimulated glucose uptake.

Discussion Insulin resistance is central to the pathophysiology of T2D, obesity, and metabolic syndrome. Insulin resistance can be identified in offspring of T2D parents many years prior to the onset of the disease and predicts disease development ( 3 ). In addition, insulin resistance is present in a substantial fraction of the general population, making these individuals at higher risk for the development of T2D and metabolic syndrome ( 6 ). One powerful approach to determining the cellular components of disease pathogenesis is the use of induced pluripotent stem cells (iPSCs) ( 40 – 42 ). iPSCs can be maintained in culture indefinitely and differentiated into almost any tissue of interest in the absence of circulating modifiers. Recently, using myoblasts generated from iPSCs (iMyos) of individuals over the range of insulin sensitivity, including patients with insulin receptor mutations ( 43 ), patients with T2D ( 15 ), and individuals without diabetes with insulin resistance ( 16 ), we found large differences in the phosphoproteome based on insulin resistance status of the donor. Many of the alterations in signaling in the iMyos of the I-Res individuals without diabetes overlap with the alterations observed in cells from patients with T2D, highlighting key steps through which to target insulin resistance ( 15 , 16 ). In addition, we found that the sex of the cell donor further modifies intracellular signaling and that these changes can be reflected in differences in downstream biological responses ( 16 ). Consistent with this, sex-specific differences in insulin sensitivity in humans have also been observed in clinical studies. Hyperinsulinemic-euglycemic clamp studies have found that healthy women are more insulin sensitive than men owing to enhanced glucose uptake and a higher proportion of type I muscle fibers in women ( 44 , 45 ). The goal of the current study was to determine how differences in gene expression might be associated with insulin resistance and sex-dependent alterations and contribute to these functional changes. Using RNA-Seq, we have identified 577 genes that are altered in their expression levels in insulin resistance in cells of both male and female donors, with 306 genes increased and 271 genes decreased in I-Res versus I-Sen iMyos. Many of these form complementary networks. For example, genes related to negative regulation of transcription are increased in insulin resistance, while those related to positive regulation of transcription are decreased, suggesting an overall decrease in transcriptional activity as a component of cell-intrinsic insulin resistance. Interestingly, among this group of genes, a subset of genes have been associated with T2D through SNPs ( 19 ), including TSHZ3 and WT1 , which are increased in expression in I-Res iMyos, and PBX2 , ZNF213 , and IRF2BP1 , which are decreased in expression in I-Res iMyos. In addition to these transcriptional regulators, 27 other genes with differential expression in I-Res iMyos have been associated with T2D through SNPs and are also functionally linked to insulin action and control of metabolism. Thus, we found increased expression of genes in I-Res iMyos for biological processes related to protein localization, Wnt signaling, lipid catabolic processes, and intracellular signal transduction, including TGFBP3 , a gene associated with adipose biology and several inflammatory diseases ( 46 , 47 ). In contrast, we found decreased expression of genes associated with cell adhesion (such as ITGA2 ), axon guidance (including WNT3 ), and calcium ion–dependent exocytosis, as well as TONSL , a negative regulator of NF-κB–mediated transcription, all of which have been linked to T2D through SNPs ( 48 ). Taken together, these cell-autonomous defects in gene expression associated with insulin resistance include potential for multisite transcriptional dysregulation and increased proinflammatory intracellular signaling. This is in agreement with our phosphoproteomics analysis using the same cellular model ( 16 ). Overlapping of the insulin sensitivity gene expression changes in iMyos with changes in gene expression observed in primary cultured myotubes from T2D/obese patients ( 49 ) showed an approximately 20% overlap in these gene signatures. Superimposed on the differences related to insulin resistance, the sex of the cell donor is a major modulator of differences in gene expression. Normal development, anthropometric traits, and disease phenotypes, such as prevalence, progression, and age of onset, have all been shown to exhibit sex-differentiated characteristics. These sex-based differences are often attributed to hormones, sex chromosomes, and environmental differences, but the full extent of these differences and their underlying molecular mechanism largely remain unknown. Indeed, a recent study of gene expression in 44 human tissues revealed that 37% of all genes showed sex-biased expression differences in at least one tissue ( 50 ). To what extent these effects were created by the hormonal milieu in vivo or were intrinsic, based on the sex of the person from whom the tissues were derived, is unclear. Here, in this ex vivo system free of added sex hormones, we found over 1,500 sex-biased genes, which are independent of the insulin sensitivity status. Only about 7% of these sex-specific gene expression differences occur in genes on the X or Y chromosome, i.e., 93% of the sex-differentially expressed genes are on autosomes. Interestingly, we identified hot spots of differential expression where multiple female-dominant genes formed clusters on chromosome 3 (20 genes) and chromosome 16 (9 genes). It is possible that these hot spots correspond to specific transcription factor binding sites, methylation sites, or interaction with long noncoding RNAs (lncRNAs) leading to downstream regulation and sex-biased gene expression. The lncRNA XIST has been shown to not only regulate gene expression on the X chromosome but also transregulate gene expression in some genes on autosomal chromosomes ( 51 , 52 ). In iMyos, all male cell lines have undetectable XIST levels, whereas in cells derived from female individuals, about half have high levels of XIST and half have very low or undetectable levels. Analysis of the data considering the varying XIST expression level in the female iMyos shows that most of the sex-biased differences in gene expression are independent of the level of XIST expression, supporting the notion that sex chromosome dosage does not play an important role in most of these sex-specific gene expression differences. Additional studies using iPSCs from patients with Turner syndrome (XO females) ( 53 ) and/or trisomy X (XXX females) ( 54 ) might help to further define the role of X chromosome dosage in these sex-specific gene expression changes. The iPSCs used for the development of the iMyos were derived from circulating blood cells taken from postpubertal adults, who had, therefore, been exposed to circulating sex hormones in vivo. However, in vitro, neither the maintenance nor the differentiation of the iPSCs involves addition of sex hormones, limiting the potential impact of hormones on these sex-specific gene expression changes. While this is a limitation of this cellular system, one of its strengths is its ability to assess cellular function in the absence of these extrinsic circulating factors, since our aim is to investigate the cell-intrinsic changes in insulin sensitivity in muscle. Interestingly, there are differences in sex hormone receptors in iMyos. While the estrogen receptor mRNA was not detected in the iMyos of either sex, the level of androgen receptor ( AR ), which coincidentally is encoded on the X chromosome, was higher in cells from I-Sen male individuals compared with both I-Res male and all female individuals. Testosterone deficiency in men has been associated with the development of obesity, insulin resistance, and T2D, in addition to its effects on erectile dysfunction ( 55 ). Likewise, men with prostate carcinoma receiving androgen deprivation therapy show a higher risk of developing insulin resistance and hyperglycemia ( 56 ), consistent with our findings of reduced AR expression level in I-Res iMyos. Indeed, normalizing AR levels in I-Res male individuals upon incubation with DHT rescues, at least in part, the impaired insulin-stimulated glucose uptake defect. In this perspective, clinical studies have shown that testosterone can promote insulin sensitivity in hypogonadal men with and without diabetes ( 57 ), and in women, androgen excess promotes insulin resistance ( 58 , 59 ). Thus, differences in AR levels could account for insulin sensitivity changes in gene expression in male iMyos; however, this is not likely to be the major driver of the sex-specific gene expression changes. Indeed, known androgen-responsive genes in the muscle ( 37 ) show minimal overlap with the sex differences in gene expression in iMyos. Likewise, in ongoing work, we find that blocking AR action in iMyos by treatment with the AR antagonist bicalutamide ( 60 ) has little impact on the sex-based differences in gene expression. In normal development, exposure to sex hormones during different developmental stages of life is known to exert epigenetic changes that can persist throughout life. Many of these are related to DNA methylation (reviewed in ref. 61 ). Similarly, alterations of the DNA methylation can contribute to differences in gene expression and provide a link between the development of metabolic diseases, genes, and environment ( 61 , 62 ). Indeed, altered DNA methylation of genes such as PDK4 and PPARGC1A has been found in skeletal muscle from patients with T2D ( 63 – 65 ). Here, we find significantly higher global DNA methylation in female iMyos as compared with male iMyos, independent of insulin sensitivity status, consistent with other studies, which have shown that in addition to methylation on the inactive X chromosome, female cells also show higher levels of autosomal methylation in muscle ( 39 ). Interestingly, this difference in global levels of DNA methylation was not observed in iMyos differentiated from iPSCs of prepubertal male and female individuals, suggesting that the sex hormone exposure, or other possible mechanisms such as aging, in the postpubertal female might lead to epigenetically mediated DNA methylation changes, some of which persist or reoccur through the reprogramming and differentiation process. Indeed, Landen et al. found that human skeletal muscle samples from female individuals had an increased number of differentially methylated regions as compared with muscle from male individuals ( 39 ). Overlapping our data with their data revealed that 39% of the genes exhibiting sex-differential expression in our data contain differentially methylated positions; however, 61% do not. Therefore, while DNA methylation may contribute to a significant fraction of the sex-specific differences in gene expression, the molecular mechanism underlying the majority of these genes involves mechanisms other than DNA methylation. Nonetheless, blocking DNA methylation with the DNMT inhibitor 5-Az reverses at least some of the sex-dependent differences in expression of male- and female-dominant genes, as well as a sexually dimorphic functional difference in RhoA activation. Despite a role in sex-specific changes, DNA methylation does not appear to be the major driving force in gene expression differences related to insulin resistance or the reduced ability of insulin to stimulate glucose uptake in I-Res cells, since neither of these were changed by 5-Az treatment. This leaves open the important question of how these insulin resistance–related changes in gene expression and glucose uptake are mediated. They could be mediated by underlying genetic effects that lead to the altered transcriptional regulation observed in these cells. In addition, there could be epigenetically mediated mechanisms involving histone modification and effects of noncoding RNAs. Histones can undergo modifications involving acetylation, methylation, phosphorylation, and ubiquitination, all of which can lead to changes in chromatin structure and alterations in gene expression ( 66 ). Histone modifiers and chromatin remodelers play an important role in iPSC reprogramming and differentiation (reviewed in ref. 67 ), but how these are regulated in redifferentiation of iPSCs is less well studied. In addition, both long noncoding and short noncoding miRNAs can modify transcription and translation ( 68 ). Indeed, XIST is an important example of how a single lncRNA can affect expression of multiple genes. The role of these other epigenetic mechanisms in insulin resistance and sex-specific gene expression changes remains to be investigated. In summary, human iPSC-derived myoblasts demonstrate a cell-intrinsic gene expression signature associated with insulin resistance. These gene expression changes are retained in cells after the reprogramming process and differentiation of iPSCs into myoblasts and appear to be independent of DNA methylation, indicating the cell-autonomous nature of these insulin sensitivity differences. Determining the mechanisms underlying these differences should provide new targets for defying insulin resistance and preventing its metabolic consequences. In addition, iPSC-derived myoblasts exhibit a large panel of differences in gene expression that are sex specific, most of which involve genes encoded by autosomal chromosomes. At least one mechanism linked to this sexual dimorphism in gene expression is differences in DNA methylation, possibly related to sex hormone exposure and epigenetic programming in the donor in vivo that either persists through iPSC reprogramming or is reintroduced during differentiation of the iPSCs into myoblasts. Understanding the impact of sex on gene expression will be important not only in insulin resistance, but also in normal physiology and pathophysiology of many diseases.

About 25% of people in the general population are insulin resistant, increasing the risk for type 2 diabetes (T2D) and metabolic disease. Transcriptomic analysis of induced pluripotent stem cells differentiated into myoblasts (iMyos) from insulin-resistant (I-Res) versus insulin-sensitive (I-Sen) nondiabetic individuals revealed that 306 genes increased and 271 genes decreased in expression in iMyos from I-Res donors with differences of 2-fold or more. Over 30 of the genes changed in I-Res iMyos were associated with T2D by SNPs and were functionally linked to insulin action and control of metabolism. Interestingly, we also identified more than 1,500 differences in gene expression that were dependent on the sex of the cell donor, some of which modified the insulin resistance effects. Many of these sex differences were associated with increased DNA methylation in cells from female donors and were reversed by 5-azacytidine. By contrast, the insulin sensitivity differences were not reversed and thus appear to reflect genetic or methylation-independent epigenetic effects. Transcriptomics of iPSC-derived myoblasts demonstrates altered gene expression in insulin resistance along with a major effect of donor sex driven by epigenetic changes.

Author contributions NH designed and performed all the experiments, analyzed all the data, designed the figures, and wrote the paper. CRK conceived the study, helped with data analysis and interpretation, reviewed and edited the manuscript, and supervised the project. Both authors read, reviewed, and edited the manuscript. Supplementary Material

We thank Kahn laboratory members, especially Thiago M. Batista and Ari Gattu, for discussions. We thank Bioinformatics Core members Hui Pan and Jonathan Dreyfuss at the Joslin Diabetes Center for help with the data analysis. Ivan Carcamo-Orive and Joshua W. Knowles provided the I-Sen and I-Res iPSC lines. This work was supported by NIH grants R01DK031036, R01DK128429 (to CRK), P30DK036836 (to Joslin Diabetes Center), and the Mary K. Iacocca Professorship (to CRK). 11/30/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e172333

oa_package/0a/47/PMC10786688.tar.gz

PMC10786689

0

Authorship note: JTH and SGC contributed equally to this work. Titin (TTN) is one of the largest and most complex proteins expressed in humans, and truncation variants are the most prevalent genetic lesion identified in individuals with dilated cardiomyopathy (DCM) or other disorders of impaired cardiac contractility. Two reports in this issue of the JCI shed light on a potential mechanism involving truncated TTN sarcomere integration and the potential for disruption of sarcomere structural integrity. Kellermayer, Tordai, and colleagues confirmed the presence of truncated TTN protein in human DCM samples. McAfee and authors developed a patient-specific TTN antibody to study truncated TTN subcellular localization and to explore its functional consequences. A “poison peptide” mechanism emerges that inspires alternative therapeutic approaches while opening new lines for inquiry, such as the role of haploinsufficiency of full-length TTN protein, mechanisms explaining sarcomere dysfunction, and explanations for variable penetrance.

TTN structure and function The sarcomere is the fundamental contractile unit of the myocyte and is commonly subdivided into Z-disc, I-band, A-band, and M-line regions, and the titin (TTN) protein spans half the sarcomere. The TTN gene has 364 exons (meta-transcript, Ensembl ID: ENST00000589042), which are differentially spliced in heart and skeletal muscle through development and disease ( 1 ). TTN cardiac expression is regulated by two promoters: a major promoter that regulates the expression of N2BA, N2B, and novex isoforms (novex-1, novex-2, and novex-3), and an internal promoter located at the I-/A-band junction that regulates the expression of Cronos, a developmentally regulated TTN isoform that is missing Z-disk and most I-band exons ( 2 ). N2BA is the longest isoform, ranging in size from approximately 3.3 to 3.8 MDa, and is the predominant isoform in the developing heart, while N2B is approximately 3 MDa and lacks many I-band exons, including those encoding PEVK (enriched with proline, glutamate, valine, and lysine residues), and other extensible segments. In heart failure, the stoichiometry of N2BA to N2B shifts higher to favor the more extensible N2BA isoform ( 3 ) that resembles the ratio identified in fetal hearts ( 4 ). TTN is required for sarcomere assembly and twitch contraction ( 5 ) through interactions with multiple partners including α-actinin ( 6 ) at the Z-disc and obscurin ( 7 ) at the M-line. Contributing to passive force, TTN’s distensible, spring-like I-band and PEVK domains can be modified by phosphorylation ( 8 ) and other factors. Moreover, TTN functions as a mechanotransduction signaling hub with stretch-dependent TTN-protein interactions ( 9 ). TTN’s complex regulation and functions as well as its enormous size have limited our understanding of its dysfunction in cardiac disorders until recently. TTN variants in health and disease Heterozygous TTN truncation variants ( TTN tvs) are the most prevalent genetic lesion identified in dilated cardiomyopathy (DCM), a disorder associated with cardiac chamber enlargement and impaired contractile function ( 10 ). DCM prevalence has been estimated at approximately 1 in 200 individuals, and TTN tvs can be identified in up to 25% of individuals with DCM ( 11 ). DCM risk may depend on TTN tv localization, as TTN tvs localized to exons encoding A-band residues have a higher pathogenicity compared with those localized in differentially spliced exons such as those encoding I-band residues ( 12 ). DCM risk is elevated up to approximately 50-fold in carriers of TTN tvs, but healthy individuals may also harbor TTN tvs, and explanations for this variable penetrance are incomplete but probably related to genetic background. Acquired risk factors are yet to be determined. TTN tvs are also reported in peripartum cardiomyopathy ( 13 ) and cardiomyopathy associated with chronic alcohol consumption, suggesting that additional stressors act in concert with TTN tvs ( 14 ). The sarcomere integrates TTN tv-generated truncated TTN protein Recent studies, including two presented in this issue of the JCI , have utilized vertical agarose gel electrophoresis (VAGE) to evaluate expression of the TTN tv allele in human DCM myocardium and model systems. An early study, relying on linkage analysis of the TTN locus on chromosome 2q31, used VAGE to evaluate human myocardial lysates with an A-band TTNtv (c.43628insAT) ( 15 ). The presence of truncated TTN protein of the c.43628insAT allele was further corroborated by the same group in a functional study characterizing a c.43628insAT knockin mouse model ( 5 ). Indeed, the truncated TTN protein was estimated to be approximately 1% of full-length TTN. Additional studies of two TTNtv DCM cohorts also validated the presence of low-level truncated TTN species from human myocardial specimens, largely corroborating earlier work from human cardiomyocytes differentiated in vitro from an induced pluripotent stem cell (iPSC) model derived from an individual with DCM who was a TTN tv carrier ( 16 , 17 ). Now in the JCI , Kellermayer, Tordai, and co-authors ( 18 ) confirmed the presence of truncated TTN protein in additional human DCM samples, while further supporting previous studies that had demonstrated truncated TTN protein within myofibril fractions isolated by biochemical approaches ( 16 ) or by colocalization microscopy ( 19 ). In the JCI study, Kellermayer, Tordai, and colleagues question the role of TTN haploinsufficiency ( 18 ). However, the findings support a dominant-negative or poison peptide genetic mechanism for some TTN tvs ( Figure 1 ). To begin studying the functional impact of TTNtvs, McAfee et al. now report an elegant study ( 20 ). The authors developed a patient-specific, custom TTNtv antibody that was designed to specifically recognize the 32 amino acid neoepitope encoded by a DCM-associated heterozygous exon 329 frameshift mutation corresponding to the A-band structural domain (termed TTNtvA). Since the custom antibody did not bind to WT TTN protein, this tool could be used to study truncated TTN subcellular localization and to explore its functional consequences. Despite the lack of an M-line domain, TTNtvA protein was identified in skinned human cardiomyocyte fragments in the sarcomere thick filament/A-band region. This location was predicted for TTNtvA, given the position of its termination codon within the mid A-band region. To further examine the functional properties of TTNtvA protein, the research group stretched cardiomyocyte fragments from short to supraphysiological sarcomere lengths and imaged TTNtvA using the custom antibody recognizing its C-terminus. If TTNtvA were to maintain its terminal A-band positioning after stretching, and not recoil to either the Z-disc or other subsarcomere region, it could be reasonably inferred that TTNtvA could bear load across the sarcomere. Indeed, McAfee and authors observed no change in TTNtvA positioning with stretch unless potassium chloride, a thick filament disruptor, was added ( 20 ). These results demonstrate how truncated TTN can integrate into the sarcomere and bear load in a human myocardial sample. While this finding was a step forward, the functional consequences of a load-bearing, truncated TTN remain completely unknown. To consider how truncated TTN protein impacts sarcomere structure and function, Kellermayer, Tordai, and colleagues ( 18 ) also report on an analysis of human TTNtv myocardial samples using super-resolution stimulated emission depletion (STED) microscopy with TTN antibodies recognizing different epitopes either common or exclusive to full-length TTN or truncated TTN. As in the study by McAfee et al. ( 20 ), imaging was performed in conjunction with mechanical stretch. In brief, Kellermayer, Tordai, and co-authors observed that truncated TTN was expressed in myofibril fractions and that mechanical stretch elicited reduced A-band extensibility and increased distance between the titin kinase domain and the M-line, suggesting putative functional consequences of TTNtvs. Their report of structural and functional consequences may need to be further validated using reagents that specifically recognize truncated TTN proteins, but it nonetheless supports a poison peptide mechanism ( 18 ). TTNtvs also reduce full-length TTN protein levels In addition to truncated TTN production from TTN tv alleles, TTN tvs have also been reported to lead to reduced full-length TTN protein levels in human myocardial samples, suggesting a haploinsufficiency genetic mechanism ( Figure 1 ). Defined as the inability of the single WT TTN allele to produce sufficient full-length TTN protein to maintain normal cardiac function , in two recent studies ( 16 , 17 ) and in the report by Kellermayer, Tordai, and colleagues ( 18 ), TTN tv DCM myocardial samples expressed approximately 15% less full-length TTN protein relative to other DCM or control samples. Similar results were observed in human iPSC–derived cardiomyocyte models composed of similar TTNtvs, although with greater reductions of approximately 50% relative to controls ( 19 ). The role of reduced TTN protein levels is less well understood in DCM pathogenesis, but potential mechanisms gleaned from functional studies implicate impaired sarcomere function ( 19 , 21 , 22 ) and cell-signaling pathways ( 21 ). Questions and future directions Secondary to TTN’s large size and complex structure, it has been a challenge to the field to delineate the functional consequences of TTNtvs. The McAfee et al. study ( 20 ), with its unique strategy for exclusive detection of truncated TTN, clarifies the presence and behavior of TTN within the sarcomere. However, future studies will be essential to understand how this protein, despite a capacity for bearing mechanical load, differs from full-length TTN. Specifically, does TTNtv’s lack of thick filament–encoding residues impair force production and disturb other protein interactions such as M-line interactions, or could it disturb cardiac function through activation of the unfolded protein response as recently reported by others ( 17 )? While McAfee et al. ( 20 ) provide some insights into localization and load capacity, more work is needed to fully understand the functionality of truncated TTNs and whether they differ for distinct truncations. Similarly, Kellermayer, Tordai, and co-authors ( 18 ) report structural alterations in DCM samples from individuals carrying TTNtvs, but made several assumptions based on indirect studies from heterogenous samples obtained from explanted human tissue. In considering all the studies together, it appears that the combination of reduced total full-length TTN and the insertion of a truncated peptide into the sarcomere is present and likely plays a role in disease pathogenesis. Models and approaches are needed to experimentally dissect the functional consequences of sarcomere-integrating TTNtv proteins as well as full-length TTN protein haploinsufficiency. One such approach may be to combine human iPSC models and genome-editing technologies such as CRISPR. Despite maturation limitations, human iPSC–derived cardiomyocytes can be developed using CRISPR genetic ablation of truncated TTN protein to explore the specific functional impact of these poison peptides. Through the use of a 3D cardiac microtissue model to study sarcomere contractile function in a biomimetic context, truncated TTN ablation was shown to partially rescue sarcomere function, thus implicating truncated TTN as a sarcomere poison. Because the rescue was partial, this study also supported a combinatorial genetic mechanism including haploinsufficiency ( 19 ). Continued efforts toward a precise understanding of how TTNtvs lead to DCM and other cardiomyopathies will catalyze the development of mechanistically precise therapies targeting one of the most important heart failure causes in the field.

We thank Gloria Fuentes for assistance with figure design. Funding supported was obtained from the NIH (R01HL165220, to JTH and SGC). 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e175206

oa_package/2f/97/PMC10786689.tar.gz

PMC10786690

0

Introduction The relationship between the immune system and cancer growth is complex; the interplay between the immune system and the tumor impacts all aspects of cancer growth, from tumorigenesis to tumor growth and eventual metastasis. Such interaction between the immune system and the tumor is based on the theory of cancer immunoediting, which involves 3 phases: elimination, equilibrium and escape ( 1 ). In the elimination phase, a competent immune system attacks and destroys tumor cells. Tumor cells that survive the initial attack may enter a state of dormancy known as the equilibrium phase before entering the escape phase, in which tumor cells acquire the ability to evade the immune system and grow unchecked ( 1 ). Tumors can escape immune surveillance through various mechanisms, including alteration or loss of antigens, upregulation of immune checkpoint molecules, and manipulation of cytokines and oncogenic signaling to create an immune-suppressive tumor microenvironment ( 2 , 3 ). Attempts to reengage the immune system to counteract cancer-induced immune evasion have resulted in the emergence of various immunotherapeutic modalities ( 4 , 5 ). Immune checkpoint inhibitors (ICIs) are one such immunotherapeutic intervention. Immune checkpoints are a group of proteins that are expressed on the surface of various cells, and they play a crucial role in modulating immune responses. However, immune checkpoints have been exploited by cancer cells for immune evasion ( 6 ). In turn, ICI therapy seeks to overcome this unfavorable immune-cancer interaction. ICIs have been approved to treat several types of cancer, including melanoma, lymphoma, renal cell, and lung cancer. Particularly, ICIs are increasingly being used as a first-line option for both early and advanced stages of non–small cell lung cancer (NSCLC) ( 7 ), a leading cause of cancer-related deaths globally ( 8 ). The use of ICIs has shown improved overall survival in NSCLC ( 9 ), but it has also been linked to immune-related adverse events (irAEs), including lung toxicity ( 10 , 11 ). Checkpoint inhibitor pneumonitis (CIP) is a major irAE associated with significant morbidity and mortality ( 12 , 13 ). Despite this clinical significance, the mechanisms underlying lung injury in CIP are not well understood, limiting the availability of specific treatments. In this Review, we outline the mechanisms of action and use of ICIs, the clinical features of CIP, and recent research aimed at understanding the biological underpinnings of this condition.

Cancer remains a leading cause of mortality on a global scale. Lung cancer, specifically non–small cell lung cancer (NSCLC), is a prominent contributor to this burden. The management of NSCLC has advanced substantially in recent years, with immunotherapeutic agents, such as immune checkpoint inhibitors (ICIs), leading to improved patient outcomes. Although generally well tolerated, the administration of ICIs can result in unique side effects known as immune-related adverse events (irAEs). The occurrence of irAEs involving the lungs, specifically checkpoint inhibitor pneumonitis (CIP), can have a profound effect on both future therapy options and overall survival. Despite CIP being one of the more common serious irAEs, limited treatment options are currently available, in part due to a lack of understanding of the underlying mechanisms involved in its development. In this Review, we aim to provide an overview of the epidemiology and clinical characteristics of CIP, followed by an examination of the emerging literature on the pathobiology of this condition.

ICI biology T cells are an integral part of the adaptive immune response; their activation by an offending antigen (e.g., an infectious agent or a tumor antigen) can propagate a series of inflammatory responses. In cancer, these antigens can arise from tumor cells; as such, T cells are known to be an important factor in immune cancer surveillance. However, T cell activation is naturally balanced by counterregulatory mechanisms in the form of immune checkpoints, which serve to curb the immune response and avoid autoimmunity ( 14 ). This natural phenomenon in turn is exploited by cancer cells to induce immune tolerance. The natural T cell–mediated immune response is a complex process that involves multiple steps and interactions with other cell types aimed at targeting specific foreign antigens or epitopes. During immune surveillance, antigen-presenting cells (APCs) play a crucial role in this process by recognizing tumor antigens and activating T cells ( 1 ). T cell activation requires two signals: a primary signal transmitted through its T cell receptor and a costimulatory signal delivered by the CD28 receptor. This costimulation occurs when APC surface proteins B7-1 (CD80) or B7-2 (CD86) interact with CD28 ( 6 ). This costimulatory signal is essential for proper T cell activation and function. Cytotoxic T lymphocyte–associated protein-4 (CTLA-4) and programmed cell death protein-1 (PD-1) receptors are critical checkpoints that function at distinct stages of the T cell activation cycle. CTLA-4 competes with CD28 for their shared ligands, CD80 or CD86 ( 15 ), limiting further T cell activation ( 16 ). This process occurs at the start of the T cell activation cycle. T cells subsequently can become unresponsive or undergo a state of anergy ( 17 ). At the same time, CTLA-4 signaling in T cells promotes their transformation into Tregs ( 18 ) that are essential in peripheral immune tolerance ( 19 ). CTLA-4 expression in lymphocytes is thought to be driven by lung cancer or the tumor microenvironment ( 20 ). While CTLA-4 regulates the early process of T cell activation within the lymph nodes, PD-1 checkpoints function at a later stage in the peripheral tissues. In contrast to that of CTLA-4, PD-1 expression is observed at the late effector phase of the activated T cell ( 21 ). Upon its interaction with programmed cell death ligand-1 or -2 (PD-L1 or PD-L2), intracellular pathways are triggered that inhibit T cell receptor signaling, decrease proliferation, and suppress effector functions ( 22 ). PD-1’s ligands are expressed on a wide variety of nonhematopoietic cells, including endothelial and epithelial cells ( 23 ). Cancer cells exploit the inhibitory function of the PD-1 checkpoint pathway by overexpressing PD-L1 or PD-L2, thereby limiting the host’s immune response ( 24 ) ( Figure 1 ). In addition to the CTLA-4 and PD-1 pathways, lymphocyte activation gene 3 (LAG-3) is a known checkpoint involved in immune regulation. LAG-3 is primarily found on lymphocytes, but it can also be expressed in NK and myeloid cells. LAG-3 activation by interaction with its primary ligand, the major histocompatibility II receptor found on APCs, can impede T cell proliferation and differentiation ( 25 ). In vivo pharmacological blockade or genetic ablation of LAG-3 has been consequently shown to restore function in exhausted T cells, leading to an enhanced antitumor response ( 26 ). However, simultaneous inhibition of LAG-3 and PD-1 function resulted in the optimal antitumor response in murine models ( 27 ). Therefore, clinically, the use of LAG-3 blocking antibodies has only been approved in combination with PD-1 blockade ( 28 ), and limited data are available on its pulmonary toxicity profile. Other checkpoint proteins, including B and T cell lymphocyte attenuator (BTLA), V-domain Ig suppressor of T cell activation (VISTA), and T cell immunoglobulin and mucin domain 3 (TIM-3) are also being investigated as potential targets for inhibition in early-stage clinical trials ( 29 ). Like PD-1 and LAG-3, these proteins all play a role in limiting T cell activity and effector function. While clearly important in cancer biology, these pathways also play a pivotal role in maintaining immune homeostasis. This is best evidenced by the emergence of spontaneous autoimmunity in CTLA-4– and PD-1–knockout murine models ( 14 , 30 , 31 ). It is noteworthy to mention that the loss of CTLA-4 inhibition in murine models leads to fatal disease soon after birth, in contrast to PD-1–knockout mice that develop nonlethal autoimmunity ( 30 , 32 ). Unsurprisingly, disruption of these pathways through therapeutic blockade, while beneficial for tumor control, also incurs the risk of autoimmune dysfunction. Such dysfunction clinically manifests as irAEs in patients with cancer undergoing ICI treatment. irAEs can involve any organ, frequently affecting the skin as dermatitis, the gastrointestinal tract as colitis, and the endocrine system in the form of thyroiditis or hypophysitis. Of particular importance to this Review, lung involvement can present in several ways, most often as pneumonitis ( 33 ). Pulmonary irAEs ICI therapy has been linked to several complications that affect the lungs. These include sarcoid-like granulomatosis ( 34 – 37 ), pleural effusion ( 38 ), exacerbation of obstructive lung disease ( 37 , 39 – 43 ), and, most notably, CIP. Among these complications, CIP is the most widely recognized and is of serious concern due to its high morbidity and mortality rates ( 12 , 44 ). In fact, CIP is the leading cause of fatal irAEs in patients receiving anti–PD-1/PD-L1 monotherapy ( 13 ). At first, CIP incidence was thought to be low; CIP incidence in clinical trials was reported to be <6% ( 38 , 45 – 49 ). However, this assertion has been challenged by real-world data, which revealed higher rates ranging between 10% and 20% ( 44 , 50 , 51 ). Various factors have been shown to increase the risk of CIP, including use of PD-1/PD-L1 agents ( 52 , 53 ) (compared with CTLA-4 inhibitors), combination immunotherapy ( 54 , 55 ) (CTLA-4 and PD-1 combination rather than monotherapy), radiotherapy ( 56 , 57 ), and the organ of tumor origin ( 58 , 59 ) (e.g., NSCLC vs. renal cell or other cancer types). More recently, the incorporation of ICIs into neoadjuvant therapy for resectable lung cancer has revealed similar rates of CIP to those witnessed with adjuvant treatment (ranging from 1.1% to 6.4%, see Table 1 ). However, these findings necessitate careful interpretation for several reasons. First, it’s crucial to acknowledge that pneumonitis rates primarily stem from phase II clinical trials, which typically involve a smaller participant pool that may not reflect the general population. Second, our experience from the adjuvant trials underscores a concern that the reported rates of pneumonitis in these controlled trial settings may underestimate the true incidence observed in real-world clinical practice. Additionally, it’s noteworthy that a significant subset of participants receive radiotherapy, a risk factor linked with elevated pneumonitis incidence. Moreover, the integration of both neoadjuvant and adjuvant treatments in certain trial protocols further complicates the overall assessment of pneumonitis risk. A summary of the trials and their reported pneumonitis rates is provided in Table 1 . The clinical presentation of CIP includes dyspnea, cough, and hypoxemia (either at rest or with exertion) ( 60 ). These symptoms may occur soon after treatment initiation, typically within days to weeks ( 61 ). The time frames reported vary among study cohorts; however, the reported median onset is approximately 2–3 months after ICI initiation ( 12 , 51 , 61 ). The common terminology criteria for adverse events ( 62 ) can be utilized to grade the overall clinical severity of CIP, as outlined in Table 2 . Pathognomonic radiological features for CIP do not exist. Instead, a range of changes have been identified on CT scans, with approximately 50% of the cases reported in the literature presenting either as ground-glass opacities or consolidative lesions ( 63 ). The diagnosis of CIP is established after excluding other potential etiologies, such as infection, tumor progression, and alveolar hemorrhage. Although bronchoalveolar lavage fluid (BALF) may be employed to exclude alternative diagnoses or infections, it is variably employed and is not considered an essential diagnostic procedure ( 64 ). Once a diagnosis of CIP is established, ICI therapy is typically terminated, and systemic glucocorticoid therapy is initiated for the majority of patients with more severe (>grade 1) disease. However, a subset of patients fails to respond to steroids and may require additional immunosuppressive interventions beyond glucocorticoids ( 61 , 65 , 66 ). Subsequent rechallenge with ICIs after CIP resolution is typically possible in only a small percentage of patients, thereby limiting future immunotherapy for a substantial proportion of affected individuals ( 12 , 67 ). Pathophysiology of CIP The clinical presentation of CIP is similar to that of acute lung injury. However, despite being a well-recognized irAE, the pathophysiology of CIP, similar to other irAEs, remains largely elusive. Nevertheless, preliminary findings suggest the involvement of several mechanistic pathways in the development of CIP, which are described below. Cellular autoimmunity/higher T cell activity. There is accumulating evidence to suggest that T cell upregulation may be involved in the pathogenesis of CIP ( Figure 2 ). Many investigators have shown an increased number of overall T cells in addition to certain T cell subsets. Notably, in our study, the BALF of patients with CIP has shown a significant increase in CD4 + T cells ( 68 ). In 12 patients with CIP, CD4 + CD45RA – CD62L + central memory T (Tcm) cells were found to be increased ( 68 ). Tcm cells are derived from either CD4 + or CD8 + lymphocytes that circulate in the blood and target secondary lymphoid tissues to quickly propagate in response to familiar antigens ( 69 ), leading to a more rapid and augmented immune response ( 70 ). Additionally, Tcm cells have previously been shown to be resistant to steroid-induced apoptosis ( 71 ), which may account for the steroid-refractory nature of CIP in some patients. Moreover, CD62L has been demonstrated to facilitate the migration of T cells to sites of inflammation ( 72 ). This notion receives further support from research conducted on peripheral blood T cell profiling in patients with melanoma who underwent ICI treatment. Among the 18 individuals studied who later suffered irAEs, those with higher pretreatment levels of circulating Tcm cells were found to develop severe irAEs ( 73 ). In another study, researchers examined 11 patients with CIP by performing BALF sampling and conducting single-cell RNA and T cell sequencing. The transcriptomic signature of these patients showed an accumulation of Th lymphocytes, specifically Th17.1 ( 74 ), a subset of Th 17 cells that produce IFN-γ and are implicated in a number of autoimmune diseases ( 75 ). Among these Th17.1 lymphocytes, a unique cluster was identified that had a distinct transcriptomic signature characterized by genes related to cytotoxicity and monocyte activation. Using trajectory inference, they demonstrated that the Th17.1 lymphocyte colony in CIP BALF samples has plasticity and undergoes pathogenic skewing toward this IFN-γ and monocyte activation phenotype. At baseline, Th17 cells play a pivotal role in upholding gut barrier defenses, facilitating granulocyte maturation and chemotaxis, and contributing to immunity against extracellular pathogens ( 76 ). The depletion of proinflammatory Th17 cells heightens vulnerability to infections caused by Staphylococcus aureus and Candida albicans , culminating in recurrent skin and pulmonary infections ( 77 ). Concurrently, an excessive Th17 cell response can precipitate autoimmune reactions. In particular, Th17.1 lymphocytes have been implicated in driving neutrophilic inflammation, granuloma formation, and provoking resistance to corticosteroid treatment ( 75 ). Elevated quantities of these cells have been demonstrated in BALF from individuals with sarcoidosis, and they are associated with active lung disease ( 78 – 80 ). It is conceivable that the equilibrium of such a cell population may be disrupted in the context of ICI therapy, as demonstrated in a murine model in which checkpoint inhibition triggered the activation of Th17 lymphocytes ( 81 ). While these findings support the hypothesis that proinflammatory T cell subsets contribute to alveolar damage in CIP, other studies have suggested that the proliferating T cells may be clonal. For example, investigators have identified identical T cell clones in tumor tissue and the site of irAEs, albeit in extrapulmonary locations. Autopsy specimens from two patients with melanoma who died from myocarditis after receiving combination therapy with CTLA-4/PD-1 antibodies revealed shared T cell clones in the tumor, heart, and skeletal muscle, without any evidence of adjacent tissue involvement, including the smooth muscle ( 82 ). Similarly, of 73 patients with NSCLC who received anti–PD-1 agents, 25 developed skin toxic effects, and nine shared T cell antigens that were detected in both cancer and skin tissue ( 83 ). A potentially similar mechanism may also exist in CIP. Analysis of T cell receptor sequencing data from four patients who had developed pneumonitis while undergoing PD-1 blockade therapy revealed the presence of overlapping T cell clones in both the lung and tumor tissue ( 84 ). These clones were absent in peripheral blood and secondary lymphoid organs. This finding suggests that the T cell response to therapy-induced lung damage may be mediated by specific tumor antigens, rather than being a result of nonspecific immune activation. There is further evidence to support the notion of a clonal T cell selection within the tumor microenvironment ( 85 ). Specifically, in an analysis of BALF T cells from 10 patients with CIP, Suzuki et al. observed an increase in the number of PD-1 + and TIM-3 + CD8 + BALF T cells in patients with pneumonitis ( 86 ). Interestingly, CD8 cells in these patients also exhibited an increased expression of T cell immunoreceptor with immunoglobulin and ITIM domains (TIGIT). Notably, both TIGIT and TIM-3 are classified as second-wave immune checkpoints and are highly expressed in tumor infiltrating lymphocytes ( 87 , 88 ). Collectively, these findings suggest that T cells present in the tumor microenvironment may relocate to other lung compartments following activation with checkpoint blocking antibodies, possibly localizing to certain coexpressed antigens (in both the tumor microenvironment and normal lung tissue) and ultimately leading to the development of pneumonitis. In fact, these cells may not be too dissimilar to the PD-1 + CD8 + T cells identified in our CIP cohort ( 68 ). The mechanisms described above could potentially account for the sporadic nature of CIP, as it relies on antigenic similarities between the tumor and lung parenchyma. To this end, there have been several findings suggestive of molecular mimicry as a possible driver for CIP, particularly in the context of tumor mutational burden (TMB). Elevated TMB levels may be linked to a higher incidence of irAEs, potentially due to the development of neoantigens or the release of tumor antigens following cell death ( 89 ). These antigens can cause a cross-reaction with healthy tissue antigens, leading to the manifestation of irAEs. However, a recent meta-analysis failed to establish a significant correlation between TMB and irAE development ( 90 ), suggesting that other factors may also be at play. It is worth noting that a higher TMB was associated with increased tumor response rates, suggestive of a better immune trigger, though this did not necessarily translate into a higher incidence of toxicity ( 90 ). This highlights the complexity of the relationship between these two factors. Unraveling the association between the TMB and CIP may be attempted by identifying any overlap between the mutational burden in both CIP and cancer tissue samples with a corresponding T cell clone response within both compartments. Most studies investigating CIP have focused on lymphocyte changes, leaving a gap in our understanding of myeloid cell alterations during the disease process. However, recent research has shed light on the involvement of proinflammatory macrophages in CIP. For example, in 37 patients with NSCLC, there was upregulation of proinflammatory macrophages in CIP identified using bulk RNA sequencing of surgical tissue specimens ( 91 ). In this research, macrophages from patients with CIP expressed higher levels of TNF and CXC chemokine ligand-10 (CXCL-10) compared with that in individuals in the control group ( 91 ). These findings align with previous flow cytometric analysis of CIP BALF specimens, which revealed distinct clusters of IL-1β hi , TNF-α hi , CD-11b hi myeloid cells that were significantly upregulated in CIP BALF ( 68 , 74 ), along with increased BALF protein levels of CXCL-10. Similarly, Franken et al. demonstrated comparable results regarding myeloid dysfunction in CIP; they identified two different clusters representing monocyte and macrophage cells ( 74 ). The first cluster consisted of IL-1β hi monocytes with high expression of CXCL-10, while the second cluster had increased expression of proinflammatory macrophage genes, including IL-1β and TNF. Taken together, these findings may indicate that the observed changes in innate immune cells, specifically monocytes and macrophages with IL-1β hi TNF-α hi CXCL-10 hi profile, may play a central role in attracting lymphocytes to the alveoli, thus contributing to the pathogenesis of CIP. Upregulated levels of autoantibodies. Humoral immunity may also play a role in the development of irAEs associated with ICI therapy. Autoantibodies, which may already be present at low levels prior to ICI therapy or produced de novo, have been implicated in the pathogenesis of some nonpulmonary irAEs. For example, in seven patients who developed bullous pemphigoid (an autoimmune blistering skin condition) after PD-L1 therapy, a unique antibody to a basement membrane protein was observed ( 92 ). Similarly, thyroid dysfunction has also been linked to ICI therapy, with many patients who developed thyroiditis having evidence of circulating antithyroid antibodies either present at baseline or developed during treatment ( 93 ). With regards to autoantibodies in CIP, in a cohort of 66 patients experiencing irAEs, of which 14 (21%) developed pneumonitis, preexisting elevation of rheumatoid factor or antinuclear antibodies was significantly associated with irAE ( 94 ). Additionally, anti-CD74 autoantibodies have recently been implicated in patients with CIP ( 95 ). CD74 functions as a chaperone molecule involved in major histocompatibility complex II intracellular trafficking, and it acts as a high-affinity receptor for macrophage inhibitory factor, inducing inflammatory mediators and cell proliferation ( 96 – 98 ). High-throughput serological analysis of recombinant cDNA expression by Tahir et al. revealed a significant median 1.34-fold increase in anti-CD74 antibody levels after ICI treatment in CIP, while no significant changes were noted in a comparison group of 20 patients without pneumonitis ( 95 ). These findings suggest a potential role for antibody-mediated mechanisms in the development of CIP. Cytokine dysfunction. Elevated levels of various cytokines have been associated with irAEs, including CIP. In fact, there are significant changes in cytokine levels in patients who develop irAEs following ICI treatment ( 99 ). Though clearly capable of contributing to lung injury, whether these cytokine increases are causally related to irAE development remains to be seen ( 33 ). Interestingly, some cytokines have shown potential as predictive biomarkers for irAEs. For example, in a discovery cohort of 58 patients with melanoma, samples taken at baseline and at the time of toxicity identified 11 signature cytokines (including granulocyte colony-stimulating factor, granulocyte macrophage colony-stimulating factor, fractalkine, basic fibroblast growth factor-2, IFN-α2, IL-12p70, IL-1a, IL-1β, IL-1 receptor agonist, IL-2, and IL-13) that strongly correlated with the development of severe irAEs, including two cases of pneumonitis ( 100 ). Another study looking specifically at 204 patients with NSCLC, of which 43 developed irAEs, found a similar proinflammatory increase in IL-1β cytokine but also elevations in IL-5, IL-8, IL-10, IL-12p70, and granzyme A and decreased G-CSF as predictors for irAEs that included pneumonitis (5 of a total of 43) ( 101 ). IL-5, IL-8, and IL-12p70 are considered proinflammatory cytokines ( 102 – 104 ). The actual mechanism as to how IL-5, a Th2 cytokine and a powerful eosinophil activator and recruiter ( 105 ), may be involved in lung injury of pneumonitis is uncertain. One plausible explanation may be in its secondary role of B cell stimulation and augmentation of immunoglobulin production ( 106 ), but how this may lead to developing CIP is unclear. IL-8 exerts its influence through a range of mechanisms, which encompass the enhancement of neutrophil activation, granule release, superoxide generation, and the expression of adhesion molecules ( 107 ). Additionally, receptors for IL-8 are present not only on neutrophils but also on Tregs, monocytes, and NK cells, indicating their potential involvement in the complex biology of CIP ( 108 ). Conversely, IL-10 is predominantly regarded as an antiinflammatory cytokine ( 109 ), but its role in autoimmune disease remains ambiguous, as illustrated by the failure of inducing an autoimmune syndrome in IL-10–deficient mice ( 109 ). While one report suggested elevated IL-6 levels beyond baseline in CIP ( 110 ), a separate study of BALF cytokines in 12 patients diagnosed with CIP demonstrated significantly elevated IL-6 levels compared with those in individuals in the control group ( 111 ). However, it should be noted that IL-6 is not universally elevated in CIP BALF ( 68 ). Yet, tocilizumab, an IL-6 inhibitor, has been shown to be effective in treatment of steroid refractory CIP in a single-center experience report ( 112 ). In a separate analysis of serum and BALF of 13 patients with CIP after PD-1/PD-L1 therapy for NSCLC, elevation in both IL-17A and IL-35 was observed in both compartments ( 113 ). Furthermore, serum IL-17A levels were found to positively correlate with the Th17 cellular subtype. IL-17A has been implicated in other autoimmune disorders ( 114 ), acute lung injury ( 115 ), and lung fibrosis ( 116 ), which implies that it may also contribute to the pathogenesis of CIP. A summary of cytokines that have been shown to be deranged in CIP are outlined in Table 3 . Although the underlying tumor histology, host factors, and disease profiles vary among patients, there is a growing body of evidence to suggest that cytokine dysregulation plays a role in the pathogenesis of pneumonitis. While a distinct cytokine signature has yet to be identified due to these differences, this may hint at the presence of multiple pathways at play in pneumonitis. Genetic predisposition. As the development of irAEs are widely believed to be associated with autoimmunity, genetic variations have been investigated as a potential contributing factor. Various genes with single nucleotide polymorphisms have been linked to different irAEs ( 117 – 119 ), indicating a complex interplay of multiple pathways. Of particular interest are the human leukocyte antigen (HLA) variations, as they are critical in the immune cell interface. In a cohort of 256 patients receiving ICI treatment, including 29 cases of CIP, HLA typing demonstrated a strong correlation between CIP rates and germline expression of HLA-B allele 35 and HLA-DRB1 allele 11 ( 120 ). These genes are also associated with other autoimmune disorders ( 118 , 121 , 122 ), highlighting the possible role of genetic factors in the pathogenesis of CIP. In more recent work, investigators looked at the T cell receptor β variable (TRBV) in the peripheral blood leukocytes of 81 individuals with different malignancies ( 123 ). Interestingly, they uncovered a certain TRBV allele haplotype that either reduced or increased the risk of severe (≥grade 3) irAEs. While TRBV polymorphism has been linked to autoimmune diseases ( 124 ), its association with irAEs has not been demonstrated previously ( 125 ). IL-7 is a critical cytokine for lymphocyte homeostasis, and it has been shown to regulate the number of circulating T cells in humans ( 126 ). In a genome-wide association study of 1,751 patients on ICIs across multiple cancer types, several significant single nucleotide polymorphisms near IL-7 were identified that associated with ICI toxicity in general ( 127 ). These germline variants of IL-7 demonstrated higher lymphocyte stability after ICI initiation that consequently increased the risk of irAEs ( 127 ). Although not yet fully understood, these findings suggest that genetic variations likely contribute to the dysregulation leading to CIP. The microbiome. Extensive work has investigated the gut microbiome where certain flora dictate both response and ICI-induced colitis rates ( 128 ). Abundance of Bacteroidetes phylum has been shown to be protective against the development of ICI-induced colitis in a cohort ( 129 ), whereas another group demonstrated that Faecalibacterium -enriched gut microbiota was associated with more frequent ICI-induced colitis ( 130 ). The increase in colitis risk in these patients was mediated by the upregulation of T cell response through higher inducible T cell costimulator induction thought to be mediated by increased circulating IL-2 after immunotherapy. A similar process may be occurring in the lungs, but the bulk of the microbiome studies related to ICI response have focused on the gut, and much of the lung microbiome work has focused on the risk of lung cancer development ( 131 ), highlighting this as an area in need of further investigation. Potential targets for future research It is worth noting that the absence of a clear link between CIP and specific biological characteristics could be due to limitations in research studies or disease-related factors. The incidence of CIP may be too low in some cancers to identify any meaningful associations. Additionally, the misclassification of pulmonary disease as CIP could confound results. Furthermore, the immune response is complex, and multiple mechanisms may be present across different patient populations. An important variable that may lead to distinct pathobiological pathways in CIP may be the underlying tumor histology. For instance, patients with NSCLC have a higher risk of CIP than other malignancies ( 58 , 59 ). This phenomenon may be dictated by tumor immunobiology that is often disparate. To gain a better understanding of the pathogenesis of CIP, several potential avenues can be explored. First, it is necessary to understand how checkpoint blockade affects the immune landscape and function within the lung in the absence of any pathological processes. Therefore, conducting studies to investigate the effect of ICI alone on immune cell subsets would be instructive. The recent interest in use of ICIs in earlier stage cancers and nonmalignant disease presents an opportunity to study the effect of ICI in conditions in which the tumor burden is either very low or absent ( 132 , 133 ). In addition, interrogating the lung microbiome could provide crucial data about potential disruptions that might contribute to CIP. An effective approach to gain further insight into the pathogenesis of CIP would be to develop an animal model of ICI pneumonitis. Several models have been attempted, but a robust prototype is still lacking. Gao et al. used humanized mice treated with collagen-specific antibodies, followed by immune checkpoint blockade drugs leading to development of arthritis and pneumonitis ( 134 ). These mice showed increased inflammation in lung tissue and elevated levels of TNF + CD4- and CD8-infiltrating T cells in the lungs and peripheral blood. Another murine model used dual checkpoint (PD-1 and CTLA-4) blockade, causing lung inflammation with systemic T cell activation, suggesting that immunotherapy-mediated peripheral activation of T cells may be the initial immune derangement leading to CIP ( 135 ). However, the model lacked specificity due to concerns of high-dose checkpoint blockade, multiorgan irAE involvement, and genitourinary developmental abnormalities in the mice used. Nonetheless, a more stringent CIP model remains important, as it could enable the evaluation of temporal changes in the clinical course of the disease that are difficult to capture in real-world cases. This would also allow a more detailed and thorough analysis of biological changes that are difficult to yield in human studies. There is another limitation that hampers progress in this field, which pertains to our incomplete understanding of the in vivo pharmacokinetics of ICIs and their true half-life. When examining patients with ICI-induced myocarditis, it was observed that levels of ICI drugs remained elevated for several months ( 136 ). Moreover, despite receiving plasmapheresis and steroid treatment, these patients experienced prolonged suppression of PD-1 detection on T cells, as measured by flow cytometry. Additionally, significant variation was noted in the duration of ICI effects among different patients, further complicating our comprehension of the intricate interactions between ICIs and lymphocytes ( 136 ). Further research of in vivo ICI pharmacokinetics is therefore warranted. In summary, checkpoint inhibitor therapy has revolutionized the landscape of cancer treatment. However, its use has been limited by related immune toxicities, especially CIP. The pathogenesis of CIP is marked by an elevated T cell–mediated immune activation, disruption in cytokine balance, autoantibody upregulation, and underlying genetic predispositions, all contributing to lung inflammation. Certain T cell subsets appear to have greater prominence within the disease context, potentially playing pivotal roles in CIP. Despite these observations, many questions regarding the pathophysiology of CIP remain unanswered. Hence, it is imperative to undertake additional research centered on comprehending the immune landscape and establishing a robust disease model. This approach is essential to enhance our insight into CIP and formulate efficacious treatment strategies.

01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e170503

oa_package/24/44/PMC10786690.tar.gz

PMC10786691

37943610

Introduction Nascent membrane or secretory proteins are synthesized and folded in the endoplasmic reticulum (ER), which is prone to misfolding. Such misfolding may have pathogenic consequences if not cleared effectively ( 1 - 5 ). The suppressor of lin-12-like–HMG-CoA reductase degradation 1 (SEL1L-HRD1) complex represents one of the most conserved quality-control mechanisms in the cell, known as ER-associated degradation (ERAD) ( 6 – 8 ). In SEL1L-HRD1 ERAD, misfolded proteins are recognized and recruited to the SEL1L-HRD1 protein complex via ER chaperones such as osterosarcoma amplified 9 (OS9) and ER lectin 1 (ERLEC1, also known as XTP3B) ( 9 – 14 ) followed by retrotranslocation and polyubiquitination by the E3 ligase HRD1 ( 12 , 15 , 16 ) and proteasome degradation in the cytosol ( 17 , 18 ). In this complex, SEL1L is an obligatory cofactor for the E3 ligase HRD1 ( 19 , 20 ), not only controlling the protein stability of HRD1 ( 6 , 19 , 20 ), but functioning as a scaffold for other ERAD components such as OS9, ERLEC1, and degradation in ER (DERLIN) proteins ( 10 , 19 , 21 – 27 ). In yeast, SEL1L homolog Hrd3p may regulate Hrd1p autoubiquitination and self-degradation ( 28 , 29 ). Global or acute deletion of Sel1L or Hrd1 in germline and adult mice causes embryonic or premature lethality, respectively ( 20 , 30 – 32 ), pointing to the requirement of SEL1L-HRD1 ERAD function at both embryonic developmental and adult stages. Subsequent studies using cell type–specific gene KO mouse models have established its vital importance in many physiological processes, including food intake, water balance, thermogenesis, energy homeostasis, gut homeostasis, β cell identity/function, immune cell development/function, and hematopoietic cell quiescence ( 2 , 3 , 5 , 33 – 51 ). Although these advances in mouse models have provided critical insights into the physiological importance of this complex, its relevance in humans remains unknown, as no disease variant has been identified in humans. Using whole-exome sequencing (WES), here we report the identification of 3 autosomal recessive variants, SEL1L p.Gly585Asp, p.Met528Arg, and HRD1 p.Pro398Leu, in 6 children from 3 unrelated families with similar neurodevelopmental disorders — termed ERAD-associated neurodevelopmental disorders with onset in infancy (ENDI). These variants are hypomorphic and attenuate ERAD function, likely via distinct mechanisms, including substrate recruitment, SEL1L-HRD1 complex formation, and HRD1 activity. Hence, this study establishes the pathophysiological importance of SEL1L-HRD1 ERAD in humans.

Methods Human subjects. Six patients from 3 families were identified and included in the study. The patient cases were gathered through the web-based tool GeneMatcher ( 83 ) ( https://genematcher.org/statistics/ ). The Saudi Arabian boy was born in 10/2009 to a gravida 2, para 1, abortion 0 29-year-old healthy mother and a 29-year-old father following an uneventful full-term pregnancy and spontaneous vaginal delivery. He presented with global developmental delay, intellectual disability, and hypotonia. MRI at 4 months of age suggested nonspecific periventricular white matter signal. The patient was officially diagnosed with short stature at 5 years of age and has shown limited response to growth-hormone therapy. The patient was diagnosed with cataract at the age of 6 and was treated with lens extraction and intraocular lens implant placement. The patient was diagnosed with hypothyroidism and has been on 25 mcg of l-thyroxine because of elevated thyroid-stimulating hormone (TSH). Other than hypotonia, neurological examination was largely normal. The last doctor visit was in May 2022. Four Moroccan siblings, 1 female (born in 06/2005) and 3 males (born in 12/2007, 02/2011, and 10/2017) displayed developmental delay, intellectual disability, speech delay, short stature, seizures, and ataxic gait (progressive with age). Brain MRIs of patient 2 (Moroccan family proband), aged 14 years, showed small cavities in the frontal periventricular area with nonspecific ventricular dilatation on coronal T2, coronal FLAIR, and axial T2 weighted images and thin corpus callosum with no anomalies of basal ganglia or at the infratentorial level of sagittal plane. Blood tests suggested vitamin D deficiency and an infection at the time of tests on 06/2022. Family history was notable for parents being first cousins. No similarly affected relatives were found in the family. The last doctor visit was in 06/2022. The Italian girl (born in 11/2001) presented with intellectual disability, speech delay, hypotonia, severe drug-resistant seizures, stereotypies, and dysmorphic features. The patient showed no autism spectrum disorder traits. The last doctor visit date was in June 2022. CRISPR/Cas9-based KO and KI HEK293T cells. HEK293T cells, obtained from ATCC, were cultured at 37°C with 5% CO 2 in DMEM with 10% fetal bovine serum (Fisher Scientific). To generate SEL1L- or HRD1-KO HEK293T cells, sgRNA oligonucleotides designed for human SEL1L (5′-GGCTGAACAGGGCTATGAAG-3′) and human HRD1 (5′-GGACAAAGGCCTGGATGTAC-3′) were inserted into lentiCRISPR, version 2 (Addgene, 52961). Cells grown in 10 cm petri dishes were transfected with indicated plasmids using 5 μl of 1 mg/ml polyethylenimine (PEI) (MilliporeSigma) per 1 μg of plasmids for HEK293T cells. The cells were cultured 24 hours after transfection in medium containing 2 μg/ml puromycin for 24 hours and then in normal growth medium. SEL1L M528R , SEL1L G585D , and HRD1 P398L KI HEK293T cells were generated using the CRISPR/Cas9 Homology-Directed Repair (HDR) system (Integrated DNA Technologies [IDT]); 5 μL of 100 μM Alt-R crRNA (IDT) with gRNA sequence was mixed with 5 μL of 100 μM Alt-R tracrRNA (IDT) containing the Cas9 interacting sequence. To anneal the oligos, the duplex mixture was heated at 95°C for 5 minutes and then cooled at room temperature for 20 minutes, and 9 μL of the guide complex was incubated with 6 μL of the 62 μM Alt-R Cas9 enzyme (IDT) at room temperature for 20 minutes; 5 μL of the ribonucleoprotein (RNP) complex, together with 1.2 μL of the 100 μM HDR donor oligo (IDT) and 1.2 μL of the 100 μM Alt-R Cas9 electroporation enhancer (IDT), was added into the 100 μL HEK293T cell suspension (about 5 × 10 5 cells) in electroporation solution (Ingenio). The mixture was transferred into a 0.2 cm cuvette, and electroporation was performed using Lonza Nucleofector IIb (Lonza). To prepare cell culture media, 3.4 μL pf 0.69 mM Alt-R HDR Enhancer V2 (IDT) was added to 2,000 μL DMEM with 10% fetal bovine serum (Fisher Scientific). After electroporation, cell suspension was added to the cell culture media, and the mixture was incubated in 4 wells of a 24-well plate (500 μL per well). The cells were cultured at 37°C with 5% CO 2 . After 5 days of incubation, the genomic DNA of the cell culture was extracted with 50 mM NaOH. DNA fragments covering the target sites were amplified by PCR using HotStart Taq 2× PCR Master Mix (ABclonal) and analyzed by Sanger Sequencing (Eurofins Genomics US) to estimate the percentage of mutant allele in the cell pool. In parallel, cells were diluted into 8 cells per mL and cultured in 96-well plates (100 μL per well) for single-cell isolation. After 10 days, 100 single-cell colonies were transferred into 24-well plates. The SEL1L M528 region of each colony was amplified by a 50 μL PCR reaction, and 25 μL of the PCR product was treated with endonuclease NsiI (NEB) in rCutSmart Buffer (NEB), incubated at 37°C overnight. PCR products that were resistant to NsiI digestion were further analyzed by Sanger sequencing. The SEL1L G585D and the HRD1 P398L regions were amplified using a 25 μL PCR reaction and sequenced. Cell colonies with homozygous SEL1L M528R , SEL1L G585D , or HRD1 P398L alleles were transferred into a 6-well plate for further experiments. Sequences were as follows: crRNA (guide sequence): SEL1L M528R : 5′-CTAGCTCAGATGCATGCCAG-3′, SEL1L G585D : 5′-TACCTCCTCCTGGCTGAACA-3′, HRD1 P398L : 5′-CACAGCCTCTCCTGAGCTGG-3′; HDR donor oligo (mutation sites are underlined): SEL1L M528R : 5′-AATTTAGCTTCTCAGGGAGGCCATATCTTGGCTTTCTATAACCTAGCTCAGA G GCATGCCAGTGGCACCGGCGTGATGCGATCATGTCACACTGCAGTGGAG-3′, SEL1L G585D : 5′-GGCGATTACAATGCTGCAGTGATCCAGTACCTCCTCCTGGCTGAACAGG A CTATGAAGTGGCACAAAGCAATGCAGCCTTTATTCTTGATCAGAGTAAGG-3′, HRD1 P398L : 5′-TGGCCCCCCATGGGCCCCTTTCCACCTGTCCCGCCTCCCC T CAGCTCAGGAGAGGCTGTGGCTCCTCCATCCACCAGTGCAGG-3′; amplification PCR primers: SEL1L M528R : F: 5′-AATCTGTATCAGTGTGTTAGCTTGTATTA-3′, R: 5′-AGACTTTCCTGCTGGGCAA-3′; SEL1L G585D : F: 5′-AAACCTGTTGACTTCTAAAGAGTAAGTGAAAACTT-3′, R: 5′-AATGTCAAATCCATTTCTACAGTCAACTCG-3′; HRD1 P398L : F: 5′-CAGTCAGTGTGACCAGTGCT-3′, R: 5′-CTCACCCCCAAGAAGAACCC-3′; and sequencing primers: SEL1L M528R : 5′-CTTACAGATGGCATTGGAGTCAAGAGA-3′, SEL1L G585D : 5′-CCCACCTCACACAGTTGTTTAAGAATGT-3′, HRD1 P398L : 5′-CCTCCGTCTTCTCTCTGCAG-3′. Plasmids. The following plasmids were used in the study (h denotes human genes; m denotes mouse genes): pcDNA3-h-proAVP(G57S)-HA (described previously, ref. 47 ); mSel1L cDNA (cloned from mouse liver cDNA and inserted into the pcDNA3 to generate pcDNA3-mSEL1L[WT]-FLAG). Point mutations of SEL1L in this study were generated using site-directed mutagenesis. The SEL1L-FLAG mutants G585D and M528R were generated using the plasmid pcDNA3-mSEL1L(WT)-FLAG as the template. All plasmids were validated by DNA-Seq. The mutagenesis primers were as follows: mSEL1L-FLAG-F: 5′-CGCGGATCCACCATGCAGGTCCGCGTCAGGCTGTCG-3′, R: 5′-CGCTCTAGACTATTTATCATCATCATCTTTATAATCTCCGCCCTGTGGTGGCTGCTGCTCTGG-3′. G585D-F: 5′-TGGCTGAGCAGGACTACGAGGTGGC-3′, R: 5′-GCCACCTCGTAGTCCTGCTCAGCCA-3′. M528R-F: 5′-CCTCGCACAGAGGCACGCCAGCGGC-3′, and R: 5′-GCCGCTGGCGTGCCTCTGTGCGAGG-3′. hHRD1 cDNA was cloned from pcDNA3-hHRD1(WT)-Myc-His (a gift from Y. Ye, National Institute of Diabetes and Digestive and Kidney Disease, Bethesda, Maryland, USA) and inserted into the pcDNA3 to generate pcDNA3-hHRD1(WT)-FLAG. Point mutations of HRD1 in this study were also generated using site-directed mutagenesis. The HRD1-FLAG mutants P398L, C2A(C291A/C294A), P397L, and P396L were generated using the plasmid pcDNA3-hHRD1(WT)-FLAG as the template. Sequences were as follows: hHRD1-FLAG-F: 5′-GGCGGTACCATGTTCCGCACGGCAGTGATGATG-3′, R: 5′-GGCGGATCCTCATTTATCATCATCATCTTTATAATCTCCGCCGTGGGCAACAGGAGACTC-3′; P398L-F: 5′-GTCCCGCCTCCCCTCAGCTCAGGAGAG-3′, R: 5′-CTCTCCTGAGCTGAGGGGAGGCGGGAC-3′; P397L-F: 5′-CCTGTCCCGCCTCTCCCCAGCTCAGGAG-3′, R: 5′-CTCCTGAGCTGGGGAGAGGCGGGACAGG-3′; P396L-F: 5′-CCACCTGTCCCGCTTCCCCCCAGCTC-3′, R: 5′-GAGCTGGGGGGAAGCGGGACAGGTGG-3′; C2A-F: 5′-ATGGACAATGTCGCCATCATCGCCCGAGAAGAGATG-3′, R: 5′-CATCTCTTCTCGGgcGATGATGgcGACATTGTCCAT-3′. Western blot and antibodies. Cells were harvested and snap-frozen in liquid nitrogen. The proteins were extracted by sonication in NP-40 lysis buffer (50 mM Tris-HCl at pH 7.5, 150 mM NaCl, 1% NP-40, 1 mM EDTA) with protease inhibitor (MilliporeSigma), DTT (MilliporeSigma, 1 mM), and phosphatase inhibitor cocktail (MilliporeSigma). Lysates were incubated on ice for 30 minutes and centrifuged at 16,000 g for 10 minutes. Supernatants were collected and analyzed for protein concentration using the Bio-Rad Protein Assay Dye (Bio-Rad); 20–50 μg of protein was denatured at 95°C for 5 minutes in 5× SDS sample buffer (250 mM Tris-HCl pH 6.8, 10% sodium dodecyl sulfate, 0.05% bromophenol blue, 50% glycerol, and 1.44 M β-mercaptoethanol). Protein was separated using SDS-PAGE or Phos-tag gel (as described previously, refs. 84 , 85 ), followed by electrophoretic transfer to PVDF (Fisher Scientific) membrane. The blots were incubated in 2% BSA/TBST with primary antibodies overnight at 4°C: anti-HSP90 (Santa Cruz Biotechnology Inc., sc-13119, 1:5,000), anti-SEL1L (home-made, ref. 33 ; 1:10,000), anti-HRD1 (Proteintech, 13473-1, 1:2,000), anti-OS9 (Abcam, ab109510, 1:5,000), anti-CD147 (Proteintech, 11989-1, 1:3,000), anti-IRE1α (Cell Signaling Technology, 3294, 1:2,000), anti-ERLEC1 (Abcam, ab181166, 1:5,000), anti-UBE2J1 (Santa Cruz Biotechnology Inc., sc-377002, 1:3,000), anti-DER2 (gift from Chih-Chi Andrew Hu, Houston Methodist Hospital, Houston, Texas, USA, ref. 86 , 1 :1,000), anti-VCP (Proteintech, 10736-1, 1:3000), anti-FAM8A1 (Proteintech, 24746-1, 1:3000), anti-FLAG (MilliporeSigma, F1804, 1:1,000), anti-HA (MilliporeSigma, H3663, 1:5,000), anti-PERK (Cell Signaling Technology, 3192, 1:5000), anti-eIF2α (Cell Signaling Technology, 9722, 1:5000), anti–p-eIF2α (Cell Signaling Technology, 9721, 1:1,000), anti-GRP78 BiP (Abcam, ab21685, 1:5000), and anti-PDI (Enzo Life Sciences, ADI-SPA-890-D, 1:5000). Membranes were washed with TBST and incubated with secondary antibodies, either HRP conjugated (Bio-Rad, 1:10,000), anti-rabbit IgG TrueBlot HRP (Rockland, 18-8816-33, 1:500), or anti-mouse IgG TrueBlot-HRP (Rockland 18-8817-31, 1:500), at room temperature for 1 hour for ECL chemiluminescence detection system (Bio-Rad) development. Band intensity was determined using Image lab (Bio-Rad) software (verison 6.1). IP. For SEL1L-FLAG and HRD1 IP, HEK293T cells transfected with the indicated plasmids or KI HEK293T cells were snap-frozen in liquid nitrogen and whole-cell lysate was prepared in the IP lysis buffer (150 mM NaCl, 0.2% Nonidet P-40 [NP40], 0.1% Triton X-100, 25 mM Tris-HCl pH 7.5) at 4°C, supplemented with protease inhibitors, protein phosphatase inhibitors, and 10 mM N -ethylmaleimide. A total of approximately 5 mg protein lysates were incubated with 15 μl anti-FLAG agarose (MilliporeSigma, A2220) or 2 μl anti-HRD1 antibody (Proteintech, 13473-1) overnight at 4°C with gentle rocking. HRD1 IP lysates were incubated with 10 μl protein A agarose (Invitrogen, 20333) at 4 o C for 2 hours after incubation. Incubated agaroses were washed 3 times with the IP lysis buffer and eluted in the SDS sample buffer at 95 o C for 5 minutes followed by SDS-PAGE and immunoblot. Denaturing IP for ubiquitination assay. HEK293T cells were transfected with proAVP(G57S)-HA plasmids for 24 hours and then treated with 10 μM MG132 for 2 hours. The cells were snap-frozen in liquid nitrogen, and whole-cell lysate was prepared in the NP-40 lysis buffer (50 mM Tris-HCl at pH7.5, 150 mM NaCl, 1% NP-40, 1 mM EDTA) with 1% SDS and 5 mM DTT, denatured at 95°C for 10 minutes, and centrifuged at 16,000 g for 10 minutes. Subsequently, supernatants were diluted 1:10 with NP-40 lysis buffer and incubated with 15 μl anti-HA agarose (Thermo Fisher,26182) overnight at 4°C with gentle rocking. The incubated agaroses were washed 3 times with the NP-40 lysis buffer and eluted in the SDS sample buffer at 95°C for 5 minutes, followed by SDS-PAGE and immunoblot. Chemical treatment. Cells were treated with 50 μg/ml cycloheximide for the indicated times followed by Western blot analysis or treated with 10 μM MG132 followed by denaturing IP. WT HEK293 cells treated with 100 nM thapsigargin for 4 hours were included as positive controls for UPR. Statistics. Statistics tests were performed using GraphPad Prism, version 8.0 (GraphPad Software). Unless indicated otherwise, values are represented as means ± SEM. All experiments were repeated at least 2 to 3 times and/or performed with multiple independent biological samples from which representative data are shown. All data sets passed normality and equal variance tests. Statistical differences between the groups were compared using unpaired 2-tailed Student’s t test for 2 groups or 1-way ANOVA or 2-way ANOVA for multiple groups. P < 0.05 was considered statistically significant. The intensities of the Western blot bands between different samples in some experiments were also statistically compared using 1-way ANOVA with post hoc Tukey-Kramer test in the R environment. The input data were first examined for homoscedasticity using the Breusch-Pagan test implemented in the ncvTest function in the R car package. In our experience, data that do not satisfy a constant variance usually display log-normal distribution. Therefore, the log-transformed data were used as input in those cases. Study approval. Study protocols and written, informed consent protocols were approved by the institutional review boards at the Research Advisory Council (RAC) (King Faisal Specialist Hospital and Research Centre, KFSHRC RAC 2080006); the APHP-Délégation Interrégionale à la Recherche Clinique (DIRC) Assistance Publique-Hôpitaux de Paris, Paris, France (2015-03-03/DC 2014–2272); the Ethical Committee of the University of Naples Federico II (48/16), the University of Michigan Medical School (IRBMED, HUM00227482), and Health Sciences Research (IRB-HSR, University of Virginia, HSR230351). The patients and/or the parents provided written, informed consent prior to participation in the study. Written, informed consent was received for the use of the photographs. Data availability. The materials and reagents used are either commercially available or are available upon request. All data and materials for the manuscript are described in Methods. Values for all data points in graphs are reported in the Supporting Data Values file.

Results Identification of biallelic SEL1L and HRD1 variants in humans. Six patients from 3 unrelated families in Saudi Arabia (patient 1), Morocco (patient 2–5), and Italy (patient 6) were suspected of inherited genetic disease during clinical visits ( Figure 1, A–C ). Patients 1, 2, 3, and 6 were subjected to WES of DNA samples ( Figure 1, D–F ). WES results were stringently filtered for novel variants by excluding variants with low sequencing quality, in the noncoding region, with high frequency in the population, or likely to be benign in silico ( Figure 1, D–F ). We failed to identify any known variants linked to inherited neurological or metabolic disorders, but rather noted 3 variants linked to the same protein complex/pathway, namely SEL1L-HRD1 ERAD: SEL1L p.Gly585Asp (NM_005065: exon 17: c.1754G>A) in the Saudi Arabian patient (patient 1), SEL1L p.Met528Arg (exon 16: c.1583T>G) in the Moroccan patients (patient 2 and 3), and HRD1 p.Pro398Leu (NM_172230: exon 12: c.1193C>T) in the Italian patient (patient 6) ( Figure 1, D–F , and Table 1 ). Indeed, homozygous SEL1L p.Met528Arg was the only variant shared between patients 2 and 3, but not found in the healthy parents. Although additional variants including homozygous, heterozygous, compound heterozygous, and de novo mutations were identified in the Saudi Arabian (patient 1) and Italian (patient 6) patients ( Figure 1, D and F , and Supplemental Table 1 ; supplemental material available online with this article; https://doi.org/10.1172/JCI170054DS1 ), SEL1L p.Gly585Asp and HRD1 p.Pro398Leu variants were considered as potential candidates based on their biological relevance as reported in mice ( 2 , 3 , 5 ) and also according to the American College of Medical Genetics (ACMG) and the Association for Molecular Pathology (AMP) 2015 guidelines for clinical interpretation of genetic variants ( 52 ). Indeed, neither of the SEL1L variants was found in public genetic variants databases, such as 1000gp3 ( https://www.internationalgenome.org/ ), ESP6500 ( https://esp.gs.washington.edu/drupal/ ), ExAC ( https://gnomad.broadinstitute.org/ ), and gnomAD ( https://gnomad.broadinstitute.org/ ), and both variants were consistently predicted to be damaging using various prediction tools, such as Combined Annotation Dependent Depletion (CADD) ( https://cadd.gs.washington.edu/ ), PolyPhen2-HVAR ( http://genetics.bwh.harvard.edu/pph2/ ), Sorting Intolerant from Tolerant (SIFT) ( https://sift.bii.a-star.edu.sg/ ), LIST-S2 ( https://list-s2.msl.ubc.ca/?session=838295C8507C38640C29677123B49248 ), Mendelian Clinically Applicable Pathogenicity (M-CAP) ( http://bejerano.stanford.edu/mcap/ ), BayesDel addAF ( https://fenglab.chpc.utah.edu/BayesDel/BayesDel.html ), DEOGEN2 ( http://babylone.3bio.ulb.ac.be/MutaFrame/ ), Functional Analysis through Hidden Markov Models (FATHMM-MKL) ( http://fathmm.biocompute.org.uk/ ), MutationAssessor ( http://mutationassessor.org/r3/ ), MutationTaster ( https://www.mutationtaster.org/ ), and PrimateAI ( https://github.com/Illumina/PrimateAI ) ( Table 1 ). Similarly, allele frequency of HRD1 p.Pro398Leu was low in the population and was predicted to be benign to damaging in various prediction databases ( Table 1 ). SEL1L and HRD1 genes are located on chromosomes 14 and 11, respectively ( Figure 1, G and H ). Using Sanger sequencing, we confirmed that the 3 variants were found to be homozygous in all patients, but heterozygous in all parents ( Figure 1, G and H ). Three siblings for patient 1 were either heterozygous or WT for the SEL1L allele ( Figure 1, G and H ). Clinical features of the patients. Patient 1, a 14-year-old boy born to healthy consanguineous Saudi Arabian parents (with 3 healthy siblings) showed hypotonia (poor sucking and floppiness) and microcephaly at 4 months of age. Three siblings, 2 WT and 1 heterozygous at the SEL1L G585 locus, were healthy. The patient presented with global developmental delay (sat at 2 years of age, walked at 5, uttered mama/dada at 6, and is not at this writing toilet trained), moderate-severe intellectual disability (limited 2-word sentences and cannot count to 3, with an IQ of 35), and central hypotonia (with brisk deep tendon reflexes, wide-based gait with hands held on the side to support his balance) ( Figure 2A , Table 1 , Supplemental Table 2 , and Supplemental Video 1 ). Physical examination performed at the age of 13 years revealed that he was underweight (–3.9 SD), and showed short stature (–3.9 SD), microcephaly (–4.2 SD), subtle facial dysmorphism (downslanting palpebral fissures and overbite), pectus excavatum ( Figure 2B ), a moderate degree of joint hyperlaxity, and shawl scrotum. The patient had a total of 3 seizures at 8 years of age with largely normal electroencephalogram (EEG). His medical history was notable for frequent airway infections, although his workup did not suggest immunodeficiency. Other than hypotonia, neurological examination was largely normal. Patients 2, 3, 4, and 5, 4 Moroccan siblings, 1 female (the proband, born 2005) and 3 males (born 2007, 2011 and 2017), born to healthy consanguineous parents ( Figure 1B ), presented since a few months of age with developmental delay, intellectual disability, speech delay, short stature, seizures, and ataxic gait (progressive with age). The 2 older siblings (patients 2 and 3) had single seizure history, and the 2 younger siblings (patients 4 and 5) showed microcephaly. The proband showed severe spastic and ataxic gait, falls, wide-based gait, pes cavus and equinus, mild dystonia, paraparesis with pyramidal signs of lower limbs, with brisk, diffused tendon reflexes and clonus, pyramidal extension of the first toe, and bilateral positive Babinski sign ( Figure 2C , Table 1 , Supplemental Table 2 , and Supplemental Video 2 ). The proband showed varus equus, scoliosis, and arched palate. The 4 patients also shared facial dysmorphism, including downslanting palpebral fissures and overbite ( Figure 2D ), and were diagnosed with unilateral maculopathy, pallor of temporal poles, and severe corneal dystrophy. MRI of the proband showed small cavities in the frontal periventricular area with nonspecific ventricular dilatation ( Supplemental Figure 1 ). EEG of the proband showed generalized discharges of polyspikes and slow waves. Patient 6, an Italian girl born to healthy nonconsanguineous parents in 2001, presented since her first months of age with hypotonia and severe drug-resistant seizures that were resolved by the age of 14 years ( Figure 1C ). She exhibited intellectual disability, speech delay, stereotypic movements, a clumsy gait ( Figure 2E , Table 1 , Supplemental Table 2 , and Supplemental Video 3 ), and dysmorphic facial features ( Figure 2F ). Physical examination at 16 years of age revealed that she was underweight (37.5 kg body weight, <5th percentile, z score = –3.45), and showed short stature (height 139.5 cm, <5th percentile, z score = –3.9), and microcephaly (head circumference 51.8 cm, <5th percentile, z score = –2.6). Brain MRI performed in the first year of life revealed a cerebellar cyst without other notable findings. When repeated at the age of 21 years, an abnormal signal was detected in the globus pallidum and substantia nigra (not shown). Cardiac and abdomen ultrasounds were both normal. The patient has been on risperidone ( Table 1 ) since the COVID-19 outbreak because of worsening of her behavior with agitation and aggression. In summary, all 6 patients were symptomatic at infancy and presented with neurodevelopmental disorder, developmental delay, intellectual disability, and facial dysmorphisms ( Table 1 and Supplemental Table 2 ). Four out of six patients had microcephaly. Two patients showed hypotonia, with floppiness, unsteady and clumsy gait, and difficulty in walking, but no frank ataxia, while the other 4 patients from the Moroccan family exhibited severe ataxia that progressed with age. Interestingly, hypotonia in the 2 patients did not progress with age, and in fact the Italian patient was no longer hypotonic at the most recent evaluation in June 2022. No notable abnormalities were observed in routine blood chemistry tests (glucose, electrolytes, blood urea nitrogen, creatinine, aspartate aminotransaminase, alanine aminotransaminase, and albumin) and complete blood counts in all 6 patients. SEL1L and HRD1 variants are hypomorphic with impaired ERAD function. To investigate whether and how these disease variants affect SEL1L-HRD1 ERAD function, we generated knockin (KI) HEK293T cells carrying the biallelic variants using the CRISPR/Cas9 system and verified by Sanger sequencing ( Supplemental Figure 2 , A–E). We then tested to determine whether these disease variants affect ERAD function by measuring protein stability and levels of known endogenous ERAD substrates such as inositol-requiring enzyme 1α (IRE1α) ( 41 ), OS9 ( 53 ), and cluster of differentiation 147 (CD147) ( 54 ) as well as the disease mutant of proarginine vasopressin (proAVP) Gly57Ser (Gly-to-Ser at residue 57) ( 47 ). Indeed, endogenous substrates became accumulated in all 3 KI HEK293T cell lines ( Figure 3, A and B ; see complete unedited blots in the supplemental material) due to protein stabilization, similar to that in SEL1L- or HRD1-KO HEK293T cells ( Figure 3, C and D , and Supplemental Figure 3 ). Similarly, proAVP (Gly57Ser) accumulated in transfected KI cells, forming much more extensive high molecular weight (HMW) aggregates than those in WT cells ( Figure 3, E and F ). A couple of points are worth noting here: first, although all variants caused substrate accumulation, the extent of substrate accumulation differed among the variants, the highest being the SEL1L M528R variant and the lowest being either the SEL1L G585D or HRD1 P398L variant, which is consistent for both endogenous and model substrates ( Figure 3, B and F ). Second, the extent of substrate accumulation and HMW aggregation in KI cells was modest compared with that in KO cells ( Figure 3, B and F ), pointing to the hypomorphic nature, rather than loss of function, of these variants. Taken together, these data suggest that 3 variants are hypomorphic with moderate to severe ERAD dysfunction. Lack of an overt unfolded protein response in KI cells. Intriguingly, we did not observe an overt unfolded protein response (UPR) in these KI cells, as demonstrated by the lack of IRE1α phosphorylation and X-box–binding protein 1 ( XBP1 ) mRNA splicing as well as phosphorylation of protein kinase R-like ER kinase (PERK) and eukaryotic initiation factor-2α (eIF2α) ( Supplemental Figure 4 , A–D). ER chaperones, such as immunoglobulin heavy chain–binding protein (BiP) and protein disulfide isomerase (PDI), were accumulated in KI HEK293T cells ( Supplemental Figure 4 , E and F). These data point to a cellular adaptive response in cells expressing these disease variants. Sequence and structural analyses of SEL1L and HRD1 variants. We next asked how these variants affect ERAD function. We first performed in silico conservation and structural analyses. These variants affect conserved residues from yeast or drosophila to humans, with the exception of HRD1 Pro398, which is absent in yeast ( Figure 4, A–C ). Position-specific scoring matrix (PSSM) analysis ( 55 ) showed that all 3 variants replaced evolutionarily selected aa and that the mutations may be detrimental to SEL1L and HRD1 function ( Figure 4, D–F ). The SEL1L variants, p.Met528Arg and p.Gly585Asp, affect residues in the Sel1-like repeat–middle (SLR-M) and the linker region between SLR-M and -C, respectively ( Figure 4G ). To visualize these variants in the ERAD complex, we performed AI-based AlphaFold2 prediction network analysis ( 56 ) to model the structure of the human SEL1L (107–723 aa)-HRD1 (1–334 aa)-OS9 (33–655 aa)-DERLIN1 (1–213 aa) protein complex ( Figure 4H ). Structural modeling of the human SEL1L-HRD1 complex showed a great similarity to the cryogenic electron microscopy (Cryo-EM) structure of the yeast Hrd1p-Hrd3p complex (PDB ID 6VJZ) ( 12 ) ( Supplemental Figure 5A ). ConSurf conservation analysis ( 57 ) of SEL1L confirmed that both Met528 and Gly585 residues were in highly conserved patches ( Supplemental Figure 5B ). SEL1L Met528 is predicted to be a part of the α-helix facing outward from the putative substrate binding groove and OS9 ( Figure 4I ). Mutation of Met528 to Arg is expected to seriously disrupt the α-helical structure and destabilize the protein. On the other hand, Gly585 is located on a loop between the 2 helices in the putative substrate-binding groove ( Figure 4J ). While mutation of Gly585 to Asp is not predicted to disrupt the α-helical structure, it is located in the substrate-binding groove in close proximity to the substrate(s) and lectins (OS9 and ERLEC1). Moreover, Pro398 of HRD1 is located in the proline-rich region (~50 Pro in a stretch of 140 aa) of its cytosolic domain, C-terminal to the really interesting new gene–finger (RING-finger) domain ( Figure 4K ). This proline-rich region is disordered based on IUPred2 prediction ( 58 ) ( Figure 4L ), with no predictable structure. SEL1L and HRD1 variants impair ERAD function via distinct mechanisms. We next explored how these variants cause ERAD defects using KI HEK293T cells. We first measured protein levels of the SEL1L-HRD1 ERAD complex. Noticeably, SEL1L M528R KI HEK293T exhibited reduced SEL1L and HRD1 protein levels, by approximately 80% and 60%, respectively ( Figure 5, A and B ), uncoupled from their gene transcription ( Figure 5, C and D ). Indeed, both SEL1L and HRD1 proteins were unstable in SEL1L M528R KI cells ( Figure 5, E and F ). In contrast, SEL1L G585D exhibited a modest reduction of SEL1L and HRD1 protein levels, by approximately 20% to 30% ( Figure 5, A and B ), without changes in mRNA levels ( Figure 5, C and D ). SEL1L G585D had a subtle effect on the stability of SEL1L protein, but not HRD1 protein, in KI HEK293T cells ( Figure 5, E and F ). This reduction in ERAD protein levels was unlikely to explain the ERAD defects associated with SEL1L G585D -expressing cells, as heterozygosity of SEL1L or HRD1 is sufficient for ERAD function ( 20 , 34 , 43 , 44 ). On the other hand, HRD1 P398L had no effect on either protein levels or stability of SEL1L and HRD1 ( Figure 5, A, B, E, and F ). Hence, the SEL1L M528R variant causes ERAD dysfunction by reducing protein stability and levels of the SEL1L-HRD1 complex, but not SEL1L G585D and HRD1 P398L . Given the location of SEL1L G585 residue, we next asked whether SEL1L G585D affects substrate recruitment. During ERAD, substrates are recruited by lectins such as OS9 and ERLEC1 to the SEL1L-HRD1 complex, which also includes ubiquitin-conjugating E2 enzyme J1 (UBE2J1) and DERLIN proteins ( 18 ). To circumvent the confounding issue of reduced SEL1L and HRD1 protein levels in SEL1L KI cells, we used an overexpression system in SEL1L –/– HEK293T cells. Surprisingly, the SEL1L G585D variant significantly reduced its interactions with 2 lectin proteins (ERLEC1 and OS9), by approximately 70% to 80%, and with HRD1 by approximately 50% compared with that of WT SEL1L ( Figure 6, A and B ). In contrast, SEL1L interaction with UBE2J1 and DERLIN2 was not affected in SEL1L G585D -expressing cells ( Figure 6, A and B ). In contrast, SEL1L M528R did not affect the interaction between SEL1L and HRD1 or other ERAD components in transfected SEL1L –/– HEK293T cells ( Figure 6, A and B ). Hence, unlike SEL1L M528R , the SEL1L G585D variant impairs ERAD function by attenuating substrate recruitment. This conclusion is in line with the prediction that SEL1L G585 faces the substrate-binding groove and is in close proximity to OS9 ( Figure 4J ). In HRD1 P398L KI HEK293T cells, the interactions of HRD1 with other ERAD components, such as SEL1L, UBE2J1, DER2, valosin-containing protein (p97/VCP), and family with sequence similarity 8 member A1 (FAM8A1), were comparable to those in WT HEK293T cells ( Figure 7, A and B ). However, substrate ubiquitination was significantly attenuated in HRD1 P398L KI HEK293T cells, similar to that in the other 2 SEL1L variants ( Figure 7, C and D ), providing further support for ERAD dysfunction. Given that HRD1 P398L is close to the RING domain ( Figure 4K ), we next asked whether HRD1 P398L may affect HRD1 activity by modulating its ubiquitination using denaturing immunoprecipitation (IP) followed by Western blot. For unknown reasons, we failed to detect ubiquitination of endogenous HRD1 proteins in WT and KI cells even with the treatment of MG132 ( Supplemental Figure 6 ). Upon transfection in HRD1 –/– HEK293T cells, P398L mutation attenuated HRD1 ubiquitination compared with those in WT cells (lanes 3 and 9 versus 2 and 8, Figure 7, E and F ). The effect of P398L on HRD1 ubiquitination was similar to that of HRD1 C2A mutation (lanes 4 and 10), a mutation in the RING domain known to abolish HRD1 E3 activity ( 8 ). Interestingly, mutation of the neighboring HRD1 Pro to Leu (P397L or P396L) had a much milder effect on HRD1 ubiquitination (lanes 5–6 and 11–12, Figure 7, E and F ). MG132 treatment enhanced HRD1 ubiquitination in all samples (lanes 8–12 versus 2–6, Figure 7, E and F ), suggesting that HRD1 ubiquitination may contribute to its turnover. These data suggest that HRD1 P398L affects HRD1 ubiquitination, which may contribute to its dysfunction. Taking these data together, we conclude that these 3 variants cause ERAD dysfunction at distinct steps of ERAD, including substrate recruitment ( SEL1L G585D ), SEL1L-HRD1 protein stability and complex formation ( SEL1L M528R ), and HRD1 activity ( HRD1 P398L ).

Discussion This study reports 3 variants in SEL1L and HRD1 genes in 6 patients from 3 unrelated families. These patients manifest similar clinical features, including developmental delay, microcephaly, intellectual disability, facial dysmorphism, hypotonia, and ataxia. Using KI HEK293T cells expressing individual variants, we further show that these variants impair ERAD function at distinct steps of ERAD, including substrate recruitment, SEL1L-HRD1 protein stability and complex formation, and HRD1 activity ( Figure 8 ). We speculate that the phenotypic variations among these patients may reflect different levels of ERAD dysfunction associated with the variants and/or less likely, possible effects of other non-ERAD variants. A few additional variants were identified from the Saudi Arabian (patient 1) and Italian (patient 6) patients ( Supplemental Table 1 ). Most of the heterozygous and compound heterozygous variants were predicted to be benign by the pLI score (the intolerance of the gene to loss of function) and variant effect prediction tools (CADD, PolyPhen-2 HVAR, SIFT), except for the heterozygous Furry-like (FRY-like) transcription coactivator ( FRYL) variant ( FRLY c.7490C>G, p.T2497R) identified in patient 1; however, Fryl heterozygous mutant mice were found to be normal compared with WT littermates, while homozygous mutant mice showed lower birth rate and renal defects (hydronephrosis) if they survived ( 59 ), suggesting that the FRYL variant may not be disease relevant in patient 1. Similarly, although several additional homozygous variants were identified, the reported functions of these proteins are not biologically relevant to the symptoms observed in our patients, e.g., Ras-associated protein rab17 ( RAB17), which encodes a GTPase to enable GDP-binding activity ( 60 ), von Willebrand factor A containing 5B2 domain ( VWA5BA ), which belongs to the family of von Willebrand factors crucial for primary platelet and collagen adhesion function ( 61 ), and Solute carrier family 25, member 53 ( SLC25A53 ), which is predicted to be an integral component of the mitochondrial inner membrane (Alliance of Genome Resources, https://www.alliancegenome.org/ ) with unknown function. Among them, although mutations of other RAB family proteins have been linked to neurological disorders ( 62 ), RAB17 is an epithelial cell–specific GTPase ( 63 ) and is expressed at a very low level in the central nervous system (GTExPortal, gtexportal.org). Similarly, 2 other variants identified in patient 6, membrane-spanning 4-domains, subfamily a, member 12 ( MS4A12 ) and protein phosphatase 1 regulatory subunit 32 ( PPP1R32 ), are associated with colon cancer ( 64 , 65 ) and ciliary movements ( 66 ), respectively. Given the (patho-)physiological importance of SEL1L-HRD1 ERAD ( 2 , 3 , 5 , 67 ) and given that the M528R variant is the only variant shared among the affected siblings from the Moroccan family, we believe that SEL1L-HRD1 ERAD variants are most likely to be disease causing in these patients. Comparing disease-variant KI HEK293T cells to ERAD KO HEK293T cells, our biochemical analyses showed that these variants attenuated ERAD function. Since all the parents and some healthy siblings were heterozygous for the variant, we propose that all these variants cause more than a 50% reduction in ERAD function. Further comparisons among the 3 variants showed that SEL1L M528R may be the most severe one. This may account for the differences in clinical features between patient 2 to 5 with the SEL1L M528R variant (ataxia and microcephaly) and the other 2 patients, 1 and 6, with the SEL1L G585D and HRD1 P398L variants (hypotonia). While the underlying molecular mechanisms are distinct for these variants in causing ERAD dysfunction, they all invariably cause ERAD dysfunction, leading to the stabilization and accumulation of endogenous ERAD substrates. Hence, these studies suggest that there is a threshold requirement for SEL1L-HRD1 ERAD function essential for normal neuronal function in humans. We reported that these human SEL1L-HRD1 variants compromised ERAD via distinct mechanisms. Specifically, in HRD1 P398L KI HEK293T cells, HRD1 ERAD function was impaired. Following overexpression in HRD1-deficient HEK293T cells, we found that the HRD1 P398L variant impaired HRD1 ubiquitination. While this finding is potentially interesting, as it may reflect HRD1 autoubiquitination as reported by Baldridge et al. ( 28 ), we are aware that overexpression of HRD1 likely alters the stoichiometric ratios of the ERAD components that do not accurately reflect those of the endogenous HRD1 complex. The “ubiquitinated HRD1” result from (partially) unassembled and misfolded HRD1 that are targeted for proteasome-dependent degradation. Studies are underway to explore whether HRD1 P398L affects autoubiquitination of the RING domain specifically related to channel gating or other lysine residues. In the accompanying paper ( 68 ), we reported an additional 5 patients carrying another SEL1L variant ( SEL1L p.Cys141Tyr) identified from a Slovakian Romani family. This group of patients exhibited not only similar neurological disorders, but severe agammaglobulinemia resulting from the lack of B cells. This difference in clinical manifestation is likely due to the fact that SEL1L p.Cys141Tyr is the most severe variant among the four. Moreover, a SEL1L mutation ( p.Ser658Pro) was previously identified in Finnish hounds with cerebellar ataxia (also known as cerebellar ataxia Finnish hound type [CAFH]) ( 69 ), further suggesting that SEL1L may play an important role in maintaining normal neurological function or neurodevelopment. These findings provide strong experimental support for the notion that hypomorphic SEL1L-HRD1 variants are pathogenic in humans. With that said, how these variants are linked to neurological defects in these patients remains to be investigated and is of great interest going forward. Although this lacks substantial evidence in humans, we speculate that SEL1L-HRD1 ERAD variants cause disease via substrate-dependent and cell-type–specific manners, as none is associated with an overt UPR. Other mechanisms, such as organellar dysfunction, may also contribute to this pathological process. This study reports what we believe is the first set of human patients carrying variants in the core components of a key protein degradative machinery, providing key evidence for its pathophysiological importance in humans. It is worth noting that several variants have been identified in p97/VCP, another key component of the ERAD machinery involved in protein retrotranslocation from the ER. However, unlike SEL1L-HRD1 ERAD variants, these p97/VCP variants cause multisystem disorders ( 70 – 72 ). Differences in clinical features between these 2 sets of patients are likely due to the fact that, in addition to ERAD, p97/VCP is involved in a wide variety of other cellular functions, including genomic stability, translational stress response, and RNA biology ( 71 ). Moreover, a number of variants in genes involved in protein glycosylation have been reported to cause congenital disorders of glycosylation (CDG), manifestations of which also largely include neurodevelopmental delay and variable facial dysmorphism ( 73 – 78 ) — clinical consequences similar to those of our ENDI patients described in this study. This similarity is not surprising, as glycosylation is intimately associated with ER protein folding, maturation, and degradation ( 79 ). However, CDG patients also exhibit multisystemic symptoms, including hypoglycemia and liver, skin, gastrointestinal, and coagulation abnormalities ( 75 ), which were not observed in ENDI patients. Hence, identifying Mendelian disorders caused by mutations in core ERAD components is essential in delineating the importance of ERAD in humans. ENDI is a rare neurodevelopmental disorder associated with SEL1L-HRD1 ERAD and characterized by infantile-onset developmental delay, intellectual disability, microcephaly, facial dysmorphisms, hypotonia, and/or ataxia. Intellectual disability affects about 1% to 3% of the population ( 80 , 81 ), while ataxia has an estimated overall prevalence of 26 in 100,000 in children ( 82 ). Our data suggest that evaluating SEL1L-HRD1 ERAD has diagnostic values for those with intellectual disability, developmental delay, and ataxia. While it is currently rare, we expect that more SEL1L-HRD1 ERAD variants will surface as evidence grows for its importance in humans. Options for treating patients with ENDI are currently very limited, but this study provides a framework for our future effort to target this important ERAD complex.

Authorship note: HHW, LLL, and ZJL contributed equally to this work. Recent studies using cell type–specific knockout mouse models have improved our understanding of the pathophysiological relevance of suppressor of lin-12-like–HMG-CoA reductase degradation 1 (SEL1L-HRD1) endoplasmic reticulum–associated (ER-associated) degradation (ERAD); however, its importance in humans remains unclear, as no disease variant has been identified. Here, we report the identification of 3 biallelic missense variants of SEL1L and HRD1 (or SYVN1 ) in 6 children from 3 independent families presenting with developmental delay, intellectual disability, microcephaly, facial dysmorphisms, hypotonia, and/or ataxia. These SEL1L (p.Gly585Asp, p.Met528Arg) and HRD1 (p.Pro398Leu) variants were hypomorphic and impaired ERAD function at distinct steps of ERAD, including substrate recruitment (SEL1L p.Gly585Asp), SEL1L-HRD1 complex formation (SEL1L p.Met528Arg), and HRD1 activity (HRD1 p.Pro398Leu). Our study not only provides insights into the structure-function relationship of SEL1L-HRD1 ERAD, but also establishes the importance of SEL1L-HRD1 ERAD in humans. ER-associated degradation-associated neurodevelopmental disorder with onset in infancy is associated with hypomorphic variants of SEL1L and HRD1.

Author contributions HHW and LLL designed and performed most experiments. ZJL generated all KI cells. XW performed structural analysis. OA, GC, MOH, LH, AM, MA, FSA, CB, KP, and NBP performed exome sequencing WES analysis, identified variants, and acquired clinical data. QP, SS, and MB provided insightful discussions. YL provided assistance on statistical analysis. SS and LQ directed the study, designed experiments, and wrote the manuscript with help from HHW, LLL, and ZJL. HHW, LLL, and ZJL wrote the Methods and figure legends. All authors commented on and approved the manuscript. Supplementary Material

We are deeply grateful to all patients and their families for their consent and willingness to participate in this study; Chih-Chi Andrew Hu (Houston Methodist Hospital) for reagents; members of the Qi, Arvan, and Pletcher laboratories; and Jiwon Hwang and Ryan Baldridge at the University of Michigan Medical School for technical assistance and insightful discussions and constructive comments on the manuscript. This work was supported by R01DK128077, R01DK132068 (to SS), the Telethon Foundation, the Telethon Undiagnosed Diseases Program GSP15001 (to NBP), R01DK120330, R35GM130292, and the Michigan Protein Folding Disease Initiative (to LQ). The authors extend their appreciation to European Reference Network ITHACA (to NBP). LLL and ZJL are supported in part by National Ataxia Foundation post- and predoctoral fellowships (NAF 918037 and 1036307). XW was/is supported in part by Pandemic Research Recovery Grant U078128 at the University of Michigan Medical School and American Society of Nephrology Postdoctoral Fellowship. 11/09/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e170054

oa_package/5f/4c/PMC10786691.tar.gz

PMC10786692

37934611

Introduction ATR (ataxia telangiectasia and Rad3-related) is a critical kinase in the DNA damage response (DDR) ( 1 , 2 ). Preclinical data have identified multiple cancer-related phenotypes sensitizing tumor cells to monotherapy ATR inhibition (ATRi) ( 3 ). Additionally, ATRi potentiates DNA-damaging therapies, including chemotherapy, radiotherapy ( 4 ), and targeted therapies such as poly ADP-ribose polymerase (PARP) inhibitors ( 5 ), making it a promising combination partner. Emerging evidence suggests ATRi may also modulate antitumor immune responses ( 6 – 8 ). ATR is activated by diverse DNA lesions causing exposure of expanses of single-stranded DNA ( 2 ). This replication stress is a frequent consequence of oncogene activation and impaired G1 checkpoint control and can be secondary to exogenous and endogenous sources of DNA damage and repair. Activated ATR phosphorylates targets, including checkpoint kinase 1 (Chk1), leading to stabilization of replication forks, activation of DNA repair, and activation of cell-cycle checkpoints ( Figure 1B ). Hence, monotherapy ATRi is predicted to affect tumors with high levels of replication stress, reduced DNA repair, or nonfunctional cell-cycle checkpoints, leading to accumulation of DNA damage and cell death. In preclinical models, ATRi kills tumor cells with loss of ataxia telangiectasia mutated (ATM) ( 9 ), AT-rich interactive domain-containing protein 1A (ARID1A) ( 10 ), and specific components of the DDR pathway ( 11 – 13 ) or those driven by oncogenes such as cyclin E and Myc ( 14 – 16 ). Emerging data suggest that increasing the DNA damage load in cells could promote an antitumor immune response, for example, through interferogenic nucleic acid–sensing pathways ( 17 ). Ceralasertib (AZD6738, AstraZeneca) ( 18 ) is a potent, selective, orally bioavailable, ATP-competitive ATR inhibitor, with antitumor activity demonstrated in multiple preclinical models ( 19 ). We report the results of the p hase 1 study of ATR i nhibition al o ne or with radiation t herapy (PATRIOT) study ( 20 ), a first-in-human dose-finding study that determined safety, tolerability, recommended dose and schedule, pharmacokinetics (PKs), and antitumor activity of ceralasertib monotherapy and explored potential predictive biomarkers of response to ATRi.

Methods Patient population. Patients were 18 years and over, with advanced solid malignancy, without standard anticancer treatment options. All had Eastern Cooperative Oncology Group (ECOG) ( https://ecog-acrin.org/resources/ecog-performance-status/ ) performance status 0–1, life expectancy of at least 3 months, and adequate organ function. Key inclusion criteria are provided in the Supplemental Methods . Study design. This was a multipart, multicenter, open-label phase I study. Part A comprised a dose escalation and part B a dose expansion. During dose expansion, participants were selected based on the presence or absence of putative biomarkers of response to ATRi. Part C (combination with radiotherapy) will be reported separately. Patients in this study started ceralasertib between July 2014 and October 2020. The data cutoff was in October 2022, when 4 patients were still on study medication, all for at least 2 years. The primary objective was to determine the safety and feasibility of administration of ceralasertib monotherapy in patients with advanced solid tumors. The secondary objectives were to identify a dose and schedule for further studies of ceralasertib and to assess antitumor responses and PK. Exploratory objectives included PD studies in tumor and normal tissue and the potential value of putative markers of sensitivity to single-agent ATRi, including measures of immune activation. Study treatments. Ceralasertib was administered orally, twice daily. For part A (dose escalation; Figure 1, A and B ), the starting dose of 20 mg was selected based on animal toxicity studies. Dosing was continuous, and escalation used a modified Fibonacci method. Initial dose escalation was planned in single patient cohorts, changing to 3+3 design after the first grade-2 toxicity was seen. This occurred in the first patient. Cohorts of 3 to 6 patients were assessed for toxicity during a DLT window of 28 days (1 cycle), with a nontolerable dose defined as 2 or more of 6 patients experiencing a DLT. DLT definitions are in the Supplemental Methods . For part B, all patients received the RP2D as defined for part A. Part B allowed for different schedules (continuous/intermittent) to be assessed. Initially, a continuous dosing schedule was used for the first 6 patients. Subsequently, the safety review committee authorized the assessment of an intermittent schedule, 14 days on and 14 days off. Pretreatment biopsy was mandatory in part B. DNA-Seq of archival tumor material or review of external tumor sequencing was used to enrich for patients with putative genomic markers of sensitivity to ATRi, based on preclinical data, including oncogene amplification or driver mutation, ATM/G1 pathway defects (alteration in ATM, CHK2, or other components of the pathway causing G1 cell-cycle arrest after DNA damage), SWI/SNF (switch/sucrose nonfermentable, a chromatin-remodeling complex) pathway defect, genomic instability/homologous recombination deficit (HRD), or defect in a gene synthetically lethal with ATRi in published data ( 46 , 47 ) ( Figure 1B and Supplemental Table 3 ). Study assessments. Patients were assessed weekly during cycle 1 and twice weekly thereafter, with safety assessments including blood hematology and biochemistry, physical examination, and toxicity scoring. Safety and tolerability were assessed using Common Terminology Criteria for Adverse Events (CTCAE), version 4.03. Participants had ECG and urinalysis at the start of each cycle of treatment and assessment of left ventricular ejection fraction every 8 weeks. Response assessment imaging was conducted according to RECIST 1.1 within 28 days of starting ceralasertib and every 8 weeks. PKs. Intensive PK sampling in part A occurred after a single dose from predose up to 24 to 72 hours and again at day 15 and day 29 of continuous dosing. Participants fasted for 1 hour before and 2 hours after dosing for PK assessment. In part B, sampling coincided with day 15 PD assessments. Full details are given in Supplemental Methods . PDs and histology. PD sampling took place at baseline (within 7 days prior to dosing) and between days 15 and 22 of dosing (day 14 for intermittent dosing cohorts). PD samples included PBMCs and tumor biopsies. PBMCs were analyzed by immunofluorescence for γ (S139) H2AX, p- (S345) Chk1, and total Chk1, as described in Supplemental Methods . Paired tumor biopsies were formalin fixed and analyzed for nuclear p- (S635) Rad50 and γH2AX by IHC. Translational methods. DNA-Seq of tumor and matched buffy coats was either by whole-exome sequencing or a custom-designed panel targeting all exons of genes of interest for 173 genes, including potential markers of sensitivity to ATRi. ACK-lysed whole blood or PBMCs were stained for flow cytometry using 8 multicolor panels. Plasma cytokines were assessed using the Bio-Plex Pro 27-Plex Panel (Bio-Rad). See Supplemental Methods for full information. Statistics. Simple descriptive statistical data analysis methods were used to summarize the data. Categorical data used numbers and percentages of patients in the categories/groups; where appropriate, 95% CIs were reported. Continuous nonnormally distributed data as assessed by visual inspection were described using median, IQR, and minimum and maximum values. Statistical analyses were conducted using STATA, version 17.0. Additional genomic and laboratory data were plotted and analyzed using Prism 8 (GraphPad), and ggplot2 in R version 4. For comparison of biomarkers at baseline and on treatment, paired t tests (2 tailed) or their nonparametric equivalents were used. When comparing fold-change data normalized to baseline, Wilcoxon’s signed rank test with a hypothetical median of 1 (no change from baseline) was used unless otherwise stated. Study approval. This study was conducted in accordance with protocol requirements, good clinical practice (GCP), and the Declaration of Helsinki. All participants provided written, informed consent. The protocol was approved by the local ethics committee (NRES Committee London–City and East, reference 14/LO/0465). Data availability. Deidentified individual participant data that underlie the figures in this article will be made available to researchers who provide a methodologically sound proposal and complete the data access agreement. Tumor profiling, flow cytometry, and clinical annotations can be provided. PK and PD data cannot be provided. Values for all data points in graphs are reported in the Supporting Data Values file.

Results Patient characteristics A total of 26 patients were enrolled and started ceralasertib in the dose-escalation phase across 3 centers between July 2014 and July 2016. In the dose-expansion phase, 43 patients were enrolled, of whom 41 received at least 1 dose of study drug (2 progressed prior to treatment start) between December 2016 and October 2020 ( Figure 1A ). Patient and tumor characteristics are given in Table 1 . Dose escalation and toxicity A total of 67 patients received a dose of ceralasertib and were evaluable for safety ( Figure 1A ). Twenty-six patients were treated with continuous dosing schedule during the dose-escalation phase, at doses from 20 to 240 mg BD ( Figure 1C ). At the maximum administered dose of 240 mg BD, 3 of 6 patients had dose-limiting toxicities (DLTs). There were no DLTs at 160 mg BD and 1 at 80 mg BD (grade 3 [G3] thrombocytopenia with epistaxis, Table 2 and Supplemental Table 2 ; supplemental material available online with this article; https://doi.org/10.1172/JCI175369DS1 ). The maximum tolerated dose was 160 mg BD. DLTs were thrombocytopenia (G4, n = 2 at 240 mg, G3 with epistaxis, n = 1 at 80 mg) and elevated amylase (G3, n = 1 at 240 mg, Supplemental Table 2 ). Dose-expansion participants received 160 mg BD, either continuously or on a 2-week-on, 2-week-off schedule ( Figure 1C ). This was investigated after the development of toxicity beyond the DLT window in continuously dosed patients, leading to dose modifications. The intermittent schedule was chosen based on modeling of bone marrow recovery and was better tolerated, with incidence of G3 or greater anemia, 33% on continuous versus 9% on intermittent schedule, and 8% versus 0 % G3 leukopenia. Platelets and other hematological parameters were also more favorable with an intermittent schedule, recovering in the treatment break ( Figure 2A , Table 2 , and Supplemental Figure 1 ) ( 21 ). Six of 12 (50%) patients on the continuous schedule (including those in part A) versus 10 of 35 (29%) on the intermittent required dose reduction or interruption for toxicity. Four patients in the dose-escalation and 1 in the dose-expansion phase withdrew due to toxicity. There were no treatment-related deaths. Four deaths occurred on study medication: 2 from disease progression, 1 from pneumonia, and 1 from adult respiratory syndrome assumed to be COVID-19 related (no leukopenia observed for the latter 2 participants). The recommended phase 2 dose (RP2D) for the intermittent schedule was 160 mg BD, although other doses were not evaluated on an intermittent schedule. Serious adverse events related to study treatment are shown in Supplemental Table 1 . PKs Ceralasertib was rapidly orally absorbed across all doses following single and multiple dose administration (median time to peak drug concentration [t max ] 0.5 to 4 hours), with mean terminal elimination half-life of 5.3 to 7.7 hours at the 40 and 80 mg dose levels and 11.2 to 12.8 hours at the 160 and 240 mg dose levels. Following single dosing, ceralasertib exposure increased approximately proportionally with increasing doses between 80 to 240 mg ( Figure 2B ). There was some evidence for accumulation after repeated dosing with higher predose and maximum concentration (C max ) levels at days 15 and 29 compared with day 0. Accumulation ratios based on C max and AUC were between 1.6- and 2.2-fold higher ( Supplemental Figure 2 ). Pharmacodynamics Paired PBMCs were available for the majority of study participants. PBMCs were analyzed for p-Chk1, the downstream phosphorylation target of ATR. There was variation in p-Chk1 levels with treatment, but this was not consistent ( Supplemental Figure 3 ). p-Chk1 has been described as decreasing with ATRi in the presence of exogenous DNA damage ( 4 ) and as increasing with ATRi reflective of replication stress and DNA damage ( 22 ). Increased γH2AX positivity was observed in PBMCs after treatment at the RP2D for most subjects ( Figure 2C ), likely reflecting DNA damage in proliferating bone marrow cells due to ATRi. Four paired tumor biopsies were available for IHC. These tumor biopsies showed upregulation of p-Rad50, a marker of ATM pathway activation, after treatment with ceralasertib ( Figure 2, D and E ), as well as an increase in the number of γH2AX-positive cells ( Figure 2, F and G ). Response At data cutoff, 4 patients remained on study; all had received a minimum of 24 cycles. Sixty-six patients were evaluable for response assessment, 26 in the dose-escalation and 40 in the dose-expansion phases. The best overall responses were 5 (8%) confirmed partial responses (PR), 34 (52%) stable disease (SD), including 1 unconfirmed PR, and 27 (41%) progressive disease, including clinical progression ( Figure 3, A–C ). Of those with SD or better, 25 of 39 (68%) had duration on study of at least 4 months, with many showing a slowing of tumor growth ( Supplemental Figure 4 ). For those taking 160 mg BD or more, 4 of 49 (8%) had PR, 30 (61%) SD, and 15 (30%) progressive disease. Patients with Response Evaluation Criteria in Solid Tumors (RECIST) ( 23 ) PR were dosed at 40, 240 (continuous schedule), and 160 (intermittent schedule) mg BD. Median duration of response was 46.7 weeks (IQR, 14.9–251.0). Responding histologies were as follows: (a) ovarian clear cell carcinoma with ARID1A mutation and high mutational load (160 mg BD, remains on study, 251 weeks at data cutoff; Figure 3D , Supplemental Table 4 ), (b) head and neck squamous cell carcinoma (HNSCC) with CDKN2A and MRE11A frameshift (160 mg BD, 170 weeks, remains on study; Figure 3E ), (c) esophageal squamous cell carcinoma with homologous recombination (HR)/Fanconi pathway deficiency due to BRIP1 frameshift mutation and PALB2 deletion (160 mg BD, 47 weeks; Figure 3F ), moderate mutational load (12.4 mutations/Mb), and APOBEC mutational signature ( 24 – 26 ); (d) nasopharyngeal carcinoma with NRAS activating mutation (240 mg BD, 14 weeks; Figure 3G ), and (e) HNSCC with APC frameshift and TP53 mutation (40 mg BD, 15 weeks; Figure 3H ). One participant had an unconfirmed PR; this patient had TP53 mutant pancreatic adenocarcinoma with no other mutation (160 mg BD, 15 weeks; Figure 3I ). Patients with durable RECIST SD included those with HNSCC with ARID2 frameshift (99 weeks), HNSCC with no sequencing available (48 weeks), HNSCC with CCND1 amplification (49 weeks), and digital papillary adenocarcinoma with TP53 mutation (51 weeks). Genomic and molecular correlates Sequencing data were available for 5 of 26 patients in the dose-escalation and 36 of 41 in the dose-expansion phases. Patients with durable responses all had an alteration that may sensitize to ATRi ( Supplemental Table 3 and Supplemental Figures 5 and 6 ). For patients dosed at 40 mg BD or more, out of 11 patients with no mutation of interest, 1 had a PR (9%) and out of 30 with a mutation of interest, 4 had a PR (13%). Of those with PR or SD, median duration of response was 105 days for those without a mutation of interest and 185.5 days for those with a mutation of interest. Unless otherwise stated, participants were taking 160 mg BD of intermittent ceralasertib. Durable responses in tumors with SWI/SNF loss The most durable response was in a patient with clear cell ovarian carcinoma and an ARID1A mutation ( E21763fsX ) with loss of protein expression ( Figure 3, D and J ). Seven participants had aberrations in the SWI/SNF pathway, of whom 6 derived clinical benefit. One other patient had a clear ARID1A loss on IHC: a patient with eccrine adenocarcinoma with ARID1A stop-gain mutation ( R693X , resulting in truncated protein expression) and H score of 0 ( Figure 3K ) with CDKN2A deletion (240 mg BD; SD, 34 weeks). A patient with an ARID2 frameshift-bearing HNSCC had tumor shrinkage of 29% and remains on study at this writing after 99 weeks. Other SWI/SNF aberrations are described in Supplemental Table 4 . Notably, all other ARID1A mutants showed high protein expression ( Figure 3, L–N ), and the responding patient also had a high tumor mutational burden (TMB); there was no clear difference in TMB between patients with or without clinical benefit ( Supplemental Figure 7 ). ATM pathway There was no relationship between ATM expression and response or duration on study ( Figure 3A ). Twenty patients had ATM protein assessed: 4 were defined as ATM-low, with 25% or less ATM nuclear positivity (10%, 10%, 5%, and 0%; Figure 3O ). Out of these, median duration on treatment was 13 weeks (range 8–29) with 3 of 4 experiencing SD and 1 progressive disease. One patient had a pathogenic ATM mutation ( R1898fsX ) with some protein expression (50% nuclear positive) and a coexisting ARID1A mutation (see above), remaining on study for 39 weeks with SD. Another had MRE11 stop-gain mutation (R633X), together with CDKN2A stop-gain, and remains on study at this writing after more than 32 months with a confirmed PR ( Figure 3E ). MRE11, a component of the MRN complex, activates ATM after DNA damage. Other aberrations Oncogene amplification. We identified 11 patients with oncogene-driven tumors (5 NRAS, 2 HRAS, 1 KRAS activation, 2 CCNE1 amplification, 1 CCND1 amplification), of whom 3 derived clinical benefit ( Supplemental Figure 5 ). Of those with CCNE1 amplification, 1 (peritoneal carcinoma, 20 mg BD; Figure 3Q ) had a best response of progressive disease, 1 (serous endometrial carcinoma; Figure 3R ) SD, on study for 12 weeks, and 2 others had increased cyclin E1 expression by IHC without gene amplification: 1 with serous endometrial carcinoma ( Figure 3S , with germline BRCA1 mutation) and the other with cervical adenocarcinoma ( Figure 3P ), both with SD for 16 and 29 weeks, respectively. P53. We have previously demonstrated no relationship between p53 functionality and ATRi sensitivity in a panel of cell lines ( 4 ). This was confirmed by the lack of difference in clinical benefit and duration on study between p53 WT and p53-mutant/p53-deleted tumors ( Supplemental Figure 6 ). ATRi modulates the tumor-immune microenvironment We have previously shown preclinically that ATRi can affect the immune tumor microenvironment (TME), particularly when combined with radiotherapy ( 6 , 27 ). In support of this, paired biopsies from a responding patient (40 mg BD, HNSCC, RECIST PR) showed an increase in immune-cell infiltration and programmed death-ligand 1 (PD-L1) staining on immune cells at 2 weeks ( Figure 4A ). Therefore, we profiled, in detail, the immune response to ATRi in the peripheral blood of 8 participants (best responses of 5 SD, 2 progressive disease, and 1 nonevaluable [NE], all treated with 160 mg BD intermittent schedule) and in paired tumor biopsies (on treatment versus baseline) from 8 participants (5 SD, 2 progressive disease, and 1 NE). In the peripheral blood, we observed a reduction in Tregs and a trend toward increased CD8 + T cells, leading to an increased CD8/Treg ratio after ATRi ( Figure 4B ). All were on an intermittent schedule, allowing assessment of changes after 2 weeks of ceralasertib (day 14) and a 2-week break (day 29). Proportions of T cell subsets changed after ATRi, with increased naive and central memory CD8 + and CD4 + T cells after ATRi ( Figure 4C ). Importantly, there were increased frequencies of memory CD4-TEMRA (effector memory reexpressing CD45RA) cells at day 29 ( Figure 4D ). Detailed profiling revealed a reduction in PD-1–positive CD8 + T cells and NK cell activation, with a trend toward increased NKG2A- and CD69-positive NK cells with ceralasertib, which normalized after the 2-week break ( Figure 4E ). The circulating myeloid compartment was also altered by ATRi, with a reduction in classical and intermediate monocytes and a change in circulating myeloid-derived suppressor cells (MDSCs), with increased granulocytic MDSCs (gMDSC) and reduced monocytic MDSCs (mMDSCs) after ceralasertib, again trending to baseline after treatment break ( Figure 4F ). Circulating cytokine levels were modulated on ceralasertib therapy, with an increase in CCL2 and decrease in CCL4 and CCL5 levels observed after 2 weeks of treatment ( Figure 4G ). Responders to ATRi have inflamed tumors RNA-Seq of paired tumor biopsies was performed to assess differential gene expression after 2 weeks of ceralasertib treatment. Eight paired tumor biopsies were analyzed from 3 patients with PR, 4 with SD, and 1 NE, treated at various dose levels. Additional baseline samples were also available for 1 PR and 1 SD. When all samples were considered together, there were few differences in differential gene expression between baseline and on-treatment biopsies ( Figure 5A ). However, when responders (PR) were compared with nonresponders (SD), there were marked differences in both baseline and on-treatment gene expression ( Figure 5, B–E ) with clustering of a number of differentially expressed genes according to response ( Supplemental Figure 8 ). The most common genes that were differentially expressed were immune related, with adaptive, innate, and cytokine-related genes highly represented ( Figure 5E ). Pathway analysis of the most differentially expressed genes found that these were predominantly immune related ( Supplemental Figure 9 ). Gene-set enrichment analysis (GSEA) revealed enrichment of inflammatory response genes between baseline and on-treatment samples. When responders were compared with nonresponders, responding patients had more inflamed tumors both at baseline and on treatment, with significantly higher transcript levels for multiple immune-related genes ( Figure 5F ). Expression of cell-type–specific genes was different between responders and nonresponders. Responders had a significantly higher expression of PTPRC (CD45) at baseline and on treatment than nonresponders; they also had an increase in ITGAX (CD11c) with treatment. Several other genes showed similar elevation in responders compared with nonresponders, but this did not reach statistical significance ( Figure 5G ). When gene expression data were used for cell-type deconvolution, some differences were observed with treatment, particularly in neutrophil and macrophage populations ( Supplemental Figure 10 ). Baseline expression of macrophage, antigen-processing, and cytokine-related genes was generally higher in responding tumors ( Figure 6, A–C ), with clustering by response. T and NK cell signatures were increased in responders ( Supplemental Figure 11 , A and B). When on-treatment biopsies were analyzed, there was clustering of responders in cytotoxicity ( Figure 6D ) as well as cytokine and T cell signatures ( Supplemental Figure 11 , C and D). When plotted together, baseline and on-treatment samples tended to cluster by patient rather than by treatment, indicating a strong effect of baseline tumor inflammation on response. However, interferon-stimulated genes did appear to be upregulated in both baseline and on-treatment biopsies in responders ( Supplemental Figure 12 ). We counted stromal tumor-infiltrating lymphocytes (TILs) in H&E-stained sections at baseline and for 4 paired samples ( Figure 6E ). Those patients who derived clinical benefit from ceralasertib, defined as PR or greater than 16 weeks on study, had a trend to higher numbers of TILs than those who did not ( Figure 6F ). Stromal TILs appeared to increase in a responding patient, but not in 3 nonresponders ( Figure 6, E and G ).

Discussion Our study is the largest to date, to our knowledge, of ATRi monotherapy. We have shown that ceralasertib monotherapy is tolerable, with predominantly hematological toxicities reduced by an intermittent schedule. Ceralasertib has predictable PK and plasma levels at the RP2D that compare favorably with observed preclinical monotherapy IC 50 values (ATR IC 90 of 0.666 μM and GI 50 of approximately 1 μM, comparable to between 270–420 ng/mL; refs. 4 , 22 ). We have shown target modulation in tumor tissue and increased DNA damage in surrogate tissues. We found durable clinical benefit in diverse tumor types, with evidence suggesting multiple potential biomarkers of response to ATRi, including loss of ARID1A, genomic instability, ATM/G1 pathway abnormalities, and high tumor inflammation. Conversely, we did not find clear signals that oncogene drivers sensitize to ATRi. Other published studies of ATRi have demonstrated similar, predominantly hematological, toxicities ( 28 ). ATRi has been combined with carboplatin ( 28 ) and with paclitaxel ( 29 ) in early phase studies. The only previously published ATRi monotherapy study of BAY1895344, to our knowledge, also found durable responses in DDR-defective tumors — 4 of 11 patients with ATM protein loss or deleterious mutation and 1 with BRCA1 mutation had prolonged SD ( 30 ). We identified several patients with ATM loss, all without objective responses. Responses were associated with other factors involved in G1 cell-cycle checkpoint control, including MRE11 : loss will result in defective ATM activation, and a previous Chk1 inhibitor study observed a durable response associated with loss of another component of this complex ( 31 ). Alternative ATR inhibitors are administered intravenously, and the duration of enzyme inhibition may differ between these 2 modes of administration. This may result in differential effects on efficacy and immunomodulation. As well as convenience, oral administration with an intermittent schedule allows bone marrow recovery between dosing periods and may allow more effective tailoring of dose exposure. The introduction of a modified schedule after emergence of toxicity outside the DLT window highlights a limitation of the 3+3 study design, and alternative designs may have been able to integrate such toxicities into dose-escalation decisions. Differences in efficacy between continuous, lower-dose and intermittent, higher-dose regimens should be examined in future studies. ARID1A is a critical component of the SWI/SNF chromatin remodeling complex, which modulates the accessibility of DNA to transcription and repair machinery and is frequently mutated in cancers ( 32 ). ARID1A is important for ATR activation after double-stranded DNA breaks ( 33 ); it helps topoisomerase-II prevent DNA tangling during mitosis (decatenation) ( 34 ). Without this activity, cells activate a G2/M decatenation checkpoint ( 35 ). This is abolished by ATRi, leading to massive DNA damage ( 10 ). There are 2 main protein complexes in the SWI/SNF family: ARID1A, a critical component of the cBAF complex, and PBRM1 and ARID2, which are components of the PBAF complex ( 36 ). Only ARID1A loss has been described preclinically as an ATRi sensitizer ( 10 ), but distinct functions of different SWI/SNF complexes are unclear ( 36 ). ARID1A loss is particularly common in ovarian clear cell and uterine carcinomas ( 37 ). Of 2 patients with protein loss, 1 responded and 1 had SD, suggesting other factors may also be involved. Durable responses have been reported in patients with ARID1A loss in ongoing clinical studies ( 38 , 39 ). Other components, such as ARID2, may also be associated with clinical benefit, as suggested by the durable SD in a participant with ARID2 loss. Intriguingly, ceralasertib responders in this study had more inflamed tumors at baseline. We saw ATRi-induced changes in the immune TME. Prior studies have found that ATRi can cause marked modulation of the TME, thought to be secondary to increased DNA damage and activation of cytoplasmic DNA-sensing machinery ( 6 – 8 ). Here, we have confirmed that treatment with ceralasertib modulates the immune response, with a more favorable CD8/Treg ratio, activation of NK cells, increased frequencies of effector memory RA CD4 + T cells, and modulation of cytokines and circulating MDSCs as well as increases in TILs and inflammatory gene expression in responding patients. A recently published combination study of ceralasertib and immune-checkpoint blockade (ICB) with durvalumab in advanced gastric cancer found a benefit in those patients with ATM loss or HR deficiency and found that responders had changes in their immune TME ( 40 ). However, the specific contribution of ATRi cannot be determined from those data. Our study of ATRi monotherapy allows an opportunity to observe the immunomodulatory effects of these agents without immunotherapies and provides the first data, to our knowledge, showing that ATRi (and other DDR inhibitors) may modulate the immune TME in their own right. The possibility that inflamed tumors may respond better to ATRi suggests that (a) these tumors have preexisting DDR defects that make them more inflamed and more likely to respond to ATRi ( 41 ), with the inflammation being a phenomenon independent of the response to ATRi; and/or (b) there is modulation of antitumor immunity by the administration of ATRi. Although we noted that there may be increased TILs in the tumors of patients who benefitted from ATRi, the difference was modest and more substantial changes were seen between responders and nonresponders on the gene expression level. Notably, ATRi seem to increase responses to ICB in patients who have previously failed ICB alone ( 42 ), adding further weight to our hypothesis that ATRis have independent immunomodulatory effects. We suggest immune analyses in ongoing ATRi studies focusing on both baseline immune status and changes with therapy to uncover rational immunotherapy partners for ATRi. In particular, the effect we have observed on NK and myeloid cells should be further investigated, particularly in light of preclinical data suggesting NK cells may have a role in ATRi responses ( 27 ). The results from this study provide the first evidence, to our knowledge, that ceralasertib monotherapy is tolerable, with antitumor activity in a number of genetic backgrounds. We have recommended a 160 mg BD 2-week-on, 2-week-off dosing schedule for further evaluation. Phase I–III studies are proceeding as monotherapy or in combination with poly (ADP-ribose) polymerase (PARP) inhibitors in advanced solid tumors (ClinicalTrials.gov NCT02264678), ATM - or ARID1A -mutant tumors ( 43 ) (NCT03682289), DDR-deficient tumors (NCT03462342), and in combination with ICB ( 44 , 45 ) (NCT02664935, NCT05061134, NCT05450692). Tumor inflammation, ARID1A loss, and genome instability are among the most promising areas for future study.

Authorship note: MDF and KJH are co–senior authors. BACKGROUND P hase 1 study of ATR i nhibition al o ne or with radiation t herapy (PATRIOT) was a first-in-human phase I study of the oral ATR (ataxia telangiectasia and Rad3-related) inhibitor ceralasertib (AZD6738) in advanced solid tumors. METHODS The primary objective was safety. Secondary objectives included assessment of antitumor responses and pharmacokinetic (PK) and pharmacodynamic (PD) studies. Sixty-seven patients received 20–240 mg ceralasertib BD continuously or intermittently (14 of a 28-day cycle). RESULTS Intermittent dosing was better tolerated than continuous, which was associated with dose-limiting hematological toxicity. The recommended phase 2 dose of ceralasertib was 160 mg twice daily for 2 weeks in a 4-weekly cycle. Modulation of target and increased DNA damage were identified in tumor and surrogate PD. There were 5 (8%) confirmed partial responses (PRs) (40–240 mg BD), 34 (52%) stable disease (SD), including 1 unconfirmed PR, and 27 (41%) progressive disease. Durable responses were seen in tumors with loss of AT-rich interactive domain-containing protein 1A (ARID1A) and DNA damage–response defects. Treatment-modulated tumor and systemic immune markers and responding tumors were more immune inflamed than nonresponding. CONCLUSION Ceralasertib monotherapy was tolerated at 160 mg BD intermittently and associated with antitumor activity. TRIAL REGISTRATION Clinicaltrials.gov: NCT02223923, EudraCT: 2013-003994-84. FUNDING Cancer Research UK, AstraZeneca, UK Department of Health (National Institute for Health Research), Rosetrees Trust, Experimental Cancer Medicine Centre. The PATRIOT in-human study shows the ATR inhibitor Ceralasertib is tolerated and associates with anti-tumor activity.

Author contributions Study conceptualization, design and protocol writing were by MTD, SAS, MDF and KJH. Study safety review was by KJH, UB, MPS, JS, and MDF. JG was the study manager and KM the statistician. PP and KES were responsible for genomic and PD assessments. IHC staining and analysis were supported by GNJ, SEW, KT, LM, PM, IR, and AW. PK analysis was performed by MP and CS. Immune assays and analysis were performed by ECP, MM, and DM. Authors contributing through participant support and site-level investigation were MTD, JG, ECP, GNJ, SEW, MP, CS, KT, IR, PN, AW, MM, AJL, SB, GN, VK, LG, MPS, PP, PM, LM, JS, MDF, and KJH. The original draft of the manuscript was written by MTD, KM, and KJH, and it was reviewed and edited by MTD, JG, KM, ECP, SAS, ED, GNJ, MP, CS, CB, PN, AW, PP, KES, UB, MPS, JS, MDF and KJH. All authors approved the final manuscript. Supplementary Material

We would like to thank The Breast Cancer Now histopathology core facility and the Institute of Cancer Research Genomics Facility. We acknowledge Magnus Hallin and Martine Roudier for pathology support and assistance with scoring IHC slides. We would like to thank Michael Hubank, Clinical Genomics at The Royal Marsden NHS Foundation Trust, for assistance with panel design and validation and the Institute of Cancer Research genomics facility and Anton Patrikeev and Ritika Chauhan for bioinformatic analysis. This study was cosponsored by The Royal Marsden and The Institute of Cancer Research. Financial and drug support were provided by AstraZeneca and Cancer Research UK through the CRUK Combinations Alliance. The authors acknowledge additional financial support from the UK Department of Health and Cancer Research UK via Experimental Cancer Medicine Centre and National Institute for Health and Care Research (NIHR) Biomedical Research Centre grants to Institute of Cancer Research/Royal Marsden Hospital, King’s Health Partners/Guy’s and St. Thomas’ NHS Foundation Trust, and the University College London/UCL Hospital NHS Trust. The authors acknowledge support from the ICR/RM CRUK RadNet Centre of Excellence and the ICR Centre for Translational Immunotherapy (CTI). This project represents independent research supported by the NIHR. The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care. Funding was received from Cancer Research UK C7224/A23275, CRUKD/14/007 (to KJH); Cancer Research UK C347/A18077, C309/A25144, CTRQQR-2021\100009 (to KES); Cancer Research UK (to MTD); AstraZeneca; UK Department of Health (NIHR) NIHR202438 (to KJH); UK Department of Health (NIHR) (to MTD and MDF); Rosetrees Trust (to KJH and MTD); CRIS Cancer Foundation (to PN); and the Experimental Cancer Medicine Centre. 11/07/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e175369

oa_package/b1/60/PMC10786692.tar.gz

PMC10786693

0

Introduction Childhood interstitial lung disease (chILD) is a heterogenous group of genetic lung disorders affecting infants and children who typically present with a range of symptoms and signs such as respiratory distress, hypoxemia, and diffuse pulmonary infiltrates ( 1 – 4 ). While many causal genes have been identified for chILD ( 5 – 8 ), the most common are autosomal recessive mutations in the gene encoding ATP-binding cassette A3 (ABCA3) ( 1 , 3 , 9 ), a lamellar body–associated (LB-associated) phospholipid transporter that is expressed in the lung within alveolar type II epithelial cells (AEC2) ( 10 – 12 ). Bronchoalveolar lavage samples from patients carrying biallelic ABCA3 mutations have decreased or altered composition of pulmonary surfactant ( 13 ), a mixture of approximately 90% phospholipids and 10% proteins that is secreted by AEC2s ( 14 , 15 ) and reduces surface tension in the alveoli of air breathing organisms. While deficiency in the functionality of ABCA3 lipid transport results in altered phospholipid composition of surfactant and may contribute to disease pathogenesis, ABCA3 mutation-mediated disruption of AEC2 phenotypes by other cellular pathways has not been systematically evaluated in humans. To date, progress in understanding chILD pathogenesis has been impeded both by, first, an inability to easily access primary human AEC2s from affected children and, second, limited methods to maintain AEC2s from explanted tissue of affected children. Additionally, compound heterozygous mutant ABCA3 genotypes with novel or rare (minor allele frequency under 1%) variants in affected children have further complicated attempts to understand mutant-specific disease pathogenesis ( 16 – 18 ). Due to these challenges, prior studies have relied on genetically engineered ABCA3 overexpression models in cancer or immortalized cell lines, such as the human embryonic kidney 293 cell line (HEK293) or the lung adenocarcinoma–derived A549 cell line ( 18 – 27 ), neither of which express surfactant genes, produce functional surfactant, or express the NKX2-1 –driven gene regulatory network that defines all known endogenous lung epithelia in vivo ( 28 – 30 ). Studies utilizing these heterologous cell lines have classified ABCA3 mutations into 2 broad subtypes: type 1 mutations resulting in mistrafficking of the mutant ABCA3 protein, or type 2 mutations resulting in decreased ATPase-mediated phospholipid transport ( 19 , 21 , 23 ). However, cell lines engineered for constitutive overexpression of ABCA3 do not recapitulate normal regulation of the endogenous ABCA3-encoding locus, potentially undermining the accuracy of these findings. Methods to establish patient-specific induced pluripotent stem cells (iPSCs), advances in gene-editing technologies, and significant progress in directed differentiation to generate iPSC-derived AEC2s (iAEC2s) — cells that transcriptomically and functionally resemble primary AEC2s ( 31 – 33 ) — have recently enabled human AEC2 disease modeling ( 34 , 35 ). These edited iPSCs have facilitated functional characterization of the consequences of surfactant processing defects caused by SFTPB and SFTPC mutations ( 32 , 36 ), other causes of chILD. Here, we apply these techniques to study the downstream effects of ABCA3 mutations in chILD patient-specific iAEC2s as well as in iAEC2s derived from iPSCs engineered to carry identical homozygous mutations by gene editing. Because most patients carry compound heterozygous ABCA3 mutations, making the study of individual mutations difficult, we focused on identifying infants and children carrying homozygous ABCA3 type 1 and type 2 mutations along with documented chILD clinical phenotypes. By differentiating these ABCA3 mutant patient iPSC-lines and engineered knock-in iPSCs into iAEC2s in parallel with their gene-corrected syngeneic control lines, we found that patient-specific iAEC2s recapitulate clinically observed surfactant dysfunction through attenuated secretion of surfactant phospholipids and altered lamellar body morphology and function. Unexpectedly, they also exhibit upregulation of a variety of signaling pathways, such as NFκB, in affected iAEC2s, regardless of ABCA3 mutation type, implying an epithelial-intrinsic aberrant phenotype that results in the secretion of proinflammatory cytokines.

Methods Patient iPSC derivation and maintenance. All human iPSCs were maintained in feeder-free conditions, cultured on Matrigel-coated (Corning) plates in mTeSR media (StemCell Technologies), and passaged using Gentle Cell Dissociation Reagent (StemCell Technologies). Reprogramming of the BU3 human iPSC line was previously reported in Kurmann et al. ( 65 ), and editing of this line to target an ABCA3:GFP fusion cassette to the endogenous ABCA3 locus (BU3 ABCA3:GFP ) was previously reported in Sun et al. ( 39 ). For derivation of ABCA3 mutant patient-specific iPSC lines, dermal fibroblasts from each individual were received from Washington University School of Medicine. Genetic evaluation found no mutations in other genes associated with surfactant production, such as SFTPC or SFTPB genes. Reprogramming of dermal fibroblasts from individuals with homozygous E690K or homozygous W308R ABAC3 mutations was performed as detailed in the Supplemental Methods . Gene editing of human iPSC lines. E690K and W308R ABCA3 mutant patient-derived iPSC lines were monoallelically targeted with a tdTomato reporter at the ATG start site of the endogenous SFTPC locus using TALENS gene editing tools following the same methods detailed in Jacob et al. ( 32 ). Subsequent biallelic, footprint-free ABCA3 gene correction of patient iPSC lines by CRISPR/Cas9 gene editing is detailed in the Supplemental Methods . To introduce ABCA3 mutations and the ABCA3:GFP reporter by knockin to normal BU3 iPSCs, we used CRISPR/Cas9 single nucleotide E690K and W308R mutagenesis of the WT BU3 ABCA3:GFP iPSC line, the same guide RNAs used for gene-correction of patient-specific iPSC lines were used in conjunction with new ssODN donors to introduce the mutated sequences, as detailed in the Supplemental Methods . For the L101P mutagenesis of BU3 ABCA3:GFP iPSC line, additional gRNA and mutant carrying donor template sequences were employed, as detailed in the supplement. Directed differentiation and maintenance of iAEC2s. Directed differentiation of iAEC2s were performed as detailed in our previously published protocol ( 32 , 33 ). In brief, day 0 PSCs were differentiated into definitive endoderm (day 0–3) using StemDiff Endoderm Kit (Stem Cell Technologies), followed by anterior foregut endoderm (day 3–6) using DS/SB media (2 μM dorsomorphin, Stemgent; 10 μM SB431543, Biotechne), then further specified into NKX2-1+ lung epithelial progenitors using CBRa media (3 μM CHIR99021, Biotechne; 10 ng/mL rhBMP4, BioTechne; 100 nM retinoic acid, Sigma-Aldrich; day 6–15). On day 15, NKX2-1 expressing lung epithelial progenitors were sorted either by NKX2-1 GFP (BU3-NGST line) or using CD47 hi /CD26 lo sorting to enrich for NKX2-1+ cells ( 41 ). Sorted cells were plated in 3D Matrigel cultures and fed with distalizing CK+DCI medium (3 μM CHIR99021 [Tocris Biosciences], 10 ng/mL KGF [R&D], 50 nM dexamethasone, 0.1 mM cyclic AMP, and 0.1 mM IBMX [all from Sigma-Aldrich]), as detailed in Jacob et al ( 33 ). Additionally, to increase the frequencies of SFTPC tdTomato or ABCA3:GFP-expressing iAEC2s, withdrawal and addback of CHIR99021 to distal lung progenitor cells was conducted as previously published in Jacob et al. ( 33 ), first by plating day 30 CPM sorted cells in 3D matrigel and feeding with CK+DCI and RI for 48 hours, followed by refeeding with KGF+ DCI and RI (KDCI+RI; i.e., CHIR withdrawal) for 5 days, followed by refeeding with the standard CK+DCI media for the duration of the experiment indicated in the text. 2D monolayer culture of iAEC2s. 2D monolayered iAEC2 cultures were prepared by plating either day 15 CD47 hi /CD26 lo sorted lung progenitors, or day 43+ alveolospheres as indicated in the text, after treating with trypsin to prepare a single-cell suspension. Single-cell suspensions were then plated on Matrigel-coated 48-well tissue culture plates (Corning) at 300,000 to 600,000 cells per well. 2D cultures were fed with CK+DCI with RI every other day until confluent. Quantification of surfactant secretion by mass spectroscopy and gene expression by either RT-qPCR or immunofluorescence microscopy are all detailed in the Supplemental Methods . ABCA3:GFP Co-IP and Mass Spectrometry analyses. To identify potential protein binding partners for ABCA3, we prepared iAEC2 cell pellets as indicated in the text, immunoprecipitated ABCA3:GFP fusion protein using a monoclonal anti-GFP antibody (Invitrogen GF28R) versus IgG control and identified candidate coprecipitated peptides and proteins, by mass spectroscopy (< 1% FDR cutoff) as detailed in the Supplemental Methods . Bulk RNA-Seq and bioinformatic analysis. Triplicate differentiations of each indicated line were performed and RNA extracts of iAEC2s ( n = 3 per line) profiled by bulk RNA-Seq with bioinformatics analyses, including GSEA, performed as we have previously published in Sun, et al. ( 39 ), and as further detailed in the Supplemental Methods . Data sets are available for free download from the gene expression omnibus ( https://www.ncbi.nlm.nih.gov/geo/ ; GSE205319; GSE205318) or through the bioinformatics portal at www.kottonlab.com EdU incorporation and CFE assays. CFE assays and EdU uptake scores after 24 hours of incubation in cultured iAEC2s were both prepared as detailed in the Supplemental Methods . Measurement of NFκB pathway activity in patient iAEC2s. Bioluminescence quantification of p50/65 heterodimer binding activity was performed in AEC2s using our published lentiviral NFκB signaling reporter vector ( 51 ) with methods for transduction of iAEC2s, sorting, and bioluminescence measurements detailed in our prior publication ( 36 ). Day 258 W308R and cW308 patient iAEC2s, and day 158 E690K and cE690 iAEC2s were transduced for biolumescence profiling. To quantify cytokine secretion, supernatants were harvested from iAEC2s after 2D culture and analyzed through a human magnetic Luminex assay (R&D Systems) on Bio-Plex 200 multiplexing analyzer system (Bio-Rad), with cytokine list and methods detailed in Supplemental Materials. Statistics. Unless otherwise specified in the text of Figure legends, statistical comparisons between 2 groups was performed using the student’s t test with significance determined by 2 tailed P value < 0.05. For comparisons across more than 2 groups, -way ANOVA with Tukey’s multiple comparisons testing was used with P < 0.05 used to determine statistical significance, as indicated in each figure legend. Study approval. Maintenance, editing, and directed differentiation of all cell lines was performed under regulatory approval of the Boston University Institutional Review Board (IRB; protocol H-33122). For derivation of ABCA3 mutant patient-specific iPSC lines, dermal fibroblasts from each individual were received from Washington University School of Medicine after review and approval by the Human Research Protection Office of Washington University School of Medicine. Data availability. A Supporting Data Values file with all reported data values is available as part of the supplemental material.

Results E690K and W308R autosomal recessive ABCA3 mutations cause severe respiratory disease in newborns. To develop an in vitro human model system able to study the downstream consequences of mutations in ABCA3 , we sought to generate iPSCs from individuals carrying rare homozygous ABCA3 mutations ( Figure 1 and Supplemental Figure 1 , A–E; supplemental material available online with this article; https://doi.org/10.1172/JCI164274DS1 ). We identified 2 individuals, 1 homozygous for ABCA3 W308R (c.922T>C) — which was predicted by in silico algorithms to be a type 1 mutation resulting in disrupted ABCA3 trafficking — and another homozygous for ABCA3 E690K (c.2068G>A) — which was predicted to be a type 2 mutation resulting in disrupted phospholipid transport. Both individuals underwent lung transplantation in the first few years of life, allowing access to lung explant histological analyses, which revealed diffuse AEC2 hyperplasia, alveolar septal thickening, and immune — neutrophilic and lymphoid — infiltrates ( Figure 1A , Supplemental Figure 1, A–E , and Supplemental Methods ). Ultrastructural examination of the E690K mutant lung explant AEC2s revealed dysmorphic, small, dense lamellar bodies ( Figure 1, B and C ), as have been reported in prior studies of lung tissue from individuals with biallelic ABCA3 mutations ( 1 ). Derivation of E690K and W308R ABCA3 mutant and gene-corrected patient iAEC2s. We reprogrammed dermal fibroblasts procured from these 2 individuals to generate patient-specific iPSC lines from each, carrying either homozygous E690K or homozygous W308R ABCA3 mutations (hereafter E690K-iPSCs and W308R-iPSCs, respectively) ( Figure 1D and Supplemental Figure 1F ). Next, using TALENS-mediated monoallelic targeting of the endogenous SFTPC locus, we introduced a heterozygous SFTPC tdTomato reporter into each line, as we previously described ( 32 ). Because iPSC lines generated from distinct individuals may exhibit different differentiation efficiencies and phenotypes due to differences in genetic backgrounds, we generated syngeneic gene-corrected control iPSCs lines using footprint free CRISPR/Cas9 gene editing to correct both mutant alleles for each corresponding ABCA3 mutation into WT (corrected lines hereafter referred to as cE690-iPSCs and cW308-iPSCs, Figure 1D and Supplemental Figure 1, G–I ). Differentiation of each patient-specific iPSC line according to our previously published distal lung differentiation protocol ( 32 , 33 ), followed by serial enrichment of NKX2-1 expressing lung epithelial progenitors using surrogate cell surface markers ( Figure 1E and Supplemental Figure 2 and Supplemental Methods ) resulted in the emergence of SFTPC tdTomato + cells in all 4 lines ( Figure 1, E–H ). After a brief period of CHIR99021 (hereafter referred to as CHIR) withdrawal to mature iAEC2s ( Figure 1E and Supplemental Figure 2D ) as previously published ( 33 ), tdTomato expression was quantified on day 43–44 with no significant differences in reporter expression detected comparing mutant to corrected iAEC2s produced from each line: 33.47% ± 1.53% for E690K cells versus 30.53% ± 1.83% for cE690 cells, and 28.63% ± 1.68% for W308R cells versus 24.13% ± 2.21% in cW308 cells ( Figure 1, G and H ). Further, there were no statistically significant differences in mRNA expression of a panel of AEC2 markers, SFTPC, SFTPB, NAPSA, PGC, and LPCAT1 between mutant and corrected iAEC2s (FDR > 0.05; n = 3 samples per genotype, each purified on day 43–44 based on tdTomato+ sorting and analysis by bulk RNA sequencing; Figure 1, G and H and Supplemental Figure 2 , and Supplemental Methods ), suggesting no appreciable impact of these ABCA3 mutations on distal lung epithelial–directed differentiation at this stage. ABCA3 mutant iAEC2s have attenuated surfactant lipid secretion. Clinically, biallelic ABCA3 mutations are associated with altered surfactant composition in affected infants and children, evident as decreased levels in bronchoalveolar lavage fluid of surfactant phospholipid species such as phosphatidylcholine (PC) and dipalmitoyl PC (DPPC, PC32:0) ( 13 , 37 , 38 ). Hence, we sought to measure levels of surfactant phospholipids secreted by the iAEC2s generated from each iPSC line. Because our distal lung cultures described previously ( 32 , 33 ) are typically conducted in a 3D format where iAEC2s orient apically internally toward each sphere lumen, it is difficult to readily access and measure surfactant secretion. To better access secretion from iAEC2s, we converted 3D cultured iAEC2 spheres into a 2D iAEC2 monolayer culture system, simultaneously enabling easy access to apically secreted material and monitoring of ABCA3 expression in real time by high-resolution imaging ( Figure 2, A–C ). To determine the impact of a submerged 2D culture on maintenance of iAEC2 identity, including ABCA3 expression, we utilized a normal iPSC line (BU3) engineered with CRISPR/Cas9 gene editing to carry a biallelic ABCA3:GFP knock-in fusion reporter cassette, targeted to the endogenous ABCA3 locus at the endogenous stop codon ( Figure 2A ). We recently published complete characterization of this iPSC line ( 39 ), hereafter referred to as BU3 ABCA3:GFP , demonstrating sensitive and specific GFP-based readouts of: (a) ABCA3 locus activity, (b) expression and intracellular localization of ABCA3 protein via fluorescence microscopic imaging of the ABCA3:GFP fusion protein, and (c) the formation and localization of lamellar bodies in iAEC2s through GFP labelling of the outer limiting membrane of these organelles. To determine whether this line could be adapted to 2D culture ( Figure 2B ) for real-time imaging of surfactant secretion, which is known to occur through regulated exocytosis of lamellar bodies at the AEC2 apical membrane ( 40 ), we differentiated BU3 ABCA3:GFP iPSCs into primordial NKX2-1+ lung progenitors and purified them by CD47/CD26 sorting, using methods we previously published ( Figure 2C ) ( 33 , 41 ). After transferring sorted progenitors into 2D monolayered versus 3D sphere culture conditions from day 15–32 ( Figure 2B ), we observed that 2D culture, compared with 3D culture, resulted in brighter and significantly more frequent expression of the ABCA3:GFP reporter (33.27% ± 4.29% versus 13.04% ± 3.43%, respectively) and augmented mRNA expression of NKX2-1 , ABCA3 , and NAPSA , while minimizing mRNA expression of nonlung endoderm genes CDX , AFP , or TFF1 , consistent with augmented iAEC2 differentiation in 2D conditions ( Figure 2, A–E ). We also observed that iAEC2s could be similarly transitioned from 3D to 2D cultures at multiple time points during iAEC2 epithelial sphere passaging (such as day 75; see below) with successful maintenance or augmentation of ABCA3 expression, suggesting that 2D culture adaptation for distal iPSC-derived lung epithelia is feasible at a variety of time points or developmental stages. Thus, our data indicate that 2D monolayer–cultured cells retain their AEC2 programs and readily express ABCA3 proteins, consistent with our recent report detailing methods for adapting iAEC2s to 2D and air-liquid interface culture conditions ( 42 ). We next sought to determine whether 2D cultured iAEC2s are capable of surfactant phospholipid secretion through regulated ATP-stimulated lamellar body exocytosis, as is known to regulate secretion from AEC2s in vivo. We plated day 75 ABCA3:GFP+ iAEC2s in our 2D culture condition and tested responses to treatment with secretagogues (ATP, 100 μM; Phorbol 12-myristate 13-acetaten, 300 nM) previously shown to result in exocytosis of LB contents in cultured primary AEC2s ( 43 ). We monitored potential secretagogue-induced LB exocytosis by culturing cells in a medium containing FM4-64, a lipophilic dye, for live-cell confocal imaging. Within 15 minutes of secretagogue addition, FM4-64 fluorescent signal arose from the GFP+ intracellular vesicles upon their fusion with the apical cell membrane, whereas no signal was detected in the absence of secretagogues ( Figure 2, F and G ). Thus, ABCA3:GFP+ iAEC2s readily respond to secretagogue signals resulting in the exocytosis of lipophilic contents from within the LB through the apical surface of the cell, into the culture supernatants, resembling the physiological secretion of surfactant from primary AEC2s into alveolar lumens. Having developed a 2D system able to monitor apical surfactant secretion from iAEC2s, next we assessed whether iAEC2s derived from our patient-specific iPSCs harboring ABCA3 mutations have attenuated surfactant secretion. We performed secretagogue-induced lipidomic analysis of apical supernatants from patient and syngeneic gene-corrected iAEC2s plated in our 2D culture system ( Figure 2H ). Consistent with previously published clinical findings of decreased PC measured in the bronchoalveolar lavage fluid samples of patients with biallelic ABCA3 mutations, we observed attenuated secretion of overall PC species, including surfactant-specific DPPC (also PC 32:0), in ABCA3 mutant iAEC2s (W308R and E690K), both at baseline and after secretagogue treatment, compared with their syngeneic gene-corrected control iAEC2s (cW308 and cE690, respectively, Figure 2, H and I ). While E690 corrected (cE690) iAEC2s did not have as robust a response to secretagogues as corrected W308 (cW308) cells, potentially due to the slightly higher baseline (presecretagogue) secretion from cE690 cells, regardless, ABCA3 gene correction resulted in higher secretion in both mutant iAEC2 genotypes, both at baseline and after secretagogue exposure ( Figure 2I ). These results indicate that reduced surfactant secretion in these patients resulted from the ABCA3 mutations and demonstrate the feasibility of functional restoration of ABCA3-dependent surfactant lipid transport through ABCA3 gene editing. Further, we conclude that the surfactant dysfunction phenotype observed in these patients is captured by our in vitro patient-specific iPSC-derived iAEC2 disease model. iAEC2s expressing E690K and W308R mutant ABCA3 proteins contain smaller lamellar bodies. We next sought to determine whether ABCA3 mutations are sufficient to result in either ABCA3 protein mistrafficking or altered LB size and function if expressed in iAEC2s of an independent genetic background. We thus performed CRISPR/Cas9–based biallelic single nucleotide mutagenesis of the ABCA3 locus to introduce each ABCA3 mutation (E690K and W308R) into a normal BU3 ABCA3:GFP line, enabling real-time visualization of mutant ABCA3:GFP protein trafficking. To see how these mutations of interest compare against a known, previously characterized, severe mistrafficking (type 1) ABCA3 mutant (L101P) ( 19 , 21 , 23 ), we also engineered a syngeneic line to carry L101P ABCA3:GFP. Thus, in total, we generated 4 parallel syngeneic iPSC lines able to express WT or the 3 mutagenized variants of ABCA3:GFP fusion proteins from the endogenous human ABCA3 locus upon differentiation to iAEC2s (henceforth referred to as WT-AG, E690K-AG, W308R-AG, and L101P-AG, Figure 3A ). To compare A549 cell lines, which have been published in the past to study ABCA3 biology, to our iAEC2 model, we also generated 4 A549 cell lines using lentiviral constructs similarly encoding GFP fused to the WT, E690K, W308R, or L101P ABCA3 mutant coding sequences. Each lentiviral ABCA3:GFP fusion cassette was driven by a constitutively and ubiquitously active EF1aL promoter ( 44 ), resulting in overexpression of these ABCA3:GFP variants in A549 cells ( Figure 4A ). We differentiated all 4 BU3 ABCA3:GFP iPSC lines to iAEC2s using our distal lung differentiation protocol ( Figure 1E ). We observed decreased frequency and intensity of ABCA3:GFP expression in iAEC2s at day 43–44 across all 3 mutant lines compared with the syngeneic WT line ( Figure 3, B–D ), indicative of possibly lower levels of ABCA3:GFP intracellular protein or incomplete protein folding or trafficking as a direct or indirect result of ABCA3 mutations. By harnessing the utility of the ABCA3:GFP fusion reporter construct, we next sought to study whether mutant ABCA3:GFP protein trafficking and LB morphology were affected within either iAEC2s or A549 cells expressing identical mutant ABCA3:GFP proteins. Using high resolution confocal imaging, we first saw an expected vesicular localization pattern of the WT ABCA3:GFP fusion proteins in both the iAEC2 and A549 cells, consistent with our prior profiles of this reporter system ( 39 ). We also saw mislocalization of L101P mutant protein, seen intracellularly in a nonvesicular, diffuse cytoplasmic pattern in both iAEC2s and A549 cells ( Figure 3E and Figure 4B ), confirming the severe type 1 mutant-specific ABCA3 protein mistrafficking phenotype previously only demonstrated in heterologous cell line models ( 12 , 20 , 23 ). Similar to the WT protein, both the E690K and W308R mutant proteins appeared to localize to intracellular vesicles in the iAEC2 and A549 cell models ( Figure 3E and Figure 4C ). However, vesicles outlined by mutant ABCA3:GFP proteins were significantly smaller in size compared with those outlined by WT proteins with average diameters of 0.72 μm ± 0.04 for the E690K mutant and 0.67 μm ± 0.04 for the W308R mutant compared with 1.53 μm ± 0.09 in WT vesicles ( Figure 3, E and F and Figure 4, C and D ). These smaller ABCA3 mutant LB sizes are consistent with actual electron micrographs from patients with the E690K mutation ( Figure 1B ) and previously published findings ( 1 , 9 , 27 , 45 ) describing small lamellar bodies in patients with ABCA3 mutations, regardless of mutation type, presumably due to decreased functional phospholipid transport into these vesicles. Despite these significant intracellular staining differences compared with WT control iAEC2s, gross cellular morphologies were similar between WT controls and all 3 mutant genotypes. To confirm our confocal findings of loss of function but seemingly intact ABAC3 trafficking in both E690K and W308R mutants, we performed Western blot analysis of intracellular protein extracts prepared from normal versus mutant iAEC2s generated from each line. Immunoblotting with an anti-GFP antibody identified the various forms of proteolytically processed ABCA3:GFP proteins from a 220 kDa to 180 kDa product, a finding that indicates successful post-Golgi ABCA3 trafficking and processing necessary for subsequent localization to LB precursor vesicles ( 19 , 20 ). Consistent with our confocal observations, in both iAEC2 and A549 models, we found complete absence of the 180 kDa band in the L101P type 1 mutant, consistent with loss of normal processing and trafficking. We also found presence of the 180 kDa band in the E690K and W308R mutants, qualitatively indicating some degree of successful post-Golgi ABCA3 processing and trafficking ( Figure 3, G and H and Figure 4E ) in both model systems. In contrast, we detected a quantitative difference between the 2 model systems with a small but statistically significant decrease in mutant E690K protein cleavage compared with WT protein. This decrease was detectable in the iAEC2 model ( Figure 3H and Supplemental Figure 5 ), but not in the A549 model system ( Figure 4, E and F ). Co-IP/MS analyses of WT and mutant ABCA3:GFP proteins suggests potential ABCA3 binding partners. After seeing the severe mistrafficking phenotype of the L101P mutant ABCA3:GFP protein in our iAEC2 model, we next sought to interrogate whether protein-protein interactions of the WT ABCA3 protein would be disturbed by the most severe L101P mutant ABCA3 protein. We first began by using proteomics to characterize potential interacting protein partners of the WT ABCA3 protein, as previous studies have been limited by lack of access to reliable ABCA3 antibodies for protein pull down. To do so, we performed co-IP using a monoclonal antibody against the GFP portion of the WT ABCA3:GFP fusion protein in iAEC2 cell lysates, followed by trypsinization and mass spectrometry analysis ( Supplemental Figure 3A ). We identified 20 unique candidate binding partners of WT ABCA3 ( Supplemental Figure 3B ), classified into proteins that have been reported to facilitate transmembrane protein localization, folding, and vesicle trafficking, such as TMED10, RAB7A, and GIPC1 ( 46 – 48 ), as well as additional proteins that potentially facilitate intracellular vesicle and lamellar body exocytosis, such as the SNARE protein assembly regulator, STXBP2. Moreover, we identified proteins related to solute and phospholipid transporters associated with ABCA3:GFP, such as the glucose transporter SLC2A1 and members of the P4-ATPase phospholipid transporter/flippase complex TMEM30A, TMEM30B, and ATP11A, which are responsible for fine tuning surfactant lipid speciation and content ( Supplemental Figure 3B ). Finally, ABCA3:GFP preferentially associated with the IL-6 signal transducer, IL6ST, and TRAP, a TNF1-associated chaperone, suggesting possible alternate roles or binding partners for ABCA3. After characterizing the WT ABCA3 potential interacting partners, we next sought to interrogate the protein binding partners of the severely mistrafficked L101P mutant ABCA3:GFP fusion protein. Some protein-protein interactions were retained — particularly those associated with scaffolding factors such as GIPC1 — at a higher normalized affinity level compared with the WT protein. However, the L101P mutant protein showed significantly attenuated binding affinity for lysosomal protein RAB7A; lipid transporting partners, such as the ATP11A; and the TMEM30A, TMEM30B flippase proteins; as well as the vesicular exocytosis regulator STXBP2. We also found L101P had a unique set of protein-binding signatures consisting of various ER-based chaperone proteins responsible for unfolded protein binding, including protein products such as CCT2, CCT3, CCT5, and CCT6A ( Supplemental Figure 3B , C). While these results require further validation of protein binding to ABCA3, the proteomic findings are consistent with our previous immunoblot findings and previous studies for L101P in heterologous cell lines ( 19 , 20 , 22 , 23 , 27 ), suggesting mislocalization of the mutant protein in the ER. Transcriptomic analyses of ABCA3 mutant versus gene-corrected iAEC2s reveal upregulation of inflammatory pathways. Although ABCA3 mutations have been assumed to result mainly in defective surfactant secretion, the altered intracellular binding partners suggested by our co-IP studies and known mistrafficking effects also raise the possibility that ABCA3 mutations might also cause generalized perturbations in the AEC2 itself. To provide an unbiased assessment of potential epithelial-intrinsic responses that might result from ABCA3 mutations, we performed transcriptomic analyses comparing mutant to normal or gene-corrected iAEC2s in both the patient-specific genetic backgrounds and in our syngeneic BU3 model system, where we had knocked in the same ABCA3 mutations into normal cells of a distinct genetic background. We performed bulk RNA-Seq on patient-specific SFTPC tdTomato + iAEC2s (E690K and W308R mutants) and their gene-corrected syngeneic control iAEC2s as well as the single WT and 3 CRISPR/Cas9 mutagenized ABCA3:GFP+ iAEC2s ( n = 3 per genotype × 4 genotypes, day 43–44 of differentiation; Figure 5A ). Pairwise comparisons between each pre- versus post-gene–corrected patient-specific iAEC2 sample revealed 1,384 differentially expressed genes (DEGs) for the E690K mutant and 254 genes for the W308R mutant (FDR < 0.05, absolute log 2 FC > 1, Supplemental Table 1A and B. Among the top 50 genes enriched in the E690K iAEC2s were NFκB signaling factors such as TNFRSF10B, IL23A, and NFKB1, protein chaperones such as HSPA1A, and gene products, such as TGFβ, which have been previously published as playing a role in interstitial lung diseases. Notable top enriched genes in the W308R iAEC2s included known cell death regulators CFLAR and GPC3, and toll like receptor TLR2, possibly suggestive of a diseased cell. Analyses of all DEGs by gene set enrichment analysis (GSEA) using MSigDM Hallmark gene sets ( 49 ) revealed inflammatory pathways enriched in the E690K iAEC2s, including TNFα signaling via NFκB, inflammatory response, IL6 JAK STAT3 signaling, interferon responses, complement, IL2 STAT5 signaling, and additional noteable pathways related to TGFβ signaling, the unfolded protein response, and apoptosis (FDR < 0.05, Figure 5B ). GSEA analysis of W308R mutant iAEC2 versus corrected cW308R iAEC2s revealed similar enrichment of inflammatory pathways, including TNFα signaling via NFκB, inflammatory response, IL6 JAK STAT3 signaling, IL2 STAT5 signaling, and interferons ( Figure 5C , FDR < 0.05). In contrast, proliferative and metabolic pathways such as E2F targets, G2M checkpoint, mitotic spindle, and glycolysis were downregulated (FDR < 0.05) in the E690K mutant cells with proliferative pathways also downregulated in the W308R mutant cells. Pairwise comparisons between corresponding engineered knock-in mutant and WT ABCA3:GFP+ iAEC2s revealed 1,096 DEGs in the E690K mutant (E690K-AG) and 816 DEGs in the W308R mutant (W308R-AG) compared with the WT iAEC2s (WT-AG, FDR < 0.05, absolute log 2 FC > 1, Supplemental Table 2 , A and B). Analyses of DEGs by GSEA revealed enrichment of similar pathways shared with the patient-specific iAEC2 data sets ( Figure 5, D and E ) including TGFβ signaling, hypoxia, epithelial mesenchymal transition, and IL2_STAT pathways. Consistent with gene-corrected patient iAEC2s, surfactant-related pathways such as adipogenesis, peroxisome, and fatty acid metabolism were differentially regulated in the WT-AG versus mutant cells ( Figure 5, D and E ). Most notably, the TNFα signaling via NFκB pathway was consistently upregulated across all 4 mutant lines in the 2 distinct genetic backgrounds (both patient-intrinsic iPSC backgrounds and BU3 iPSC knock-in background, Figure 5, B–D ). Proliferative differences between mutant and gene-corrected iAEC2s. We functionally assessed whether both ABCA3 mutations negatively impacted iAEC2 proliferation as predicted by our bulk RNA-Seq results, which showed diminished Myc targets, E2F targets, or cell cycle–associated pathways in the patient-specific mutant iAEC2s ( Figure 5 ). Proliferative capacity is a particularly important characteristic of AEC2s in vivo since they serve as the predominant facultative progenitors tasked with maintaining or replenishing the alveolar epithelium, and diminished AEC2 progenitor and regenerative capacity resulting from ABCA3 deficiency has been suggested in vivo in genetic mouse models ( 50 ) but is unstudied in humans. Hence, we performed EdU incorporation assays to formally quantify proliferation in all patient iAEC2 lines. Comparing 24-hour EdU incorporation by SFTPC tdTomato + iAEC2s within each syngeneic iAEC2 paired sample, we found no proliferative difference between W308R and cW308 iAEC2s, but we found lower proliferation in E690K iAEC2s with proliferating cells making up 32.83% ± 3.21% of total cells compared with a higher 41.73% ± 2.20% in the corrected cE690 iAEC2s ( Figure 6A ). To see whether these proliferative differences functionally translated to changes in clonogenicity, we performed colony forming efficiency (CFE) assays on day 57 epithelial sphere outgrowths from day 43 plated SFTPC tdTomato + patient iAEC2s. Consistent with no differences in the EdU incorporation, we found no CFE changes in the W308R compared with the cW308 iAEC2s, but we observed an approximately 3-fold lower CFE in the E690K mutant compared with cE690 iAEC2s, consistent with reduced progenitor activity (E690K CFE = 2.38% ± 0.23% versus cE690 CFE = 6.97% ± 0.33%; Figure 6B ). E690K and W308R ABCA3 mutations increase iAEC2 NFκB signaling resulting in secretion of inflammatory cytokines. Since transcripts related to NFκB signaling were upregulated in common across all ABCA3 mutant iAEC2s regardless of genetic background, we next sought to validate this finding by quantifying the activity of NFκB signaling in iAEC2s. To quantify the level of canonical NFκB pathway activity, we transduced both mutant and syngeneic gene-corrected iAEC2s ( n = 3) with a lentivirus we have previously published ( 51 ) that carries a dual gene expression cassette encoding a constitutively active GFP reporter and a luciferase reporter driven by a minimal promoter with adjacent enhancer consisting of multiple repeats of the p50/p65 heterodimer consensus binding sequence ( Figure 6C ). Analysis of luciferase activity in GFP+ transduced iAEC2s validated significantly increased NFκB pathway activity in both E690K and W308R mutant iAEC2s ( Figure 6D ), findings consistent with our prior report of canonical NFκB activation in iAEC2 models of another chILD related mutation, SFTPC I73T ( 36 ), and possibly suggesting a common signaling pathway augmented in diseased iAEC2s. Analysis of supernatants from 2D monolayer-cultured W308R mutants versus gene-corrected patient iAEC2s (but not E690K mutants) revealed significantly increased secretion of selected proinflammatory cytokines, such as increased CX3CL1, a potent chemoattractant for lymphocytes and macrophages and a known contributor to the pathogenesis of other ILDs ( 52 – 54 ), and MMP-10, a recently proposed biomarker for idiopathic pulmonary fibrosis ( 55 ) ( Figure 6E and Supplemental Figure 4 ), further suggesting that changes in iAEC2s beyond mere surfactant secretion defects, may occur as a result of specific types of ABCA3 mutations.

Discussion Individuals carrying pathological variants of ABCA3 mainly present with either neonatal respiratory distress (NRDS) in full-term infants or chILD in children, or both. Incidentally, many patients with chILD do not present with NRDS, suggesting surfactant dysfunction alone in neonates is not the primary pathogenic mechanisms triggered by ABCA3 mutations to disrupt AEC2 metabolism for chILD pathogenesis. Thus, current therapies for patients with lung disease due to ABCA3 mutations are nonspecific and ineffective, with the only known definitive treatment option being lung transplant, as was the case for the children homozygous for E690K or W308R ABCA3 mutations profiled in our studies ( Figure 1, A and B and Supplemental Figure 1I ). Development of therapeutic alternatives has been challenging due to an inability to access and culture primary AEC2s from patients with chILD. In the present study, we overcame these hurdles by developing and extensively characterizing an in vitro disease model using iAEC2s derived from patient-specific and ABCA3:GFP reporter–containing iPSC lines, enabling the detailed analysis of ABCA3 mutant phenotypes compared with syngeneic gene-corrected controls. Through functional, proteomic, lipidomic, and transcriptomic analyses, we find our iAEC2 model recapitulates the clinically observed phenotype of surfactant dysfunction resulting from diminished secretion and predicts a previously unappreciated epithelial-intrinsic aberrant AEC2 phenotype caused by ABCA3 mutations leading to increased canonical NFκB signaling, and, in some genotypes, to diminished progenitor capacity. Whether this altered cellular phenotype results from a toxic gain of function effect of mutant ABCA3 protein or secondary effects from loss of ABCA3 function will require further research. Our iAEC2 disease model has the potential to advance an understanding of chILD pathogenesis and may provide a foundation for the development of pathway-specific therapeutics designed to reverse the readouts of aberrant iAEC2 function detailed in this report. The introduction of a variety of ABCA3 mutations into a normal iPSC line resulted in similar perturbations to those observed in patient-specific mutant lines, thus confirming that surfactant dysfunction results from abnormal ABCA3 rather than from associated modifying gene mutations present only in affected patients’ genetic backgrounds. Applying our model system, we confirmed that, similar to previous studies using HEK293 and A549 cells ( 19 – 21 , 23 ), the E690K mutant protein mostly retained WT protein trafficking to LBs, with only a small, though statistically significant, decrease in proteolytic cleavage efficiency compared with WT cells. Unexpectedly, we also found W308R to traffic to LBs similar to WT cells by both fluorescence microscopy and Western blotting, contrary to in silico predictions using sequence-based protein conformational algorithms ( 56 – 58 ). While both mutant proteins retained some degree of proteolytic cleavage and trafficking patterns, we observed smaller LBs formed by the mutant ABCA3 fusion proteins similar to previous studies ( 27 , 45 ). Combined with measurements of decreased surfactant lipids found in patient iAEC2s, this phenotype further supports our current understanding of ABCA3 as the major surfactant lipid transporter in AEC2s ( 10 – 12 ) and lends support to LB size as a useful readout for small molecule screens to restore ABCA3 function. The finding of smaller lamellar body size in our model is consistent with the abnormally small and dense lamellar body morphologies found in the lung tissues of patients with ABCA3 mutations ( Figure 1 ) ( 1 , 2 , 9 , 59 ) and suggests that these smaller lamellar bodies are a direct consequence of dysfunctional mutant ABCA3 rather than the epiphenomena of other perturbations, such as lung infections or ventilator induced lung injuries, which frequently occur in patients with chILD. Encouragingly, based on similarities between CFTR (ABCC7) — itself an ABC transporter protein ( 60 ) and the causal gene for cystic fibrosis ( 61 – 63 ) — and ABCA3, recent studies have repurposed drugs developed to modulate CFTR function for restoration of ABCA3 function in cancer cell line models. These prior studies used readouts of vesicle size and ABCA3 mistrafficking in the context of forced overexpression of mutant ABCA3 proteins in A549 cells ( 24 , 25 ). While A549 cells — an adenocarcinoma cell line thought to be derived from AEC2s — lack surfactant protein or NKX2-1 gene expression and thus do not recapitulate normal AEC2 physiology, we were surprised to find that A549 cells do model many of the intracellular features of our iAEC2 mutant ABCA3 model system. For example, when examined head-to-head, A549s overexpressing identical ABCA3 mutant proteins to those expressed from the endogenous locus in our iAEC2s demonstrated qualitatively similar (though not quantitatively identical) protein trafficking phenotypes, along with smaller LBs in E690K and W308R mutants. This finding lends support to the use of A549 cells, in tandem with iAEC2s, for future mechanistic and therapeutic studies. For example, high throughput drug screening could be performed in A549 cells, with potential hits confirmed in a more physiologically relevant cell model, such as iAEC2s or primary AEC2s, if available. Our ABCA3 disease model leveraging footprint-free CRISPR/Cas9 gene editing tools, enabled the pairwise comparison of syngeneic mutant versus corrected iAEC2s while controlling for the rest of the genetic background. We employed models that carried the patient’s own genetic background, including any genetic susceptibility to the disease, as well as mutagenized WT iPSC lines (BU3) engineered to have the same ABCA3 mutations, allowing for stringent identification of mutant-specific pathways in both diseased and healthy genetic backgrounds. Transcriptomic analyses by bulk RNA-Seq across all 3 genetic backgrounds (E690K individual, W308R individual, and knockin mutant BU3 iPSC lines) revealed enrichment of a wide range of inflammatory pathways with the NFκB signaling pathway shared across all ABCA3 mutant iAEC2s ( Figure 5 ). This implied aberrant AEC2 phenotype resulting from ABCA3 mutations in AEC2s, is consistent with the published conditional mouse ABCA3 knock out model, which showed increased inflammation in mouse lungs in vivo resulting from loss of ABCA3 function ( 50 ). Importantly, Rindler et al. also observed that after tamoxifen induced ABCA3 deletion in a portion of AEC2s in these mice, those AEC2s with intact ABCA3 protein appeared to survive, proliferate, and reconstitute the AEC2 compartment with ABCA3 expressing cells, thus implying a survival advantage to AEC2s with normal ABCA3 compared with those lacking intact ABCA3. Interestingly, we also found proliferative pathways enriched in our gene-corrected patient iAEC2s with the cE690 gene-corrected iAEC2s, demonstrating improved proliferation and CFE ( Figure 6 ). A proliferative advantage for AEC2s containing normally functioning ABCA3, if confirmed, may provide the foundation for delivery of gene correcting vectors as a future therapeutic option, as recently demonstrated in mouse lungs following in utero gene-correction of mutant SFTPC I73T by Alapati et al. ( 64 ). An important question to consider is how NFκB signaling is activated in mutant iAEC2s. It is possible that some of the enriched pathways upregulated in mutant iAEC2s are upstream of NFκB signaling, such as the unfolded protein response (UPR) or dysregulation of overall intracellular lipid homeostasis without adequate ABCA3 function. Future studies focused on validation of these potential upstream pathways will likely further inform disease mechanisms. Moreover, our recent report detailing the downstream impact of SFTPC I73T mutations in iAEC2s also revealed upregulation of the NFκB signaling pathway ( 36 ). This suggests a potential common response downstream of differing types of perturbed pathways resulting from disparate mutations that alter AEC2 function. Our model serves as a reductionist, epithelial-only model system, allowing delineation of potential disease activating mechanisms that initiate in the cell type carrying the AEC2-specific mutation of interest, without the presence of additional neighboring lineages that can confound this understanding. On the other hand, a complete understanding of chILD will require more complex models introducing these lineages that likely respond to and interact with the initiating dysfunctional cell type. As a result of increased NFκB pathway activity, we found secretion of proinflammatory cytokines MMP10 and CX3CL1 in the W308R mutant iAEC2s. Future studies with animal models and more physiologically complex culture systems such as those incorporating coculture with relevant immune and mesenchymal cell types, will be needed to facilitate a more comprehensive understanding of the paracrine effects of diseased AEC2s as a result of NFκB upregulation caused by ABCA3 mutations. Thus, our study reports a human disease model system based on iAEC2s carrying pathogenic ABCA3 mutations and suggests epithelial-intrinsic contributions to chILD pathogenesis. This model should now facilitate the development of effective targeted therapies able to prevent or reverse AEC2 dysfunction in patients with chILD carrying ABCA3 mutations.

Mutations in ATP-binding cassette A3 (ABCA3), a phospholipid transporter critical for surfactant homeostasis in pulmonary alveolar type II epithelial cells (AEC2s), are the most common genetic causes of childhood interstitial lung disease (chILD). Treatments for patients with pathological variants of ABCA3 mutations are limited, in part due to a lack of understanding of disease pathogenesis resulting from an inability to access primary AEC2s from affected children. Here, we report the generation of AEC2s from affected patient induced pluripotent stem cells (iPSCs) carrying homozygous versions of multiple ABCA3 mutations. We generated syngeneic CRISPR/Cas9 gene-corrected and uncorrected iPSCs and ABCA3-mutant knockin ABCA3:GFP fusion reporter lines for in vitro disease modeling. We observed an expected decreased capacity for surfactant secretion in ABCA3-mutant iPSC-derived AEC2s (iAEC2s), but we also found an unexpected epithelial-intrinsic aberrant phenotype in mutant iAEC2s, presenting as diminished progenitor potential, increased NFκB signaling, and the production of pro-inflammatory cytokines. The ABCA3:GFP fusion reporter permitted mutant-specific, quantifiable characterization of lamellar body size and ABCA3 protein trafficking, functional features that are perturbed depending on ABCA3 mutation type. Our disease model provides a platform for understanding ABCA3 mutation–mediated mechanisms of alveolar epithelial cell dysfunction that may trigger chILD pathogenesis. A human pluripotent stem cell-derived model mirrors ABCA3 mutation-mediated alveolar type 2 cell dysfunction.

Author contributions YLS, EEH, and DNK designed the project, performed experiments, and wrote the manuscript. HH performed the lipidomic analyses. PY performed Western blots. JAW and FSC provided patient tissue, clinical information, analyzed data, and edited the manuscript. CVM performed RNA-seq analyses. JK and AE performed co-IP/MS experiment. MJ performed iPSC reprogramming and archiving. KG performed cell differentiation and CFE measurements. All authors reviewed and approved the final manuscript. Supplementary Material

This study was supported by NIH grants F30 HL142169 and TL1 TR001410, PCTC JumpStart Award U24HL134763 to YLS, R01 HL149853 to JAW, U01HL134745, U01HL134766, U01HL152976, and R01HL095993, to DNK, and a grant from the Novartis Institutes for Biomedical Research, Cambridge, MA to DNK. iPSC derivation and cell sharing was supported by N01 75N92020C00005 and U01TR001810. The authors thank all members of the Kotton Laboratory for insightful discussions. They are grateful for the technical assistance of Jyh-Chang Jean in the CReM iPSC Core, Ashley LeClerc and Yuriy Alekseyev of the Boston University School of Medicine (BUSM) Microarray and Sequencing Core Facility and Brian R. Tilton of the BUSM Flow 16 Cytometry Core. For facility management, they thank Greg Miller, CReM Laboratory Manager, and Marall Vedaie and Olivia Hix, Kotton laboratory managers. 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e164274

oa_package/e8/7f/PMC10786693.tar.gz

PMC10786694

37988169

Introduction Neuronal inclusions consisting of the microtubule-associated protein tau are found in various neurodegenerative diseases known as tauopathies. Alzheimer’s disease (AD) is the most prevalent tauopathy and is characterized by the accumulation of extracellular β-amyloid (Aβ) plaques and intracellular misfolded tau, which forms neurofibrillary tangles, causing microglial activation and synaptic loss ( 1 , 2 ). Most AD cases are sporadic, while a small portion of patients with AD carry genes that cause increased Aβ production and subsequently induce progressive neurodegeneration and microglial activation ( 3 ). Many drugs that target Aβ plaques have proven ineffective in the treatment of AD in clinical trials. Recently, the FDA approved Aducanumab and Lecanemab, both are monoclonal antibody–based therapies targeting Aβ and show moderate improvements in cognitive decline for patients with early AD ( 4 ). However, the efficacy and potential side effects of these treatments remain to be monitored. While Aβ plaque accumulation is frequently observed in aging brains, it doesn’t necessarily lead to the development of AD. For example, a neuroimaging study of 513 participants demonstrated that individuals with both amyloid and tau pathologies exhibited a 27% prevalence of AD. In contrast, those with only amyloid pathology had just 6% prevalence ( 5 ), underscoring the significant role of other factors, like tau, in AD development. Moreover, genetic and epigenetic studies of patients with AD reveal that many genes associated with AD risk are enriched in microglia and are related to immune responses ( 6 , 7 ). Microglia are immune cells in the brain that play important roles in brain development, homeostasis maintenance, and diseases ( 8 ). Accumulating evidence shows that dysregulated microglial activation and altered homeostatic functions contribute to neurodegeneration ( 9 , 10 ). Earlier studies demonstrated that microglia and neurons interact through multiple routes, such as direct contact, the formation of tunneling nanotubes, and the secretion of exosomes or soluble factors (such as cytokines and chemokines) ( 11 ). Despite the tremendous efforts that have been devoted to the investigation of microglial-neuronal interactions, the current understanding of the role of microglia in tauopathy remains incomplete. This is complicated by the fact that microglia exhibit multidimensional phenotypes, including immune activation and suppression, during the course of neurodegenerative diseases ( 12 ). Moreover, microglia express multiple types of lectins (e.g., galectins) that serve as critical checkpoints for microglial activation and contribute to neurodegeneration ( 13 , 14 ). The role of Galectin-3 (Gal3), encoded by the Lgals3 gene, in neurodegenerative diseases has recently attracted much attention ( 15 ). Gal3 is a proinflammatory autocrine mediator that can bind toll-like receptor 4 (TLR4) and the receptor expressed on myeloid cells-2 (TREM2) ( 1 , 14 ). In a model of Huntington’s disease, microglial Gal3 has been shown to accumulate on damaged lysosomes, interfering with their clearance activity and facilitating inflammation ( 16 ). A large-scale proteomic analysis of the brains of individuals with AD recently identified Gal3 in a microglial metabolism module enriched with genetic risk factors of AD ( 17 ). Furthermore, comparison analysis of models representing AD, amyotrophic lateral sclerosis (ALS), and multiple sclerosis showed that microglia in neurodegenerative contexts displayed elevated levels of genes associated with the Trem2-Apoe pathway, including Lgals3 and Clec7a , which contribute to the development of detrimental microglial phenotypes ( 18 ). Moreover, Gal3 knockout was demonstrated to reduce the plaque burden and improve cognitive behaviors in 2 transgenic mouse models of AD (APP/PS1 and 5xFAD) ( 1 , 19 ). Nevertheless, the roles of Gal3 in tauopathies (including AD and frontotemporal lobar dementia [FTLD]), which involve amnestic cognitive impairment and neuroinflammation, remain unclear.

Methods Specific details on the research materials and portions of the experimental protocols can be found in the Supplemental Methods . Animals. Tau22 mice (B6. Cg-Tg(Thy1-MAPT)22Schd) express mutated human tau (G272V and P301S) associated with FTLD (1N4R tau) driven by a neuron-specific promoter (Thy1.2), and were maintained in the C57BL/6J background ( 33 ). APP/PS1 mice (B6C3-Tg(APPswe,PSEN1dE9)85Dbo/Mmjax) were originally obtained from The Jackson Laboratory. Gal3 knockout mice ( Lgals3 –/– ) in a C57BL/6J background ( 60 ) were kindly provided by Fu-Tong Liu from the Institute of Biomedical Sciences, Academia Sinica. Tau22/ Lgals3 +/+ and APP/PS1/ Lgals3 +/+ mice were initially crossed with Gal3 knockout mice (WT/ Lgals3 –/– ). The resultant offspring were then crossed to obtain Tau22/ Lgals3 –/– and APP/PS1/ Lgals3 –/– mice. Mice were housed at a density of no more than 5 per cage in individually ventilated cages, with water and a chow diet (LabDiet), under a 12:12 hour light/dark cycle at the Institute of Biomedical Sciences Animal Care Facility (Taipei, Taiwan). Cells. The neuronal-like cell line SH-SY5Y-eGFP-tau-P301L (SY5Y-tau) ( 26 ) was a gift from Yung-Feng Liao at the Institute of Cellular and Organismic Biology, Academia Sinica. SY5Y-tau cells were grown in DMEM (HyClone) supplemented with 10% heat-inactivated FBS, 5 μg/mL blasticidin (Gibco) and 200 μg/mL hygromycin (Thermo Fisher Scientific) at 37 °C in a humidified 5% CO 2 -containing atmosphere. All SY5Y-tau cells were induced to express tau for subsequent experiments with 1 μg/mL doxycycline hyclate (Santa Cruz). Human tissue samples. Postmortem human cortical and hippocampal specimens from 4 normal individuals, 6 patients with FTLD-Tau (CBD, PSP, and Pick’s) and 4 patients with AD were obtained from the UCD Alzheimer’s Disease Center ( Supplemental Table 1 ). Human iMGL. Human iPSCs were routinely maintained in Hung-Chih Kuo’s laboratory at the Institute of Cellular and Organismic Biology, Academia Sinica, Taiwan, as previously described ( 61 ). The sex and APOE genotype of human iPSCs are listed in Supplemental Table 2 . Microglial differentiation was conducted according to an established protocol elsewhere with slight modification ( 20 ) as detailed in Supplemental Methods . pTau preparation. Recombinant pTau (1N4R) was routinely prepared as previously described except that the size exclusion FPLC chromatography step was omitted ( 21 ). Briefly, pTau and tau were expressed using the constructs pMK1013-GSK-tau(1N4R) and pMK1013-tau(1N4R), respectively. The constructs were transformed into the E . coli BL21 codon plus strain and cultured overnight. Cells were cultured to an OD 600 between 0.3 and 0.5 before the addition of IPTG. After 2 hours, cells were pelleted and resuspended in 10 mL of cold purification buffer (20 mM Tris-HCL pH 5.8, 100 mM NaCl, 1 mM PMSF, and 0.2 mM orthovanadate). Subsequently, 1 mg/mL of lysozyme was added and the mixture was incubated at 30°C for 30 minutes. The mixture was then sonicated (Branson Digital Sonifier 450; 30% amplitude; Pulse-ON time 5 seconds and Pulse-OFF time 5 seconds for a total of 3 minutes) and centrifuged at 17,000 g for 40 minutes at 4°C. The supernatant was boiled and cooled down on ice for 30 minutes, respectively. The mixture was then subjected to centrifugation at 7,000 g for 50 minutes at 4°C. The supernatant was transferred to a new tube and added with 0.5 mM DTT and 1 mM EDTA. For digestion, TEV protease was added (1 OD280 TEV: 100 OD280 sample) and incubated at 4°C overnight. The mixture was centrifuged at 7,000 g for 30 minutes at 4°C, and the resulting supernatant was transferred to a new tube. Gel filtration was performed using a spin column (Amicon Centrifugal Filter Unit, Ultra-15, 10 kD) at 5,000 g for 20 minutes at 4°C to reduce the volume to 1–2 mL. A total of 4 mL of gel filtration buffer (20 mM Tris-HCL pH 7.4, 100 mM NaCl) was added to the spin column, and the buffer exchange step was repeated twice. Finally, the supernatant was transferred to a new tube and aliquoted for storage at –80°C. Size of recombinant proteins were analyzed with SDS-PAGE and TOOL Start Blue Staining Reagent according to the manufacturer’s protocol (TTD-BS1L, TOOLS). Endotoxin in the recombinant proteins were detected by using an endotoxin ELISA kit (abx514093, Abbexa) and Pro-Q Emerald 300 LPS gel stain kit (P20495, Thermo Fisher Scientific). All recombinant proteins had negligible endotoxin levels below detection ranges (0.015—1.0 EU/mL) at the working concentration (50 nM). All batches of pTau were produced at Michigan State University in the laboratory of Min-Hao Kuo and were validated to meet quality control standards. Specifically, the purified pTau was shown to form thioflavin S-reactive amyloid under inducer-free conditions and caused at least 50% cell death in SH-SY5Y cells within 24 hours at a concentration of 1 μM. Cell death was quantified by propidium iodide and fluorescein diacetate differential staining. EV isolation and preparation. The conditioned medium (15 mL) harvested from iMGL was centrifuged at 2,000 g for 30 minutes to exclude cell debris. For concentration, the supernatants were transferred into Amicon Ultra15 Centrifugal Filter Units (3K) devices and centrifuged at 4,000 g for 45–55 minutes. The concentrated culture media (1 mL) were transferred into a sterile 1.5 mL microcentrifuge tube, and 0.5 volumes of Total Exosome Isolation reagent (Invitrogen) were added. The concentrated culture media and the reagent were mixed by pipetting to achieve a homogenous solution, followed by incubation at 4°C overnight. After incubation, the samples were centrifuged at 10,000 g for 1 hour at 4°C. Supernatants were carefully removed without disturbing the EV pellets at the bottom of the tubes. EV pellets were either resuspended in PBS for NTA and cell treatments or in RIPA lysis buffer for immunoblotting. To compare the levels of Gal3 in EV and non-EV secreted forms, EVs were prepared following a protocol for EV isolation detailed elsewhere ( 62 ), with the addition of a concentration step using Amicon Ultra15 Centrifugal Filter Units (3K) devices as described above. EVs were pelleted from 150 μL of concentrated medium, and the supernatant was carefully collected. The pelleted EVs were then resuspended in 150 μL of RIPA lysis buffer. We compared Gal3 levels by using equal 50-μL volumes of both the supernatant and EV lysates for Western blot analysis. The Proteinase K protection assay was conducted as detailed elsewhere ( 29 ). EVs were treated with 1 mg/mL proteinase K for 30 minutes at 37°C. A control aliquot, without proteinase K, was incubated under the same condition. Treatment of microglia and neurons. iMGL were pretreated with 10 μM TD139 (B2266-5, BioVision) or 0.1% DMSO as a vehicle control for 24 hours before treatment with pTau (50 nM) or fibrillar Aβ (1 μM) for 6 hours, followed by the collection of the iMCM and cell samples for RNA preparation. SY5Y-tau cells were treated with iMCM for 24 hours before the assay. For neutralizing antibody treatments, 3 μg/mL B2C10 (an anti-Gal3 neutralizing antibody that binds Gal3 at the first 18 amino acids of N-terminus ( 14 , 63 ), 556904, BD Pharmingen) was coincubated with iMCM and SY5Y-tau cells for 24 hours. SY5Y-tau cells were treated with recombinant Gal3 (1 μM) or EVs (10 μg/mL) for 24 hours before assays. Recombinant Gal3 was conjugated with Atto-565 fluorescence dye ( Atto565 rGal3) by using a Lightning-Link Rapid Atto 565 Conjugate system (351-0030, Innova BioSciences) and used to treat SY5Y-tau cells for 4 hours to evaluate the ability of Gal3 to enter cells. RNA extraction, cDNA synthesis, and quantitative PCR. Total RNA was extracted from hippocampal tissues and iMGL by using the GENEzol TriRNA Pure Kit (Geneaid Biotech) and subsequently reverse transcribed into cDNA by using Superscript III (Invitrogen) according to the manufacturer’s protocol. RT–qPCR was performed by using SYBR Green PCR Master Mix (Life Technologies) with amplification on a LightCycler 480 Real-Time PCR System (Roche Life Science) or ABI QuantStudio 5 Real-Time PCR System (Applied Biosystems). Relative expression of genes of interest were determined by using the ΔΔCT method. For normalization, Gapdh and 18S served as the housekeeping genes for mice and humans, respectively. The expression levels of genes of interest were normalized to those of the respective reference gene, and relative expression levels were calculated. Primers used for the quantification of genes of interest are listed in Supplemental Table 8 . scRNA-Seq of microglia. Adult microglial isolation was performed according to the protocol detailed elsewhere ( 39 ). For single-cell gene expression library construction, FACS-sorted cells were immediately captured in droplets that were emulsified with gel beads by using the 10× Genomics Chromium Next GEM Single Cell 3’ GEM, Library & Gel Bead Kit v3.2 reagent (10× Genomics) according to the manufacturer’s protocols. These libraries were sequenced using a NovaSeq 6000 (Illumina). The details of the bioinformatics analysis of microglia are provided in the Supplemental Methods . Statistics. Data in histograms and dot plots are expressed as the mean ± SEM, while violin plots show the medians with the 25th and 75th percentiles. Statistical comparisons were performed with GraphPad Prism 9 software using the 2-tailed unpaired Student’s t test (for 2 comparison groups with normal distribution criteria) or 1- or 2-way ANOVA with Tukey’s multiple comparison test (for groups across variables, comparing all groups to each other). For all tests, a P value < 0.05 was considered significant. All experiments were conducted in at least triplicate unless otherwise indicated. No statistical methods were applied to predetermine the sample size. Study approval. Human brain samples were obtained from the UCD Alzheimer’s Disease Center. All animal studies were conducted following protocols approved by the Institutional Animal Care and Utilization Committee (IACUC, Academia Sinica, Taiwan). Data availability. All data presented in this study are either included in this article and its Supplemental Information or are available upon reasonable request to the corresponding author. Uncropped Western blots can be found in the Supplemental Material. Supplemental data and Supporting Data Values are provided with this paper. RNA data SRA files have been deposited in the NCBI’s Sequence Read Archive ( https://www.ncbi.nlm.nih.gov/sra ; PRJNA1017614 and PRJNA1018595).

Results Upregulation of microglial Gal3 in tauopathy. To characterize the role of Gal3 in patients with tauopathy, we analyzed cortical and hippocampal samples from individuals with FTLD and AD ( Supplemental Table 1 ; supplemental material available online with this article; https://doi.org/10.1172/JCI165523DS1 ). As expected, AT8 immune-positive signals were found only in diseased brains. In addition, Gal3 expression was elevated in microglia, as indicated by IbaI-positive cells ( Figure 1, A–C , and Supplemental Figure 1 , A and B). Some Gal3-positive microglia were observed in areas surrounding AT8-positive soma and/or dystrophic neurites ( Supplemental Figure 1A ). To assess whether exposure to pathogenic tau from degenerating neurons activates microglia in tauopathy brains, we first differentiated 3 independent human induced pluripotent stem cell (iPSC) lines into microglia (iMGLs, ( 20 )) ( Supplemental Figure 2 and Supplemental Table 2 ). Treatment of these iMGL with a very low level of recombinant hyperphosphorylated tau (pTau, WT, 1N4R ( 21 ); 50 nM; Supplemental Figure 3 ) for 6 hours significantly upregulated Gal3 protein and transcript levels ( Figure 1, D–G ). No obvious intracellular Gal3 puncta were observed in these Gal3-containing iMGL ( Figure 1E ). Importantly, treating iMGLs with pTau, but not with tau or LPS, for 6 hours recapitulated the abnormal upregulation of genes (APOE, PILRA, ATG7, ANP32A, and GPR141) ( Figure 1H ) and downregulation of genes (PRKCA, ANPS1A, MEF2C, and CECR2) ( Figure 1I ) observed in the microglia of patients with AD ( 10 , 22 ). This suggests that our in vitro model using pTau-treated iMGL is an appropriate model of the AD context. Gal3 early responsive genes. To assess the critical roles of Gal3 in the initial microglial response to pathogenic tau, and its subsequent activation, iMGL prepared from 3 independent control iPSC lines were treated with recombinant pTau and subjected to RNA-Seq ( Figure 2A ). The function of Gal3 was inhibited by treatment with TD139, a cell-permeable Gal3 inhibitor ( 16 ). In total, 7,989 differentially expression genes (DEGs; 3,507 upregulated and 4,482 downregulated) were identified between the pTau and Control treatments, and 758 DEGs (179 upregulated and 579 downregulated) were identified between the pTau-plus-TD139 and pTau treatments ( Figure 2, B and C and Supplemental Table 3 ). Transcriptomic profiling further revealed that pTau-induced canonical pathways in iMGLs closely resembled those observed in AD, as seen in other experimental models of the disease ( Supplemental Figure 4 , A and B). Pathway analysis showed that treatment with pTau elevated multiple pathways (including the activation of protein polyubiquitination, NIK/NFκβ signaling, and inflammatory responses) ( Figure 2D , Supplemental Figure 4D ), and suppressed a few other pathways (including cell division, DNA replication, and metabolic pathways) ( Supplemental Figure 4 , C and E). Gal3 was among the immediate early genes upregulated in iMGL in response to pTau treatment. Inhibition of Gal3 by TD139 normalized a subgroup of DEGs (green dots, Figure 2C ) triggered by pTau. Specifically, 758 DEGs were identified between the pTau-plus-TD139 and pTau treatments and were designated as Gal3-early responsive genes (Gal3-ER genes) ( Figure 2B and Supplemental Figure 5 ). These Gal3-ER genes were enriched in the pathways of extracellular matrix organization, cell adhesion, and immune response ( Figure 2E and Supplemental Figure 4 , F and G). Consistent with the upregulation of Gal3 in iMGL in the initial phase of pTau exposure ( Figure 1G ), the mRNA levels of Gal3 and several proinflammatory cytokines were increased in the iMGL-derived conditioned medium (iMCM) ( Figure 2, F–H ), all being markedly reduced after treatment with TD139, suggesting that Gal3 plays a role in mediating the inflammatory response evoked by pTau in iMGL ( Figure 2, H and I ). Previous studies have identified a cluster of disease-associated microglia (DAM) in an AD mouse models, as well as a group of microglia known as MGnD ( 18 , 23 ). The latter exhibit conserved microglial properties across experimental models of AD, ALS, and multiple sclerosis. Microglia exhibiting upregulated levels of Apoe and Gal3 are known to exhibit neurodegenerative phenotypes, and Gal3 is known to stimulate the TREM2-DAP12 signaling during microglial activation ( 1 , 24 ). Furthermore, a recent study examining the relationship between TREM2 and ApoE4, both risk factors for AD, found exacerbated neurodegeneration in P301S tau mice with TREM2 knockout and ApoE4 expression. The findings indicate that TREM2-independent microgliosis could facilitate tau-mediated neurodegeneration when ApoE4 is present ( 25 ). In the current study, our analysis indicates that, among the DEGs induced by pTau in iMGL, 15.2% of the upregulated DEGs and 16.6% of the downregulated DEGs are shared within DAM genes ( Supplemental Figure 4H ). Similarly, 32.1% of the upregulated DEGs and 52.9% of the downregulated DEGs are shared with MGnD genes ( Supplemental Figure 4I and Supplemental Table 4 ). Through qPCR analysis, we found that pTau upregulated the level of APOE and downregulated the level of TREM2 , while TD139 treatment did not reverse these effects of pTau ( Supplemental Figure 4 , J and K). As a control, treatment with TD139 alone for 6 hours caused 205 upregulated genes and 132 downregulated genes ( Supplemental Figure 4L ). Nevertheless, these genes did not exhibit significant enrichment in specific biological processes as observed in the gene ontology (GO) pathway analysis using DAVID Bioinformatics Resources ( https://david.ncifcrf.gov/ ). Principal component analysis of the overall samples is shown in Supplemental Figure 4M . To assess the expression profiles of Gal3-ER genes, we selected 8 genes based on RNA-Seq analysis of iMGLs treated with pTau for 6 hours ( Figure 2C ). We then examined their expression patterns in iMGLs using quantitative real-time PCR (RT–qPCR) after treating them with pTau for 6, 24, 48, or 72 hours, in either the absence or presence of TD139. The upregulation of all 8 Gal3-ER genes by pTau was sensitive to TD139 treatment following a 48-hour exposure to pTau ( Supplemental Figure 6 ). Half of these Gal3-ER genes ( AQP9 , MMP1 , MMP13 , and TNFSF15 ) remained sensitive to TD139 after 72 hours of treatment with pTau ( Supplemental Figure 6A ), while the other half ( ADGRE1 , SLC1A3 , CCL8 , and TNFSF11b ) were no longer sensitive to TD139 following a 72-hour treatment with pTau ( Supplemental Figure 6B ). These results suggest that Gal3 likely plays a critical role in the early phase following treatment with pTau. Additional pathways may be activated to further regulate the pTau-mediated changes in the transcriptomic profile of iMGLs. pTau-activated microglia release Gal3 in free and EV-associated forms, both contributing to the development of tauopathy. To evaluate the function of microglial Gal3, iMCM collected from pTau-treated iMGL was added to a neuroblastoma cell line harboring pathogenic tau (SH-SY5Y-eGFP-tau-P301L, SY5Y-tau) ( 26 ) ( Figure 3A ). SY5Y-tau represents a neuronal-like cell line expressing neuronal markers ( Supplemental Figure 7 , A and B). The treatment of SY5Y-tau cells with iMCM harvested from iMGL treated with pTau, but not pTau plus TD139, resulted in increases in the levels of misfolded tau (MC1-positive) and the activity of GSK-3β (a dominant kinase for tau) in the cells ( Figure 3B and Supplemental Figure 7 , C and D). Since TD139 is a cell-permeable Gal3 inhibitor, it is likely that TD139 entered iMGL to suppress the activation of iMGL and resulted in iMCM that is less detrimental to SY5Y-tau cells. Notably, the above effect of iMCM was dependent on the presence of iMGL. Without iMGL conditioning, iMGL medium containing pTau alone did not elevate MC1 levels in SY5Y-tau cells ( Supplemental Figure 7 , E and F). To investigate the roles of Gal3 detected in iMCM ( Figure 2H ), SY5Y-tau cells were treated with iMCM collected from pTau-treated iMGL in the absence or presence of a Gal3-neutralizing antibody (Gal3 Ab) that inhibited Gal3 ( Figure 3C ). Interestingly, the inclusion of Gal3 Ab reduced the misfolded tau level without affecting the activity of GSK-3β in SY5Y-tau cells ( Figure 3D and Supplemental Figure 7 , G and H), suggesting that Gal3 released from iMGL may enhance the amount of misfolded tau in SY5Y-tau cells. Consistent with this hypothesis, the addition of Atto-565-conjugated recombinant Gal3 ( Atto565 rGal3) to SY5Y-tau cells significantly upregulated the level of misfolded tau ( Supplemental Figure 7 , I and J). In addition, Atto565 rGal3 was found to colocalize with misfolded tau (MC1-positive) in the cells ( Supplemental Figure 7 , K and L). The inclusion of the Gal3 Ab, but not a control IgG1 antibody, significantly normalized the upregulation of misfolded tau mediated by rGal3 ( Figure 3, E and F ), suggesting that Gal3 in its free form can enter SY5Y-tau cells and plays a direct role in facilitating the accumulation of misfolded tau. Microglia have been reported to facilitate the spreading of tau through the release of exosomes containing pathogenic tau ( 27 , 28 ), we next investigated the role of Gal3 in tau transmission via microglial extracellular vesicles (EVs), including exosomes. The results of nanoparticle tracing analysis (NTA) revealed that pTau markedly enhanced the numbers of EVs released by iMGL, but this effect was not observed in the presence of TD139 ( Figure 3, G and H ), suggesting that Gal3 may participate in EV biogenesis. Consistent with a recent study showing that Gal3 can be recruited into exosomes ( 29 ), treatment with pTau greatly increased the amounts of Gal3 and misfolded tau (MC1-positive) in the EVs of iMGL, which were markedly reduced by TD139 ( Figure 3, I and J ). Most intriguingly, pTau treatment altered not only the numbers but also the protein contents of EVs. Specifically, the level of CD63 was significantly upregulated, while that of CD81 was markedly downregulated by pTau. CD63 and CD81 are membrane proteins that belong to the tetraspanin family and are commonly associated with EVs, particularly exosomes ( 30 ). Analysis of additional EV markers revealed that Alix, but not Tsg101, was also upregulated by the pTau treatment ( Supplemental Figure 8 , A–C). TD139 treatment normalized the pTau-induced upregulation of CD63 and downregulation of CD81 ( Figure 3, I and J ), without affecting the upregulation of Alix. Collectively, Gal3 appears to play an important role in regulating EV biogenesis. To determine whether pathogenic tau and Gal3 are localized inside or outside of the EVs, we conducted a proteinase K resistance assay. Proteinase K treatment effectively cleaved CD11b, which is on the surface of the EVs, while there were no significant reduction on the levels of misfolded tau (MC1-positive) and Gal3 ( Supplemental Figure 8 , D–F). This suggests that both misfolded tau and Gal3 are mainly located inside the EVs. Most intriguingly, a filter assay analysis revealed that the EVs isolated from iMGLs treated with pTau contained significant amounts of prefibrillar oligomers detected with an A11 antibody ( 31 ) ( Figure 3K ). Treatment with TD139 markedly reduced the levels of these oligomeric proteins, suggesting a Gal3-dependent mechanism. We therefore hypothesized that the coexistence of misfolded tau (MC1-positive) and Gal3 in EVs from pTau-treated iMGL might also facilitate the formation of oligomeric aggregates and contribute to the signals detected by A11. To assess the possibility that Gal3 directly interacts with pTau and facilitates the accumulation of pathogenic tau, we employed a thioflavin-S fluorescence assay. Full-length rGal3, but not its N-terminal domain (NTD) or C-terminal carbohydrate recognition domain (CRD), greatly increased the aggregation of pTau into β-pleated–sheet structures ( Figure 3L and Supplemental Figure 9 , A–C). Further, mutations in the aromatic residues (WY/G) of Gal3 ( 32 ), which are critical for its liquid-liquid phase separation (LLPS) of Gal3, did not significantly affect its enhancement of pTau aggregation. Conversely, lactose — but not sucrose — which binds with Gal3 in the CRD, reduced the effect of Gal3 on pTau aggregation ( Supplemental Figure 9 , D–F). Thus, while the CRD is required for this function of Gal3, it is insufficient on its own, as CRD alone did not increase the aggregation of pTau. Further analysis revealed that, during the lag phase of pTau aggregation, the addition of Gal3 significantly increased the aggregation signal. This indicates that Gal3 plays a critical role in facilitating pTau aggregation ( Supplemental Figure 9 , C and F and Supplemental Table 5 ). To assess the effect of these microglial EVs on recipient cells, SY5Y-tau cells were exposed to EVs released by iMGLs treated with the specified conditions. EVs isolated from iMGLs treated with pTau markedly augmented the levels of misfolded tau (MC1 positive) in SY5Y-tau cells, while TD139 treatment to iMGL attenuated this pTau-induced effect ( Figure 3, M and N ). We next analyzed the levels of Gal3 released by EVs and in its free form. Treatment with pTau elevated Gal3 levels in both the EV-associated and free forms. In all 4 conditions tested, more Gal3 was released from iMGLs in its free form than in EVs ( Supplemental Figure 8 , G and H). Apart from the coexistence of both Gal3 and pTau in EVs, our study does not specify whether free-form Gal3 also interacts with pTau in other cellular contexts, such as the extracellular spaces between neurons and microglia, or within recipient cells themselves. Collectively, these results indicate that, upon pTau stimulation, microglia release Gal3 in both its free and EV-associated forms, subsequently exacerbating tauopathy ( Figure 3O ). Gal3-associated microglia in THY-Tau22 mice. In line with our observations in human AD and FTLD brains, we detected an upregulation of Gal3 in microglia located adjacent to neurons containing misfolded tau (MC1-positive) and aggregated tau (AT100-positive) in the CA1 region of THY-Tau22 mice (Tau22; Figure 4, A–C and Supplemental Figure 10A ). Tau22 is an animal model of tauopathy. It is based on the overexpression of a human tau transgene carrying mutations associated with FTLD, and it progressively develops hippocampal tau pathology ( 33 ). Additionally, we found that exogenously added recombinant Gal3 labeled with Atto565 (designated Atto565 rGal3) preferentially bound to the MC1-positive neurons of hippocampal slices of Tau22 mice over those of WT mice ( Supplemental Figure 10B ). These findings suggest a potential involvement of extracellular Gal3 in mediating abnormal interactions between microglia and neurons in the context of tauopathy. To investigate the transcriptomic profile of Gal3-positive microglia by single-cell RNA-Seq (scRNA-Seq), hippocampal tissues of WT and Tau22 mice (12.5 months old, n = 8 in each group) were harvested and subjected to mechanical dissociation. After myelin removal, the cell suspensions were stained with an anti-CD11b antibody — a myeloid cell marker — and isolated by FACS. The CD11b-sorted cells were subjected to scRNA-Seq by using the 10× Genomics platform ( Figure 4D ). We conducted quality control ( Supplemental Figure 11 , A–D), principal component analysis, and dimensionality reduction using uniform manifold approximation and projection (UMAP) ( Supplemental Figure 11 , E and F). The isolated cells, mostly microglia, were assigned to each cluster based on the expression of established markers from the PanglaoDB database ( 34 ). In total, we identified 12 distinctive clusters of microglia, 3 clusters of monocytes, and 1 cluster of macrophages, granulocytes and epithelial cells ( Figure 4E ). Compared with WT mice, Tau22 mice had more microglia in 7 Clusters ( 1 , 2 , 5 , 6 , 9 , 11 , 12 ) and fewer microglia in 2 Clusters ( 3 , 10 ). 3 Clusters ( 3 , 7 , 8 ) had no major change in their numbers (i.e., the change in microglia population was less than 1%; Supplemental Figure 11G ). The DEGs (log 2 -fold–change threshold of at least 0.25) identified by Seurat 4.0 were subjected to GO analysis by using DAVID Bioinformatics Resources ( 35 , 36 ) to highlight the major biological processes and marker genes associated with each microglial cluster ( Supplemental Figures 12–14 ). Interestingly, 6 of the Clusters ( 2 , 5 , 6 , 9 , 11 , 12 ) with increased cell numbers are enriched with translational processes. For pseudotime analysis to identify the longest trajectory of cell clusters based on the difference on gene profile from their origin, Cluster 1 and Cluster 2 with the highest cell number were appointed as the origins. The longest pseudotime distance for Cluster 1 and Cluster 2 are Clusters 10 and 9, respectively ( Supplemental Figure 11H ). Gal3 ( Lgals3 in mice)-associated microglia (GAM) were enriched in Cluster 9 and could also be observed in several other clusters to a lesser extent ( Figure 4, F and G ). We next investigated the involvement of these microglial clusters in the pathway associated with EVs, such as exosomes, multivesicular bodies, and endosomal pathways. Interestingly, we found that Clusters 4 and 9 were enriched with cellular components related to exosomes ( Supplemental Figure 15A ). Among the exosome components, CD63 and CD9 were previously identified as DAM genes ( 23 ) and notably exhibited high expression levels in Cluster 9 ( Supplemental Figure 15B ). The finding prompted us to compare microglial clusters with DAM genes using hierarchical clustering analysis. This revealed that Cluster 9 exhibited the closest similarity to DAM, followed by Clusters 7 and 8 ( Supplemental Figure 15C ). Additionally, when comparing microglial clusters with microglia of neurodegenerative diseases (MGnD) genes ( 18 ), we found that Cluster 9 also displayed highest similarity to MGnD, followed by Clusters 8 and 12 ( Supplemental Figure 15D ). These findings suggest that the Lgals3- enriched Cluster 9 may play a pivotal role in pathways associated with EVs as evidenced by the iMGL study, and may potentially contribute to the intricate interplay between microglial activity and tau transmission. Because not all microglia express Lgals3 , we aimed to specifically characterize the Lgals3 -positive and Lgals3 -negative microglia. We define Lgals3 -positive microglia as cells with Lgals3 expression levels greater than or equal to 1 average log Unique Molecular Identifier (UMI) count, which serves as a reference point. Compared with Lgals3 -negative microglia in Tau22 mice, which had a zero average log UMI count, Lgals3 -positive cells exhibited 812 DEGs, designated as GAM genes ( Figure 4H , Supplemental Figure 16A , and Supplemental Table 6 ). Consistent with the findings in human iMGLs ( Figure 3I ), we found that Cd63 and Cd81 were both GAM genes in the microglia isolated from Tau22 mice, and were upregulated and downregulated, respectively ( Figure 4H ). Moreover, GO analysis revealed enrichment of GAM genes in cellular components such as exosome, multivesicular body, and late endosome ( Supplemental Figure 16 , B and C), further supporting the role of Gal3 in the regulation of EV-related pathways. Further examination via Ingenuity Pathway Analysis (IPA) showed that 19 of the 396 upregulated GAM genes and 13 of the 416 downregulated GAM genes were direct downstream target genes of Lgals3 ( Figure 4, H and I ). Consistent with this prediction, immunofluorescence staining demonstrated that CD68, a gene directly downstream of Gal3 ( Figure 4I ) was expressed at a higher level in Gal3-positive microglia compared with that of the Gal3-negative microglia, in Tau22 mice ( Figure 4, J and K ). Further analyses of GAM genes using the Reduce Visualize Gene Ontology (REVIGO) tool ( 37 ) highlighted the activation of translation and ribosome activities, inflammation, and the immune system, followed by the processes of ATP production, apoptosis, chemotaxis, protein folding, and p53 signaling ( Figure 4L ). Interestingly, the 812 GAM genes shared 172 genes with the AD DAM genes ( 23 ) ( Supplemental Figure 17A ), including upregulated microglial activation genes (such as Clec7a , Cd68 , Csf1 , Apoe , and Cybb ) and downregulated microglial homeostatic genes (such as P2ry12 , TMEM119 , Csf1r , Hexb, and Slc2a5 ). Collectively, GAM is a subset of microglia with several features similar to those of DAM (e.g., the enhanced inflammatory responses and protein translation processes) and some unique only to GAM (including the protein folding process, energy metabolism, transcription, and specific translation processes) ( Figure 4L and Supplemental Figure 17 , B–F). Additionally, we conducted a comparative analysis between GAM in Tau22 mice and the DEGs in pTau-induced iMGLs, as well as the effects of Gal3 inhibition with TD139, to explore their potential relevance for future cross-species studies. Among the identified conservation of canonical pathways, pathways such as hepatic fibrosis signaling, Rho family GTPases, neuroinflammation, integrin, and IL8 signaling, were suppressed in the presence of TD139 ( Supplemental Figure 18 , A and B). Loss of Gal3 protects against tauopathy. To assess the importance of Gal3 in tauopathies in vivo, we crossed Tau22 mice with Gal3 knockout mice (Tau22/ Lgals3 –/– , Figure 5A ). The knockout of Gal3 reduced the levels of misfolded tau (MC1-positive), aggregated tau (AT100-positive), and phosphorylated tau (AT8-positive) in the hippocampal CA1 region of Tau22, as assessed by immunofluorescence ( Figure 5, B–E and Supplemental Figure 19 , A and B) and Western blot analyses ( Figure 5, F and G ). We next investigated the major phosphatase, PP2A, and kinases that regulate the abnormal hyperphosphorylation of tau ( 38 ). Compared with Tau22/ Lgals3 +/+ mice, Tau22/ Lgals3 –/– mice exhibited reduced levels of the inactive/demethylated form of PP2A and decreased kinase activities of GSK-3β and CaMKII-α ( Figure 5, F and G ). Such alterations in PP2A and kinase activities may result in the reduction of tau phosphorylation. Importantly, the loss of Gal3 also prevented the learning and memory deficits present in Tau22/ Lgals3 +/+ mice ( 33 ) to a great extent, as assessed by the Morris water maze test ( Figure 5, H and I ). Consistent with the GAM gene analysis ( Figure 4, H and I ), the number and level of CD68-positive microglia in Tau22/ Lgals3 –/– mice were indeed lower than those in Tau22/ Lgals3 +/+ mice ( Figure 5, J and K ), suggesting a key role of Gal3 in microglial activation. Given that synaptic loss is a feature of tauopathy that is also presented in Tau22 mice ( 33 ), we performed immunofluorescence staining of synapses at the CA1 region using VGLUT1 and Homer1 as presynaptic and postsynaptic markers, respectively. Our data showed that Gal3 knockout rescued the number of synapses assessed by the colocalization of VGLUT1 and Homer1 ( Figure 5, L and M and Supplemental Figure 20 , A and B). This finding suggests that GAM mediates the loss of synapses in Tau22 mice. To further delineate the protective role of Gal3 depletion in tauopathy, we analyzed the gene expression profiles of the hippocampi of Tau22/ Lgals3 –/– mice and corresponding controls using bulk RNA-Seq. In total, 3,770 DEGs were identified between Tau22/ Lgals3 +/+ and WT mice, and 868 DEGs were identified between Tau22/ Lgals3 –/– and Tau22/ Lgals3 +/+ mice ( Supplemental Figure 21 , A–D). In particular, the knockout of Gal3 normalized 348 DEG genes between Tau22/ Lgals3 +/+ and WT mice ( Figure 6, A and B and Supplemental Table 7 ). Further GO enrichment analysis revealed that the upregulated DEGs of Tau22/ Lgals3 +/+ versus WT mice were enriched in multiple processes, including metabolic processes, oxidative reduction processes, and immune system processes ( Figure 6, C and D ). Importantly, the downregulated DEGs by Lgals3 deletion within the context of Tau22 were primarily enriched in immune responses and the production of cytokines and chemokines ( Figure 6E ). Conversely, the downregulated DEGs in Tau22/ Lgals3 –/– versus Tau22/ Lgals3 +/+ mice were enriched in processes including nervous system development, protein phosphorylation, synapse assembly, and learning ( Supplemental Figure 21 , E and F). No specific enriched processes were identified for the upregulated DEGs in Tau22/ Lgals3 –/– versus Tau22/ Lgals3 +/+ mice. These findings are consistent with what were observed in human iMGLs, confirming that Gal3 plays a principal role in governing the microglia-mediated immune response in tauopathy. We next categorized these DEGs by their enriched cell type based on the Tabula Muris Consortium database ( 39 ) ( Supplemental Figure 21D ), and, as predicted, we found that the largest population of DEGs identified between Tau22/ Lgals3 –/– and Tau22/ Lgals3 +/+ mice was enriched in microglia (21.3%; Supplemental Figure 21D ). The expression of 4 microglia-enriched DEGs (i.e., Clec7a , Tlr2 , Cd68, and Cxcl16 ), which were also identified as GAM genes by scRNA-Seq analyses ( Figure 4H ), was validated by RT–PCR ( Figure 6F ). Accordingly, the expression of the Dectin-1 protein, encoded by Clec7a , a direct downstream target of Gal3 ( Figure 4I ), was significantly downregulated by the deletion of Gal3 in Tau22 mice ( Figure 6, G and H ). Nonetheless, although the levels of Apoe and Trem2 were upregulated in Tau22/ Lgals3 +/+ mice, the deletion of Gal3 did not reverse their expression ( Supplemental Figure 21 , G and H). Importantly, when compared with the Gal3-enriched Cluster 9 microglia from the scRNA-Seq analysis ( Figure 4G ), bulk RNA-Seq analysis showed that 19 upregulated genes and 1 downregulated gene exhibited reversed expression levels in Tau22/ Lgals3 –/– mice ( Supplemental Figure 21 , I and J). Interestingly, the removal of Gal3 also ameliorated the increase in the number of GFAP-positive astrocytes in Tau22 mice ( Supplemental Figure 22 , A and B). This observation supports our hypothesis that Gal3 plays an important role in microglia activation because previous studies had demonstrated that activated microglia may secrete immune mediators (such as IL1α, TNFα, and C1q) that contribute to the transformation of neurotoxic reactive astrocytes (A1 astrocytes) ( 40 ). Additionally, astrogliosis has been reported in tauopathy ( 41 ). Transcriptomic analysis further revealed that the levels of several genes associated with A1 astrocytes (including Fbln5 , Ugt1a1 , Gbp2 , C3, and Srgn ) were reduced in Tau22/Lgals3 –/– mice ( Supplemental Figure 22C ). Gal3 mediates Aβ-induced tau pathology. Ample evidence suggests that amyloid plaque formation precedes tau pathology in AD. Dystrophic neurites with phosphorylated tau (AT8-positive) have been detected in mouse models of amyloid pathology, such as APP/PS1 and APP-knockin mice ( 42 , 43 ), but the involvement of microglia remains largely unclear. Here, we showed that the treatment of iMGL with fibrillar Aβ for 6 hours significantly upregulated Gal3 ( Figure 7, A–D ). Additionally, we conducted a morphological analysis of iMGL and assessed the expression levels of Gal3. While a portion of Gal3-expressing iMGL appeared to display a round morphology, we found no significant correlation between Gal3 expression and round-shaped morphology ( Supplemental Figure 23 , A–C). Conversely, we observed a positive correlation between Gal3 expression and the size of iMGL in the presence of fibrillar Aβ ( Supplemental Figure 23D ). Furthermore, fibrillar Aβ also triggered the secretion of Gal3 ( Figure 7E ). Additionally, we observed the upregulation of proinflammatory genes as an early response to fibrillar Aβ by iMGL ( Figure 7F ). To investigate the effects of Gal3 on Aβ-induced tau pathology, we collected iMCM from iMGL exposed to fibrillar Aβ with or without TD139. iMCM from fibrillar Aβ-exposed cells induced a higher level of misfolded tau (MC1-positive) in SY5Y-tau cells, which was prevented by inhibiting Gal3 in iMGL using TD139 ( Figure 7, G and H ). We next crossed APP/PS1 mice with Gal3 knockout mice and found that the levels of phosphorylated tau present in the dystrophic neurites of the cortex and hippocampus of APP/PS1/ Lgals3 –/– mice were lower than those of APP/PS1/ Lgals3 +/+ mice (11 months old, symptomatic stage; Figure 7, I–K ). Moreover, the level of Aβ plaques were also lower in APP/PS1/ Lgals3 –/– mice ( Supplemental Figure 24 ), supporting the conclusion that Gal3-positive microglia are essential for the Aβ-tau axis in AD.

Discussion Increasing evidence shows that pathological tau is transmitted and propagated between neurons ( 44 – 46 ). The secreted tau in its physiological state is dephosphorylated, while the pathological tau is phosphorylated and misfolded ( 47 , 48 ). Nonetheless, both pTau and nonphosphorylated tau have been found in the CSF of patients with AD, indicating that both forms of tau can be secreted ( 49 , 50 ). As the primary phagocytic cells in the brain, microglia play a major role in the clearance of extracellular tau. An early study demonstrated that microglia clear pTau slightly faster than nonphosphorylated tau ( 51 ). However, whether microglia exhibit a preference for taking up pTau remains unclear. Here, we present evidence demonstrating distinctive effect of pTau in recapitulating the dysregulation of microglial genes that have been observed in patients with AD ( Figure 1, H and I ). Interestingly, such dysregulation was not observed under experimental conditions involving nonphosphorylated tau and LPS. Our findings collectively suggest that Gal3-associated microglia facilitate the transmission and aggregation of misfolded tau to recipient cells ( Figure 8 ). When encountering pathological tau, microglia become active and release Gal3 into the extracellular space either directly or via EVs ( Figure 3O ). Under the tested conditions, we found that Gal3 directly facilitated the aggregation of pTau into β-pleated–sheet structures ( Figure 3L ). This interaction between pTau and Gal3 may occur in EVs and/or the extracellular space between microglia and neurons. It is plausible that Gal3 may interact with pTau and serve as an opsonin to enhance the uptake of pTau by other brain cells ( 52 , 53 ). Most importantly, we showed that Gal3 released by activated microglia in either free form or EV-associated form facilitated the accumulation of misfolded tau in SY5Y-tau cells ( Figure 3, D and N ). Moreover, the deletion of Gal3 markedly reduced the extent of tauopathy in THY-Tau22 mice ( Figure 5, A–E ). Taken together, the results indicate that the upregulation of Gal3 in microglia plays a critical role in the propagation of tauopathy. Another function of Gal3 reported in the present study is that Gal3 may contribute to the altered release and properties of EVs derived from pTau-treated microglia ( Figure 3, G–J ). This is of great interest because EVs are a route of tau transmission ( 45 ). A previous study demonstrated that the deletion of Trem2 enhances the transmission of tau via exosomes ( 27 ). Our findings indicate that pTau treatment significantly reduces the level of TREM2 in iMGL, and these iMGL did release more EVs. Intriguingly, Gal3 inhibition suppressed the abnormal release of EVs triggered by pTau, without affecting the levels of TREM2 ( Supplemental Figure 4K ). This suggests that Gal3 might act either as a key mediator downstream of TREM2 or within an independent pathway parallel to TREM2 in regulating the EV pathway. Consistent with the results of our IPA of GAM genes indicating that CD63 (a marker of EVs resulting from endosomes ( 30 )) may be a direct downstream target of Gal3 ( Figure 4, H and I ), the level of CD63 in EVs derived from pTau-treated iMGL was markedly elevated ( Figure 3, I and J ). Furthermore, scRNA-Seq analysis showed enriched levels of EV-associated components, including exosomes, late endosomes, and multivesicular bodies, in GAM ( Supplemental Figure 16 , B and C). Because we employed a low concentration of pTau (50 nM) to induce Gal3 upregulation in iMGLs and microglial activation in the present study, immunofluorescence staining could not reliably detect pTau signals in iMGLs after a 6-hour treatment. It is possible that at this low concentration, any pTau taken up by iMGLs has been either secreted out or digested within the 6-hour treatment period. Interestingly, although we could not observe pTau in iMGL cells using the immunofluorescence staining method, we successfully detected pTau in the EVs isolated from pTau-treated iMGL, as shown in Figure 3 I. This suggests that iMGL may secrete pTau through EVs. These observations together suggested that microglia enhanced the release and production of EVs in a Gal3-dependent manner when encountering pTau. Because Gal3 regulates the intracellular trafficking of cellular protein(s) by interacting with Alix (an accessory protein associated with the endosomal sorting complex required for transport [ESCRT] machinery) ( 54 ) and Alix plays a key role in apical exosome release ( 55 ), it is plausible that cytosolic Gal3 may alter the transport of different cellular proteins to their destinations, including exosomes and EVs, by binding to Alix in pTau-treated microglia. Further investigation is warranted to delineate the underlying mechanism. The function of Gal3 in EV biogenesis has not been clearly defined. A recent study showed that HIV infection significantly increases the level of Gal3 in exosomes derived from DCs and that Gal3 mediates HIV transmission to T cells by DC exosomes ( 56 ). Our data showed that the presence of Gal3 in the same EV compartment as pTau facilitated the formation of pathogenic tau oligomers and increased the accumulation of misfolded tau in the recipient neuronal-like cells ( Figure 3, I–N ). Given that TD139 is membrane-permeable, it may inhibit not only intracellular and extracellular Gal3, but also EV Gal3. The inhibition of Gal3 by TD139 in tauopathy not only prevented the EV-mediated spread of misfolded tau by microglia but also reduced the amount of EV Gal3, which might target neighboring neurons to promote the accumulation of pathogenic tau. It is a long-held view that activated microglia and innate immune factors contribute significantly to the pathogenesis of AD and tauopathy. Although microglial depletion has been shown to reduce AD pathology ( 57 ), microglial activation is complex and can produce both beneficial and detrimental effects on neuronal functions. In the present study, we identified a group of Gal3-associated microglia (GAM) in an animal model of tauopathy (THY-Tau22) that facilitate the progression of tauopathy ( Figure 4 ). Notably, these GAM were located in close proximity to neurons containing misfolded tau in the hippocampi of Tau22 mice ( Figure 4B ). It appears that microglia respond to signals (e.g., released pathogenic tau) from pTau-containing neurons by upregulating Gal3 and activating microglia. Accordingly, the addition of pTau to iMGL evoked Gal3 upregulation and the production of proinflammatory cytokines ( Figure 2, G and H ). In addition, GAM may also increase the phosphorylation of tau via the secretion of proinflammatory factors that regulate the activities of kinases and phosphatases in neurons, as reported elsewhere ( 38 ). Collectively, the results indicated that the genetic deletion of Gal3 in Tau22 mice ameliorated major disease-related symptoms ( Figure 5 ), supporting the importance of GAM in tauopathy. The results of the present study and many other reports suggest that Gal3 plays a critical role in a wide variety of neurodegenerative diseases, including AD, Huntington’s disease, and Parkinson’s disease ( 1 , 15 , 16 , 58 ). It is likely that Gal3 mediates a conserved activation program and a stimulus-specific program in microglia while encountering various stimuli. A comparison of GAM genes and AD DAM genes revealed 172 common genes between these groups ( Supplemental Figure 17 , A and B). Both the GAM and DAM genes highlighted translation, ribosome assembly and biogenesis, TNF production, and the inflammatory response, mitochondrial electron transport, and antigen processing and presentation ( Supplemental Figure 17D ) ( 23 ). Interestingly, GAM genes were also enriched in pathways related to apoptosis, protein folding, ATP metabolic processes and IFN-γ signaling ( Supplemental Figure 17E ), indicating that the GAM genes reported here may regulate a subset of microglial activities specific to pTau. Notably, abnormal activation of the ribosomal pathway has also been reported in amyloid plaque–associated microglia ( 59 ), which are activated with Gal3 signals ( 1 , 19 ). These findings support our hypothesis that microglial Gal3 is an important regulator of the promotion of tauopathy by Aβ ( Figure 7 ). In summary, our findings demonstrate that Gal3 plays a critical upstream role in tauopathy by controlling microglial activation and the propagation of tauopathy. Herein, we demonstrated that the exposure of microglia to pathological tau from neurons led to the immediate upregulation of Gal3 in a subset of microglia, GAM, and a network of genes initiating the activation of microglia, which subsequently caused synaptic loss and neurodegeneration in tauopathy. Most intriguingly, microglia were shown to release Gal3 in both its free and EV-associated forms, facilitating the accumulation of misfolded tau in recipient neuronal cells. Direct interaction between Gal3 and misfolded tau may occur in EVs and/or the extracellular space and could greatly enhance the formation of pathogenic pTau. The removal of Gal3 markedly reduced tau pathology in mouse models of tauopathy and AD. The lack of a Gal3-dependent TREM2/APOE response indicates that Gal3 affects EV dynamics and tauopathy without directly intersecting with the TREM2/APOE signaling pathways. By focusing on Gal3 inhibition, therapeutic strategies could potentially modulate EV dynamics and microglial activation without altering the TREM2/APOE pathway, which could be advantageous in cases where modulation of TREM2/APOE is neither desirable nor effective.

Alzheimer’s disease is characterized by the accumulation of amyloid-β plaques, aggregation of hyperphosphorylated tau (pTau), and microglia activation. Galectin-3 (Gal3) is a β-galactoside–binding protein that has been implicated in amyloid pathology. Its role in tauopathy remains enigmatic. Here, we showed that Gal3 was upregulated in the microglia of humans and mice with tauopathy. pTau triggered the release of Gal3 from human induced pluripotent stem cell–derived microglia in both its free and extracellular vesicular–associated (EV-associated) forms. Both forms of Gal3 increased the accumulation of pathogenic tau in recipient cells. Binding of Gal3 to pTau greatly enhanced tau fibrillation. Besides Gal3, pTau was sorted into EVs for transmission. Moreover, pTau markedly enhanced the number of EVs released by iMGL in a Gal3-dependent manner, suggesting a role of Gal3 in biogenesis of EVs. Single-cell RNA-Seq analysis of the hippocampus of a mouse model of tauopathy (THY-Tau22) revealed a group of pathogenic tau-evoked, Gal3-associated microglia with altered cellular machineries implicated in neurodegeneration, including enhanced immune and inflammatory responses. Genetic removal of Gal3 in THY-Tau22 mice suppressed microglia activation, reduced the level of pTau and synaptic loss in neurons, and rescued memory impairment. Collectively, Gal3 is a potential therapeutic target for tauopathy. Gal3 participates in the regulation of extracellular vesicles, the promotion of pathogenic tau fibrillation, and the activation of microglia.

Author contributions JJS designed, performed experiments, analyzed data and wrote the manuscript with the assistance of CWL, HLL, and TNAN. HMC bred and maintained animals used in the experiments. YMC, HTL, TMK, and SYC designed, performed, and analyzed scRNA-Seq. FLC and HCK maintained and provided iPSCs used in the experiments. ML, HTH and MHK synthesized and provided pTau used in the study. FTL provided Lgals3 –/– mice, conceptual advice, and materials. DB and LB provided THY-Tau22 mice used in the study. LWJ provided human brain samples. HLC, YCS, and JRH provided recombinant Gal3 proteins in the study. YC designed and supervised the study and edited the manuscript. Supplementary Material

This research was supported by Academia Sinica and Ministry of Science and Technology, Taiwan [AS-GC-110-MDO5, AS-GC-110-05, MOST 107-2320-B-001 -013 -MY3]. The “Alzheimer & Tauopathie” laboratory is supported by programs d’investissements d’avenir LabEx (excellence laboratory) DISTALZ (Development of Innovative Strategies for a Transdisciplinary approach to ALZheimer’s disease). ML, HTH, and MHK were supported by grants from the National Institutes of Health (R41AG057274 and R01AG062435). LWJ was supported by UCD Alzheimer’s Disease Research Center – US NIH grant P30 AG10129. We thank Academia Sinica Core Facility and Innovative Instrument Project (AS-CFII108-113) for cell sorting service. We are grateful to Yung-Feng Liao (Institute of Cellular and Organismic Biology, Academia Sinica) for the SH-SY5Y-eGFP-tau-P301L cell line. We thank UCD Alzheimer’s Disease Center for providing human brain specimens. We thank Peter Davies (The Feinstein Institute for Medical Research, New York, USA) for sharing the MC1 antibody. We thank Amanda McQuade (University of California, Irvine, USA) for technical advices on the preparation of iMGL, Christina Ising (University Hospital of Bonn, Germany), Ruei-Yu He, and Yung-An Huang (Institute of Chemistry, Academia Sinica, Taiwan) for technical advice on the biochemistry assay. Graphical abstract was created with BioRender.com. 11/21/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e165523

oa_package/c1/eb/PMC10786694.tar.gz

PMC10786695

37988162

Introduction One in six pregnancies is affected by some form of gestational diabetes (GD), a prevalence that is steadily rising in the context of the worldwide epidemics of obesity and early diabetic onset ( 1 ). Of these cases, 80% are classified as gestational diabetes mellitus (GDM), which present an episode of glucose intolerance associated with the metabolic disruptions occurring during the second or third trimester of the pregnancy. The remaining cases arise earlier in pregnancy and are referred to as diabetes in pregnancy. They include women with prediabetes who become diabetic during pregnancy or women with overt type 1 or type 2 diabetes who have difficulty regulating their glycemia during pregnancy. When untreated, this condition has direct short-term consequences on pregnancy, with poor maternal and offspring outcomes, including increased risk of retinopathy, nephropathy, and preeclampsia for the mother and congenital cardiac malformation, macrosomia, and jaundice for the newborn ( 2 , 3 ). Early diagnosis and advances in maternal glucose control have greatly mitigated the perinatal consequences of GD for the mothers and the offspring ( 2 – 4 ). However, these advances have only marginally affected its long-term consequences. Exposure to hyperglycemia in utero remains associated with long-term morbidities in adult offspring, including increased risk of metabolic diseases, atherosclerosis, and cardiovascular diseases ( 5 – 9 ). The mechanisms driving this pathological transmission across generations remain unknown. Here we establish two reliable genetic and pharmacological mouse models that emulate the adverse perinatal and long-term consequences associated with GD in human. In these models, we reveal the existence of a lasting hematopoietic memory associated with GD that contributes to the increased susceptibility to atherosclerosis of the adult offspring. Mechanistically, we show that the acquisition of the hematopoietic memory during diabetic gestation relies on the activity of the advanced glycation end product receptor (AGER) pattern recognition receptor and the nucleotide binding and oligomerization domain-like receptor family pyrin domain-containing 3 (NLRP3) inflammasome to promote placental inflammation. Finally, we show that this memory is associated with epigenetic changes in hematopoietic stem and progenitor cells (HSPCs) and progenitor cells. Together, our results highlight important pathways that connect maternal and fetal health to adult pathologies. Notably, they demonstrate that the hematopoietic system can acquire a lasting memory of gestational events and that transgenerational hematopoietic memory could be an important “effector” of pathologies in adulthood.

Methods Mice. Wild-type C57BL/6J (B6.SJL-Ptprca, strain 000664), Pepcb/BoyJ (strain 002014), Ins2 Akita/J (strain 003548), Ager –/– (strain 032771), Apoe –/– (strain 002052), and Nlrp3 –/– (strain 021302; gift from R. Marsh, CCHMC) mice were purchased from The Jackson Laboratory. Depending on the experimental design, mice were maintained on a Ctrl diet with 13 kcal% fat (5010; Lab Diets) or switched to HFD with 60 kcal% fat (D12492; Research Diet Inc.). In the pharmacological model, females were fasted for 6 hours before STZ treatment (60 mg/kg body weight). Animal breeding was performed over the weekend, leading to approximation of the gestational age of ±1 day. Reagents and resources. A list of reagents and resources is presented in Supplemental Table 2 . Blood glucose measurement. Blood was sampled from the tail, and blood glucose was measured using an Accu-Chek Performa Glucometer. Nonfasting glucose was measured at fixed times between 10:00 am and 12:00 pm. For intraperitoneal ITT and oral GTT, adult offspring were fasted 6 and 12 hours, respectively, before receiving insulin (0.5 U/kg body weight) by intraperitoneal injection or glucose (2 g/kg body weight) per oral gavage. Blood glucose was measured before treatment to determine fasting glucose levels and every 15 minutes after insulin/glucose treatment. In vivo treatment. BM from 8-week-old Ctrl and WT STZ offspring along with diabetic Ins2 Akita/+ males was analyzed by flow cytometry 3 days after intraperitoneal treatment by LPS (35 μg/mouse) or pIC (10 μg/g body weight). 5-azadC intraperitoneal injection (0.25 mg/kg body weight) was performed 3 times a week (Monday, Wednesday, Friday) starting at 5 weeks of age. BM transplantation. For competitive BM transplantation assay, 250 FACS-purified HSCs (HSC-SLAM) were mixed with 3 × 10 5 unfractionated whole BM cells from Pepcb/BoyJ mice (CD45.1 + ) and retroorbitally injected into lethally irradiated CD45.1 + Pepcb/BoyJ recipient mice (11 Gy, delivered in split doses 3 hours apart). Hematopoietic reconstitution was monitored by flow cytometry on peripheral blood collected retroorbitally. HSPC donor chimerism was assessed in the BM by flow cytometry 20 weeks after transplantation. For noncompetitive transplantation, 5 × 10 6 whole BM cells were retroorbitally transplanted into lethally irradiated CD45.1 + Pepcb/BoyJ or Apoe –/– recipients. Flow cytometry. BM preparation and cell-surface staining were performed as described previously ( 16 ). In brief, BM cells were isolated by crushing long bones and hips, treated with red blood cell lysis buffer (150 mM NH 4 CL and 10 mM KHCO 3 ), and washed with staining media (SM) (HBSS with % fetal bovine serum); 8 × 10 6 unfractionated BM cells were stained for flow cytometry analysis. For sorting analysis, pools of unfractionated BM cells were purified by Ficoll separation (Histopaque-11191, MilliporeSigma 11191), then enriched for c-Kit + cells using magnetic beads/autoMACS separation (Miltenyi Biotec). BM cells were stained with unconjugated rat lineage-specific antibodies (Ter119, Mac1, Gr-1, B220, CD5, CD3, CD4, CD8) or biotinylated lineage-specific antibodies (Gr-1, Mac-1, B220, Ter119, and CD3e), followed by staining with goat anti-rat PE-Cy5 secondary antibodies or streptavidin–eFluor 6450 secondary antibodies, respectively. Cells were then stained using c-Kit-APC-eFluor 780, Sca-1-PB, CD48-BV711 or CD48-PerCP-Cy5.5, CD150-PE or CD150-BV650, Flt3-biotin or Flt3-PE, CD62L-BV510, FcgR-BV510 or FcgR-PerCP-eFluor 710, CD34-FITC, EPCR-PerCP-eFluor 710, CD41-BV605, CD105-APC, and CD127-BV785 antibodies. Secondary staining was performed with streptavidin-BV711/ streptavidin-PE-Cy7. The Zombie NIR Fixable Viability Kit (BioLegend) was used for dead cell exclusion. For Hoechst 33342/Ki67 staining, surface-stained cells were fixed and permeabilized using a Cytofix/Cytoperm kit (BD) before staining with a PE-conjugated anti-Ki67 antibody and Hoechst 33342 (5 μg/ml). Flow cytometry reagents are presented in Supplemental Table 3 . Cell sorting was performed on a FACSAria II (BD Biosciences). Analyses were performed on a 5 laser (16 ultraviolet [355 nm] channels, 16 violet [405 nm] channels, 14 blue [488 nm] channels, 10 yellow-green [561 nm] channels, 8 red [635 nm] channels) Aurora Spectral Flow Cytometer (Cytek Biosciences). Data analysis was performed using FlowJo (BD Biosciences) software. Tissue preparation for histology and flow cytometry. Hearts were removed, fixed in 10% buffered formalin, treated on sucrose gradient, and embedded in OCT before being serially sectioned and stained with H&E dyes. Aortic valves were evaluated and graded by one investigator blinded to the study protocol. Prior to aorta harvest, hearts were perfused with 10 mL of PBS to limit blood contamination. Isolated aortas were mechanically dissociated before being incubated in an enzyme cocktail (450 U/mL collagenase type I, 125 U/mL collagenase type XI, 60 U/mL hyaluronidase type I-S, and 60 U/mL DNase I grade II) at 37°C for 50 minutes ( 62 ). Single-cell suspensions were filtered through a 70 μm filter and stained for flow cytometry analysis. For placenta studies, placentas were harvested from dams at gestation days 17–18 and mechanically dissociated for 1 to 2 minutes in 1 mL of cold StemPro Accutase Enzymatic Solution before incubation at 37°C for 35 minutes ( 63 ). After filtration through a 70 μm strainer and treatment with red blood cell lysis buffer, cells were washed and stained for flow cytometry. Confocal immunofluorescence microscopy. From 500 to 1,000 HSCs were directly sorted on Polysine Microscope Slides (Thermo Fisher Scientific, P4981-001). Cells were allowed to adhere to the glass slide for 10 minutes, fixed using 4% paraformaldehyde (BD Cytofix, 554655) for 15 minutes, and permeabilized with 0.2% Triton X-100 (Sigma-Aldrich, T9284) for 20 minutes at room temperature. After blocking with 10% donkey serum (Sigma-Aldrich, D9663) for 20 minutes, slides were successively stained with anti-FOXO3 (EMD Millipore, 07-1719, dilution: 1/200) and Alexa Fluor 568–conjugated goat anti-rabbit (Thermo Fisher Scientific, A11011, dilution: 1/500) IgG (H+L) antibodies. After wash, slides were mounted using Gold Antifade Mounting Media with DAPI (Invitrogen, 536939) and analyzed with a LSM 710 confocal microscope system equipped with an inverted microscope (Observer Z1, Zeiss) using ×60 magnification at 1.2 AU pinhole 564 to give an optical section thickness of 0.39 μm images. Image analyses were performed using Imaris (Oxford Instrument, version 9.5) and NIS-elements (Nikon, version 5.20.02) software. BMDMs. One million whole BM cells were plated on a 24-well plate in complete DMEM media (supplemented with 10% FBS, 1% penicillin/streptomycin in the presence of recombinant mouse macrophage colony-stimulating factor [M-CSF]; 20 ng/mL, BioLegend). After 6 days in culture, BMDMs were activated with recombinant murine IFN-γ, (20 ng/mL) and 20 ng/mL LPS (20 ng/mL) or media-only Ctrl. Cell numbers were determined 3 hours and 24 hours after activation using a hematocytometer and trypan blue for dead cell exclusion. Immunoblot analysis. FACS-sorted LSK (5 × 10 4 ) or AutoMACS-enriched c-Kit + cells (2 × 10 5 ) were used for immunoblotting analyses. Whole-cell lysates were prepared in 1× RIPA lysis buffer containing protease and phosphatase inhibitor cocktails. Cell lysates were resuspended in Laemmli buffer, boiled for denaturation, electrophoresed through 4%–15% SDS-PAGE gradient gel, and transfered to a PVDF membrane. Western blot analysis was performed with anti-DNMT1, anti-DNMT3A, and anti-ACTB antibodies and revealed with anti-mouse or anti-rabbit IgG secondary antibodies tagged with horseradish peroxidase (HRP). qRT-PCR analysis. Quantitative reverse-transcription PCR (qRT-PCR) was performed using total RNA isolated from approximately 1 × 10 5 cells. RNA was treated with DNase I and reverse transcribed using the SuperScript III Kit and random hexamers (Invitrogen). The cDNA equivalent of 200 cells per reaction was analyzed in triplicate on a 7900 HT Fast Real-Time PCR System (Applied Biosystems) using SYBR green reagents (Applied Biosystems) and gene-specific primers. DNA methylation analysis. LSK cells were isolated by FACS, and DNA was extracted using the QIAamp DNA Mini Kit (QIAGEN) per the manufacturer’s instructions. Sequencing was performed at the CCHMC DNA Sequencing and Genotyping Core, with subsequent bioinformatics analyses performed by the CCHMC Bioinformatics Collaborative Services. Briefly, 60 ng of genomic DNA was bisulfite converted with the EZ DNA Methylation Kit (Zymo Research). Libraries were sequenced on the Illumina NovaSeq 6000 with 100 bp paired end reads and a depth of at least 30 million reads per sample. Reads were aligned to the mm10 reference genome by Bismark(version 0.18.2). CpG methylation calls were extracted from the alignment using the Bismark Methylation Extractor. MethylKit, version 1.12.0, in R, version 3.6.1, was used for methylation analysis. CpG sites covered in all samples of each group were considered for the downstream analysis. Methylation information was summarized over a tiling window of 1,000 bp in length across the whole genome. Differentially methylated regions (DMRs) with a percentage methylation difference larger than 0% and a q value of less than 0.25 were identified using the χ 2 with overdispersion correction. DMRs were annotated using the genomation, version 1.18.0, and GenomicRanges, version 1.38.0, packages in R, version 3.6.1. DMRs in promoter, exonic, and intronic regions were subjected to Gene Set Enrichment Analysis (GSEA) using GSEA, version 3.0., for genes with multiple DMRs. The DMR with the highest percentage of methylation change was selected for GSEA analysis. Percentages of methylation change values for genes were used as a rank score to run the GSEAPreranked module in GSEA. ATAC-Seq analysis. ATAC-Seq assays were performed as previously described on isolated nuclei from 50,000 sorted LSK cells ( 64 ). After the nuclei preparation, the transposase reaction was performed for 60 minutes at 37°C. The transposed DNA was purified using a QIAGEN MinElute Kit, and library fragments were amplified using 1× NEBNext PCR Master Mix. The libraries were purified with the SPRI Beads Double Size Selection (0.4/1.2×) (Beckman Coulter) and then sequenced on the Illumina NovaSeq X Plus with PE150, aiming at greater than 120M read pairs per sample. ATAC-Seq reads in the FASTQ format were subjected to quality control using FastQC, version 0.11.7, Trim Galore!, version 0.4.2, and cutadapt, version 1.9.1. The trimmed reads were aligned to the reference mouse genome version GRCh38/mm10 using Bowtie2, version 2.3.4.1, with parameters “--very-sensitive-local -X 2000”. Aligned reads were stripped of duplicate reads using sambamba, version 0.6.8. Peaks were called with MACS2, version 2.1.2, using the parameters “-g mm -p 0.01--shift -75 --extsize 150 --nomodel -B --SPMR --keep-dup all --call-summits”. Consensus peaks among all samples were obtained in 2 steps by selecting called peaks present in at least 75% of the biological replicates and by merging selected peaks at 50% overlap using BEDtools, version 2.27.0. The resulting sets of peaks were converted from BED format to a gene transfer format (GTF) to enable fast counting of reads under the peaks with the program featureCounts, version 1.6.2 (Rsubread package) (Bioconductor). Differential open chromatin regions (DOCs) between groups of samples were assessed with the R package DESeq2, version 1.26.0. Peaks were associated to nearest or overlapping gene and genomic features. GSEA was carried out using the GSEAPreranked script and hallmark gene sets. RNA-Seq analysis. Ctrl, WT STZ , and Nlpr3 STZ placenta were dissected at G17 and placental CD45 + cells isolated by FACS sorting. Total RNA was purified using a RNeasy Micro Kit Column System (QIAGEN). RNA quality was controlled using an Agilent Bioanalyzer before processing for retrotranscription, linear amplification, and cDNA library generation. The whole transcriptome was amplified using the SMARTer Ultra Low RNA Kit for Illumina Sequencing (Clontech). cDNA libraries were prepared using Nextera XT DNA Sample Preparation Reagents. Fragmented and tagged libraries were pooled and were sequenced on an Illumina NovaSeq 6000 platform using a paired end 150 bp sequencing strategy. RNA-Seq reads in FASTQ format were subjected to quality control using FastQC, version 0.11.7, Trim Galore!, version 0.4.2, and cutadapt, version 1.9.1. The trimmed reads were aligned to the reference mouse genome, version mm10, with the program STAR, version 2.6.1e, and stripped of duplicate reads with the program sambamba, version 0.6.8 ( 5 ). Gene-level expression was assessed by counting features for each gene, as defined in the NCBI’s RefSeq database. Read counting was done with the program featureCounts, version 1.6.2, from the R subread package. Raw counts were normalized as transcripts per million (TPM). Differential gene expressions between groups of samples were assessed with the R package DESeq2, version 1.26.0. Gene list and log 2 fold change were used for GSEA analysis using the Gene Ontology (GO) pathway data set. Multiplex cytokine analysis. The concentration of cytokines was measured in duplicate in serum isolated from LPS-treated adult Ctrl and GD WT STZ offspring using the Mouse High Sensitivity T-Cell 18-plex Discovery Assay (Eve Technologies, MDHSTC18). Statistics. All results are expressed as means, with error bars reflecting SD. Statistical analyses were performed using GraphPad Prism, version 9. Ordinal variables (atherosclerosis grade) were analyzed using the χ 2 test for trend. Differences between 2 groups were assessed using unpaired, 2-tailed Student’s t test. Data involving more than 2 groups were assessed by 1- or 2-way ANOVA with Tukey’s or Šidák’s post hoc test. P < 0.05 was considered significant. Study approval. Mice were bred and housed in the AAALAC-accredited animal facility of CCHMC. All animal experiments were approved by the CCHMC IACUC. Data availability. Next-generation sequencing data are available at the NCBI’s Gene Expression Omnibus database (GEO GSE244698). Values for all data points in graphs are reported in the Supporting Data Values file.

Results Two independent genetic and pharmacological models of diabetes in pregnancy. We conducted our studies in 2 independent mouse models of diabetes in pregnancy. First, we used a mouse genetic model of type 1 diabetes ( Ins2 Akita/J ) under a rigorous breeding scheme ( Figure 1A ). C57BL/6-backcrossed Ins2 Akita/J mice carry a spontaneous mutation in the Ins2 gene that induces the toxic folding of the insulin protein, leading to reduced β cell mass and impaired insulin secretion ( 10 ). Mice heterozygous for the Ins2 Akita mutation displayed normal weight and a diabetic profile with hyperglycemia, but not hyperinsulinemia. The colony was maintained by crossing Ins2 Akita/+ males to C57BL/6 females to limit intergenerational GD and potential germline alterations. To reinforce the hyperglycemia phenotype in females, F1 heterozygote Ins2 Akita/+ and WT littermate females were fed with high-fat diet (HFD) (60 kcal% fat) for 3 weeks before breeding with C57BL/6 males ( Figure 1A ). In this condition, Ins2 Akita/+ dams displayed high nonfasting blood-glucose levels during gestation (measured at gestational days G10±1 and G17±1) ( Figure 1C ). To independently validate our results, we established a pharmacological C57BL/6 mouse model, based on a 2-hit approach that combines HFD with 3 injections of streptozotocin (STZ) (60 mg/kg/d) to target maternal pancreatic β cells and induce diabetes ( Figure 1B ) ( 11 ). STZ administration was completed 1 week prior to mating, precluding any direct impact of STZ on the offspring. In this model, dams developed hyperglycemia during gestation with a late-stage peak at G17±1 ( Figure 1D ). Consistent with human pathology, both animal models showed an increase in perinatal adverse events (for ~15% of the dams), including cases of stalled labor, dystocia, and pup mortality (data not shown) ( 2 ). Offspring (F2) were maintained under regular 13 kcal% fat diet from birth into adulthood and analyzed as adults at 8 to 12 weeks of age. Analyses were performed on WT offspring that did not carry the diabetogenic mutation (denoted hereafter as WT Akita ) or were not directly exposed to STZ treatment (denoted hereafter as WT STZ ). GD offspring were compared with control (Ctrl) mice born to normal pregnancy generated through a similar breeding scheme. As expected, GD offspring showed no gross metabolic abnormalities at weaning (4 weeks) and adulthood (8 weeks) when assessed by (a) body weight, (b) nonfasting or fasting blood glucose, and (c) glucose-tolerance test (GTT) or insulin-tolerance test (ITT) ( Figure 1, E and F , and Supplemental Figure 1 , A and B; supplemental material available online with this article; https://doi.org/10.1172/JCI169730DS1 ). Together, our results describe 2 independent GD mouse models with gestational hyperglycemia and perinatal adverse features that recapitulate human pathology. They show that offspring born to these models of diabetic pregnancy do not present major metabolic abnormalities or glycemic alterations when analyzed during early adulthood. Diabetic pregnancy promotes atherosclerosis development in adult offspring. In human populations, GD has been associated with early onset of atherosclerosis development in offspring ( 8 , 9 ). To assess whether the mouse models mimic human pathology, we induced GD in atherosclerosis-prone Apoe -KO mice using the STZ protocol and challenged the offspring with HFD to favor atherosclerosis development ( Figure 2, A and B ). As described in WT mice, adult GD Apoe –/– offspring did not display any changes in body weight or gross glycemic alterations ( Figure 2C ). However, aortic valve histological examinations showed accelerated atherosclerosis development in GD offspring compared with Ctrls ( Figure 2D ). Notably, we observed increased inflammation, lipid deposition, and cartilaginous metaplasia in the aortic valves ( Figure 2E and Supplemental Figure 2 , A and B). We then used hematopoietic transplantation to assess the contribution of the hematopoietic system to the atherosclerotic phenotype. Saturating amounts of BM cells (5 × 10 6 ) isolated from Ctrl or GD WT Akita offspring were transplanted into Apoe –/– irradiated recipient mice that were subsequently challenged with HFD ( Figure 2F ). Transplanted mice showed no overt glycemic alterations ( Figure 2G ). The severity of the atherosclerotic features was reduced in Apoe –/– recipient mice, consistent with previous reports ( 12 , 13 ). However, even in this context, recipient mice transplanted with GD offspring BM showed an increase in aortic valve atherosclerotic lesions ( Figure 2, H and I , and Supplemental Figure 2C ). These lesions in the Apoe –/– recipients were associated with exacerbated monocytosis and accumulation of aortic myeloid cells ( Supplemental Figure 2 , D and E) ( 14 , 15 ). Together, these results validate the use of the 2 GD mouse models to mimic the pathological conditions described in human cohorts born to diabetic pregnancy. They also suggest that the hematopoietic system contributes, at least to some extent, to atherosclerosis development in adult GD offspring. Long-term alterations of the steady-state hematopoiesis in offspring born to diabetic pregnancy. We next performed a comprehensive analysis of the hematopoietic compartments present in the BM of 8-week-old WT Akita mice born to diabetic pregnancy ( Figure 3A ) ( 16 ). While BM cellularity was not dramatically affected, we found that steady-state hematopoiesis was skewed toward the myeloid lineage ( Figure 3B ). We analyzed the immature Lineage – c-Kit + Sca-1 + (LSK) BM compartment subfractionated based on the expression of the marker Flt3, CD150, and CD48 ( Figure 3C and Supplemental Figure 3A ). GD did not affect the frequency and phenotype of the hematopoietic stem cells (HSCs) defined as LSK/Flt3 – CD48 – CD150 + (HSC-SLAM). In contrast, we observed an expansion of the short-term multipotent progenitors MPP3 and MPP4, defined as LSK/Flt3 – CD48 + CD150 – and LSK/Flt3 + CD48 + CD150 – , respectively. This was associated with an expansion of the myeloid committed progenitors (granulocyte/macrophage progenitor [GMP]: Lineage – c-Kit + Sca-1 – CD34 + FcγR + ) ( Figure 3D ). Functionally, a decrease in the quiescence of the HSC compartment was detectable by intracellular Hoechst 33342/Ki67 staining ( Figure 3E ). HSC activation was further confirmed by reduced nuclear localization of FOXO3, a key regulator of HSC quiescence ( 17 ) ( Figure 3F ). This activated phenotype was associated with a loss of HSC fitness in competitive transplantation assay ( Figure 3, G and H ) ( 18 ). Similar BM myeloid skewing was observed in the pharmacological STZ model with an expansion of the MPP4 and GMP compartments ( Supplemental Figure 3 , B–D). Although we did not observe a major alteration of the HSC quiescence at steady state, competitive transplantation assay revealed a similar loss of HSC fitness ( Supplemental Figure 3 , E–H). Together, these results suggest a long-lasting effect of GD on the offspring hematopoietic system that persists into adulthood. We observed marks of activation of the HSC compartment and an expansion of key hematopoietic myeloid progenitors (MPP and GMP). Although these alterations did not lead to any overt hematological pathologies at the time of analysis, they signal the persistence of a latent dysregulated hematopoietic state in offspring born to diabetic pregnancy. GD offspring display altered hematopoietic response to inflammatory cues. We assessed the functional effect of GD on the ability of the offspring hematopoietic system to respond to inflammatory challenge ( Figure 4A ). We used LPS to mimic bacterial infection in Ctrl and GD WT STZ offspring and in adult diabetic Ins2 Akita/+ males (nonfasting blood glucose: 390.7 ± 57.7 mg/dL, n = 10). As expected, LPS treatment led in all conditions to a decrease of the BM cellularity and a reduction of the number of BM myeloid cells ( Figure 4B ). In contrast, we observed an altered hematopoietic stress response in WT STZ GD offspring and diabetic Ins2 Akita/+ mice, phenotypically characterized by limited MPP3 expansion and a slow recovery of the GMP compartment ( Figure 4C ). This phenotype in GD offspring was associated with a reduced inflammatory cytokine response, particularly for IL-6, IL-12p70, and, to a lesser extent, IFN-γ, TNF-α, MCP1, and MIP2 ( Figure 4D ). It is noteworthy that similar alterations were found in a model of viral infection using polyinosinic:polycytidylic acid (pIC), suggesting that these functional characteristics are not linked to a specific inflammatory pathway ( Supplemental Figure 4 , A–D). These results indicate that GD leads to lasting disruption of the hematopoietic stress response in the offspring. Although GD offspring do not display any gross metabolic defects, we observed that they mimic the hematopoietic features found in diabetic mouse models. Thus, these results suggest the existence of a long-term functional glycemic memory in adult offspring born to diabetic pregnancy. To confirm these observations, we generated BM-derived macrophages (BMDMs) from adult Ctrl, GD WT Akita , and WT STZ offspring, along with diabetic Ins2 Akita/+ mice ( Figure 4E ). We did not observe any qualitative or quantitative defects in BMDM generation based on cell number and immunophenotype (data not shown). As expected, all BMDMs acquired an inflammatory phenotype upon treatment with LPS (10 ng/mL) and IFN-γ (10 ng/mL), as assessed by the acquisition of the CD86 marker and the expression of inflammatory cytokines such as Il6 , Il1a , and Tnf ( Supplemental Figure 4E and data not shown). However, we found that BMDMs generated from WT Akita and WT STZ GD offspring were reduced in number after activation compared with Ctrl ( Figure 4F ). This loss of cellularity was detectable as early as 3 hours after activation ( Supplemental Figure 4F ). Consistent with the maintenance of a functional glycemic memory, this property of BMDMs generated from GD offspring mimicked the behavior of BMDMs derived from diabetic Ins2 Akita/+ mice. Importantly, this property was maintained when GD offspring BM cells were transplanted into normal congenic mice ( Figure 4G ). Thus, BMDMs generated from recipient mice 6 months after BM transplantation maintained a heightened sensitivity to inflammation ( Figure 4H ). Together, these data show that BMDMs generated from GD offspring and diabetic mice share functional properties that are distinct from those of Ctrl BMDMs. Results in the BM transplantation setting demonstrate that the diabetic memory generated in GD offspring is an intrinsic hematopoietic property supported by alterations in the HSC compartment. Sterile inflammation contributes to the induction of the GD hematopoietic memory during pregnancy. We hypothesized that damage-associated molecular patterns (DAMPs) linked to hyperglycemia could contribute to the in utero induction of the GD hematopoietic memory ( 19 ). AGER is a receptor for several metabolic stress signals, such as advanced glycation end products (AGEs), HMGB1, and S100 proteins ( 20 ). Previous reports have demonstrated the role of AGER in GD-associated fetal alterations ( 21 , 22 ). We used a loss-of-function approach to determine the contribution of AGER to the development of a GD hematopoietic memory. We treated Ager -deficient ( Ager –/– ) dams with STZ to generate GD mutant offspring (Ager STZ ) ( Figure 5A ). Mutant dams did not show any alterations of the diabetic phenotype during pregnancy compared with their WT counterparts ( Supplemental Figure 5A ). Adult GD offspring were evaluated by BM phenotypic analysis at steady state and functional BMDM assessment, two defining criteria of GD hematopoietic memory in WT mice ( Figure 5B ). Based on these readouts, we observed that the disruption of the AGER pathway blocks the acquisition of the GD hematopoietic memory in offspring ( Figure 5C ). Targeted gene invalidation in the dam (by crossing male WT with female Ager –/– ) or in the fetus (by crossing male Ager –/– with female Ager +/– and selecting Ager –/– offspring) demonstrate that this pathway is specifically required in the mother for the induction of the GD hematopoietic memory in the offspring ( Supplemental Figure 5 , B and C). These results show that the maternal AGER pathway is a primary inducer of the diabetic hematopoietic memory and suggest the existence of secondary signals that affect the hematopoiesis of the offspring. We hypothesized that sterile inflammation could be central to these secondary signals. We tested the NLRP3 inflammasome, which is a known regulator of sterile inflammation and which has been associated with GD and pregnancy complications ( 23 , 24 ). We assessed NLRP3 as previously described for AGER ( Figure 5A ). Global NLRP3 targeting did not affect the dam GD ( Supplemental Figure 5D ), but did prevent the acquisition of the GD hematopoietic memory in offspring (Nlrp3 STZ ) ( Figure 5D ). Unlike AGER, NLRP3 was required in both the dam and the fetus ( Supplemental Figure 5 , E and F). Furthermore, we found that the GD hematopoietic memory correlates with placental inflammation. We observed an expansion of the macrophage population in placenta of WT STZ offspring that develop a hematopoietic memory, but not in Ager STZ and Nlpr3 STZ offspring ( Figure 5E ). As expected, accumulation of placental macrophage was associated with expression of inflammatory cytokine genes, such as Il6 , Ccl2 , Il1b , and Tnf . ( Figure 5F and Supplemental Figure 5G ). These results were strengthened by RNA-Seq analyses performed on CD45 + cells isolated from placenta, which confirmed the link between GD and placental inflammation and its dependence on NLRP3 ( Figure 5G , Supplemental Figure 5H , and Supplemental Table 1 ). Together, these results show that induction of the GD hematopoietic memory in offspring requires the AGER/NLRP3 pathways and is associated with sterile placental inflammation. DNA methylation contributes to the maintenance of the GD hematopoietic memory during adulthood. Next, we investigated the mechanisms that support the persistence of the GD hematopoietic memory in adult offspring. Diabetes has been associated with epigenetic changes ( 25 ). Particularly, hematopoietic alterations in diabetic mice have been linked to alterations in the DNA methylation landscape and increased expression of DNA methyltransferase 1 (Dnmt1) ( 26 ). In LSK cells, we confirmed a specific increase of DNMT1 but not DNMT3a protein in adult diabetic Ins2 Akita/+ and STZ-treated mice ( Figure 6A ). Interestingly, we found that LSK cells from GD adult offspring show a similar increase of DNMT1 protein, even as the models do not display overt hyperglycemia. Using reduced representation bisulfite sequencing (RRBS) analysis, we found that DNMT1 upregulation correlates with methylome alterations in HSPCs of adult GD offspring. It particularly affected genes that respond to reactive oxygen/nitrogen species, which are important features of diabetes ( 27 ), and gene pathways previously found differentially methylated in cord blood cells from diabetic pregnancy (e.g., cell-cell adhesion, MAPK signaling, cytosolic transport) ( 28 ) ( Supplemental Figure 6A and data not shown). These results were reinforced by transposase-accessible chromatin sequencing (ATAC-Seq) assay performed in LSK cells isolated from adult Ctrl and GD offspring ( Supplemental Figure 6 , B–D). Despite some degree of variability between replicates, differential analysis of the accessible sites showed differences in the chromatin structure in Ctrl and GD LSK compartments. Particularly, we observed a reduced accessibility in GD offspring that affects genes involved in metabolism, oxidative stress, and inflammation pathways. Although limited in scope, these analyses are consistent with the idea of epigenetic alterations in GD offspring. To assess the contribution of DNA methylation to the GD hematopoietic memory, we used a low dose of the DNA methyltransferase inhibitor 5-aza-2′-deoxycytidine (5-azadC) to reset the methylome profile in GD offspring ( 29 , 30 ) ( Figure 6B ). GD WT STZ offspring were analyzed immediately after treatment or following a 2-week recovery period. We observed that 5-azadC treatment affects BM cellularity without dramatically disturbing the hematopoietic hierarchy assessed by flow cytometry ( Figure 6C and Supplemental Figure 6E ). Hematopoietic parameters were normalized after the end of treatment ( 30 ). As expected, DNMT1 and DNMT3a protein levels in HSPCs were reduced by 5-azadC treatment, but were restored following treatment cessation ( Figure 6D ). By assessing the BMDM function, we found that hypomethylating treatment limits the loss of BMDM cellularity following inflammatory activation, consistent with a loss of the GD hematopoietic memory ( Figure 6E ). However, we found that this effect was temporary, as the BMDM phenotype remerges after treatment cessation ( Figure 6E ). Consistent with the idea of a GD hematopoietic memory, these results show that offspring born to diabetic pregnancy maintain into adulthood high expression of DNMT1, a molecular feature associated with overt diabetes. In addition, these results suggest that upregulation of DNMT1, and the associated changes in DNA methylation, is one of the factors supporting the GD memory in the hematopoietic system.

Discussion Our results demonstrate the existence of a hematopoietic memory associated with GD. It is tempting to speculate that this phenomenon is related to the hematopoietic memory recently established in the context of an immune response ( 31 ). Referred to as trained immunity, this concept proposes that the hematopoietic system not only directly responds to inflammatory signals but also can “remember” the inflammatory events, therefore heightening or dampening the response to secondary challenges ( 32 , 33 ). This property, uncovered in short-lived innate immune cells, such as monocytes and macrophages, was expanded to hematopoietic progenitor cells ( 34 , 35 ). Its persistence through BM transplantation suggests that it originates in HSCs ( 36 , 37 ). Consistent with this finding, our work indicates that the HSC compartment is altered by GD, as shown by the persistence of the GD phenotype for over an 8-week growth period, from the neonatal period to adulthood. In adult GD offspring, we found an increased HSC activation at steady state and a loss of fitness upon transplantation. We observed that the abnormal BMDM function in GD offspring could be maintained in the BM transplantation setting for over a 6-month period. These results confirm that the hematopoietic system, and particularly the HSC compartment, are reactive to organismal metabolic stresses ( 17 , 18 , 38 ). They are also consistent with recent findings showing that endogenous “sterile” signals associated with Western diet or hyperglycemia can induce a memory in innate immune cells and their precursors ( 39 – 41 ). Mechanistically, our work describes a scenario in which GD-associated stress signals promote the induction of the hematopoietic memory through activation of the AGER pathway. While AGER expression is low in most cell types in healthy conditions, it is upregulated in several disease states, including diabetes ( 42 ). In adults, AGER signaling has been shown to promote myelopoiesis in a hyperglycemic environment ( 43 ). AGER and its ligands S100A8/A9 also play a dominant role in myelopoiesis in the context of intermittent glycemic fluctuation ( 44 ). During pregnancy, AGER contributes to preeclampsia, preterm birth, and different congenital malformations associated with diabetic pregnancy ( 21 , 22 ). Our data indicate that the AGER pathway is an essential upstream inducer of the GD hematopoietic memory. AGER belongs to a class of pattern-recognition receptors that recognize broad common features. We notice that Toll-like receptors (TLR4/TLR2), which share common ligands and signaling pathways with AGER, were not able to compensate for AGER deficiency ( 45 , 46 ). This specific requirement of AGER may rely on specific ligands, which remain to be determined. It may also be related to a specific pattern of expression. Differential dam/offspring loss-of-function experiments demonstrate that maternal AGER is the contributor of this effect while fetal AGER is dispensable. This maternal specificity reveals the existence of secondary signals able to promote hematopoietic alterations in offspring. Consistent with this idea, we found that the maternal and fetal NLPR3 inflammasome are required for the induction of the GD hematopoietic memory. Although our results do not formally link AGER and NLRP3 activity, our work indicates that these signaling pathways affect the level of placental inflammation. We envision that the developmental window conducive to the acquisition of the diabetic memory occurs in late gestation during the HSC transition from a fetal to an adult identity, a stage that has been shown to be sensitive to inflammatory signals ( 47 ). We speculate that some of the mechanisms revealed in our study could also contribute to the persistent immune alterations recently described in the context of infection occurring during pregnancy ( 48 ). Together, our results are consistent with the well-established role of AGER and NLRP3 in controlling the physiological and pathological inflammatory processes of pregnancy ( 23 , 49 ). They are also in line with the hematopoietic impact of these signaling pathways in situations of metabolic stress triggered by either oxidized low-density lipoprotein or hyperglycemia ( 40 , 41 , 43 ). Finally, our results highlight that AGER-dependent “sterile” inflammation processes occurring during pregnancy are not only key determinants of the immediate pregnancy outcome, but also affect the long-term health of the offspring ( 50 ). Epigenetic alteration acquired in utero may constitute the link between GD and its effects on the health of the adult offspring ( 51 ). Human studies using placenta, umbilical cord, or adult peripheral blood as well as skeletal muscle and adipose tissue suggest persisting epigenetic alterations in the offspring, including DNA methylation, histone modifications, and miRNA expression ( 28 , 52 – 55 ). However, these studies in adults were limited by the intrinsic diversity of the human populations, the multiple confounding environmental factors affecting the epigenome, and the difficulty in experimentally assessing the clinical relevance of these alterations. Mouse models bypass these limitations and therefore could be important tools for studying transgenerational transmission ( 56 , 57 ). While our results in mouse models correlate DNMT1 upregulation with the persistence of a GD hematopoietic memory during adulthood, the full impact of DNMT1 overexpression on the DNA methylation landscape and chromatin structure in HSPCs remains to be elucidated. We found that treatment with a DNA methyltransferase inhibitor disrupts the maintenance of the hematopoietic memory. However, we notice that this pharmacological approach is temporary, as GD hematopoietic memory is rapidly restored after treatment. This may be linked to the limited ability of the pharmacological approach to effectively target and reprogram the immature HSCs that sustain this phenotype. Alternatively, this may suggest the existence of other epigenetic mechanisms able to restore increased DNMT1 expression and contribute to the maintenance of GD hematopoietic memory. As DNMT1 expression emerges as a marker of the impact of prenatal and adult glycemia on the hematopoietic system, a deeper analysis of the regulation of the Dnmt1 locus in diabetes could improve our understanding of the mechanisms controlling the maintenance of the hematopoietic memory in offspring born to diabetic pregnancy ( 26 ). Our work shows that GD hematopoietic memory is associated with the alteration of the hematopoietic response to acute inflammatory stress. In adult GD offspring, we observed a decrease in production of inflammatory cytokine in response to LPS. Similarly, BMDMs derived from GD offspring displayed an increased susceptibility to inflammatory signals that led to a reduced cellularity in culture. This inflammatory dampening contrasts with the apparent contribution of the GD hematopoietic memory to the development of atherosclerosis in offspring. The connection between these seemingly contradictory phenotypes remains to be established. Previous studies showed that myeloid cells isolated from diabetic patients and BMDMs cultured in diabetic conditions display a heightened activation of the NLRP3 inflammasome ( 58 , 59 ). Here, we show that the NLRP3 inflammasome is required for the acquisition and/or the manifestation of the hematopoietic GD memory in offspring. In this context, we speculate that the GD hematopoietic response to acute inflammatory stimulation could be linked to NLRP3 hyperactivation, leading to an inflammatory form of cell death known as pyroptosis ( 60 ). Interestingly, pyroptosis is a key promoter of the inflammatory phenotype fueling the initiation and the progression of atherosclerosis ( 61 ). We propose that, in chronic atherogenic conditions, heightened NLRP3 activation in hematopoietic cells may promote the atherosclerosis development associated with GD. Further studies are needed to fully evaluate the status of the NLRP3 pathway in GD offspring and test its contribution to the pathological consequences of GD hematopoietic memory. This work exemplifies how prenatal health can have broad and lasting consequences on adult health. It particularly highlights the unappreciated contribution of the hematopoietic system in the transgenerational transmission of cardiovascular pathologies, such as atherosclerosis. By controlling the production of immune cells, the hematopoietic system is at the center of multiple pathological conditions and diseases. In this context, our work suggests that the induction and the lasting maintenance of a hematopoietic memory in offspring born to diabetic pregnancy alter their inflammatory stress responses and contribute to the development of chronic disease by creating vulnerability to lifestyle and environmental factors.

Gestational diabetes is a common medical complication of pregnancy that is associated with adverse perinatal outcomes and an increased risk of metabolic diseases and atherosclerosis in adult offspring. The mechanisms responsible for this delayed pathological transmission remain unknown. In mouse models, we found that the development of atherosclerosis in adult offspring born to diabetic pregnancy can be in part linked to hematopoietic alterations. Although they do not show any gross metabolic disruptions, the adult offspring maintain hematopoietic features associated with diabetes, indicating the acquisition of a lasting diabetic hematopoietic memory. We show that the induction of this hematopoietic memory during gestation relies on the activity of the advanced glycation end product receptor (AGER) and the nucleotide binding and oligomerization domain-like receptor family pyrin domain-containing 3 (NLRP3) inflammasome, which lead to increased placental inflammation. In adult offspring, we find that this memory is associated with DNA methyltransferase 1 (DNMT1) upregulation and epigenetic changes in hematopoietic progenitors. Together, our results demonstrate that the hematopoietic system can acquire a lasting memory of gestational diabetes and that this memory constitutes a pathway connecting gestational health to adult pathologies. Adult offspring born from diabetic pregnancy acquire a lasting hematopoietic memory that contributes to long-term adult pathology.

Author contributions VG performed and analyzed all the experiments with the help of SG, AA, and M Solomon. M Sakabe and NC performed all cardiac histopathology preparations, which were independently evaluated by AK. XZ and HLG provided expertise for ATAC-Seq analysis. AK and MX provided expertise and critical insights for the development of the project and manuscript. VG and DR designed, interpreted, and analyzed the studies and wrote the manuscript. Supplementary Material

This work was supported by NIH grants R01HL141418 and R01DK133145 (to DR), R01DK121062 (to HLG), and R01HL132211 and American Heart Association grant 23TPA1070041 (to MX). This work benefitted from funding from the Cooperative Center for Excellence in Hematology (award number U54DK126108). We acknowledge the assistance of the CCHMC Comprehensive Mouse and Cancer Core, Pathology Research Core, Research Flow Cytometry Core (supported by an NIH S10OD025045 grant), Single Cell Genomics Core, DNA Sequencing and Genotyping Core, and Bioinformatics Collaborative Services. We particularly acknowledge the assistance of Aditi Paranjpe for the bioinformatics analyses. The author would like to thank Jose Cancelas, Jane C. Khoury, and Katherine Bowers (CCHMC) for their support and insights on the project. 11/21/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e169730

oa_package/95/8b/PMC10786695.tar.gz

PMC10786696

37971879

Introduction Hematopoietic stem cell transplantation (HCT) is an essential and often the sole curative treatment strategy for high risk hematologic malignancies ( 1 ). Graft-versus-host disease (GVHD), the foremost complication of allogeneic HCT, is a major limitation of this procedure, accounting for deleterious effects on quality of life and increased mortality from HCT ( 2 , 3 ). Current diagnosis of acute GVHD (aGVHD) and chronic GVHD (cGVHD) in patents who have undergone a bone marrow transplant is based on inaccurate, operator-dependent clinical markers, and less often on biopsies. These methods are time consuming, costly, invasive, and yield late-stage diagnoses that negatively affect morbidity and mortality. In addition, current practice lacks accurate biomarkers for prediction of disease occurrence, identification of disease onset, prediction of disease response to treatment, and accurate assessment of the actual response to treatment ( 4 ). Multiple prognostic and diagnostic biomarkers for cGVHD have been proposed, including IL2Rα, aminopeptidase N (CD13), IL4, IL6, TNFα, ST2, OPN, chemokine ligands such as CXCL9, CXCL10, and CXCL11 ( 5 – 14 ), cellular biomarkers including immune cells subpopulations ( 15 – 18 ), miRNA ( 19 ), and others. However, none of these biomarkers have been clinically validated. In addition, all these markers are indicative of immune system derangement, lacking information on the damaged tissue targeted by the alloimmune process. Thus, there is an unmet need for simple objective tools that can aid the treating physician in easier identification and scoring, and can assist with personalization of management in patients suffering from cGVHD. Classic liquid biopsies analyze circulating cell-free DNA (cfDNA) via genetic variations or mutations in the DNA of a fetus, a tumor, or a transplanted solid organ. However, these approaches are blind to DNA released from cells with a normal genome, as would occur in organs damaged by pathologies such as GVHD. We and others have previously shown that tissue-specific DNA methylation patterns can provide powerful, universal biomarkers for detecting the tissue origins of cfDNA, reflective of elevated turnover or damage in specific organs and regardless of the underlying pathology ( 20 – 22 ). For example, we showed that genomic loci specifically unmethylated in lung epithelial cells or in hepatocytes can serve as cfDNA biomarkers to detect specific lung or liver injury ( 20 – 26 ). The aim of this study was to establish a set of affordable, yet highly specific and sensitive methylation markers for cell types relevant to patients at risk for developing cGVHD, to examine their utility for detection of damage to specific organs in patients with clinically suspected cGVHD, and to create a cfDNA-based model that can assist the treating physician in surveillance and treatment decisions.

Methods Methylation analysis. We prepared cfDNA and measured its concentration (in nanograms per milliliter plasma), then treated with bisulfite to expose the status of methylation and performed multiplex PCR as described ( 36 ) to amplify marker loci. PCR products were sequenced on a NextSeq machine (Illumina), and the fraction of molecules carrying a tissue-specific pattern of methylation was determined. We used this information, averaged over the markers for each tissue, to assess the relative contribution of each tissue to cfDNA. In addition, by multiplying the proportion of cfDNA from each tissue by the total concentration of cfDNA in a sample, we calculated the absolute concentration of cfDNA from each tissue to plasma, expressed in genome equivalents per milliliter plasma, as described ( 27 , 36 ). Primer sequences of all markers, as well as clinical and methylation data for all samples are provided in Supplemental Tables 1 and 2 . Clinical assessment of patients undergoing HCT in the chronic setting. To assess the utility of cfDNA for detecting organ damage in cGVHD, we prospectively collected 101 plasma samples from 101 individuals more than 100 days after allogeneic stem cell transplantation, arriving for planned routine clinical follow up, at the Bone Marrow Transplant (BMT) day care unit at Hadassah medical center. Upon each visit blood was drawn for regular blood tests (extra 10 mL of blood was drawn for cfDNA analysis) and the patient underwent a full assessment by the treating physician which included cGVHD grading according to the 2014 NIH criteria. During the course of this 38-month study, 65 patients were diagnosed at any point with clinically evident cGVHD, while 36 were not found to have clinical signs of cGVHD. Patient characteristics. The median age of patients was 47 years. A total of 65% of the patients were men and 35% were women. The majority of patients underwent transplant due to acute myeloid leukemia (57%), had a matched sibling (63%), were treated with a myeloablative conditioning regimen (64%) and received stem cells withdrawn from peripheral blood (PBSC, 92%). Most of the patients (55%) received a transplant from a matched sex, while 25% were transplanted from a mismatched donor sex, in a female to male direction. Of the 101 samples, 57% were collected from patients with a history of aGVHD. None had signs of overlap (both acute and chronic) GVHD at the time of sampling. A single patient developed liver GVHD 1 month after Donor Lymphocyte Infusion (DLI). The median time from transplantation was 783 days (range 101–7,878 days). Half of the samples ( n = 49 samples) were taken from patients receiving 1 or more immunosuppressive agents at the time of collection. Only 7 patients had evidence of CMV viremia at the time of collection. One patient had biopsy-proven colitis, which did not show CMV inclusion bodies, while none of the remaining 6 had any evidence for CMV disease. Four patients were treated for CMV infection. Eleven patients had a positive EBV-PCR in peripheral blood (with a median of 300 copies/mL), none of which was clinically significant. One patient was positive in the upper respiratory tract for RSV and another for influenza. One patient had staphylococcus epidermis bacteremia. Chimerism levels were routinely monitored. 98% of the samples were obtained from patients with a blood driven STR assay indicating 100% donor-derived hematopoietic cells. Two samples exhibited a donor chimerism ranging from 88%–92%, precluding analysis of the relationship between degree of chimerism, cfDNA methylation profiles, and a potential relapse. None of the samples were taken at the time of relapse. Statistics. Assessment of cfDNA plasma levels in healthy controls versus allogeneic transplanted patients with and without clinical signs of cGVHD was performed using nonparametric, unpaired, Mann Whitney test. Analyses were performed using GraphPad Prism (version 10.0.1), and results were considered statistically significant for P values of less than 0.05. We used machine learning to evaluate the predictive power of both cfDNA and biochemical measurements in relation to clinical evident cGVHD. We compared multivariate logistic regression (MLR), XG boost and random forest (RF) classifiers on our data set. MLR, XGboost, RF had an average accuracy of 0.74, 0.67, and 0.65, respectively, by Repeated-K-fold cross-validation (K = 5) with a SD of 0.23, 0.22, and 0.3, respectively. As the MLR model had both higher accuracy with similarly robust results by cross validation, we applied MLR for further analyses. Furthermore, MLR emerges as the most fitting estimator based on the following considerations: (a) We anticipate that GVHD will consistently increase the levels of measured markers, signifying increased cell death. Hence, a monotonous model that consistently increases in response to changes in its features should be appropriate. (b) The size of our data does not support models with a large number of parameters — MLR bears a single parameter per measurement, reducing the risk of overfitting. (c) MLR inference naturally provides a probability score. We hypothesize that measurements of cfDNA and blood biochemical values possess significant predictive potential for the presence of cGVHD. We leveraged Shapley values to gauge the magnitude of the predictive capability of each feature. This latter technique offers a principled approach to feature selection, promoting enhanced performance with reduced overfitting. Since we expect higher cfDNA levels to indicate cGVHD, we constrained the parameter space of the model to be nonnegative for all coefficients and compared the performance to an unconstrained optimization in order to explore the overfitting potential of the model. A total of 93 samples (for which data was available for all parameters) were used for the analysis. We employed Shapley analysis ( 28 ) on a collection of 17 features (comprising of GGT, ALP, ALT, AST, TBil, Total cfDNA level [presented in ng/mL], and organ specific cfDNA: cfSkin, cfLung, cfGI, cfLiver, cfNeutrophils, cfMonocytes, cfEosinophils, cfB cells, cfT cells, cfCD8 cells, and cfTregs cell). Next, to robustly validate the predictive potential of the features, we utilized Repeated-K-Fold cross-validation ( 37 ). We conducted repeated 5-fold cross-validations across the feature sets given a positive coefficient (constrained optimization). Each set, labeled, consists of the highest-ranking features, meaning, set =1 is the single top-ranking feature, set =2 consists of the 2 top ranking features, set =3 of the 3 top ranking features and so on. The metrics (Specificity, NPV, PPV, AUC, and Precision) for each were calculated. The selection of the best feature set was determined based on those achieving the highest AUC and demonstrating favorable performance across other metrics. We calculated recall, specificity, AUC, NPV, and PPV of logistic regression models trained using only the best feature set. A comparison of cfDNA features compared with blood biochemical features and with the combination of both (meaning the entire set) was performed. All analyses were performed using Python 3.10. Using the model, accuracy ([(TP+TN)/Total testing samples] x100%), specificity ([TP/(TP+FN)] x100%), sensitivity([TN/(TN+FP)] x100%), PPV ([TN/(TN+FN)] x100%) and NPV/precision ([TP/(TP+FP)] x100%) were measured. Graphical representation of the tradeoff between specificity and sensitivity was done using the receiver operating characteristics curve (ROC). AUC was calculated in order to determine the ability of the classifier to distinguish positive and negative results. Spearman rank correlation was used to determine the significance of correlation between each pair of variables and other parameters. Study approval. The study was approved by the Hadassah Medical Center IRB committee and is consistent with the declaration of principles of Helsinki. Written informed consent was received prior to participation. Data availability. Primer sequences of methylation markers, as well as clinical and methylation data values for all samples are provided in Supplemental Tables 1 and 2 . Values for all data points in graphs can be found in the Supplemental Table 5 : Supporting Data Values file.

Results DNA methylation markers for targeted assessment of cGVHD-relevant tissue damage. We compared publicly available methylomes of specific human tissues ( 21 ) and identified genomic loci containing CpG sites that are uniquely unmethylated in specific tissues or cell types, relevant to cGVHD. These included hepatocytes (5 markers), skin (5 markers), lung epithelial cells (10 markers), and intestinal epithelial cells (8 markers). We designed multiplex PCR cocktails to amplify all these loci from genomic DNA after bisulfite conversion and sequenced the products to determine the fraction of unmethylated DNA molecules present in the starting material. Figure 1 shows the fraction of methylation blocks from each marker locus that were unmethylated in the indicated samples. As we have shown previously, molecules containing multiple unmethylated CpG sites could be assigned with extreme specificity to a given tissue of origin. We also spiked genomic DNA from specific tissues into genomic DNA of leukocytes to determine assay sensitivity and linearity and found that as little as 0.5% of DNA from the target tissue could be robustly identified when present in a mixture (not shown). These findings establish a cocktail of DNA methylation markers that can be used to identify DNA derived from the liver, skin, lungs, and intestine with extreme specificity and sensitivity. We also used methylation markers specific to selected immune and inflammatory cell types: neutrophils, eosinophils, monocytes, B lymphocytes, and T lymphocytes (including CD8 + and regulatory T cells); all of which showed extreme specificity and sensitivity ( 27 ). Elevated cfDNA levels in patients undergoing HCT with and without clinical GVHD. The overall scheme of the experiment is shown in Figure 2. We recruited a total of 101 patients who underwent HCT, obtained blood samples, and recorded clinical cGVHD scores as well as blood counts and standard blood biochemistry. We determined plasma cfDNA concentration and methylation patterns, compared findings to clinical and biochemistry data, and then developed and validated a model for inference of cGVHD based on cfDNA parameters combined with blood biochemistry markers ( Figure 2 ). The characteristics of the 101 recruited patients and samples are detailed in Supplemental Tables 2–4 and in the Methods section; supplemental material available online with this article; https://doi.org/10.1172/JCI163541DS1 We compared total and tissue-specific cfDNA concentration in samples from healthy individuals (median age 37 years old (range 24–68), 58% women and 42% men), samples from patients who underwent HCT and had no evidence of cVHD from patients undergoing HCT defined by the treating physician as having clinically evident cGVHD. The NIH 2014 criteria were used for defining disease severity (mild, moderate, and severe) and organ scoring (range 0–3). Analyzing 101 samples from 101 patients; patients undergoing HCT with cGVHD (in any organ) had statistically significant higher concentrations of total cfDNA compared with patients undergoing HCT with no clinical evidence of cGVHD ( P < 0.0001) ( Figure 3A ). Total cfDNA levels in patients undergoing HCT were similar to those in healthy controls ( P = 0.63). cfDNA signals from skin ( P = 0.0188), intestine ( P = 0.009), liver ( P = 0.0023), and lungs ( P = 0.0050) were also significantly higher in the group with clinically evident cGVHD compared with the group of patients who did not meet the NIH 2014 criteria for cGVHD ( Figure 3, B–E ). In addition, the concentration of cfDNA originating from GI, liver, and lung was significantly higher in patients who underwent HCT with no evidence of clinical GVHD compared with people in the healthy control group ( P = 0.0002, P = 0.0003, and P < 0.0001, respectively) ( Figure 3, C–E ). Moreover, cfDNA originating from skin and liver significantly correlated with organ-specific clinical GVHD presence (score 0 versus score 1–3) ( P = 0.0022, P = 0.0003, Supplemental Figure 1 , A and C, respectively). Interestingly, patients undergoing HCT with and without lung cGVHD (score 0 versus 1–3) had significantly higher levels of lung cfDNA compared with healthy controls ( P < 0.0001), but lung cfDNA did not correlate with the presence of clinical lung score ( Supplemental Figure 1B ). Analysis of immune-derived cfDNA showed a significantly higher concentration of cfDNA originating from neutrophils, monocytes, eosinophils, and B and T lymphocytes in patients undergoing HCT diagnosed with clinical cGVHD compared with patients who were not diagnosed ( Figure 4 ). cfDNA from neutrophils, T cells, and CD8 + T cells was elevated in patients undergoing HCT who have no clinical cGVHD compared with people in the healthy volunteer group ( Figure 4, A, E, and F ). We next sought to identify correlations between cfDNA parameters and cGVHD clinical scores among patients undergoing HCT. We produced a correlation matrix for all 101 plasma samples for which all tested parameters were available. cfDNA parameters were highly correlated internally — for example, samples with high concentration of total cfDNA tended to also have high levels of organ specific cfDNA ( Figure 5 and Supplemental Figure 2 ) — and there was a significant internal correlation among cGVHD clinical scores, between cGVHD severity assessment and specific organ grading ( Supplemental Figures 3 and 4 ). Moreover, we found a significant correlation between clinical cGVHD severity assessment and total cfDNA as well as organ specific cfDNA levels ( Figure 5 and Supplemental Figures 3–5 ). A combined score for blood-based detection of cGVHD. We wished to create a model that could aid the treating physician to predict the likelihood that a patient has active cGVHD. Employing Shapley analysis ( 28 ) on 17 clinical and cfDNA features (see Methods) yielded positive Shapley values for 7 features, including alanine transaminase (ALT), total cfDNA, cfDNA of monocytes, cfDNA of skin, GGTp, cfDNA of neutrophils, and cfDNA of eosinophils. Figure 6A shows distribution graphs for these features, and Figure 6B shows the average absolute Shapley values for each individual feature. We conducted repeated 5-fold cross-validations across these 7 feature sets, starting from the feature having the highest value, ALT, and sequentially adding the next feature in line (e.g., ALT and total cfDNA; ALT, total cfDNA, cfDNA of monocytes, etcetera). The metrics (specificity, negative predictive value [NPV], positive predictive value [PPV], AUC, and precision) for each number of features selected are illustrated in Figure 7. Notably, the 3 first features maximize the AUC, as well as displaying favorable behavior across the other metrics. Therefore, we opt for these 3 features (consisting of ALT, total cfDNA,and cfDNA of monocytes) as the optimal feature set. Recall, specificity, AUC, NPV, and PPV of logistic regression models trained using only ALT, only cfDNA features (total cfDNA and cfDNA of monocytes), and all 3 features are shown in Figure 8 A. The ROC curves of these models are shown in Figure 8B . Finally, we compared the performance of our models to the exact equivalent set of models, where, instead of using a constrained optimization, we used an unconstrained optimization (allowing negative coefficients). Shapley values are shown for all 17 features ( Supplemental Figure 6 ). The metrics (specificity, NPV, PPV, AUC, and precision) for each are illustrated in Supplemental Figure 7 . Favorable behavior across all metrics was reached at (ALT, γ glutamyl transpeptidase (GGTp), total cfDNA, cfMonocytes, cfEosinophils, and alkaline phosphatase (ALP)) and repeated 5-fold cross-validation was performed to compare the recall, specificity, AUC, NPV, and PPV of logistic regression models trained using these features ( Supplemental Figure 8A ), as well as the ROC curves of these models ( Supplemental Figure 8B ). Evidently, both constrained and unconstrained optimization techniques demonstrate comparable performance, suggesting minimal overfitting with either optimization technique. Moreover, the findings emphasize the high predictive capability of a small set of features, consisting of biochemical and cfDNA measurements. This aligns with our hypothesis that cGVHD leads to increased cell death, consequently elevating the levels of the observed markers.

Discussion Our study shows that tissue-specific DNA methylation patterns can serve as plasma biomarkers for detection of tissue turnover in cGVHD. We demonstrated a general elevation in cfDNA concentration in patients with cGVHD, and an elevation of cfDNA from specific organs as well as immune and inflammatory cells. Combining cfDNA markers with standard biochemical markers allowed us to discriminate patients with and without cGVHD with good sensitivity, accuracy, and precision, suggesting feasibility of a blood-based objective assessment of disease. In agreement with our findings, it has been demonstrated that mitochondrial cfDNA (COX1 DNA) is higher in patients undergoing SCT compared with people who are in the normal control group and correlates with the presence of cGVHD ( 29 ). To our knowledge, this is the first report of potential tissue-specific cfDNA utility in the context of chronic GVHD. Our findings are consistent with, and expand upon, recent studies that focused on the distinct setting of aGVHD ( 30 ). Cheng et al. used a smaller number of patients ( n = 27) and performed shallow whole genome bisulfite sequencing followed by deconvolution, to assess the levels of cfDNA from different recipient sources. Their key finding was that aGVHD — within the first 3 months after HCT — was associated with elevated levels of cfDNA from solid organs (multiple tissues combined). Waterhouse et al. ( 31 ) demonstrated substantial differences in the concentration of 1 colon-specific and 1 liver-specific cfDNA marker in 10 and 14 patients with liver and colon aGVHD, respectively. Moreover, they have demonstrated a decline in these markers in patients who were successfully treated. The clinical condition that we studied — cGVHD — is more challenging, as clinical manifestation is typically less abrupt and therefore tissue damage/turnover, which might be reflected by elevated cfDNA levels, is likely to increase gradually. In addition, our approach differs from Cheng et al. in that we use a PCR-Seq of targeted methylation markers to assess the contribution of specific cell types — a method that gives up breadth, i.e., information obtained is limited to a preplanned subset of tissue sources — for specificity, depth, simplicity and low cost. In contrast to the Waterhouse study, our methodology probes multiple indicators for each organ, allowing us to examine a broader spectrum of damaged end organs. Altogether, these 3 studies support the notion that organ damage/turnover and immune deregulation in GVHD are amenable for a cfDNA methylation–based analysis, and that liquid biopsies can be developed into an objective, quantitative, clinically useful tool aiding the treating physician in diagnosing chronic as well as acute GVHD. An important observation of our study was that most patients undergoing HCT with no clinically detected GVHD had an elevated concentration of the cfDNA derived from donor T cells and recipient intestine, liver, and lung. We propose that this is a reflection of inflammation and increased cellular turnover in patients undergoing HCT, which are taking place even while clinically evident organ function remains in the normal range. This idea is consistent with the model proposed by Cooke et al., whereby cGVHD develops through early inflammation and tissue injury, chronic inflammation, dysregulated immunity, and, eventually, aberrant tissue repair leading to fibrosis ( 32 ). Our study was not designed to test if elevated cfDNA from a given tissue source is predictive of future cGVHD. Additionally, larger cohorts will be needed to assess the prognostic potential of methylation-based biomarkers. Such studies will also be able to test the provocative idea that elevated tissue-specific cfDNA takes place chronically even without an overt clinical manifestation, reflecting a low level of allogeneic damage to host tissues that is offset by organ regeneration. Our study also provides a distinct angle on the nature of immune processes taking place in cGVHD. Extensive studies have revealed the involvement of reactive donor T cells (mainly Th/Tc17), thymic dysfunction, reduced memory B cell formation concomitant with enrichment of alloreactive B cells, reduced levels of T follicular helper cells, macrophage tissue sequestration and activation, and more ( 3 , 16 , 17 ). These studies have typically not characterized immune cell turnover. Our findings suggest that allogeneic HCT causes high turnover of donor T cells (including CD8 and, to a lesser extent, regulatory T cells), at any point after HCT and more so during cGVHD. We hypothesize that the elevated turnover of donor adaptive and innate immune cells, even years after HCT, results from the continuous interaction with allogeneic host tissues. The mechanisms and implications of this phenomenon remain to be elucidated. At a practical level, it is possible that combining cfDNA biomarkers of solid tissues with cfDNA biomarkers of inflammatory / immune cells may increase the specificity of liquid biopsies, e.g. will allow us to differentiate organ damage due to immune attack from damage due to other etiologies ( 33 ). Further studies are needed to examine this intriguing possibility. Implementation of tissue-specific cfDNA biomarkers in clinical GVHD will require additional studies to optimize specificity and sensitivity and to understand how cfDNA dynamics relate to and predict clinical phenotype. We note that the nature of DNA methylation offers a tremendous potential for refining cfDNA analysis. For example, methylation atlases ( 21 , 34 ) allow us to develop new methylation markers that are specific to hepatocytes from different zones in the liver, alveolar or bronchial epithelial cells, epithelial cells of different segments of the intestine, as well as subsets of immune cells such as tissue-specific macrophages. With regard to sensitivity, emerging lessons from cfDNA-based early cancer detection suggest that parallel assessment of multiple specific markers in the same plasma sample can boost sensitivity by increasing the chance of identifying cfDNA from the tissue of interest. Such refinements of cfDNA assays may facilitate discrimination between suggested biological subgroups of GVHD (resolved GVHD, active late aGVHD, active cGVHD, inactive cGVHD and no GVHD), which is challenging in the current clinical setting. In particular, it will be important to search for cfDNA biomarkers that distinguish active chronic from inactive chronic disease, given the relevance for treatment decisions ( 32 ). We note that while our results reveal a good correlation between cfDNA and clinical overt cGVHD, there are outliers showing high cfDNA with no clinical disease and low counts with clinically graded disease. We propose that this discrepancy partly results from the inability of the current clinical grading system to accurately assess the active versus inactive state of cGVHD. Longitudinal studies assessing this hypothesis are warranted. Using multivariate logistic regression, we pinpointed a trio of pivotal features (ALT, total cfDNA concentration, and monocyte/macrophage cfDNA concentration) yielding compelling results: specificity of 86%, a positive predictive value (PPV) of 89%, and a robust AUC value of 0.8. Notably, what we believe sets our approach apart is its efficiency in capitalizing on a balanced selection of a single biochemical parameter and a pair of distinct cfDNA parameters. This pragmatic strategy not only streamlines the diagnostic process but also markedly enhances the ability to accurately discern cases of cGVHD. Conceptually, we believe that combined measurements of classical markers such as liver enzymes and cell counts with cfDNA biomarkers is expected to synergize. The reason is that cfDNA provides distinct biological information; it reveals cell turnover, which is different from cell counts; it is a definitive marker of cell death (while cytoplasmic proteins may be released to blood upon transient cell injury); and it is cleared rapidly, revealing information about acute tissue damage. Particularly intriguing is the role of cfDNA derived from monocytes and macrophages, suggesting a pivotal role of macrophage turnover in the context of cGVHD. This observation is consistent with current understanding of the contribution of macrophage infiltration and activation within affected organs to the pathobiology of the disease ( 35 ). Our study has several limitations. First, we acknowledge that the assay used in this study, based on massively parallel sequencing, may pose challenges to implementation in a standard clinical setting. However, as we have shown before, the small number of target loci makes it possible for translation into a simpler version, based on quantitative PCR ( 20 ). Such a version will have the advantage of delivering results faster (same day), at a low cost and in a point-of-case setting. Second, the process underlying elevated tissue-specific cfDNA is not fully understood; for example, it could reflect an increased rate of cell death in the tissue of origin, enhanced turnover rate, or disruption of local removal of debris from dying cells. Regardless, elevated cfDNA appears to correlate well with clinical cGVHD. Third, our study was designed for diagnostic purposes only and not for predictive, prognostic, or response to treatment purposes. These need to be explored in separate, well-designed, prospective, longitudinal studies including a large cohort of patients who have undergone transplants. This research should delve into the changes in cfDNA patterns over various time points, commencing before the conditioning protocol, spanning the transplantation phase, and encompassing the periods of acute and chronic GVHD. Fourth, this is a single center study and further validation studies using independent cohorts from additional centers are needed. In conclusion, we demonstrate the potential utility of tissue-specific methylation markers for objective and clinically useful detection of cGVHD. We envision that cfDNA biomarkers can transform GVHD treatment into a highly personalized process, where patients are monitored by liquid biopsy many times after transplant and during treatment to monitor disease and adjust treatment.

Authorship note: BA and DN are co–first authors. BACKGROUND. Accurate detection of graft-versus-host disease (GVHD) is a major challenge in the management of patients undergoing hematopoietic stem cell transplantation (HCT). Here, we demonstrated the use of circulating cell-free DNA (cfDNA) for detection of tissue turnover and chronic GVHD (cGVHD) in specific organs. METHODS. We established a cocktail of tissue-specific DNA methylation markers and used it to determine the concentration of cfDNA molecules derived from the liver, skin, lungs, colon, and specific immune cells in 101 patients undergoing HCT. RESULTS. Patients with active cGVHD showed elevated concentrations of cfDNA, as well as tissue-specific methylation markers that agreed with clinical scores. Strikingly, transplanted patients with no clinical symptoms had abnormally high levels of tissue-specific markers, suggesting hidden tissue turnover even in the absence of evident clinical pathology. An integrative model taking into account total cfDNA concentration, monocyte/macrophage cfDNA levels and alanine transaminase was able to correctly identify GVHD with a specificity of 86% and precision of 89% (AUC of 0.8). CONCLUSION. cfDNA markers can be used for the detection of cGVHD, opening a window into underlying tissue dynamics in patients that receive allogeneic stem cell transplants. FUNDING. This work was supported by grants from the Ernest and Bonnie Beutler Research Program of Excellence in Genomic Medicine, The Israel Science Foundation, the Waldholtz/Pakula family, the Robert M. and Marilyn Sternberg Family Charitable Foundation and the Helmsley Charitable Trust (to YD). Tissue-specific DNA methylation markers allow the monitoring of tissue damage in chronic graft-versus-host disease via circulating cell-free DNA.

Author contributions BA, BG, SJS, RS, and YD designed the research study. BA, DN, OGR, SG, IFF, and SP conducted experiments. BA, ES, OGR, IA, BG, SJS, RS, YD, and JM analyzed data. BA, SG, MK, AF, PS, TZ, and ALS provided reagents. BA, ES, IA, BG, RS, YD, and JM wrote the manuscript. BA and DN are co–first authors. The order of names in the paper, as well as the order of co–first authors, was determined based on the relative contributions, which were in different areas (study design, patient access and data interpretation, methylation analysis, and statistical analysis). Supplementary Material

This work was supported by grants from Israel Innovation authority (to YD and BA), the Ernest and Bonnie Beutler Research Program of Excellence in Genomic Medicine, The Israel Science Foundation, the Waldholtz/Pakula family, the Robert M. and Marilyn Sternberg Family Charitable Foundation, and the Helmsley Charitable Trust (to YD). YD holds the Walter and Greta Stiel Chair and Research grant in Heart studies. We thank Abed Nasereddin and Idit Shiff from the Interdepartmental Unit of the Hebrew University of Medicine for their support with DNA sequencing. 11/16/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e163541

oa_package/dd/32/PMC10786696.tar.gz

PMC10786697

0

Conclusions and future directions The analysis in this Review indicates that metabolic reprograming is a reasonable strategy for enhancing antitumor immunity in TME. In a general sense, characterization of the immune component of the TME could be considered when deciding the metabolic approaches to alter the TME and could inform an appropriate strategy to directly manipulate local inflammation signals to achieve better outcomes. For example, the TME of glioblastoma is largely myeloid in nature and lacks infiltrating CD8 + T cells; thus, targeting myeloid metabolism to shift the TME to a more inflamed phenotype could unlock the therapeutic benefits of anti-PD1 or other checkpoint blockades. Additionally, these metabolic pathways may overlap, likely due to the nutrient-scarce TME; therefore targeting one metabolic pathway may also inhibit tumor cell activity while abrogating immunosuppressive myeloid cell functions. However, current drugs are lacking specificity, such as those with myriad off-target effects (metformin, etomoxir, and others); lacking tumor specificity; or, in the case of CNS tumors, lacking BBB penetrance. These drug design limitations in the metabolic field make them unusable in a precision medicine setting and more so in most brain tumor settings. In general, careful consideration of the location of the tumor, its metabolic niche, and immune cell component are all factors to consider when studying and establishing novel metabolic targets to improve host-mediated immune rejection of the tumor. One strategy would be the use of myeloid cell–targeting lipid-coated nanoparticles ( 77 , 160 ). In this Review, we described numerous instances in which metabolic modulation led to improved outcomes both in mono- and combinatorial therapeutic settings. Of the described metabolic pathways within TAMs, arginine metabolism can be prioritized as a pathway for disruption, especially when paired with an immune checkpoint blockade or other immune-based therapy. However, the timing and targeting for this approach would need to be carefully considered, because limiting arginine metabolism within T cells would inhibit their cytotoxic functions and induce cell cycle arrest ( 114 , 161 ). Aside from arginine metabolic inhibitors, glutaminase inhibitors have the potential to target both immunosuppressive TAMs and the tumor, as glutaminolysis is essential for both cells. Glycolytic inhibitors are also a therapeutic candidate given their effect on TAMs and direct effect on tumor growth. Overcoming the protumoral role of TAMs, while sparing adaptive immunity, is a significant challenge that remains to be overcome in tumors, and with properly designed drug delivery methods, such as the packaging of therapies within TAM-targeting nanoparticles, proper cell-level specificity may be achieved ( 77 , 162 , 163 ). To highlight the challenges of using therapies to target immunometabolism in tumors we have included tables to illustrate key considerations ( Table 1 ) and other prerequisite for metabolic drug discovery ( Table 2 ) that are essential when considering these approaches. Once these goals are achieved, a new era of precision medicine may include metabolic phenotyping of the patient tumors for targeted metabolomic strategies.

Immunometabolism is a burgeoning field of research that investigates how immune cells harness nutrients to drive their growth and functions. Myeloid cells play a pivotal role in tumor biology, yet their metabolic influence on tumor growth and antitumor immune responses remains inadequately understood. This Review explores the metabolic landscape of tumor-associated macrophages, including the immunoregulatory roles of glucose, fatty acids, glutamine, and arginine, alongside the tools used to perturb their metabolism to promote antitumor immunity. The confounding role of metabolic inhibitors on our interpretation of myeloid metabolic phenotypes will also be discussed. A binary metabolic schema is currently used to describe macrophage immunological phenotypes, characterizing inflammatory M1 phenotypes, as supported by glycolysis, and immunosuppressive M2 phenotypes, as supported by oxidative phosphorylation. However, this classification likely underestimates the variety of states in vivo. Understanding these nuances will be critical when developing interventional metabolic strategies. Future research should focus on refining drug specificity and targeted delivery methods to maximize therapeutic efficacy.

Therapeutic potential of targeting macrophage metabolism in cancer Immunometabolism is a rapidly growing area of study that explores how immune cells employ various nutrients to support their growth and functionality. The metabolic programming of immune cells has wide-ranging effects on different disease processes, and a true understanding of these processes is critical to the future of immunotherapies for diseases. In solid tumors, tumor-associated macrophages (TAMs) are abundant and control multiple aspects of tumor growth ( 1 , 2 ), including immune suppression and evasion through mechanisms such as TGF-β and IL-10 ( 3 – 8 ); the promotion of angiogenesis through the secretion of VEGF ( 9 – 12 ), which mediates resistance to chemotherapy by protecting tumors from oxidative stress; and the promotion of tumor growth after radiation ( 13 – 18 ). Despite their central role in tumor biology, we still lack a proper understanding of how TAM metabolism influences tumor growth and antitumor immune responses. Understanding the specific metabolic functions in the tumor microenvironment (TME) is critical for developing tumor-extrinsic immune-mediated therapies to improve cancer outcomes. TAMs coordinate an immunosuppressive TME Most circulating myeloid cells arise from stem cells in the bone marrow (termed hematopoietic lineage cells), and they include neutrophils, basophils, mast cells, dendritic cells, macrophages, and monocytes. In solid tumors, these immune cells play diverse roles in tumor progression, metastasis, and immunosuppression. In this Review, we will be focusing on TAMs and their immature precursors, monocytes ( 19 ). TAMs are distinct from the resident microglia population, which is derived from yolk sac progenitors as opposed to hematopoietic lineages ( 20 ). Tissue-resident cells, such as microglia in the brain, Langerhans cells in the skin, and liver-resident Kupffer cells, also play roles in malignancies ( 21 – 25 ), but they will not be discussed in this Review. The immunological status of TAMs can be aligned to proinflammatory M1 or immunosuppressive M2, but this is an oversimplification, and TAM immune phenotypes exist in a continuum. Because of this, there are conflicting results regarding the attribution of one metabolic pathway ascribed to a specific TAM immune phenotype. Notably, there is no direct association of a specific metabolic pathway with a TAM polarization state. As illustrated in Figure 1 , we have provided an overview of TAM metabolic processes and their relative contribution to pro- or antiinflammatory phenotypes. Aerobic glycolysis versus mitochondrial glucose oxidation in TAMs Metabolism is required for the proper function of all cells, and it can be categorized into catabolic (from the Greek root of “breaking down”) or anabolic (from the Greek root word of “upward”) metabolism. Most studies on immunometabolism in tumors have focused on catabolic processes, as nutrient scarcity typically drives catabolism in immune cells ( 26 ). The term “glycolysis” typically refers to the breakdown of glucose to lactate, even though it truly refers to the generation of pyruvate. The conversion of pyruvate to lactate in the presence of oxygen is defined as “aerobic glycolysis” or the “Warburg effect.” The importation of glucose-derived pyruvate into the mitochondria is called glucose oxidation or aerobic respiration. In this Review, we will use aerobic glycolysis to refer to lactate production and oxidative phosphorylation (OXPHOS) or tricarboxylic acid (TCA) cycle when discussing glucose oxidation. Analysis of macrophages demonstrates an upregulation and dependence on aerobic glycolysis in response to inflammatory stimuli such as TLRs ( 27 ). Activation of TLR4 by LPS or other pathogen-associated molecular patterns results in elevated aerobic glycolysis through the PI3K/AKT signaling cascade; an increase in lactate production; expression of activation surface markers, such as CD40, CD80, and CD86; and elevated GLUT1 glucose transporters — effects that can be reversed in the presence of immunosuppressive IL-10 signaling ( 27 , 28 ). Subsequent studies have shown that there is a metabolic “break” of glucose oxidation centered within the TCA cycle. This interruption typically occurs at the isocitrate dehydrogenase (IDH) step of the TCA cycle. Specifically, it involves the conversion of isocitrate to α-ketoglutarate (α-KG), a critical point in the cycle in which citrate normally feeds into subsequent oxidative OXPHOS processes. The break, known as “reverse TCA cycle flux” or “TCA cycle rewiring,” involves the diversion of citrate away from the traditional energy production pathway and its conversion into itaconate, a metabolite with immunomodulatory properties. This metabolic rewiring is predominantly observed in macrophages and has been linked to their inflammatory responses. While this metabolic rewiring is prominently recognized in inflammatory macrophages and dendritic cells, further investigation is required to determine whether this phenomenon extends to other cell populations ( 27 , 29 , 30 ). Inflammatory processes induce a break in the TCA cycle so that intermediates can be used as anabolic intermediates in other pathways that support the inflammatory response ( 27 , 30 , 31 ). Examples of this include (a) the malonylation of glyceraldehyde 3-phosphate dehydrogenase preventing its binding to TNF-α transcripts; (b) the shunting of succinate to stabilize HIF-1α via inhibition of prolyl hydroxylase domain (PHD) enzymes; and (c) the accumulation of citrate and itaconate to inhibit OXPHOS, thereby allowing the macrophages to assume a more glycolytic phenotype ( 32 – 35 ). Succinate is shunted out of the TCA cycle to stabilize HIF-1α, further promoting glycolysis and the production of IL-1β ( 35 ). Citrate can be transported out of the mitochondria to be used as an alternate source of cytosolic NADPH via the IDH1 and IDH2 shuttle ( 36 ). Indeed, the downstream product of IDH1 and IDH2, α-KG, can shift inflammatory activation by suppressing HIF-1α activity and functioning as a substrate for the demethylation of H3K27 ( 37 , 38 ). Thus, the ratio of α-KG to succinate correlates with the inflammatory activity of myeloid cells ( 30 , 39 ). Additionally, decreased rates of OXPHOS are thought to be due to the elevated expression of NOS, a key enzyme producing NO that is capable of reversibly binding complex I in the electron transport chain thereby inhibiting downstream OXPHOS ( 31 , 40 , 41 ). These studies highlight both a reduction in TCA activity and a perturbation in OXPHOS in inflammatory macrophages, as they are both dependent on each other to function. Notably, this break in the TCA cycle is not as straightforward among all inflammatory processes. For example, mitochondrial OXHPOS is required for activation of the NLRP3 inflammasome ( 42 , 43 ). Inhibition of mitochondrial electron transport chain complexes I, II, III, and V effectively blocks the activation of the NLRP3 inflammasome. The introduction of exogenous enzymes capable of restoring the function of mitochondria, without inducing the production of ROS, successfully rescues NLRP3 inflammasome activation in the absence of native mitochondrial complex I or complex III activity. In another critical study, authors identified that TCA metabolism is necessary for an inflammatory response ( 44 ). In this study, shortly after TLR ligation, macrophages rapidly generate acetyl-CoA from TCA-generated citrate needed to fuel histone acetylation promoting the expression of inflammatory genes. How can aerobic glycolysis and a broken TCA cycle be critical for inflammatory macrophage activation, while intact TCA metabolism is also essential? To untangle this apparent paradox, one needs to consider the temporal component of these processes. Upon LPS and IFN-γ stimulation, a two-stage remodeling of the TCA cycle occurs: an early stage with a temporary accumulation of intermediates, like succinate and itaconate, and a late stage in which these metabolites diminish, resulting in a progressive breakdown in TCA/OXPHOS, which accompanies inflammatory cell activation ( 29 ). When put into a broader context, the early stages of inflammatory macrophage activation require intact TCA/OXPHOS, but this breaks down over time as inflammatory cells become reliant on aerobic glycolysis. Therefore, the simplified notion that aerobic glycolysis is preferential for inflammatory activation comes with the caveat of longitudinal kinetics. In contrast to inflammatory cells, immunosuppressive myeloid cells have been historically considered less dependent on aerobic glycolysis and more mitochondrial dependent; however, this is an oversimplification. The emergence of pyruvate dehydrogenase kinase 1 (PDK1) has been shown to be a key regulatory step in macrophage polarization, promoting proinflammatory outcomes by restricting commitment to the TCA cycle by inhibition of pyruvate dehydrogenase, whereas loss of PDK1 promotes antiinflammatory outcomes ( 45 ). Under metabolite-restricted conditions, the glycolysis inhibitor 2-deoxyglucose (2-DG) has been used to demonstrate that glycolysis is necessary to fuel immunosuppressive myeloid functions, further contributing to the complexity of metabolic requirements for immunosuppressive myeloid cells in tumors. More specifically, 2-DG can inhibit TAM polarization after immunosuppressive IL-4 treatment but only in glucose-limited conditions and not in conditions sufficient to maintain TCA cycle function, such as high galactose or glutamine supplementation. These data suggest that immunosuppressive TAMs prefer glucose as a fuel for TCA cycle support but have the means necessary to maintain TCA function in nutrient-restricted conditions such as glycolysis inhibition ( 46 , 47 ). Reinforcing these findings, a seminal study identified that mTORC2 and IRF4 work in parallel to upregulate glucose metabolism and promote immunosuppression in tumors ( 48 ). Based on these findings, it is evident that immunosuppressive TAMs prefer to use glucose metabolism to fuel TCA cycle turnover and thereby promote OXPHOS, but they are also capable of adapting their metabolic programing to utilize other sources. Research by the Reinfeld group serves as an example of the highly glycolytic nature of TAMs. Their work showed that in colorectal cancer TAMs consume more glucose than any other cells within the TME, surpassing even the tumor cells ( 49 ). This study indicated that glutamine is the preferred substrate for tumors in vivo, calling into question how much the Warburg effect is essential to tumor growth in vivo. Indeed, two landmark studies have shown the necessity of OXPHOS in tumor growth ( 50 – 53 ). Collectively, the findings from these studies indicate that, in an in vivo setting, the target of glycolytic inhibition may be primarily the TAMs rather than tumor cells. The pentose phosphate pathway supports inflammatory myeloid activation The pentose phosphate pathway (PPP) is critically associated with glycolytic metabolism, but it is understudied due to a lack of pathway-specific reagents. The PPP is an anabolic pathway that shares several enzymes with the glycolytic pathway and is necessary for shunting glucose-6-phosphate away from glycolysis to rapidly restore the depleting NADPH pool in inflamed myeloid cells ( 30 ). Although inflamed myeloid cell metabolism relies on PPP replenishment of NADPH, immunosuppressive TAMs upregulate the sedoheptulose kinase CARKL, suppressing flux to the PPP ( 54 ). NADPH, in this context, serves as a pool of reductive power to replenish rapidly depleting glutathione pools, thus protecting the cell from ROS. Myeloid cell reliance on PPP to produce NADPH further illustrates the complex coordination of metabolic intermediates in myeloid cell inflammation. Fatty acid metabolism Fatty acid oxidation (FAO) is another essential input into the mitochondria and a robust source of metabolic acetyl CoA, FADH 2 , and NADH when glycolysis is insufficient for the energetic needs of a cell. Fatty acids can be imported via CD36 and FATP1 or generated de novo via lipolysis ( 55 , 56 ). FAO occurs after fatty acids are transported through the mitochondrial membrane via the carnitine palmitoyltransferase (CPT) system ( 57 – 59 ). CPT1 is located on the outer mitochondrial membrane and transports long-chain fatty acids but not medium- or short-chain fatty acids ( 60 ). CPT1 converts acyl-CoA to acyl-carnitine, which then can be converted to acyl-CoA by inner mitochondrial membrane-bound CPT2. It was historically thought that immunosuppressive myeloid cells increase their reliance on FAO to maintain their high energetic needs; however, this interpretation assumed that etomoxir was an FAO inhibitor. The studies that demonstrated that FAO is important for immunosuppressive myeloid function used etomoxir at concentrations that inhibited mitochondrial metabolism, resulting in changes to acetyl-CoA pools in macrophages or induced severe oxidative stress ( 61 , 62 ). Additionally, in a subsequent study, CPT2 knockout in bone marrow–derived macrophages did not increase oxygen consumption rates in response to the addition of palmitate in the presence of IL-4 while maintaining immunosuppressive gene expression ( Arg1 , Mgl2 , Retnla ) and immunosuppressive markers (CD206 and CD301), demonstrating that FAO, as assessed by oxygen consumption rates, is not required for IL-4–induced immunosuppressive polarization ( 63 ). Another study using concentrations of etomoxir that have more selective effects demonstrated that FAO is dispensable for M2 polarization ( 64 ). As such, the role of FAO on myeloid immunosuppression may be more dispensable than previously thought. Bone marrow–derived myeloid cells activated with LPS tend to increase fatty acid anabolism, in contrast to the fatty acid catabolism increase in immunosuppressive myeloid cells. Increased fatty acid uptake and de novo synthesis are hallmarks of LPS stimulation and revolve around SREBP1 signaling in late-phase activation of inflammatory myeloid cells ( 65 , 66 ). TLR4 signaling, which upregulates aerobic glycolysis, is dependent on the enzyme fatty acid synthase to generate palmitate and is attenuated when fatty acid synthase is inhibited ( 67 ). LPS-mediated increases in fatty acid synthesis result in accumulation of lipid droplets; increased expression of PGE2, IL1-β, and NOS; and the resulting production of NO, while reducing the ability of the TCA cycle to perform FAO ( 68 – 70 ). Further complicating the paradigm, studies with etomoxir in LPS-stimulated myeloid cells have shown an attenuation of inflammatory outputs, specifically mitochondrial ROS and NLRP3 inflammasome pathways, which could be attributed unknown off-target effects of the drug or an unknown dependence on some level of FAO by inflammatory myeloid cells ( 71 , 72 ). Targeting glycolysis and mitochondrial metabolism in TAMs Given the pivotal influence of glucose metabolism on TAM functions, reconfiguring these metabolic processes may offer a promising pathway to enhance current immunotherapy treatments. When considering combinatorial approaches to treating cancer, the metabolic phenotype and tumor-associated immune cell compartment should be considered. A general trend across most immunologically “cold” tumors is a hypoxic and metabolite-scarce TME due to resource limitations, leading to an exclusion of cytolytic CD8 + T cells and promotion of the activity of immunosuppressive TAMs ( 73 ). Thus, restricting the ability of TAMs to support this phenotype via metabolic inhibition may be a novel approach to controlling the TME, with the added benefit of also inhibiting tumor cell growth directly, as tumor cells tend to prefer high metabolic rates that rely on increased glucose consumption, increased glycolysis, and elevated TCA cycle turnover. Metformin, an antidiabetic drug that has numerous off-target effects, including inhibiting mitochondrial complex 1, has been used to target TAM metabolism by modulating the AMPK/mTOR/NF-κB signaling axis ( 74 – 78 ). These studies indicate that metformin can promote the antitumor phenotype of TAMs. However, the specific target of metformin that is responsible for proinflammatory skewing is not well defined. 2-DG is another metabolic inhibitor used in tumor treatment with effects that can be partially attributed to TAM reprogramming. In preclinical models of melanoma, 2-DG decreases TAM expression of Arg1 , Fizz , Mrc1 (encoding CD206), and Vegf , indicating that 2-DG diminishes immunosuppression within the TME ( 46 ). In another study, conditioned media from pancreatic tumor cells increased the aerobic glycolysis in macrophages, which resulted in prometastatic/angiogenic phenotypes, whereas treatment with 2-DG could reverse these effects ( 79 ). Such evidence further suggests that TAMs can rely on glucose metabolism to support both protumor and antiinflammatory phenotypes in vivo. Notably, 2-DG treatment has several off-target effects ( 80 , 81 ). Therefore, caution should be used when interpreting the metabolic effects of these compounds on TAM activities in tumors. Phosphofructokinase (PFK) inhibition has also been shown to abrogate inflammatory phenotypes in macrophage cultures in vitro. However, tumor-conditioned media is also capable of inducing expression of PFK in immunosuppressive TAMs ( 82 ). Pyruvate kinase M2 (PKM2) exerts control over TAM expression of PD-L1 in pancreatic cancer, while also supporting tumor cell growth, exemplifying the dual benefit of glycolytic perturbation ( 83 , 84 ). Indeed, lactylation of PKM2 has been shown to prevent inflammatory activation of macrophages ( 85 ). An inhibitor of HIF-1α, PX-478, halts glycolysis in tumor cells, leading to better responses to radiotherapy in prostate cancer, prevention of metastases in small cell lung cancer, and enhanced immunogenic cell death in pancreatic cancer ( 86 – 88 ). Inhibition of HIF-1α also triggers a shift from immunosuppressive phenotypes and decreased PD1/PD-L1 expression ( 89 , 90 ). Therefore, targeting glycolysis (and FAO) may exert therapeutic impact on immunosuppressive TAMs ( 91 ). For many brain tumor malignancies, a translational barrier for metabolic manipulation is sufficient blood-brain barrier (BBB) penetration. There are a variety of strategies for overcoming the BBB that do not rely on medicinal chemistry strategies, such as BBB-opening ultrasound or conjugating long-carbon chains to targeting RNA linkers, but these have not been leveraged yet for metabolic therapeutic approaches ( 92 – 95 ). CNS penetrant drugs, while scarce, have been developed, such as the 2-DG mimetic dichloroacetate ( 96 ). The 2-DG prodrug WP1122 has direct cytotoxic effects on glioblastoma cell survival when combined with histone deacetylase inhibitors ( 97 ). Dichloroacetate is being evaluated in an ongoing phase II clinical trial of patients with glioblastoma (NCT05120284) ( 91 ). Another metabolic modulation strategy that is in clinical trials is tamoxifen, which is primarily used as an estrogen receptor mimetic but also has electron transport chain inhibitory properties ( 98 – 100 ). An ongoing phase II clinical trial using tamoxifen has demonstrated safety (NCT04765098), but results of efficacy have not yet to be shown. In glioblastoma, nutrient limitations result in the glioma cells relying on FAO to support cellular processes and proliferation, which can be targeted with the use of etomoxir ( 101 , 102 ). These findings in glioblastoma need to be carefully interpreted, as the heterogeneity of the tumor and TME, along with off-target effects, may lead to an overinterpretation of the reliance of tumor cells on fatty acid metabolism ( 103 ). Further work to produce brain-penetrant glycolysis– and OXPHOS–modulating drugs is needed to adequately target the metabolic phenotypes of TAMs within the TME. An overview of the strategies used to perturb glycolysis, oxidative phosphorylation, and fatty acid oxidation in TAMs can be found in Figure 2 . Arginine metabolism The amino acid arginine provides the most direct link between metabolites and myeloid functions in tumors. Arginine is an essential nutrient in tissues, serving as a key component in protein synthesis. Studies in pancreatic cancer identify it as the most depleted metabolite in the interstitial milieu (pyruvate, tryptophan, and cysteine were also significantly depleted) ( 104 ). Our own data show similar depletion in glioblastoma, along with several other limiting amino acids, such as glutamine and l -aspartic acid, which is likely secondary to consumption by myeloid cells ( 105 , 106 ). Arginine is by no means the only metabolite depleted in tumors, as other studies have shown that serine, exogenous fatty acids, aspartate, glutamine, and serine can also be limiting ( 107 – 110 ). As tumors gain biomass, there is a requirement for arginine as a proteinogenic amino acid ( 111 ). Arginine metabolism has been at the center of macrophage/myeloid cell biology for over two decades. The original M1/M2 description of macrophage polarization hinges largely on the split between metabolic selection of iNOS activity versus arginase-1 (Arg1) activity, respectively ( 112 ). While it is recognized that strict M1/M2 phenotypes are an oversimplification in vivo, it holds true that myeloid cells consume large amounts of arginine ( 113 ). This myeloid cell consumption of arginine outcompetes T cells, which requires this metabolite for T cell proliferation ( 114 ). Arginine deprivation perturbs T cell activation, and L-arginine supplementation enhances T cell survival and effector functions in vivo ( 114 ). This consumption also has effects on tumor cells. A recent study showed that pancreatic tumors must upregulate arginine biosynthesis to compensate for this competition for local arginine ( 115 ). The byproducts of arginine metabolism, polyamines, have long been associated with both increased malignancy and immunosuppression ( 116 ). In the context of glioblastoma, previous work from our group showed that TAMs actively produce polyamines to buffer themselves within the acidic TME ( 117 ). We also found that blockade of polyamine synthesis was sufficient to enhance survival in preclinical models of glioblastoma, an effect dependent on host immunity. Our group has recently found that TAMs produce creatine from arginine, which is being fed to tumor cells to allow their survival in the hypoxic niche ( 118 ). Arginine is also a substrate for proline, which promotes the production of collagen fibrils and subsequent fibrosis, which promotes immune exclusion from tumors ( 119 – 121 ). Finally, there is evidence that polyamines can directly promote macrophage alternative activation. The hypusination of EIF5a, facilitated by the polyamine spermidine, exerts direct regulation over the transcription and translation of mitochondrial proteins in macrophages ( 122 ). The polyamines generated by IL-4–induced macrophages promote alternatively activated macrophage activation by stimulating OXPHOS and the TCA cycle. Other byproducts of arginine metabolism produced by myeloid-derived suppressor cells (MDSCs) or macrophages, such as NOs and peroxynitrites, directly inhibit several aspects of T cell activity and function through nitrosylation ( 123 , 124 ). iNOS/NOS2 is an enzyme that is expressed in cells in response to inflammatory activity or hypoxia. It produces NO by converting l -arginine to l -citrulline ( 125 , 126 ). NO reprograms macrophage mitochondrial metabolism by generating nitroxyl, which inhibits the pyruvate dehydrogenase complex. This causes macrophages to enter a glycolytic state, thereby reducing ATP production by the TCA cycle ( 127 – 129 ). These changes, which are mediated by LPS, promote inflammatory polarization, which leads to increased cytokine and NO production by macrophages ( 128 ). This process is kept in balance by the immunosuppressive cytokine IL-10, which is modulated by a commitment to aerobic glycolysis and, in turn, modulates NO-mediated suppression of OXPHOS ( 130 ). Proinflammatory macrophage polarization can have a variety of pro-oncogenic effects, including promoting pathways for angiogenesis, the epithelial-mesenchymal transition, cancer cell survival, proliferation, and metastasis ( 125 , 126 ). iNOS/NOS2-targeting therapies have been studied in several in vitro and in vivo models ( 131 – 133 ). However, clinical trials of NOS inhibitors are scarce because of concerns about off-target effects ( 126 ). Therapeutic targeting of arginine metabolism Due to the apparent central importance of arginine catabolism in promoting immunosuppression, several strategies have been developed to target this pathway to promote antitumor immunity ( Figure 3 ). The small-molecular inhibitor CB-1158 that blocks Arg1 enhances immunotherapy in murine models of colorectal tumors, melanoma, and breast cancer and has now been advanced to clinical trials ( 134 ). Although it was demonstrated safe in the context of immunotherapy, efficacy studies have not yet been reported ( 135 ). Several other Arg1 inhibitors are being tested or are in development for future therapeutic use ( 136 , 137 ). The inhibition of polyamines is another strategy for modulating arginine metabolism. Several studies have shown that polyamine inhibition using difluoromethylornithine (DFMO) can be an effective way to enhance immunotherapy ( 138 – 141 ). In an initial study, combination treatment involving DFMO and a synthesis inhibitor (AMXT-1501) effectively reversed immune suppression in melanoma mouse models ( 142 ). Subsequently, in a preclinical model of melanoma and colon carcinoma, inhibition of polyamine activity led to increased tumor infiltration of granzyme B + IFN-γ + CD8 + T cells, while concurrently reducing immunosuppressive TAMs ( 143 ). These findings align with those of our recent investigation, wherein we observed that polyamine blockade facilitated the infiltration of CD8 + T cells and synergistically enhanced animal survival when combined with anti-PD1 or anti–PD-L1 immunotherapy in murine models of glioblastoma ( 117 ). Furthermore, the combination of polyamine blockade and immune checkpoint blockade exhibited a synergistic effect in murine models of mammary carcinogenesis ( 144 ). Collectively, these studies demonstrate that polyamine blockade has the potential to modify the TME in a manner that promotes T cell–dependent antitumor responses. Finally, blocking iNOS-mediated immunosuppression has also shown effects in promoting an antitumor response. For example, phosphodiesterase-5 (PDE5) inhibitors, such as sildenafil, which interfere in cGMP-dependent iNOS signaling, have been shown to prevent MDSC-mediated immunosuppression ( 145 – 147 ). PDE5 inhibition can reduce iNOS and Arg1 activity in MDSCs, thereby triggering antitumor response and T cell infiltration in preclinical models of colon cancer and breast cancer ( 145 ) or in patients with end-stage multiple myeloma ( 146 ). PDE5 inhibition also prevents MDSC-induced NK suppression, increasing NK cytotoxicity in murine models and in humans with abdominal malignancy ( 147 ). A phase II study has been initiated to investigate the effects of combining nivolumab, tadalafil, and oral vancomycin in patients diagnosed with refractory primary hepatocellular carcinoma or liver-dominant metastatic cancer originating from colorectal or pancreatic cancers ( 148 ). The results of this combinatorial therapy have not yet been reported. Other amino acid oxidizing pathways in TAMs TAMs also express other immunosuppressive amino acid–oxidizing enzymes, such as IDO and IL4i1. IDO is an intracellular, tryptophan-metabolizing enzyme that functions through its catalysis of the rate-limiting step of the kynurenine pathway. IDO is highly expressed in TAMs and MDSCs in the TME and plays a role in tumor immune escape ( 149 , 150 ). Tryptophan metabolism and subsequent depletion via IDO expression in macrophages has been shown to inhibit antigen-specific T cell proliferation and activation ( 151 ). Additionally, kynurenine has been shown to directly activate the aryl hydrocarbon receptor, and this activation leads to the generation of immunosuppressive FoxP3 + regulatory T cells ( 152 ). In patients with breast cancer, increased IDO-expressing MDSC populations are correlated with increased amounts of these FoxP3 + regulatory T cells and a poor prognosis ( 153 ). Several IDO inhibitors in combination with other immune therapeutics are being actively investigated in a number of clinical trials across a broad array of cancers. IL4i1 is a l -amino acid oxidase secreted by APCs that primarily functions to oxidize phenylalanine ( 154 ). Expression of IL4i1 in TAMs has been shown to prevent T cell proliferation and cytokine production, decrease the CD8 + T cell response, and promote the differentiation of naive CD4 + T cells into FoxP3 + regulatory T cells ( 155 , 156 ). In addition to suppressing the antitumoral T cell response, IL4iL expression has also been shown to recruit immunosuppressive MDSC populations to the TME in a mouse model of melanoma ( 157 ). Glutamine metabolism Glutamine is the most abundant amino acid found in the human body, and it is conditionally essential in times of catabolic stress. Once transported into a cell, glutamine is lysed into ammonium ion and glutamate by mitochondrial glutaminases in a process known as glutaminolysis. The resultant glutamate can subsequently be converted into α-KG by glutamate dehydrogenas or aminotransferases, allowing anapleurotic reactions to support TCA function. As a major source of carbon and nitrogen, glutamine is essential for production of amino acids, purine, pyrimidines, and lipids. Additionally, glutamine-derived glutamate can be utilized in synthesis of glutathione, which is used to neutralize ROS and maintain redox balance. As highly proliferative cells with large anabolic requirements, tumor cells exhibit particularly high levels of glutamine uptake and dependence. Glutamine metabolism plays a pivotal role in macrophage activation and polarization. Glutamine-derived α-KG is required for the differentiation of macrophages to an antiinflammatory, immunosuppressive phenotype, and induction of endotoxin tolerance ( 30 ). Glutamine deprivation and glutaminolysis inhibition using a GLS1 inhibitor decreased expression of immunosuppressive genes while upregulating inflammatory genes such as IL-1 β and Tnf in bone marrow–derived macrophages ( 37 ). The α-KG derived from glutamine limits the activation of an inflammatory macrophage phenotype via suppression of the NF-κB pathway by functioning as a substrate for PHD inhibition of HIF-1α and suppression of IKKβ ( 37 ). Furthermore, α-KG directly supports epigenetic changes, specifically JMJD3-dependent H3k27me3 demethylation, allowing for the transcription of key immunosuppressive genes like Il10 , Tgfb , and Arg1 , and it can reverse chronic inflammatory phenotypes in alveolar macrophages ( 37 , 38 ). It is important to note that while α-KG can be produced by IDH2/3 within the TCA cycle, or IDH1 within the cytosol, prior studies have exclusively focused on α-KG derived from glutaminolysis ( 37 – 39 ). Extracellular glutamine, along with arginine, is required to produce NO in murine macrophages. Thus, glutamine contributes to the polarization of macrophages toward an immunosuppressive phenotype like those of TAMs in the TME. Despite the pivotal role for glutamine metabolism in macrophage polarization and function, very little is known about how these processes work within the TME. Glutamine metabolism is a promising target to sensitize tumors and their immunosuppressive microenvironments toward immunotherapy. JHU083, a prodrug version of the glutamine antagonist 6-diazo-5-oxo-L-norleucine, is a glutamine metabolism inhibitor that is selectively activated in the TME to mitigate toxicity; it has been shown to inhibit tumor growth and promote survival in tumor-bearing mice, particularly in combination with immunotherapy ( 158 ). Administration of this prodrug augmented endogenous antitumor immunity, as it promoted activation, proliferation, and memory in tumor-infiltrating lymphocytes ( 158 ). Additionally, JHU083 inhibits the recruitment of immunosuppressive MDSCs to the TME and induces MDSC apoptosis while simultaneously reprogramming MDSCs and TAMs to a proinflammatory antitumor phenotype ( 159 ). Notably, glutamine inhibition via JHU083 increased the effectiveness of anti-PD1 and anti-CTLA4 checkpoint blockade in tumors that did not benefit from monotherapy ( 159 ). An outline of the glutamine metabolic pathway and current targeting strategies can be found in Figure 3 .

01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e175445

oa_package/21/03/PMC10786697.tar.gz

PMC10786698

37962965

Introduction Adolescent idiopathic scoliosis (AIS) is a condition in which the spine is deformed with a lateral curvature exceeding 10 degrees in otherwise healthy adolescents ( 1 , 2 ). With a prevalence of 0.47% to 5.2% in adolescents worldwide, AIS is the most common pediatric skeletal disorder and usually worsens during the pubertal growth spurt ( 1 – 5 ). In severe cases, AIS can cause cardiopulmonary difficulties, leading to shortness of breath and potential mortality ( 2 , 6 – 9 ). Despite its high prevalence and long-term physical and mental health implications, AIS lacks an agreed-upon theory of etiopathogenesis, which severely impedes the rational development of early diagnostic, preventive, and therapeutic strategies ( 1 , 2 , 4 ). Although the causes of AIS are believed to be multifactorial, population and twin studies suggest a strong contribution of genetic factors to the development of AIS ( 10 – 12 ). Many common or rare variants in coding or noncoding regions of genes (e.g., LBX1 , GPR126 , PAX1 , CHL1 , POC5 ) have been identified as being associated with AIS ( 13 – 19 ). However, the vast majority of the heritability of AIS is still unexplained, and causative mechanisms linking the susceptible genes to AIS remain unclear. Glycine is a crucial neurotransmitter that plays a role in both inhibitory and excitatory neurotransmission in the CNS ( 20 ). In the spinal cord and brain stem, glycine mainly acts as an inhibitor by binding to ionotropic glycine receptors (GlyRs), leading to postsynaptic hyperpolarization and inhibition of neural activities. Extracellular glycine levels are tightly controlled by 2 glycine transporters, glycine transporter 1 (GLYT1) and GLYT2. GLYT1 is primarily expressed by astrocytes adjacent to glycinergic neurons to facilitate the rapid clearance of glycine from the synaptic cleft ( Supplemental Figure 1 ; supplemental material available online with this article; https://doi.org/10.1172/JCI168783DS1 ) ( 21 – 23 ). Homozygous mutations in the SLC6A9 gene that encodes GLYT1 cause glycine encephalopathy, also known as nonketotic hyperglycinemia (NKH), which is a severe neurological disease caused by abnormally high levels of glycine in the cerebrospinal fluid (CSF) and characterized by respiratory failure, progressive hypotonia, and startle-like reflexes ( 24 , 25 ). Interestingly, although most glycine encephalopathy patients die before 7 months of age, those who survive show progressive early onset scoliosis as a result of apparent neurological defects ( 26 – 28 ). In this study, we investigate the genetic basis and pathogenic mechanism of AIS in a multicenter cohort of patients. Linkage analysis and genome sequencing identified a number of rare heterozygous variants in SLC6A9 , which were mostly deleterious, affecting the membrane presentation and glycine uptake of GLYT1. The AIS patients exhibited increased plasma glycine levels and aberrant paraspinal muscle activities. In the zebrafish model, disruption of slc6a9 led to an AIS-like phenotype. We showed that disturbance in the normal function of central pattern generators (CPGs) by either excessive glycine or developmental defects resulted in lateral spinal curvature. We further tested the feasibility of treating this deficiency to prevent scoliosis.

Methods Study participants. Subjects diagnosed with AIS were recruited from the DKCH, PUMCH, and SRC. Diagnosis was made by standing whole-spine radiographs. Eligible subjects were patients diagnosed with scoliosis (Cobb angle ≥10°) without manifestation of any congenital or neuromuscular defect at the time of recruitment ( Supplemental Figure 2 and Supplemental Table 1 ). Bilateral sEMG was recorded in a sample group of patients in comparison with 2 controls matched by sex and age. We recruited five 3-generation pedigrees from DKCH. We performed WGS on all families and further evaluated 2 families, families 1 and 2, with an autosomal dominant inheritance pattern because they shared a common candidate gene. The probands of family 1 and family 2 were initially identified at the time of their scoliosis surgery. Further genealogical investigation led to the identification of 2 large multiplex families ( Figure 1A ). We also consecutively recruited a total of 843 sporadic AIS cases, of which 118 patients were subjected to whole-exome sequencing (WES) and 725 patients were analyzed by targeted sequencing. Further genealogical and genetic investigations of the sporadic patients identified 3 trio families with apparent dominant inheritance (family 3-5) ( Figure 1, A–C ). We also recruited 3,219 ethnicity-matched subjects in Hong Kong with no evidence of AIS confirmed by radiographs as the general population controls. We further enrolled 2 additional AIS cohorts from PUMCH ( n = 223) and SRC ( n = 635) ( Figure 1C ). Among the cases from PUMCH, 120 and 103 individuals were analyzed by WES and WGS, respectively. All cases from SRC were analyzed by WES. Plasma glycine concentration assay. The following SLC6A9 variant-carrying AIS patients were recruited for measuring plasma glycine concentrations: 9 patients (II-1, II-3, II-5, II-7, II-8, III-2, III-3, III-6, and III-7) from family 1; 2 patients (II-7 and II-9) from family 2; I-2 from family 3; I-1 from family 5; and 2 sporadic patients (PUMCH_1 and PUMCH_2) from PUMCH. The unaffected family members without SLC6A9 variants were recruited as a control group, including 4 members (I-1, II-2, II-6, and III-1) from family 1, 3 members (II-5, II-10, and III-1) from family 2, and I-1 from family 3. An additional group of 28 adolescents without AIS were recruited and served as age-matched general controls. Fasting plasma samples of recruited individuals were isolated from fresh whole blood by centrifugation at 4000 rpm for 15 minutes. The glycine concentration was measured according to the manual of the fluorometric glycine assay kit (Abcam, ab211100). Fluorescence was read at Ex/Em 535/587 nm in endpoint mode using a microplate reader (Varioskan Flash, Thermo Fisher Scientific). sEMG. sEMG signals were detected with an amplifier of 1,000 times, sample frequency of 2,000 Hz, and filtering band of 15 to 1,000 Hz (YRKJ-A2004, Zhuhai Yiruikeji Co.). Back skin of participants was cleansed with 75% alcohol before electrode placement. Four pairs of silver/silver chloride self-adhesive surface electrodes (Noraxon Dual Electrode) were applied on bilateral paraspinal muscles at thoracic vertebra levels of T3-5 and T9-11 in III-6 and III-7 from family 1 and T3-5 and T5-7 in II-7 from family 1. Signals from T3-5 were used to remove ECG contamination ( 83 ). The recording at T9-11 or T5-7 reflected the paraspinal muscles at the apex of the spinal curvature as illustrated ( Figure 2B ). All subjects were asked to lie on a test bed for surface electrode placement, while the impedance was tested under 10 kΩ. Then they were instructed to relax and stand in an upright posture for 5 seconds as well as proceed with left and right trunk bending. The sEMG signals during left and right trunk bending were used as normalization of standing sEMG measurements. Raw sEMG signals were preprocessed with filtering, zero mean, and ECG removal. In vitro glycine uptake assay. HEK293T cells were plated onto poly- l -lysine–coated 24-well plates (Sigma-Aldrich, P6282) and grown to 50%-60% confluence. Cells were transfected with Flag-GLYT1 WT, Flag-GLYT1 variants, and pCMV-3Tag-1A backbone. The detailed procedure of glycine uptake assay was described previously ( 84 ). Briefly, prior to uptake, the cells were washed 3 times with assay buffer containing 116 mM NaCl, 1 mM NaH 2 PO 4 , 26 mM NaHCO 3 , 1.5 mM MgSO 4 , 5 mM KCl, 1.3 mM CaCl 2 , and 5 mM glucose, and then incubated for 10 minutes with 1 μCi/mL [ 3 H] glycine (60 Ci/mmol, PerkinElmer, NET004001MC) at a final concentration of 200 μM at 37°C. Glycine uptake was terminated by quick washing with ice-cold assay buffer followed by aspiration twice. Cells were digested in 0.1M NaOH, and the supernatants were subjected to scintillation counting (LS6500, Beckman Coulter) and protein concentration measurement using Bradford reagent (Pierce Coomassie Plus Assay Kit, Thermo Fisher Scientific, 23236). [ 3 H]-glycine uptake was calculated as nanomoles per minute per milligram of protein (nmol/min/mg protein) and normalized as a percentage of that in control cells transfected with WT plasmid. Zebrafish lines. Zebrafish embryos were collected from natural mating, maintained in E3 medium at 28.5°C, and staged according to dpf and morphology ( 85 ). WT zebrafish (TU) were used to generate the slc6a9 mutant, dmrt3a mutant, and Tg(elavl3-H2B-GCaMP6s) transgenic zebrafish lines. The slc6a9 mutant zebrafish were crossed into the Tg(mnx1:GFP) background to visualize the motoneurons or into the Tg(elavl3-H2B-GCaMP6s) background to measure neural activities ( 48 , 86 ). The zebrafish slc6a9 ta229g line has been described previously and was obtained from the National BioResource Project Zebrafish (Japan) ( 40 ). Zebrafish genome and transcript information were derived from the updated zebrafish genome annotation (GRCZ11) in the Ensembl database. CHOPCHOP ( https://chopchop.cbu.uib.no/ ) and CRISPRscan ( http://www.crisprscan.org/ ) were used to design sgRNAs ( 87 , 88 ). The sgRNAs targeting the last exon of slc6a9 (CCGTGGCGTATCGACCCTTG) and the first exon of dmr3a (TGCGCGCTGCAGGAACCACG) with minimal off-targeting scores were, respectively, selected. A Nikon SMZ 745T stereomicroscope with Warner Pico-Liter Injector PLI-90A and 3D Manual Micro-Manipulator Fits Micropipette with OD 1 mm platform was used for microinjection, and 1 nL drop of a mix of 100 ng/μL Cas9 mRNA (Alt-R S.p. Cas9 Nuclease V3, Integrated DNA Technologies, 1081059) and 300 ng/μL synthesized sgRNA (Synthego) were microinjected into WT zygotes at the 1-cell stage. The mutant zebrafish lines carrying a deletion of 22 bp mutation of slc6a9 and a deletion of 8 bp mutation of dmrt3a were established, respectively. The founder was bred to WT zebrafish to generate F0 and subsequent F1 mutants. The slc6a9 allele was genotyped using the following primers: slc6a9 -F: AGCACAGCAACTTTTCCAACC; slc6a9 -R: TGCTTCCTGGGATGGTCAGA. The PCR product sizes of WT and mutant slc6a9 allele were 255 and 233 bp, respectively. For genotyping the dmrt3a WT allele, dmrt3a -WT-F: CTGCAGGAACCACGGGGT and dmrt3a -R: AAGTTGCCAGTGTCAATGTT were used. For genotyping the dmrt3a mutant allele, dmrt3a -M-F: GCGCGCTGCAGGGGTGCTGT and dmrt3a -R were used. The PCR product sizes of WT and mutant dmrt3a allele were 501 bp and 498 bp, respectively. The Tg(mnx1:GFP) zebrafish line that labels motoneurons has been described previously and was obtained from YSY Biotech ( 86 ). To visualize neuronal calcium signals in slc6a9 mutant larvae, we generated a transgenic zebrafish line, Tg(elavl3-H2B-GCaMP6s) , using Tol2 construct with elavl3 promoter and human histone H2B that drive the expression of calcium indicator GCaMP6s in the nucleus of all neurons ( 48 ). The Tol2-elavl3-H2B-GCaMP6s construct was microinjected with Tol2 mRNA into zebrafish zygotes at the 1-cell stage. The founder fish were crossed with the WT fish to establish the Tg(elavl3-H2B-GCaMP6s) line, which was further mated with slc6a9 mutant fish to generate slc6a9 m/m ;Tg(elavl3-H2B-GCaMP6s) . Analysis of spinal neural activity. Spinal neural activity of Tg (elavl3-H2B-GCaMP6s) and slc6a9 m/m ; Tg (elavl3-H2B-GCaMP6s) fish was analyzed at 24 hpf. Fertilized eggs were immobilized by 100 μM d -tubocurarine chloride hydrate (Abcam, ab120073) in E3 medium for 10 minutes and embedded with 0.8% to 1% low-melting agarose gel in the lid of 6 cm confocal dishes with desired direction. Neural activity reflected by GCaMP6s calcium signals was visualized by a Nikon Ti2-E Widefield Microscope with a ×40 air LEN. Signals were recorded continuously in a single z plane, and a time series mode was used to record the changes of neuronal GCaMP6 signals within 1 minute at a speed of 10 frames per second (fps). The recorded images were analyzed by ImageJ software (NIH). The ImageJ Time Series Analyzer plugin was used to manually quantify GCaMP6s signals. To characterize calcium signals of specific regions, the regions of interest (ROIs) were defined and the GCaMP6s fluorescence intensities ( ΔF ) of the time-lapse images of each fish were automatically extracted. ΔF for ROI was calculated as ΔF = F ( t ) − F 0, where F 0 is a manually selected baseline and F ( t ) is the GCaMP6s fluorescence intensity at a given time. Relative intensity of GCaMP6s signals was normalized as a percentage of the mean value of ΔF . The left and right alternation index was defined as the number of consecutive pairs of patterned events occurring on opposite sides of the spinal cord, divided by the total number of events minus 1. To compare the alternation index in WT and mutant zebrafish, quantified intensities of total left- and right-side neuronal activities within a 1-minute recording time period were used. Frequency of the left-side neuronal activity was quantified as Hz. Micro-CT. Experimental zebrafish were euthanized with overdosage of MS222 solution (>250 mg/L, Sigma-Aldrich, 10521) and were fixed in 10% neutral-buffered formalin (Sigma-Aldrich, HT501128) overnight at 4°C. Fish were secured in the micro-CT instrument (Skyscan 1076, Burker). The parameters that we used for micro-CT scanning were as follows: source voltage, 40 kV; source current, 250 μA; pixel size, 8.6650 μM without filter. Drug treatment. To phenocopy the axial curvature observed in 7 dpf slc6a9 mutant zebrafish, a selective GLYT1 inhibitor ALX5407 (Sigma-Aldrich, SML0897) was used to treat the WT larvae ( 42 ). At 48 hpf, WT embryos were divided into a vehicle group (kept in E3 medium) and multiple treatment groups, in which the embryos were transferred to fresh E3 medium containing different dosages of ALX5407. Culture medium was changed daily. At 7 dpf, fish from vehicle and ALX5407 treatment groups were imaged for axial phenotype and tracked for swimming behaviors for over 10 minutes. To enhance the penetrance of axial curvature observed in 7 dpf slc6a9 m/+ zebrafish, low dosage of ALX5407 was used. At 48 hpf, embryos from WT and slc6a9 m/+ mating pairs were divided into a vehicle group (E3 medium) and a treatment group (E3 medium containing 10 nM of ALX5407). At 7 dpf, fish from vehicle and treatment groups were imaged for axial phenotype and then lysed for genotyping. A specific GlyR antagonist strychnine (Sigma-Aldrich, S0532) was used to prevent the axial curvature observed in 7 dpf slc6a9 m/m zebrafish. The 6 dpf fish from slc6a9 m/+ and slc6a9 m/+ mating pairs were divided into vehicle group and treatment group. In vehicle group, fish were kept in E3 medium, whereas in treatment group, fish were kept in E3 medium containing 0.5 μM strychnine. After 24 hours, all fish were imaged for axial phenotype and then lysed for genotyping. The human body can rapidly clear sodium benzoate by combining it with glycine to form hippuric acid for excretion ( 54 ). Hence, sodium benzoate (Sigma-Aldrich, B3420) was used as a neutralizer for glycine molecules in zebrafish to determine whether it can prevent the axial curvature of slc6a9 m/m zebrafish. The 2 dpf embryos from slc6a9 m/+ and slc6a9 m/+ mating pairs were divided into a vehicle group and a treatment group. In the vehicle group, larvae were kept in E3 medium, whereas the larvae from the treatment group were kept in E3 medium containing 0.5 ppm sodium benzoate. Culture medium was changed daily. At 7 dpf, all fish were imaged for axial phenotype and then lysed for genotyping. Functional enrichment analysis of AIS GWAS data set. A total of 1,387 SNPs that were significantly associated with AIS and mapped to 1,367 genes were collected from the NHGRI-EBI GWAS catalog database ( 89 ). Gene Ontology (GO) function enrichment analysis of these AIS-associated genes was performed by the clusterProfiler R package ( 90 ). Additional methodological information is provided in the Supplemental Methods . Statistics. Statistical data were analyzed using GraphPad Prism 7 (GraphPad Software). Student’s t test or 1-way or 2-way ANOVA was performed accordingly as indicated in the figure legends. Differences with P values of less than 0.05 were considered statistically significant. The n numbers for each group and group numbers are indicated in the figure or figure legends. Study approval. Ethics approvals were obtained from the Institutional Review Board of the University of Hong Kong/Hospital Authority Hong Kong West Cluster (HKU/HA HKW IRB, reference UW 08-158), PUMCH under the framework of the Deciphering Disorders Involving Scoliosis and COmorbidities (DISCO) study (JS-3545D), and the Institutional Review Board of the University of Texas Southwestern Medical Center (STU 112010-150), respectively. Written, informed consent was obtained from all participants and the participating family members. Zebrafish experiments were conducted in compliance with the Guidelines from The Committee on Use of Laboratory Animals for Teaching and Research (CULATR) of the University of Hong Kong (CULATR 5396-20). Data availability. Values for all data points in graphs are reported in the Supporting Data Values file. Large-size raw genotyping and sequencing data are available for access upon reasonable request.

Results Identification of SLC6A9 variants in AIS patients. We performed a genetic analysis to identify pathogenic variants in a multicenter AIS cohort consisting of multigeneration families, trios, and approximately 1,700 sporadic patients. We first conducted whole-genome sequencing (WGS) on 10 individuals in family 1 (II-1, II-2, II-3, II-5, II-7, II-8, III-1, III-2, III-3, and III-4) and 12 individuals in family 2 (I-1, II-1, II-2, II-3, II-4, II-5, II-7, II-8, II-9, III-3, III-4, and III-6) ( Figure 1, A and B , Supplemental Figure 2 , and Supplemental Table 1 ). Filtering the detected variants identified 3 candidate genes in each of the 2 families, which were all located within a short interval of chromosome 1p ( Supplemental Figure 3A and Supplemental Table 2 ). Linkage analysis of the 2 families also revealed a unique linkage region located on chromosome 1p34.1, with a maximum logarithm of odds (LOD) score greater than 3.0 ( Supplemental Figure 3B ). Intriguingly, each family had a nonsynonymous coding variant (c.1984C>T, p.R662W in family 1 and c.617A>T, p.Y206F in family 2) in SLC6A9 (NM_201649.4; NP_964012.2), which located in the linkage locus and segregated with the phenotype ( Figure 1A and Supplemental Table 2 ). These 2 variants were very rare in the Genome Aggregation Database ( https://gnomad.broadinstitute.org/ ) (gnomAD v3.1.2) (p.Y206F, 1.449 × 10 –4 ; p.R662W, 6.571 × 10 –6 ), and the amino acid substitutions were predicted to be deleterious or damaging by multiple algorithms ( Supplemental Table 2 ). The results from these 2 families indicate that SLC6A9 is a potential causal gene for AIS. We screened potential SLC6A9 variants in 3 sporadic AIS genome sequencing data sets consisting of 118 patients from the Duchess of Kent Children’s Hospital (DKCH, Hong Kong, China), 223 patients from PUMCH, and 635 patients from SRC. In the DKCH cohort, we identified 5 individuals harboring the same heterozygous SLC6A9 variant (c.617A>T, p.Y206F), which was first identified in family 2 ( Supplemental Table 1 ). Further clinical and genetic investigations revealed that the parents of 3 of the patients were also affected and carried the c.617A>T, p.Y206F variant (family 3-5) ( Figure 1A and Supplemental Table 1 ). We also identified several rare heterozygous missense variants of SLC6A9 in the PUMCH (2 variants in 2 patients) and SRC (6 variants in 12 patients) cohorts ( Figure 1C and Supplemental Table 1 ). All these variants altered highly conserved residues in various regions of GLYT1, the majority of which were predicted to be deleterious or damaging by multiple algorithms ( Figure 1, D and E , and Supplemental Table 3 ). We next performed targeted sequencing of SLC6A9 in the cohort of 725 sporadic AIS patients and a cohort of 3,219 ethnicity-matched participants without AIS in Hong Kong. We identified 9 out of 725 AIS patients and 7 out of 3,219 controls as harboring the c.617A>T, p.Y206F variant, whereas the c.1984C>T, p.R662W variant that was detected in family 1 was not found in any of sporadic patients or controls. The p.Y206F variant had a total allele frequency of 0.884% (15/1,696) in the Hong Kong AIS cohort (11 sporadic and 4 familial alleles out of 840 sporadic patients and 8 families), which was significantly higher compared with the local non-AIS controls (0.109%, 7/6,438, P = 2.39 × 10 –6 ), Chinese general population ( 29 , 30 ) (0.280%, P = 2.74 × 10 –4 ), and data in gnomAD (0.014%, P = 3.56 × 10 –20 ) ( Supplemental Tables 4 and 5 ). Notably, it is unclear whether the individuals included in the latter 2 data sets were examined for scoliosis. The identification of multiple rare variants in familial and sporadic patients and the strong association of the p.Y206F variant with AIS further indicate the genetic susceptibility of SLC6A9 to AIS. Plasma glycine levels and aberrant paraspinal muscle activity in SLC6A9 variant carriers. Patients with glycine encephalopathy harboring homozygous SLC6A9 mutations were reported to have increased glycine concentrations in the CSF or plasma ( 24 , 25 ). Because the AIS patients in our study did not show any discernible neurological defects, it was not ethically justified to obtain CSF for measuring glycine concentrations. We instead measured plasma glycine concentration, which was found to be higher in AIS patients carrying SLC6A9 variants ( n = 15) compared with unaffected controls ( n = 36). Moreover, if we only compared the individuals whose glycine levels were measured during adolescence, we identified a more significant difference between the SLC6A9 variant carriers and noncarriers ( Figure 2A ). Notably, plasma glycine concentrations in 2 p.R662W variant carriers in family 1 (III-6 and III-7) were higher than in the controls, although these 2 subjects had yet to reach puberty and had no scoliotic phenotype at the beginning of the study. We followed these 2 high-risk children from age 6 to age 9 (III-6) and from age 9 to age 12 (III-7), respectively. They were both later diagnosed with mild spinal curvature (Cobb angle: 10° and 13.3°, respectively) from their latest spinal x-ray images ( Supplemental Figure 2 and Supplemental Table 1 ), suggesting they were in the early stages of AIS development. As glycine functions as a spinal cord neurotransmitter, we next measured the activity of the paraspinal muscles of AIS patients by surface electromyography (sEMG). The bipolar electrodes were positioned at the paraspinal muscles along the spine ( Figure 2B ). We examined the 2 aforementioned young patients in preadolescence (III-6 at age 9 and III-7 at age 12 in family 1) and 2 sex- and age-matched controls. The sEMG signals in the two affected children exhibited irregular bursts, indicative of the aberrant paraspinal muscle activity in AIS patients, whereas the controls showed stationary sEMG signals on both sides ( Figure 2C ). Interestingly, we failed to detect abnormal sEMG signals in an adult patient carrying the same SLC6A9 variant (II-7, the father of III-6 and III-7 in family 1) ( Figure 2D ). These observations suggest that neuromuscular aberrations may be more pronounced if individuals are still in the initial stages of curvature development rather than a steady state after scoliosis has developed. The aberrant paraspinal muscle activity may adapt to the curvature progression in adult AIS patients and become coordinated with the well-established spinal curvature. Functional consequences of SLC6A9 variants on GLYT1. Given the main function of GLYT1 in transporting extracellular glycine into cells, we assessed the glycine-uptake capacity of GLYT1 variants in HEK293T cells, which have no endogenous GLYT1 expression. K687R, a ubiquitination-deficient mutant that stabilizes GLYT1 on the cell surface ( 31 ), and S407G, a known mutation that causes recessive glycine encephalopathy ( 25 ), served as negative and positive controls, respectively. We found the [ 3 H]-glycine uptake capacity of 7 variants (Y206F, F207Y, R333H, E338K, E446K, R643H, and R662W) was significantly reduced compared with that of WT GLYT1, whereas 3 other variants that have relatively higher allele frequency (G231S, V408I, and G677S) showed no significant differences ( Figure 3A and Supplemental Table 3 ). We next analyzed the subcellular localization of GLYT1 variants by immunofluorescent staining. WT GLYT1 was predominantly located on the cell surface, whereas GLYT1 variants (Y206F, F207Y, R333H, E338K, E446K, R643H, and R662W) were largely retained intracellularly. The localization of GLYT1 in cells expressing G231S, V408I, or G677S variants was minimally affected ( Figure 3B and Supplemental Figure 4A ). We further assessed the protein levels of GLYT1 variants, which showed all variants except G231S, V408I, and G677S had significantly lower levels of total and cytomembrane-associated GLYT1 ( Figure 3C ). After normalizing the glycine uptake activity of each variant with the corresponding cytomembrane protein level, we found no significant differences between the WT and variants ( Supplemental Figure 4B ), suggesting that decreased glycine-uptake activity might be a consequence of reduced levels of GLYT1 on the cell surface rather than due to the impairment of glycine-transporting function. GLYT1 is a member of the Na + /Cl – -dependent neurotransmitter transporter (SLC6A) family, and these transporter family members can form dimers or oligomers ( 32 ). Several studies have shown that GLYT1 can form dimeric protein complexes not only in intracellular compartments but also in the plasma membrane ( 33 – 35 ). As all identified SLC6A9 variants are heterozygous in AIS patients, it is possible that these variants also affect the function of WT GLYT1 in the complex. Therefore, we tested the effects of GLYT1 variants against WT GLYT1 by coexpressing them at a 1:1 ratio in cells. We found the majority of GLYT1 variants, including Y206F, F207Y, R333H, E338K, E446K, R643H, and R662W, impaired the localization and expression of WT GLYT1 ( Supplemental Figure 5 ). Together, our results demonstrate that most of the identified SLC6A9 variants from AIS patients (7 out of 10) caused loss of function and had dominant negative effects over WT GLYT1. Idiopathic scoliosis-like phenotype in slc6a9 mutant zebrafish. Zebrafish have inherent advantages over other animal models and have been widely used for modeling AIS ( 36 – 38 ). We generated an slc6a9 mutant zebrafish line, which carries a 22 bp deletion and produces a C-terminal truncated glyt1 (denoted as slc6a9 m afterwards) ( Supplemental Figure 6A ). The cellular assays indicated that slc6a9 m was a severe hypomorphic mutation. This mutant exhibited dominant negative effects over WT glyt1, recapitulating the characteristics of the SLC6A9 missense variants identified in AIS patients ( Figure 3, A and B , and Supplemental Figure 6 , B–E). Considering the crucial role of glyt1 in the survival of zebrafish ( 39 , 40 ), we first assessed the survival rate of slc6a9 m/+ and slc6a9 m/m mutant fish over time. The slc6a9 m/m larvae began to die at 7 days post fertilization (dpf), with none surviving at 18 dpf. In contrast, around 50% of slc6a9 m/+ fish survived to the juvenile stage at 30 dpf ( Supplemental Figure 7A ). The extracellular glycine levels were significantly higher in slc6a9 m/m fish compared with WT ( Supplemental Figure 7B ). The slc6a9 mutants had no discernible morphological defects in motor neuron, skeletal muscle, axonal tract formation, and calcified vertebrae ( Supplemental Figure 8 and Supplemental Figure 9A ). Given the important role of notochord in zebrafish spine formation ( 38 , 41 ), we examined the notochord sheath and the morphology of notochord vacuolated cells in slc6a9 m/m mutant, which did not exhibit any anomalies in notochord ( Supplemental Figure 9 , B and C). Notably, at 7 dpf, 65% of slc6a9 m/m fish showed an apparent lateral axial curvature (θ angle ≥10°) and approximately 8% of slc6a9 m/+ larvae exhibited a curvature phenotype ( Figure 4A and Supplemental Videos 1–3 ). The locomotion of zebrafish, as measured by swimming distance, was significantly decreased in all slc6a9 mutants ( Supplemental Figure 10 ). By 18 dpf, we observed spinal curvature in all dying slc6a9 m/m fish. As the slc6a9 m/m fish did not survive beyond 18 dpf, we followed the phenotype of slc6a9 m/+ fish at later developmental stages. We found that 11.9% (21 of 177), 13.4% (17 of 127), and 12.5% (4 out of 32) of slc6a9 m/+ zebrafish exhibited overt body curvature at 21, 35, and 100 dpf, respectively. These zebrafish exhibited severe lateral spinal curvature and occasionally kyphosis, but showed no congenital vertebral malformation ( Figure 4, B and C , and Supplemental Figure 11 ). To further verify the causal effects of slc6a9 on body curvature, we treated WT zebrafish larvae with ALX5407, a specific GLYT1 inhibitor ( 42 ). We observed an increase in curvature penetrance with the dose of ALX5407 ( Supplemental Figure 12A ), and 50.3% of WT fish treated with 1 μM ALX5407 showed obvious axial curvature and reduced free swimming distance, which is consistent with the phenotype of the slc6a9 m/m mutant ( Figure 4D and Supplemental Figure 12B ). A low dose of 10 nM ALX5407 induced axial curvature in only 4.65% of WT fish, but significantly increased the penetrance (from 10% to 34.9%) and severity of curvature in the slc6a9 m/+ mutant ( Figure 4E ). Additionally, we characterized a reported slc6a9 mutant line ta229g , which carries a G81D mutation causing disruption of transporter function ( 40 ). Intriguingly, 72.5% of slc6a9 ta229g/ta229g mutant fish exhibited pronounced lateral curvature at 7 dpf, which has not been described in previous studies ( Supplemental Figure 12C ). Together, our results further support the importance of GLYT1 in the maintenance of spinal alignment in a dose-dependent manner. To further investigate the functional consequence of the identified SLC6A9 variants, we microinjected SLC6A9 WT and AIS-associated variant mRNAs (Y206F and R662W) into the zygotes generated from slc6a9 m/+ and slc6a9 m/+ mating pairs. Notably, injection of 200 pg SLC6A9 WT mRNA efficiently rescued the axial curvature of slc6a9 m/m mutants ( Supplemental Figure 13 ), whereas the same dosage of Y206F and R662W variant mRNAs failed to rescue the phenotype ( Figure 4F ). These results further demonstrate that the identified AIS-associated variants are indeed deleterious, accounting for the scoliosis phenotype. Scoliosis caused by dysfunction of CPGs in zebrafish. CPGs are self-contained biological neural circuits that can generate tightly coupled patterns of neural activity driving rhythmic and stereotyped behaviors such as breathing, walking, and swimming independently of central commands or peripheral sensory inputs. The CPG involved in the control of locomotion is composed of spinal cord interneurons ( 43 – 47 ). As glycine is a major neurotransmitter in inhibitory interneurons of the spinal cord, excessive glycine caused by mutant GLYT1 may interfere with the normal function of CPGs. To understand the underlying mechanism that leads to the scoliosis-like phenotype in slc6a9 mutant zebrafish, we next investigated their neural activity and found that a pattern of left-right alternation was disrupted in the slc6a9 m/m spinal cord. The neural activities of WT fish are coordinated in both sides of the spinal cord or in specific pairs of neurons, as demonstrated by a calcium indicator in the elavl3-H2B-GCaMP6s transgenic background ( 48 ). However, such left-right alternating neural activity was impaired and abnormal unilateral neuronal activation was observed in the slc6a9 m/m mutant ( Figure 5A and Supplemental Figure 14A ). The left-right alternation index was significantly decreased in the slc6a9 m/m mutants, indicating that the total neuronal activities on one side are much stronger than on the other side ( Figure 5B ). Moreover, we observed that the signal frequency was significantly reduced in mutant fish ( Figure 5C ), which is consistent with the reduced swimming behaviors of slc6a9 m/m mutant zebrafish. Notably, we also detected a significant total signal reduction in activated neurons in slc6a9 m/m mutants, suggesting a strong inhibitory effect of the increased extracellular glycine ( Supplemental Figure 14B ). To further test the role of CPGs in maintaining spinal alignment, we generated a dmrt3a mutant zebrafish line (denoted as dmrt3a m afterwards, Supplemental Figure 15A ). Mutation in DMRT3 or disruption of Dmrt3/dmrt3a is known to affect the pattern of locomotion. The Dmrt3 mutant mice showed a significant decrease in commissural interneuron numbers and impaired limb coordination. The dI6 interneurons specified by transcription factor Dmrt3 are important for proper left-right alternation of the body ( 49 – 51 ). We observed that dmrt3a mutant zebrafish exhibited reduced survival rates compared with WT ( Supplemental Figure 15B ). Interestingly, 10% to 20% of dmrt3a m/m zebrafish exhibited apparent lateral spinal curvature starting from approximately 18 to 21 dpf, with intact calcified vertebrae and no discernible congenital vertebral malformation ( Figure 5, D and E , and Supplemental Figure 15C ). These results further support the role of left-right coordination of rhythmic motor activities in maintaining spinal alignment. Taken together, our findings demonstrate that disturbance of CPGs by either excessive glycine or developmental defects can cause a scoliosis-like phenotype. Pharmacologic prevention of body curvature. Given that GlyRs are the main receptors on inhibitory postsynaptic terminals activated by synaptic glycine ( 52 ), we determined whether strychnine, a GlyR antagonist, could prevent the curvature phenotype of slc6a9 mutant zebrafish ( 53 ). We found the strychnine treatment significantly reduced the number of slc6a9 m/m mutants with axial curvature (θ angle ≥ 10°), from 70.2% to 30.3% ( Figure 6A ). We also attempted to block excessive extracellular glycine in mutant fish using sodium benzoate, which is a glycine neutralizer that is clinically used to treat patients with glycine encephalopathy ( 54 ). We observed a moderate decrease in the number of slc6a9 m/m mutants showing a curvature phenotype from 62.7% to 40.0% ( Figure 6B ). Both strychnine and sodium benzoate treatments markedly reduced the severity of curvature ( Figure 6, A and B ), which suggests that neutralizing or blocking the activity of excessive glycine in the mutant zebrafish can partially rescue the idiopathic scoliosis-like phenotype.

Discussion Our work explores the genetic basis and pathogenic mechanisms that lead to the development of idiopathic scoliosis in adolescents. By inheritance mapping in 2 large families with dominant inheritance of AIS, we identified an AIS locus on chromosome 1p34.1 and the causal gene SLC6A9 . We further identified a number of missense SLC6A9 variants in diverse AIS cohorts. The extremely low frequency or lack of p.R662W variant in global or local populations supports its causal effect in family 1. Other variants identified in sporadic cases are also extremely rare in the general population. Interestingly, the p.Y206F variant was prevalent in the Hong Kong AIS cohort, presenting in 1 large family, 3 trios, and 11 sporadic patients, which accounts for 1.769% (15 of 848) of AIS cases studied in Hong Kong. The presence of the p.Y206F variant in the general populations (0.029% to 0.560%) may reflect its relatively low penetrance in causing AIS and/or a lack of diagnosis for mild scoliosis in the control groups. Our functional assays provide evidence that the majority of the identified SLC6A9 variants caused markedly decreased protein levels and impaired cell-surface presentation of GLYT1, consequently resulting in decreased glycine uptake. As complex membrane proteins are often subjected to quality control mechanisms in the endoplasmic reticulum (ER) ( 55 ), mutant GLYT1 may be degraded by proteasome via the ER-associated degradation (ERAD) pathway or be unstable on the cell surface and prone to endocytosis and lysosomal degradation. Our studies in animal models further showed that slc6a9 mutant zebrafish exhibited spinal curvature in an slc6a9 gene and GLYT1 inhibitor dose-dependent manner. These genetic and mechanistic studies collectively demonstrate a functional role of SLC6A9 in AIS etiology. Despite extensive studies on idiopathic scoliosis, the etiopathogenesis of AIS has remained obscure and controversial. The high incidence of scoliosis in children with neurological diseases (neuromuscular scoliosis) has led to the neuropathic hypothesis for AIS, in which a small scoliotic curve may initially develop due to a small defect in the nervous system, either from altered sensory input or altered neuromuscular control. This defect produces asymmetric muscle loading or loss of muscle support that directly leads to the initiation of the scoliotic deformity, which is further exacerbated by biomechanical factors during the adolescent growth spurt. Therefore, it has long been proposed that one of the likely causes of AIS is neuromuscular and that AIS may be a late-onset subtype or mild form of neuromuscular scoliosis ( 56 ). Because such neurological defects are subtle, not detectable by conventional clinical assessment, patients were diagnosed with “idiopathic” rather than neuromuscular scoliosis. However, convincing evidence for this hypothesis, especially human genetic evidence with mechanistic proof for the causal relationship, was largely lacking. Studies in mouse models showed that defects in the proprioceptive system caused by deletion of Runx3 or ablation of mechanosensor Piezo2 resulted in spinal misalignment ( 57 , 58 ). Genetic studies in zebrafish indicated that abnormalities in cilia or cilia-mediated CSF flow, dysfunction of Reissner’s fiber, or activation of proinflammatory signals within the spinal cord were associated with idiopathic-like scoliosis ( 36 , 59 – 65 ). Previous studies reported null variants of PIEZO2 in 3 families and 2 patients with neuromuscular symptoms and progressive scoliosis, indicating a connection between mechanosensory defects and development of spinal curvature ( 66 , 67 ). In the past decade, population genetic studies have identified many susceptibility genes for AIS, and the most significant one is LBX1 ( 13 , 14 , 18 , 19 , 68 – 70 ). LBX1 plays critical roles in specifying dorsal spinal neurons and hindbrain somatosensory neurons, suggesting a potential etiology through abnormal neural function ( 71 , 72 ). Here, the identification of SLC6A9 variants in many familial and sporadic cases extends the spectrum of glycinopathy manifestation and implies a role of glycine synaptic transmission in the etiology of AIS. A moderate elevation of glycine may be a strong risk factor for AIS ( Figure 6C ). Intriguingly, a functional enrichment analysis of all reported AIS susceptibility variants revealed the vast majority of the enriched pathways are associated with synaptic homeostasis ( Figure 6D and Supplemental Figure 16 ). These findings strongly suggest a neuropathic origin of AIS in a considerable proportion of patients. The neuropathic hypothesis has led to repeated attempts to identify a neuromuscular cause of AIS. One of the efforts involves measuring and comparing sEMG activities of the paraspinal muscles in patients and controls. Previous work mainly studied patients aged 12 to 19 years, which identified either an increased myoelectric response on the convex or concave side of the scoliotic curve or no differences between sides ( 73 ). It is unclear whether such change is the primary cause or a secondary phenomenon induced by the deformed spine. In this study, we found that healthy controls had stationary sEMG signals, but 2 children aged 9 and 12 carrying the SLC6A9 pathogenic variant showed aberrant sEMG bursts, which may reflect an impairment of the balance of the paraspinal muscle control at the preadolescent stage. These findings are consistent with the left-right coordination defects in slc6a9 mutant zebrafish, while zebrafish studies provide more accurate and much higher resolution of neural activities than human sEMG. Considering the normal sEMG in an adult patient carrying the same pathogenic SLC6A9 variant and the conflicting sEMG in relatively mature AIS patients ( 73 , 74 ), we argue that spinal curvature is a compensatory response that eventually corrects or adapts to the aberrant paraspinal muscle activity in older patients and therefore the findings from mature AIS patients varied greatly ( 73 ). Measuring sEMG on the paraspinal muscle of asymptomatic or early stage patients would be more informative. Our results suggest that disturbance of bilateral paraspinal muscle control might be a causal factor for AIS, which warrants further large-scale prospective studies in the preadolescent population (e.g., ages 8 to 12). sEMG screening in preadolescent children and long-term follow-up may allow us to determine whether aberrant paraspinal muscle activities can be used as a biomarker for identifying potential AIS patients for preventive treatment. Distinct CPGs located throughout the CNS mediate various biological rhythms ( 44 ). Left-right alternation of locomotion is thought to be organized by glycinergic commissural inhibitory neurons. Previous studies showed that the V0 commissural interneurons play a fundamental role in securing left-right alternation in the locomotor CPG ( 43 , 45 , 46 , 75 ). In particular, the inhibitory V0 neurons are required for left-right alternating modes during slow locomotion ( 75 ). Given the indispensable function of GLYT1 in inhibitory glycinergic neurotransmission and the defects of left-right alternation observed in the spinal cord of slc6a9 mutant zebrafish, we speculate that excessive synaptic glycine may compromise the function of CPGs, leading to an imbalance in neuromuscular activity of the paraspinal muscles and thus loss of spinal alignment ( Figure 6E ). Besides, the Dmrt3 + dI6 commissural interneurons are also reported to function in left-right alternation ( 51 ). The AIS-like phenotype induced by loss of dmrt3a in zebrafish further supports the functional role of CPGs in spinal alignment. The overall penetrance of spinal curvature in dmrt3a mutants was lower and onset was later than in slc6a9 mutants. This is likely due to compensatory effects of WT1 + dI6 interneurons ( 49 ). Notably, Dmrt3 + interneurons and WT1 + interneurons are both derived from Lbx1 + lineage of interneurons in the dorsal spinal cord ( 49 , 71 , 72 ). As LBX1 is so far the most important predisposing gene replicated in multiethnic populations for AIS, this further implicates a critical role of CPGs in causing AIS. Moreover, contralateral projection of the commissural axon is required for the development of CPGs ( 45 ). Several genes encoding commissural axon guidance molecules, including ROBO3 , EPHA4 , CHL1 , and DSCAM , are strongly associated with scoliosis ( 18 , 76 , 77 ). Mutations in ROBO3 cause horizontal gaze palsy with progressive scoliosis (HGPPS), which is a rare disorder that affects the spine and vision ( 77 ). EphA4-positive neurons constitute a critical component of locomotor CPG ( 78 ). This evidence implies that CPG dysfunction may be a common causative factor in the development of AIS. The markedly reduced left-right alternation index in slc6a9 mutant zebrafish indicates that the total level of neuronal activities on one side is much greater than on the other, suggesting that the contractions of muscles on one side are much greater than on the other. This imbalance of bilateral muscle contractions can initiate axial curvature. The mutant fish exhibited axial curvature to varied extents, and the persistent unbalanced paraspinal muscle activities continued to bend the body and make it curved most of time ( Supplemental Videos 1–3 ). However, how such curvature became fixed or permanent at the late stages is an unresolved question. It is possible that the initial curvature is exacerbated by other mechanisms, such as biomechanical factors or secondary bony structural changes during the adolescent growth spurt, that ultimately misshape the spine. This warrants more investigation in the future. Glycine has many potential health benefits, such as improving sleep, elevating mood, and lowering the risk of heart disease ( 79 – 81 ). Glycine is found in many protein-based food sources and is used as a food additive or taken as supplements. However, the long-term safety of glycine supplements, such as their effects on plasma or CSF glycine levels, has not been fully tested. We observed an increased penetrance and severity of axial curvature in slc6a9 m/+ zebrafish after administering a glyt1 antagonist, ALX5407, which specifically elevates glycine levels in the CSF ( 24 ). Our work raises the possibility that high levels of CSF glycine increase the risk of developing AIS in children and adolescents, especially in those who carry genetic susceptibility variants. It would be highly valuable to investigate whether there are associations among glycine levels, sEMG signals, and genetic variants with AIS, which could establish a new method for predicting AIS in preadolescents. Our data also suggest that pharmacologic interventions may be considered for preventing or alleviating the scoliotic phenotype in humans. Both strychnine and sodium benzoate treatments markedly reduced the severity of curvature in zebrafish. Although strychnine is highly toxic to humans ( 82 ), sodium benzoate is recognized as safe by the FDA and used as a treatment for a variety of human diseases, including urea cycle disorders, schizophrenia, and glycine encephalopathy ( 54 ). The moderate rescue of curvature in the animal model suggests sodium benzoate could be a potential preventive therapy in AIS patients with high levels of glycine. Our work lays the foundation for further investigations on the etiopathogenesis, early diagnosis, and potential pharmacological interventions for AIS.

Authorship note: Xiaolu Wang, MY, and JPYC are co–first authors. JPYC, YQS, and BG contributed equally to this work and are co–corresponding authors. Adolescent idiopathic scoliosis (AIS) is the most common form of spinal deformity, affecting millions of adolescents worldwide, but it lacks a defined theory of etiopathogenesis. Because of this, treatment of AIS is limited to bracing and/or invasive surgery after onset. Preonset diagnosis or preventive treatment remains unavailable. Here, we performed a genetic analysis of a large multicenter AIS cohort and identified disease-causing and predisposing variants of SLC6A9 in multigeneration families, trios, and sporadic patients. Variants of SLC6A9 , which encodes glycine transporter 1 (GLYT1), reduced glycine-uptake activity in cells, leading to increased extracellular glycine levels and aberrant glycinergic neurotransmission. Slc6a9 mutant zebrafish exhibited discoordination of spinal neural activities and pronounced lateral spinal curvature, a phenotype resembling human patients. The penetrance and severity of curvature were sensitive to the dosage of functional glyt1. Administration of a glycine receptor antagonist or a clinically used glycine neutralizer (sodium benzoate) partially rescued the phenotype. Our results indicate a neuropathic origin for “idiopathic” scoliosis, involving the dysfunction of synaptic neurotransmission and central pattern generators (CPGs), potentially a common cause of AIS. Our work further suggests avenues for early diagnosis and intervention of AIS in preadolescents. Genetic variants affecting synaptic neurotransmission and central pattern generators are strong causal risk factors of adolescent idiopathic scoliosis.

Author contributions JPYC, YQS, and BG conceived the project. JPYC, PWHC, JJR, WT, GQ, ZW, TJZ, SI, NW, CAW, and KDKL recruited patients. Xiaolu Wang, MY, JPYC, and PWHC curated data. Xiaolu Wang, MY, YF, MW, Xiaojun Wang, SZ, AMK, ZC, Xiwei Wang, and YH developed methodology. Xiaolu Wang, MY, YF, MW, Xiaojun Wang, SZ, AMK, and JJR performed experiments. Xiaolu Wang and MY visualized experiments. JPYC, QY, DQ, GQ, ZW, TJZ, SI, NW, CAW, YQS, and BG acquired funding. JPYC, YQS, and BG performed project administration. JPYC, WT, DC, SI, NW, CAW, YH, YQS, and BG supervised the project. Xiaolu Wang, MY, JPYC, YQS, and BG wrote the original draft. Xiaolu Wang, MY, JPYC, SZ, AMK, DC, SI, NW, CAW, YH, KDKL, YQS, and BG reviewed and edited the manuscript. Xiaolu Wang, MY, and JPYC share the first author position in the given sequence for their specific contributions based on workload and significance to the project. Supplementary Material

We thank all patients and their family members for participating in this study; Wing Ki Cheung of the Department of Orthopaedics and Traumatology, University of Hong Kong for assistance with sEMG testing; and Miao Chen and Jing Guo, University of Hong Kong, for assistance with confocal microscopy. The authors acknowledge the Texas Advanced Computing Center (TACC) ( http://www.tacc.utexas.edu ) at The University of Texas at Austin for providing computing resources that have contributed to the results related to the SRC cohort. Work in the Gao laboratory was supported by the Chinese University of Hong Kong start-up and internal funds, the Hong Kong Health and Medical Research Fund (06171406), and the Health@InnoHK, Innovation and Technology Commission (CTSCB). Work in the Cheung laboratory was supported by the Research Impact Fund (R5017-18F). Work in the Song laboratory was supported by the Hong Kong RGC General Research Fund (GRF) (17114519). Work in the Wise lab was supported by the Eunice Kennedy Shriver National Institute of Child Health and Human Development of the NIH (P01HD084387) and the SRC Research Fund. Work from the PUMCH team was supported by the National Natural Science Foundation of China (82072391 to NW, 81772299 to ZW, 82172382 to TJZ), the CAMS Innovation Fund for Medical Sciences (CIFMS, 2021-I2M-1-051 to TJZ and NW, 2021-I2M-1-052 to ZW), and the Non-profit Central Research Institute Fund of the Chinese Academy of Medical Sciences (no. 2019PT320025). Work in the Ikegawa laboratory was supported by a grant from the Japan Society for the Promotion of Science (22H03207). 11/14/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e168783

oa_package/64/bc/PMC10786698.tar.gz

PMC10786699

0

Introduction Bladder cancer is the sixth most common cancer in the United States, with an estimated 82,290 new cases and an estimated 16,710 deaths in 2023 ( 1 ). At diagnosis, approximately 75% of bladder cancers are non–muscle-invasive (NMIBC), while 25% are de novo muscle-invasive (MIBC) ( 2 ). The initial treatment for NMIBC is transurethral resection of bladder tumor that can be followed by treatment with intravesical therapy with Bacillus Calmette-Guerin (BCG) in patients deemed to be at high risk for recurrence. Additionally, the FDA recently approved pembrolizumab for patients who are unresponsive or intolerant to BCG. Other alternatives include cystectomy or treatment with intravesical chemotherapy. The genetics and transcriptomics of bladder cancer have now been well studied. Upwards of 70% of low-grade NMIBC and 15% of MIBC have genetic alterations in the fibroblast growth factor receptor 3 ( FGFR3 ) gene, including activating point mutations or gene fusions ( 3 – 6 ). FGFR3 alterations predominantly activate the MAPK pathway to promote cell proliferation and survival via ligand-independent dimerization. In bladder cancer, FGFR3 S249C is the predominant hotspot mutation, potentially related to APOBEC-induced mutagenesis ( 7 , 8 ). The transcriptomic subtypes of MIBC also show a bias in FGFR3 alteration frequency, with the luminal/luminal papillary (LumP) subtype being enriched for FGFR3 alterations in multiple subtype classification systems ( 9 – 15 ), while FGFR3 alterations are enriched in the UROMOL class 1 and class 3 NMIBC subtypes ( 16 ). While bladder tumors with a luminal subtype and/or FGFR3 mutations have both been associated with a non–T cell–inflamed tumor microenvironment, whether FGFR3 alterations functionally mediate this non–T cell–inflamed phenotype has not been directly explored ( 17 , 18 ). Previous works have examined the role of FGFR3 mutations in murine models of bladder cancer and, with the exception of a recent report ( 19 ), have consistently demonstrated that FGFR3 mutations alone are not sufficient to promote urothelial tumorigenesis. However, FGFR3 mutations are permissive for the development of carcinoma in situ or high-grade tumors when combined with SV40 large T antigen, PTEN loss, or the carcinogen N -butyl- N -(4-hydroxybutyl)nitrosamine (BBN) ( 19 – 22 ). In the context of BBN, transgenic mice overexpressing FGFR K644E had decreased neutrophil infiltration, but no other immune cell phenotypes or tumor microenvironment characterization was performed ( 22 ). The FGFR inhibitor erdafitinib is FDA approved for patients with FGFR-altered, advanced urothelial carcinoma (UC) that has progressed during or following prior platinum-containing chemotherapy ( 23 ). Despite the relatively non–T cell–inflamed tumor microenvironment, our group and Wang et al. have shown that patients with advanced UC with or without FGFR3 alterations respond equally well to immune checkpoint inhibition (ICI), perhaps due to a lower level of immunosuppression by stromal elements ( 24 , 25 ). Functionally, pharmacologic FGFR inhibition or FGFR3 knockdown in cell culture leads to an increased expression of IFN pathway genes, suggesting that FGFR3 signaling suppresses proinflammatory cytokine secretion in a cell-autonomous manner ( 24 , 26 ). Nonetheless, to our knowledge, no prior work has examined how FGFR3 alterations functionally affect the tumor microenvironment nor how acute FGFR inhibition may cooperate with ICI in FGFR3 -altered bladder cancer. In this work, we report a mouse model ( UPFL ) that demonstrates cooperativity between the activating hotspot FGFR S249C mutation and the Trp53 R270H mutation, to reliably produce high-grade NMIBC when expressed in uroplakin 3a–expressing (Upk3a-expressing) cells. Recapitulating human disease, UPFL tumors are papillary in histology and transcriptionally similar to both UROMOL class 1 and consensus LumP molecular subtypes ( 11 , 16 ). Exploration of FGFR3-driven immunobiology demonstrated that UPFL tumors have an intermediate T cell–inflamed immune contexture relative to our previously reported BBN (basal) and UPPL (luminal) models. We derived a syngenic cell line ( UPFL1 ) that allows for transplantable tumor studies to test the interaction between FGFR inhibition and ICI via PD-1 inhibition. UPFL1 syngeneic tumors were sensitive to erdafitinib and interestingly demonstrated hyperprogression with single-agent anti–PD-1 treatment. In contrast, the combination of erdafitinib and anti–PD-1 worked significantly better than either alone. Flow cytometry demonstrated that while anti–PD-1 treatment of UPFL1 tumors increased the number of regulatory T cells (Tregs), perhaps accounting for the hyperprogression seen in that model, combined treatment with erdafitinib and anti–PD-1 fully abrogated this Treg increase, suggesting that FGFR inhibition may be able to reverse anti–PD-1–induced immunosuppression. Moreover, erdafitinib treatment was sufficient to block Treg proliferation in vitro. In aggregate, our work establishes that dual FGFR3 and Trp53 alteration initiates high-grade, non–muscle-invasive, autochthonous murine bladder tumors with an intermediate T cell–inflamed phenotype and that erdafitinib cooperates with PD-1 checkpoint blockade to reverse anti–PD-1–induced Treg expansion and to block progression.

Methods Generation of ColA1-LSL-FGFR3 S249C mouse. We generated a mouse allele with Cre-inducible expression of human FGFR3 cDNA encoding the S249C hotspot mutation knocked into the Col1a1 locus. The human FGFR3 S249C cDNA coding region with the Kozak sequence (GCCGCCACC) was introduced into vector pGV at the EcoRI cloning site using blunt-end cloning. Successful generation of the mutation was confirmed by sequencing. Vectors were electroporated with plasmid expressing FLP recombinase into mouse embryonic stem (ES) cells (MESC10, Mirimus) engineered with an FLP homing cassette at the Co1A1 locus, and positive clones were identified by PCR. Positive ES clones were injected into mouse blastocysts for chimera generation. Chimeric mice were crossed with WT mice to generate mice with germline integration. The gene is expressed following Cre recombinase–mediated excision of a stop cassette flanked by LoxP sites (loxP-stop-loxP [LSL] FGFR3 S249C/+). Both male and female mice were used in the GEMMs. Genotyping primers. The AO123 (TCCAGTCTTCCTTGTGCATCC) and YL104 (GATAGGCAGCCTGCACTGGT) primers generate a 333-bp band in LSL-FGFR3 S249C mice harboring a targeted allele. The AO123 and AO129 (GATGTGGGGTCCTGTCCTTT) primers generate a 572-bp band in mice with a WT Col1A1 locus. For detecting recombination of the LSL cassette within the LSL-FGFR3 S249C allele, the primers AO107 (TTCGGCTTCTGGCGTGTG) and AO106 (CGCTGCCGAAGACCAACT) were used. The targeted allele produces a 686-bp band, while after recombination, the primers produce a 376-bp PCR product. Mouse strains. B6.129S4- Trp53 tm3.1Tyj /J (R270H) (strain 008182), B6;DBA-Tg(Upk3a-GFP/cre/ERT2)26Amc/J (Andrew McMahon, strain 015855), Gt(ROSA)26Sor tm1(Luc)Kael /J (William Kaelin, strain 034320) ( 28 ), and B6.129(Cg)-Foxp3tm3(Hbegf/GFP)Ayr/JFoxP3-DTR-eGFP (strain 032050-JAX) mice were obtained from The Jackson Laboratory. Tamoxifen dose and administration and genotyping. CreERT2 was activated by administration of tamoxifen (5 mg) every other day for a total of 3 times by oral gavage in 8- to 10-week-old mice. Mouse genomic DNA was isolated from a tail or toes following overnight digestion at 55°C in Nuclei Lysis Solution containing Proteinase K (Life Technologies). PCR was performed using primer pairs to distinguish WT and mutant alleles using genotyping of mouse strains as follows. The Trp53 R270H , Upk3aCreERT2 , and Rose26 LSL-Luc strains were genotyped per The Jackson Laboratory’s protocol. Generation of UPFL1 and UPFL3 cell lines. Bladders were harvested from male UFPL mice when tumors reached a diameter of 7 mm. A portion of the tissue was taken for pathologic evaluation and the remaining tumor was dissociated and digested with collagenase and Dispase (Roche). The dissociated tumor cells were resuspended in Georgetown Media and transferred to a plastic plate as described previously ( 40 ). Cells were passaged until they propagated in DMEM independently of feeder cells. Mycoplasma testing was performed monthly while cells were in culture. Short tandem repeat testing. Short tandem repeat (STR) testing was performed to establish a public database of STR profiles for our mouse cell lines. Samples were submitted to LabCorp. Eighteen mouse STR loci and 2 human STR markers (to detect human cell line contamination) were analyzed. STR profiles can be found in Supplemental Table 1 . Cell viability assay. Cell viability was measured using CellTiter-Glo Luminescent Cell Viability Assay (Promega) following the manufacturer’s protocols. In all cell lines, 500 cells per well were seeded into 96-well plates in triplicate, and erdafitinib (JNJ-42756493) treatment initiated after 24 hours. Cell viabilities were assessed after 96 hours. IC 50 values were derived from the 10-dose response curves using GraphPad Prism. Western blot. Whole-cell extracts were isolated using RIPA buffer supplemented with protease inhibitors and phosphatase inhibitors. The concentration of the isolated proteins was determined using Protein Assay Dye Reagent Concentrate (Bio-Rad, 5000006). Twenty micrograms of the protein were resolved in 7.5% to 10% Tris-acetate gels and electrophoretically transferred to PVDF membranes (Bio-Rad) and immunoblotted with antibodies against the following proteins: FGFR3 (C545F2) (rabbit mAb [1:1000]; Cell Signaling Technology, 4574), phospho-FGFR (Tyr653/654) (rabbit pAb [1:1000]; Cell Signaling Technology, 3471), Akt (pabbit pAb [1:1000]; Cell Signaling Technology, 9272S), phospho-Akt (Ser473) (D9E) XP (rabbit mAb [1:2000]; Cell Signaling Technology, 4060), P44/42 MAPK (Erk1/2) (rabbit pAb [1:1000]; Cell Signaling Technology, 9102S), phospho-p44/42 MAPK (Erk1/2) (Thr202/Tyr204) (E10) (mouse mAb [1:2000]; Cell Signaling Technology, 9106), β-actin (13E5) (rabbit mAb [1:1000]; Cell Signaling Technology, 5125), rabbit IgG (HRP-linked [1:1000]; Cell Signaling Technology, 7074), and mouse IgG (HRP-linked [1:1000]; Cell Signaling Technology, 7076). See complete unedited blots in the supplemental material. RNA/DNA extraction for RNA-seq and whole-exome sequencing. RNA was extracted from the primary tumors and the established cell lines using an RNeasy Kit (QIAGEN) per the manufacturer’s instructions. DNA was extracted from mouse livers and cell lines using DNeasy Kit (QIAGEN) per manufacturer’s instructions. RNA-seq analysis. RNA-seq libraries were prepared using a TruSeq Stranded mRNA Library Preparation Kit (Illumina, 20020595) according to the manufacturer’s protocol, and 75-bp paired-end reads were sequenced on a NextSeq 500 (Illumina). RNA reads were aligned to the mouse reference genome mm10 (ensembl) using STAR v2.5.3a ( https://github.com/alexdobin/STAR ) and the transcript levels were then quantified using SALMON v0.9.1 ( https://combine-lab.github.io/salmon/ ). Gene count data were extract from SALMON output using Tximport (Bioconductor), and normalized and compared using DESeq2 (Bioconductor). Consensus subtypes were determined by correlating the BBN, UPPL , and UPFL tumors to median expression per subtype using TCGA as the reference. The subtype was assigned based on the highest Pearson’s correlation to the given reference subtype. UROMOL class was predicted using the UROMOL predictor, as described in Lindskrog et al. ( 16 ). The basal, luminal, and Pparg signature scores were calculated based on the median expression of the genes for the indicated signature. For immune gene signatures, z scores were calculated based on all genes within previously published immune gene signatures on a per-sample basis. Immune cell fractions were calculated by CIBERSORTx ( https://cibersortx.stanford.edu ) using the default parameters. scRNA-seq analysis. CellRanger v6.1.1 ( https://www.10xgenomics.com/support/software/cell-ranger ) was used to demultiplex, generate FASTQ files, align reads to the mm10 reference genome, and produce a gene-cell matrix. Seurat v4.1.1 ( https://satijalab.org/seurat/ ) was used for further quality control and data processing. Cells with feature numbers smaller than 300 or larger than the mean plus 2-fold standard deviation of feature numbers, or with over 10% mitochondria-derived feature counts were considered as low-quality cells and were removed. Doublets identified by DoubletFinder v2.0.3 ( https://github.com/chris-mcginnis-ucsf/DoubletFinder ) were eliminated. The remaining gene-cell matrixes were transformed by SCTransform. To regress out potential batch effects within samples, samples were integrated using IntegrateData function in the Seurat package. In order to reduce the dimensionality of the data set, PCA was performed on the integrated data matrix using the top 4,000 highly variable genes. With the ElbowPlot function in Seurat package, the top 30 PCs were used to perform downstream analysis. Cell clusters were then visualized in t-distributed stochastic neighbor embedding (tSNE) space. Cell types were assigned according to the canonical marker gene expression in each cluster, then the cell types were confirmed using per-cluster SingleR (v1.8.1) ( 43 ). Epithelial cells were subsetted for further analysis. The gene-cell matrix of epithelial cells in each sample was transformed by SCTransform separately, then they were integrated in Seurat. To reduce the dimensionality of the data set, PCA was performed using basal ( Cd44 , Cdh3 , Krt16 , Krt14 , Krt5 , Krt6a , Krt6b , and Krt1 ) and luminal genes ( Fgfr3 , Foxa1 , Gpx2 , Cd24a , Erbb2 , Erbb3 , Krt19 , Krt18 , Krt8 , Krt7 , Krt20 , Gata3 , Pparg , Upk1a , Upk2 , Upk3a , Upk3b , and Upk1b ). Cell clusters were visualized in tSNE space. Basal, intermediate, and luminal clusters were designated according the expression levels of basal and luminal genes in each cluster. Syngeneic tumor formation. UPFL1 cells were injected subcutaneously (bilateral flank) into C57BL/6 mice, 5 × 10 6 cells in a total of 200 μL (100 μL of PBS [Gibco, 10010049] and 100 μL of Matrigel [Corning, 354234]). Tumor volume in subcutaneous mouse allografts was measured by caliper and calculated using the formula as follows: V = ( L × W × W )/2, where L is the tumor length and W is the tumor width. For efficacy studies, treatment was initiated at a tumor volume of 150 to 300 mm 3 . Treg proliferation assay. CD4 + T cells were isolated from spleens of FoxP3 + GFP + (B6) mice and magnetically enriched for CD4 + T cells through magnetic isolation (EasySep Mouse T cell Isolation kit, STEMCELL Technologies) with anti-CD8a biotin (catalog 13-0081-82, clone 53–6.7, Invitrogen). FoxP3 + GFP + cells were sorted using a MACSQuant Tyto cell sorter to a purity of greater than 99%. APCs were isolated from WT B6 splenocytes and irradiated at 30 Gy. The sorted Tregs were then stained with CellTrace Violet (C34571, Invitrogen) and plated with irradiated APCs, soluble anti-CD3 (14-0031-85, eBioscience), and IL-2 (212-12, PeproTech) with or without erdafitinib in the cell culture. Cells were cultured for 3 days, stained with Zombie NIR (423105, BioLegend) and CD4-PE antibody (catalog 12-0042-82, clone RM4-5, Invitrogen), and FACS analyzed. Compounds and therapeutic studies. Erdafitinib for in vitro studies was obtained from Selleck Chem (catalog S8401). For in vivo studies, anti–PD-1 and IgG2a isotype for in vivo studies were obtained from BioXcell (anti–PD-1: catalog BE0273, clone 29F.1A12 or IgG2a: catalog BE0089, clone 2A3). Anti–PD-1 or control IgG2a was administered 3 times a week (Monday, Wednesday, and Friday) via intraperitoneal injection. Erdafitinib (12.5 mg/kg) for in vivo studies was obtained from Janssen (JNJ-42756493). The in vivo vehicle consisted of 2-hydroxypropyl-β-cyclodextrin (MilliporeSigma, 332593). Both erdafitinib and vehicle control were administered by oral gavage twice a day from Monday to Friday. Flow cytometry. For flow cytometry experiments on UPFL1 allografts, treatment was initiated at tumor volume of 300 to 600 mm 3 to allow for sufficient tumor material. Mice were treated with vehicle, erdafitinib, anti–PD-1, or a combination of erdafitinib and anti PD-1 as per Compounds and therapeutic studies , with the exception that erdafitinib or vehicle was given by oral gavage twice a day for 7 days. Tumors were collected, minced into small pieces with a scalpel, and digested at 37°C for 30 minutes with a solution made of DNase I (Sigma-Aldrich, 10104159001), collagenase D (Sigma-Aldrich, 11088866001) and Hank’s balanced salt solution (Gibco, 14025-092). The digested tissue was filtered with a 70-μm cell strainer (Thermo Fisher Scientific, 22-363-548). The single-cell suspension was incubated with 1× RBC buffer (BioLegend, 420301) for 3 minutes at room temperature to lyse red blood cells. Cell pellets were washed with 1× PBS (Gibco, 10010049), spun down at 1,500 rpm for 5 minutes at 4°C, and resuspended in FACS buffer (2% FBS and 1× PBS). The isolated tumor-infiltrating immune cells were processed for flow cytometry analysis. Live cells were determined by using Zombie Aqua fixable viability kit (BioLegend, 423101/423102). Cells were stained with surface and intracellular markers for lymphoid and myeloid population. Examples of gating strategies can be found in Supplemental Figure 6 . Cells were imaged by using a BD Fortessa with FACSDiva software (v.9.0) and results were analyzed with FlowJo software (v.10.8.0). IHC. Chromogenic IHC was performed on paraffin-embedded tissues that were sectioned at 5 μm. This IHC was carried out using the Leica Bond III Autostainer system. Sequential tissue sections were labeled for antigens using Ki-67 (Cell Signaling Technology, 12202S) or cleaved caspase 3 (Biocare Medical, CP229C) antibodies. Slides were dewaxed in Bond Dewax solution (Leica, AR9222) and hydrated in Bond Wash solution (Leica, AR9590). Heat-induced antigen retrieval was performed at 100°C in Bond-Epitope Retrieval solution 2 pH 9.0 (Leica, AR9640). After pretreatment, slides were incubated with anti–Ki-67 at 1:400 and anti–cleaved caspase 3 at 1:400 for 60 minutes, followed by Novolink Polymer (Leica, RE7260-CE) secondary. Antibody detection with 3,3′-diaminobenzidine (DAB) was performed using the Bond Intense R detection system (Leica, DS9263). Stained slides were dehydrated and coverslipped with Cytoseal 60 (Thermo Fisher Scientific, 23-244256). A positive control tissue was included for each run. IHC-stained slides were digitally imaged in the Aperio AT2 (Leica) using a ×40 objective. For CD8/Masson’s trichrome, IHC was carried out using the Leica Bond III Autostainer system where tissue sections were labeled for CD8 (Cell Signaling Technology, 98941). Slides were dewaxed in Bond Dewax solution and hydrated in Bond Wash solution. Heat-induced antigen retrieval was performed at 100°C in Bond-Epitope Retrieval solution 1 pH 6.0 (Leica, AR9961). After pretreatment, slides were incubated with primary antibody diluted at 1:200 for 60 minutes followed by Novolink Polymer secondary. Antibody detection with DAB was performed using the Bond Intense R detection system. These slides were then stained using a Masson Trichrome Kit (Epredia, 87019). Afterward, slides were dehydrated and coverslipped with Cytoseal 60 (Thermo Fisher Scientific, 23-244256). A positive control tissue was included for this run. IHC-stained slides were digitally imaged in the Aperio AT2 (Leica) using a ×40 objective. IHC quantification. Following staining, slides were digitized on a ScanScope AT2 slide scanner (Leica Biosystems) with a ×40 objective. The final 8-bit image per channel resolution was 0.2529 μm per pixel. Images were uploaded to eSlide Manager as JPEG-compressed Aperio SVS files and visualized with ImageScope v12.4.3 (Leica Biosystems). Images were then analyzed with the Aperio Image Processing Toolbox (Leica) using the Nuclear V9 algorithm for both Ki-67 and cleaved caspase 3 assays, as well as the Positive Pixel Count algorithm for cleaved caspase 3. For the Nuclear V9 algorithm, which is based on the RGB color model, the average optical density (OD) values were determined for the red, green, and blue channels for both counterstain (hematoxylin) and biomarker (DAB chromogen). These input values were calculated by sampling relevant pixels from representative images. Additional input parameters included the following: Clear Area Intensity = 240 (scale of 0–255 for 8-bit image depth), Threshold lower and upper limits = 0 and 230 respectively, Smoothing = 1 μm or 4 pixels, Merging = 1.5, Trimming = Medium, Minimum Size = 10 μm 2 or 156 pixels, Maximum Size = 1,000,000 μm 2 or 15,635,200 pixels, Roundness = 0.1, Compactness = 0, Elongation = 0.1, Weak (1+)/Moderate (2+)/Strong (3+) Thresholds = 200/175/150. The output data for the Nuclear V9 algorithm included number and percent positive nuclei, intensity values, and histological scores (H-scores). The H-score Excel formula was ([@[(3+) Percent Nuclei]]*3)+([@[(2+) Percent Nuclei]]*2)+([@[(1+) Percent Nuclei]]*1) to obtain scores on a scale of 0–300. These factors give extra weight to the more intensely positive nuclei. The Positive Pixel Count algorithm is based on the Hue/Saturation/Intensity (HSI) color model. Its input parameters were as follows: Hue Value (Center) = 0.1, Hue Width = 0.1, Color Saturation Threshold = 0.05, Intensity Threshold ranges (Weak/Medium/Strong) = 175–225/125–175/0–125, respectively. The output data for Positive Pixel Count included number, area, and intensity values for all pixels, as well as positivity (no. positive pixels/no. all pixels) and H-scores. Compression quality = 70, Compression ratio = 15–25. Materials availability. Previously unpublished mouse models and cell lines generated in the course of this study are available through the request of the corresponding authors Statistics. All data were collected and analyzed using RStudio 2021.09.0 build 351, R v4.1.1, and GraphPad Prism v9.5, unless otherwise noted. Analysis-specific packages have been noted within the analysis-specific methods. Statistical comparisons were performed using 2-sided t tests or Wilcoxon’s rank-sum test (in cases of non-normal distribution) for continuous variables. P values of less than 0.05 were considered significant. All box-and-whisker plots are represented by the IQR and midline at the median. Error bars represent Q1/Q3 ± (1.5 × IQR). Study approval. All animals studies were approved by their respective institutional review boards: IACUC of the University of North Carolina at Chapel Hill (Chapel Hill, NC) and IACUC at NYU Langone (New York, NY). Data availability. All mouse RNA-seq and scRNA-seq data have been deposited into the NCBI Gene Expression Omnibus (GEO) under accession number GSE217093 ( https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE217093 ). Upper quartile–normalized RSEM gene expression data and mutation calls for TCGA-BLCA samples were downloaded from the GDC legacy archive ( https://portal.gdc.cancer.gov ). All additional supporting data have been provided in the supplement as the Supporting Data Values file.

Results UPFL mice develop papillary, high-grade NMIBCs. To understand the role of FGFR3 in bladder cancer biology, we knocked in the cDNA of human FGFR3 encoding the S249C mutation under control of a LoxP-Stop-LoxP cassette into the collagen type 1, α1 ( Col1a1 ) locus to generate mice harboring the Col1a1 -LSL- FGFR3 S249 allele (hereafter called LSL-FGFR3 S249C ) ( Figure 1A ). Prior studies examining the effect of mutant FGFR3 in bladder cancer have routinely constitutively expressed the gene transgenically under control of the Upk2 promoter ( 19 , 20 , 22 ). Transgenic overexpression carries the caveats of inappropriate temporal expression (i.e., during development). We wished to examine the effect of FGFR3 S249C activation in adult mice, in a spatiotemporally relevant manner. Based on the evidence that FGFR3 alterations are more frequently seen in the luminal molecular subtypes ( Figure 1B ) and single-cell RNA-seq (scRNA-seq) data showing that Upk3a is significantly more highly expressed than Upk2 in luminal/umbrella and intermediate urothelial cells of the normal mouse urothelium ( Figure 1C ), we drove Cre recombinase expression from the Upk3a promoter in a tamoxifen-inducible manner. Additionally, with the exception of a single recent study ( 19 ), prior studies have shown that FGFR3 activation alone in the urothelium is not sufficient for tumorigenesis ( 20 – 22 ); therefore, we crossed the LSL-FGFR3 S249C mice with LSL-Trp53 R270H mice ( 27 ), Upk3a -CreERT2 mice, and LSL-Luc mice ( 28 ) to generate Upk3a-Cre ERT2 ; Trp53 LSL-R270H/+ ; LSL-FGFR3 S249C/+ ; Rosa26 LSL-Luc/+ mice (hereafter called UPFL ). At 8 to 10 weeks of age, UPFL mice were administered tamoxifen to activate Cre in the urothelium by oral gavage and monitored for bladder tumor formation via ultrasound ( Figure 1D ). During the observation window, 47% of UPFL mice developed tumors, with a median time to tumor formation of 49 weeks ( Figure 1E ). Gross inspection of the bladder tumors revealed tumors to be papillary ( Supplemental Figure 1A ; supplemental material available online with this article; https://doi.org/10.1172/JCI169241DS1 ), which was verified upon histologic examination ( Figure 1F and Supplemental Figure 1 , B–E). We also found evidence of upper tract UC (UTUC) ( Figure 1F ) in 22% of mice, consistent with a known enrichment of FGFR3 alterations in UTUC ( 29 ). Histologically, 67% of bladder tumors assessed were high grade and 67% were Ta, with 11% being Tis only ( Figure 1G ). While FGFR3 and TP53 alterations are mutually exclusive in MIBC (The Cancer Genome Atlas [TCGA] data set, co-altered frequency = 3.7%, 2-sided Fisher’s exact test P < 0.001), they are not consistently mutually exclusive in NMIBC. Indeed, when TP53 is assessed as a pathway, TP53 pathway members are co-mutated with FGFR3 relatively frequently. In the UROMOL NMIBC data set, we found that 24.6% (71/288) of samples were co-altered for both FGFR3 and one of the publication-defined TP53 pathway genes, while FGFR3 and the TP53 pathway were altered alone in 51 samples (17.7%) and 123 samples (42.7%), respectively. Similarly, when evaluating an NMIBC cohort from Memorial Sloan Kettering (MSK), FGFR3 is co-altered with p53 pathway alterations in 19 out of 105 of tumors (18%), while remaining mutually exclusive in MIBC ( Figure 1H ) ( 16 , 30 – 32 ). UPFL tumors are associated with luminal gene expression patterns. It has now been repeatedly demonstrated that bladder cancer is a heterogeneous disease with multiple molecular subtypes. While subtyping schema differ, there is a broad consensus around the features defining intrinsic luminal and basal-like subtypes of muscle-invasive disease ( 9 – 15 ). Prior studies have also consistently documented an enrichment of FGFR3 alterations in the MIBC luminal (specifically LumP) molecular subtype ( 9 , 10 , 14 , 15 ), and the NMIBC UROMOL class 1 and class 3 subtypes ( 16 ). Following transcriptome profiling by bulk RNA-seq, the UPFL expression data were merged with data on BBN and UPPL tumors from Saito et al. ( 33 ) to characterize its similarity to both MIBC and NMIBC molecular subtypes. Consensus subtype calling was performed on the merged cohort of primary murine bladder tumor models ( UPFL , BBN, and UPPL ) and we found that 100% of UPFL tumors were most correlated to the consensus LumP subtype. Of the remaining tumors, the majority of BBN tumors were basal/squamous (Ba/Sq), and UPPL tumors were distributed across LumP, luminal unstable (LumU), and luminal nonspecified (LumNS) ( Figure 2A ). Because the consensus subtypes were developed for MIBC and the UPFL tumors are NMIBC, we also examined the molecular subtypes using the NMIBC UROMOL classifier ( 16 ). The UROMOL classifier assigned all UPFL tumors to UROMOL class 1, with a majority of BBN tumors identified as UROMOL class 2b, and again the UPPL tumors were heterogeneously dispersed, with representation in all 4 UROMOL classes ( Figure 2B ). Lindskrog et al. reported that FGFR3 alterations were enriched in both the UROMOL class 1 and class 3 subtypes, with class 1 tumors more likely to be Ta and having a longer interval to recurrence ( Supplemental Figure 2A ) ( 16 ). Together, these results demonstrate that FGFR3 activation can drive RNA expression patterns that reflect UROMOL class 1 and consensus LumP bladder tumor subtypes, both of which have relatively good prognosis. We next co-clustered the UPFL tumors with the BBN and UPPL primary tumors ( 33 ) using a canonical list of luminal and basal-like genes and saw that the BBN and UPPL tumors had distinct patterns of gene expression, as we have previously described ( 33 ). The majority of UPFL tumors, however, clustered alone with expression patterns that appeared equally luminal but less basal than our previously published luminal UPPL model ( Figure 2C ). Quantification of luminal and basal scores derived from Choi et al. ( 9 ) demonstrated that UPFL tumors had a similar luminal score, but a significantly lower average basal score, which in turn resulted in a higher luminal score–to–basal score ratio than UPPL tumors ( Figure 2D ), demonstrating that constitutively active FGFR3 signaling promotes luminal gene expression but also suppresses basal gene expression. Examination of canonical basal transcriptional signatures (p63, STAT3) ( 9 ) showed UPFL tumors had suppressed basal transcription signaling compared with BBN ( Figure 2E ). While Pparg expression itself was similarly elevated in UPPL and UPFL tumors, a multigene PPARG transcriptional signature ( 17 , 34 ) demonstrated that UPFL tumors had a significantly more activated PPARG pathway relative to BBN and UPPL ( Figure 2F ). The finding with Pparg suggested we look at broader transcriptional networks rather than just individual genes. To this end, we performed regulon ( 35 ) analysis comparing UPFL to both BBN and UPPL tumors. We saw a large number of differential regulons between UPFL and BBN tumors, many of which reflected established differences between basal and luminal tumors (i.e., FOXA1, PPARG, ESR2) ( Figure 2G ). In contrast, UPFL and UPPL tumors had very few differential regulons. However, we were struck by Erg being the most differentially upregulated regulon in UPFL versus UPPL tumors and that another of the ETS transcription factor family, ETV5, was also highly upregulated in UPFL tumors (relative to UPPL tumors) ( Figure 2H ). Additionally, a number of developmentally related regulons (HOX genes/TBX2) were upregulated in UPFL tumors relative to UPPL . Two HOX gene families were represented in the differently expressed regulons and included HOXA (UPFL vs. BBN) and HOXB (UPFL vs. BBN and UPFL vs. UPPL) regulons. These observations are in keeping with the Höglund group’s report of high expression of HOXA and HOXB genes in the Lum UrobasalA subtype, which is enriched for FGFR3-mutant tumors ( 36 ). scRNA-seq of erdafitinib-treated UPFL tumors confirms that oncogenic FGFR3 drives luminal gene expression across the spectrum of basal to luminal tumor cells. Our findings and work from others support the notion that FGFR3 activation promotes a luminal phenotype. However, whether this finding from bulk RNA-seq is due to an expansion of luminal cells or whether FGFR3 mutations drive a luminal expression pattern across all tumor cells remains unresolved. In order to more directly assess the role oncogenic FGFR3 was playing in subtype-specific cells, we treated tumor-bearing UPFL mice with vehicle or erdafitinib, a small-molecule inhibitor of FGFR ( n = 2 per group). Following 1 week of treatment, we harvested and dissociated the tumors for scRNA-seq using the 10× Genomics platform. A majority of the isolated cells, as expected, were computationally designated epithelial ( Figure 3A ). We next separately clustered the epithelial cells using basal and luminal gene markers ( 9 ) and identified 3 epithelial groups ( Figure 3B ). Assessment of the most differentially expressed genes for each epithelial cluster allowed us to assign clusters 0, 1, and 2 to intermediate, luminal, and basal urothelial cells, respectively ( Figure 3C ). In alignment with the cell identities, calculation of the basal and luminal scores for each cluster demonstrated that cluster 2 had the highest basal gene expression score, while cluster 1 had the highest luminal score and cluster 0 had an intermediate basal and luminal score ( Figure 3D ). We next examined the effect of erdafitinib on basal, intermediate, and luminal cell proportion. Erdafitinib-treated tumors appeared to have a larger proportion of intermediate cells at the expense of both luminal and basal cell types ( Figure 3E ). To better understand how erdafitinib was influencing the basal and luminal transcriptional programs within each epithelial cell type, we calculated the basal and luminal scores for the erdafitinib-treated cells and compared them to their matched vehicle-treated counterpart. While erdafitinib treatment upregulated the basal score, its most pronounced effect was the suppression of the luminal score, specifically within luminal cells ( Figure 3F ). We saw a similar pattern of gene expression change when specifically looking at erdafitinib’s effect on Krt5 and Upk3a expression across luminal, intermediate, and basal cell types, confirming the basal and luminal score findings ( Figure 3G ). Taken together, these data demonstrate that FGFR3 promotes a luminal expression pattern and suppresses the basal transcriptional program across all urothelial cell types, but most prominently in luminal cells. UPFL tumors are enriched for cytokine signaling relative to UPPL tumors and have an intermediate T cell–inflamed immune contexture. We next performed gene set enrichment analysis (GSEA), using fast GSEA (fgsea) ( 37 ), comparing UPFL and UPPL tumors and saw that the IFNA and IFNG pathways were significantly upregulated in UPFL tumors (relative to UPPL tumors) and the E2F and G 2 /M checkpoint pathways were significantly downregulated ( Figure 4A and Supplemental Figure 2B ). While the gene expression changes could emanate from tumor cells themselves, it is also possible that FGFR3 activation may induce a more T cell–inflamed tumor microenvironment than Pten loss in the UPPL model. We therefore examined the relative immune contexture of UPFL tumors as defined by immune gene signature expression. By co-clustering our UPFL primary tumors with previously published BBN and UPPL tumors, we saw that BBN and UPPL tumors have relatively T cell–inflamed and non–T cell–inflamed immune gene signature profiles, respectively, similar to previous findings from our group ( Figure 4B ) ( 33 ). The UPFL tumors appeared to have an intermediate inflamed tumor microenvironment. This pattern is also reflected within the UROMOL data, while class 2b tumors (similar to BBN) have the highest level of immune signal, and class 1 tumors (similar to UPFL ) have significantly higher expression of T cell signatures than class 3 ( Supplemental Figure 3 ) ( 16 ). In a per-signature analysis, UPFL tumors were significantly enriched, as compared with UPPL tumors, for numerous subsets of T cells, including CD8 + , central memory, and effector memory, while trending toward significance for T follicular helper cells and γδ T cells (Bindea) ( Figure 4, C–G ), the latter of which are important for adaptive immunity at mucosal surfaces ( Figure 4G ). Finally, we examined the level of expression of fibroblast and stromal signatures (FTBRS and EMT_Stroma), which have been shown to correlate with ICI response ( 38 , 39 ). While both signatures were significantly lower in UPPL compared with BBN tumors, UPFL tumors were only significantly lower that BBN tumors for the FTBRS signature ( Figure 4H ). UPFL1 and UPFL3 cell lines are sensitive to FGFR inhibition with erdafitinib. As genetically engineered mouse model (GEMM) tumors require a median of 49 weeks to form, we needed a more efficient and reproducible model to allow for the study of FGFR3-driven biology. To our knowledge, at present, there are no FGFR3 -mutant murine cell lines that have been used to form syngeneic tumors. To this end, we generated 2 cells lines ( UPFL1 and UPFL3 from tumors UPFL8425 and UPFL8583-1, respectively) using the conditional reprogramming of cells method described previously by Liu et al. ( Figure 5A ) ( 40 ). We confirmed in both cell lines the presence of recombination of the LSL cassette in the LSL-FGFR3 S249C allele ( Figure 5B ). We next assessed the relative sensitivity to the pan-FGFR inhibitor, erdafitinib, of the UPFL1 and UPFL3 cell lines to our previously published UPPL1541 cells ( 33 ). As expected, both the UPFL1 and UPFL3 cell lines had a low IC 50 (15 nM and 19 nM, respectively) to erdafitinib; this was in contrast with the 1.6 μM IC 50 for UPPL1541 cells ( Figure 5C ). Moreover, erdafitinib treatment of UPFL1 and UPFL3 cells suppressed the MEK/ERK pathway, in keeping with the known signaling pattern seen in human cell lines ( Figure 5D and Supplemental Figure 4 ). Therefore, the UPFL1 and UPFL3 cell lines are relevant models for the effects of FGFR3 inhibition in bladder tumors. Interestingly, however, the UPFL3 cells did have rebound upregulation of p-ERK after 30 minutes of erdafitinib exposure ( Supplemental Figure 4 ) that was not seen in UPFL1 cells ( Figure 5D ); this phenomenon is still unexplained at this time. FGFR inhibition enhances the effect of PD-1 blockade. Despite having a relatively non–T cell–inflamed tumor microenvironment profile, studies have confirmed that FGFR3 -altered tumors respond as well as FGFR3 -WT tumors to ICI ( 24 , 25 ). Nonetheless, to our knowledge, no prior work has examined how FGFR3 alterations functionally affect the tumor microenvironment, nor how acute FGFR inhibition may cooperate with ICI in FGFR3 -altered bladder cancer. To this end, we treated mice bearing syngeneic UPFL1 tumors with control, erdafitinib, anti–PD-1, or the combination of erdafitinib and anti–PD-1. We found that anti–PD-1 treatment led to significantly increased tumor growth (hyperprogression) and erdafitinib treatment had significantly decreased tumor volume, relative to control treated tumors ( Figure 6A ), while the combination of erdafitinib and anti–PD-1 had significantly decreased tumor growth relative to all other treatment arms. To better understand whether these changes in tumor size were related to decreased proliferation or increased cell death, we stained tumor sections from available UPFL1 tumors with antibodies against Ki-67 (proliferation) and cleaved caspase 3 (apoptosis) ( Supplemental Figure 5A ). While overall we did not see significant changes in either marker, there was a trend toward decreased proliferation among all treatment groups, with the greatest effect seen in the combination anti–PD-1/erdafitinib group ( Supplemental Figure 5B ). Consistent with the lack of response within the single-agent anti–PD-1 group, there was no difference in cleaved caspase 3 between control and anti–PD-1; however, erdafitinib and the combination treatment both trended toward increased apoptosis ( Supplemental Figure 5C ). Erdafitinib and anti–PD-1 combination therapy promotes highly inflamed tumors. ICI is dependent on the presence of immune cells that need to interact with tumor cells in the tumor microenvironment. To characterize the composition of the tumor microenvironment of UPFL1 tumors following anti–PD-1 with and without FGFR inhibition, we performed both immunohistochemistry (IHC) and flow cytometry on the posttreatment UPFL1 allografts. FFPE sections from the 4 control/treatment groups were co-stained with antibody against CD8 and Masson’s trichrome to assess IHC immune phenotype, as previously described ( 38 ) ( Supplemental Figure 5D ). Immune phenotype calls were then made by an expert genitourinary pathologist. While the majority of the control (6/7), anti–PD1- (3/5), erdafitinib (6/7), and erdafitinib/anti–PD-1 combination-treated (6/6) tumors were classified as inflamed, the pathologist noted variation in the CD8 staining intensity and therefore categorized inflamed tumors as CD8 high/low. Only anti–PD-1/erdafitinib–cotreated tumors (6/6) had consistently high CD8 staining, with control, anti–PD-1–, and erdafitinib-treated tumors having overwhelming low CD8 staining ( Figure 6B and Supplemental Figure 5D ). Anti–PD-1 treatment results in Treg expansion that is abrogated by concurrent FGFR inhibition. To better ascertain the cell type composition of the treated tumors, we performed flow cytometry on UPFL1 syngeneic tumors after 1 week of treatment with vehicle, anti–PD-1, erdafitinib, or the combination of anti–PD-1 and erdafitinib. As expected, anti–PD-1 inhibited the binding of PD-1–specific antibodies for flow cytometry on CD8 + and CD4 + T cells. Erdafitinib did not change levels of PD-1 on CD4 + or CD8 + T cells ( Supplemental Figure 6A ). We observed no significant changes in CD45 + cells, total T cells (CD3 + ), or CD8 + cytotoxic T cells by any treatment ( Supplemental Figure 6B ). Anti–PD-1–treated tumors, however, demonstrated increased numbers of CD4 + T cells that expressed higher levels of CTLA-4 ( Figure 6C ). Anti–PD-1–treated tumors also had a numerically, albeit not significantly, greater number of CD4 + FoxP3 + Tregs as compared with the control tumors ( Figure 6D ). The increase in Tregs is similar to prior studies suggesting that the expansion of Tregs by PD-1 blockade is a mechanism of hyperprogression ( 41 , 42 ). The increase in Tregs associated with anti–PD-1 treatment was abrogated in tumors that were cotreated with erdafitinib ( Figure 6D ). Additionally, we noticed a decrease, albeit not significant ( P = 0.16), in the Treg population in the erdafitinib-alone group. To determine whether this was an isolated trend or could be reproduced, we reanalyzed the scRNA-seq data in which UPFL tumors were treated with either vehicle or erdafitinib ( Figure 3 ). Using the cell type prediction package SingleR ( 43 ), we assigned a cell type to each of the cells present in the data set. In the 2 erdafitinib-treated tumors, T cells represented 1.2% and 10.2% of the immune cell populations, as compared with 10.4% and 17.6% for the vehicle-treated tumors (0.8% and 36% vs. 6.2% and 5.1% of total cells, respectively) ( Figure 6E ). We next wanted to determine whether this decrease was due to an overall reduction in the number of T cells or specific to the Treg population, which has been seen in our in vivo treatment experiment. To that end, we first examined expression of Ptprc , Cd3e , and Cd8a , the genes that encode CD45, CD3, and CD8, respectively. Recapitulating what we saw in the flow cytometry data, erdafitinib-treated tumors had no significant change in Ptprc (CD45) expression, but did have increased expression of both Cd3e (CD3, P = 0.05) and Cd8a (CD8, P = 0.007) ( Supplemental Figure 6C ). We next compared expression of Icos and Il1r1 (CD121a), markers of suppressive and proliferative Tregs, respectively, between the vehicle- and erdafitinib-treated tumors ( Figure 6F and Supplemental Figure 6D ) ( 44 – 46 ). In both cases, erdafitinib-treated samples, as a group, had decreased expression of Icos ( P = 1.1 × 10 –8 ) and Il1r1 ( P = 0.015), suggesting that erdafitinib treatment results in less suppressive Tregs and blocks their proliferation. In order to directly test the effect of erdafitinib on Tregs, we isolated FoxP3 + GFP + cells from the spleens of transgenic C57BL/6 mice overexpressing the diphtheria toxin receptor–eGFP (DTR-eGFP) fusion protein under control of the endogenous Foxp3 promoter ( Supplemental Figure 7 ) and simulated them in vitro with antigen-presenting cells (APCs) and anti-CD28 mAb; Treg proliferation was assessed as previously described ( 42 ) ( Supplemental Figure 8A ). At baseline, the addition of APCs and anti-CD3 to the FoxP3 + GFP + cell population induced an 8-fold increase in cell proliferation, and the addition of erdafitinib blunted the APC-induced proliferation in a dose-dependent manner, resulting in a 78% and 82% decrease at 1 μM and 3 μM concentrations, respectively ( Figure 6G ). Additionally, the reduction in proliferative Tregs translated to an overall increase in total percentage of cells alive at the end of the coculture experiment ( Supplemental Figure 8B ). Thus, these data demonstrate that FGFR inhibition reduced the proliferation of activated Tregs in a dose-dependent manner. Prior work has suggested that FGF-1 can have a direct effect on T cells in vitro by enhancing IL-2 production and nuclear translocation of NF-κB in FGFR-bearing Jurkat T cells ( 36 ). We assessed 2 publicly available RNA expression data sets to assess FGFR1 , FGFR2 , FGFR3 , and FGFR4 expression on T cell subsets and found that while T cells, in particular Tregs, had negligible expression of FGFR2 and FGFR4 ( FGFR3 did not meet expression thresholds to even be included in the data set), they did express FGFR1 ( Figure 6H and Supplemental Figure 8C ) ( 47 , 48 ). These findings in aggregate demonstrate that the combination of erdafitinib and anti–PD-1 is superior to either single agent alone and is potentially driven by the ability of erdafitinib to abrogate anti–PD-1–induced expansion of Tregs, potentially mediated through FGFR1 .

Discussion Herein, we present a mouse model of FGFR3 S249C -driven UC that when combined with the Trp53 R270H mutation results in high-grade NMIBC. The FGFR3-driven murine bladder tumors reflect their counterparts found in human UC. For example, RNA expression analysis of our UPFL tumors classify them in the consensus LumP and UROMOL class 1 MIBC and NMIBC molecular subtypes, respectively. Moreover, they have upregulated expression of luminal transcription factors such as Gata3 as well as heightened regulon activity of luminal transcription factors PPARG and FOXA1. Our immunocompetent UPFL model allowed for assessment of the immunobiology underlying FGFR3-driven tumors. We found that UPFL tumors have an intermediate T cell–inflamed immune contexture and used our syngeneic cell line, UPFL1 , to test the interaction between FGFR inhibition and ICI via PD-1 inhibition. UPFL1 syngeneic tumors were sensitive to erdafitinib but notably exhibit hyperprogression with anti–PD-1 treatment, potentially due to anti–PD-1–induced Treg proliferation. In contrast, the combination of erdafitinib and anti–PD-1 worked significantly better than either alone and erdafitinib appeared to abrogate anti–PD-1–induced Treg expansion. We saw that FGFR3 S249 activation promoted papillary histology as well as the development of upper tract UCs, which are known to be enriched in FGFR3 mutations ( 29 ). Moreover, transcriptome profiling of our UPFL tumors demonstrated an impressive enrichment in the luminal (consensus LumP and UROMOL class 1) molecular subtypes. While this observation along with prior work from others supports the notion that FGFR3 activation promotes a luminal phenotype, a previously unresolved question in the field is whether the enrichment of FGFR3 -altered tumors in luminal subtypes is driven by an expansion of luminal cells or whether FGFR3 mutations drive a luminal expression pattern across all tumor cells. Our scRNA-seq profiling of control or erdafitinib-treated UPFL tumors demonstrates that oncogenic FGFR3 alterations drive a luminal phenotype in all urothelial tumor cell types, suggesting that FGFR inhibition may impact a broad range of tumor cell types. We read with interest recent work from the Allis lab suggesting that oncogenic FGFR3 activation negatively associates with luminal genes and that FGFR inhibition increases expression of luminal genes ( 49 ). Our findings appear to be in opposition to their previously published work, although there are several technical differences worth mentioning. First, it is notable that their principal component analysis (PCA) associating FGFR3 mutations with basal gene expression is limited to TCGA LumP tumors only, rather than the entire spectrum of molecular subtypes, leaving open the possibility that their work more precisely reflects the role of FGFR3 in LumP tumors, while our studies and work from others reflects the role of FGFR3 alterations across the entire spectrum of UCs. Additionally, their work uses the small-molecule kinase inhibitor, PD-173074, to inhibit FGFR3, which in the literature has been used primarily as an FGFR1 inhibitor ( 50 , 51 ). Our work, in contrast, utilizes the clinically relevant compound erdafitinib. Finally, while their studies treated immortalized human bladder cancer cell lines, our experiments were performed on tumors from autochthonous GEMMs treated in vivo. Therefore, while at face value the work by Allis and coworkers appears to be at odds with our findings, both technical differences and the spectrum of tumors examined may account for the apparent discrepancies. A major finding from our work is the description of the UPFL1 allograft model as a potential model for hyperprogression in response to ICI and the ability of erdafitinib when combined with ICI to abrogate this hyperprogression. Anti–PD-1–induced hyperprogression has been attributed to reversal and expansion of exhausted, PD-1–expressing Tregs ( 41 , 42 ). In keeping with this notion, we found anti–PD-1–treated UPFL1 tumors had increased numbers of Tregs. Induction of Tregs by anti–PD-1 has also seen in an FGFR2 K660N ; Trp53-mutant NSCLC GEMM, although not statistically significantly ( 52 ). In contrast, there was no evidence of Treg induction in EGFR- and KRAS-driven NSCLC GEMMs ( 53 , 54 ). Therefore, the effect of anti–PD-1 on Treg abundance appears to be variable but intriguingly, the models that demonstrate anti–PD-1–induced Treg upregulation are FGFR driven. Remarkably, erdafitinib was able to completely abrogate anti–PD-1–induced Treg expansion, which was dose-dependent in an in vitro suppression assay. In mass cytometry data, we found that T cells express Fgfr1 , but not appreciable levels of Fgfr2 , Fgfr3 , or Fgfr4 and of T cell subsets, Tregs have the highest Fgfr1 expression. We therefore propose that cell-autonomous FGFR1 inhibition on Tregs may prevent their anti–PD-1–induced reinvigoration, allowing for anti–PD-1 to work effectively. This finding is highly clinically significant, as multiple studies are currently investigating the combination of pan-FGFR inhibitors with immunotherapy in bladder cancer (i.e., ClinicalTrials.gov NCT04003610 and NCT05564416) and other cancers (i.e., ClinicalTrials.gov NCT04949191 and NCT03547037). Furthermore, there are currently large-scale efforts afoot to develop FGFR inhibitors that selectively target FGFR2 and FGFR3, which are the predominant FGFRs with activating genomic alterations. These FGFR inhibitors will have minimal FGFR1 activity. Our work suggests that at least in the context of combining FGFR inhibition with ICI, it is critical to use an agent that inhibits FGFR1 or perhaps even the development of a selective FGFR1 inhibitor. Finally, our work also puts forth the notion that combined FGFR1 and PD-1 inhibition may be effective in both FGFR-mutated and non–FGFR-mutated tumors since the effect may be driven by FGFR1 inhibition on Tregs. In aggregate, our work reports a tractable model of FGFR3 S249C -driven papillary, high-grade, NMIBC with high penetrance and that reflects the cancer biology and immunobiology of FGFR3-driven human UC. High-grade noninvasive bladder tumors are an especially clinically important area, as patients with this grade and stage are at high risk of tumor progression, which can ultimately result in the morbidity of undergoing a radical cystectomy. A tractable model of FGFR3-driven, high-grade NMIBC will be useful for therapeutic development in this disease state where extension of progression-free survival allows patients to keep their bladder as long as possible. Our scRNA-seq studies demonstrate that FGFR3 promotes luminal expression patterns across all urothelial cell types, verifying that FGFR inhibition has the ability to affect the majority of bladder tumor cells. Finally, our preclinical work has uncovered a potential role for FGFR1-mediated Treg expansion and highlights the possibility for FGFR1 inhibition as a means to prevent anti–PD-1–induced hyperprogression. This is a previously unappreciated therapeutic strategy that should be considered for drug development and explored in clinical trials.

Authorship note: AO and TU contributed equally to this work. The combination of targeted therapy with immune checkpoint inhibition (ICI) is an area of intense interest. We studied the interaction of fibroblast growth factor receptor (FGFR) inhibition with ICI in urothelial carcinoma (UC) of the bladder, in which FGFR3 is altered in 50% of cases. Using an FGFR3-driven, Trp53-mutant genetically engineered murine model ( UPFL ), we demonstrate that UPFL tumors recapitulate the histology and molecular subtype of their FGFR3 -altered human counterparts. Additionally, UPFL1 allografts exhibit hyperprogression to ICI associated with an expansion of T regulatory cells (Tregs). Erdafitinib blocked Treg proliferation in vitro, while in vivo ICI-induced Treg expansion was fully abrogated by FGFR inhibition. Combined erdafitinib and ICI resulted in high therapeutic efficacy. In aggregate, our work establishes that, in mice, co-alteration of FGFR3 and Trp53 results in high-grade, non–muscle-invasive UC and presents a previously underappreciated role for FGFR inhibition in blocking ICI-induced Treg expansion. FGFR inhibition cooperates with PD-1 checkpoint blockade to reverse anti-PD-1 induced Treg expansion and block progression of murine FGFR3-mutant bladder tumors.

Author contributions AO, TU, MR, JSD, KKW, and WYK conceived and planned the study. AO, TU, MR, XZ, MZ, LDP, TC, YK, DW, HH, HL, ASG, JDR, UM, AC, FS, JS, SEW, JSD, and SCG generated, analyzed, and interpreted the data. JSS generated, analyzed, and interpreted data and wrote and reviewed the manuscript. AO, TU, MR, TLR, MIM, JSD, KKW, and WYK wrote and reviewed the manuscript. AO and TU contributed equally to this work and are listed in alphabetical order. Supplementary Material

We thank Bently Midkiff and Lauren Ralph in the Pathology Services Core (PSC) for expert technical assistance with histopathology/digital pathology, including tissue sectioning, IHC staining, and imaging. The PSC is supported in part by a National Cancer Institute (NCI) Center Core Support Grant (P30CA016086). Additionally, we thank the members of the Kim lab for useful discussions. The graphical abstract was created using Biorender.com. This work was supported by the University Cancer Research Fund (to WYK), the Waterproof Foundation (to MIM), the Thomas M. Mohr Fund for Bladder Cancer Research (to WYK), NIH grants R01-CA241810 (to WYK), KAKENHI 21K20814, and 23K14628 (to AO), and The Uehara Memorial Foundation (to TU). 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e169241

oa_package/f9/ab/PMC10786699.tar.gz

PMC10786700

37962961

Introduction Diarrhea remains an important cause of global mortality, and diarrheal illness accounts for more than 25% of deaths in children under 5 years of age in sub-Saharan Africa and South Asia ( 1 – 3 ). Secretory diarrhea, a major form of diarrhea with diverse etiologies, leads to the activation of prosecretory Cl – channels (such as cystic fibrosis transmembrane conductance regulator [CFTR]) and/or inhibition of proabsorptive ion transporters (such as sodium-hydrogen exchanger 3 [NHE3]) in the luminal membrane of the intestinal epithelial cells ( 4 ). Cholera is a severe form of secretory diarrhea, in which the key pathology is increased CFTR-mediated Cl – secretion and reduced NHE3-mediated Na + absorption due to elevated cAMP levels in intestinal epithelial cells ( 5 – 10 ). Current treatment of secretory diarrhea is primarily supportive, with fluid and electrolyte replacement by intravenous fluids or oral rehydration solution (ORS). Commonly used glucose-based ORS formulations rely on intact sodium-glucose cotransporter activity in cholera for electrolyte repletion ( 11 , 12 ). However, ORS has no effect on stool output, which is considered a key contributor to the low ORS use rates globally ( 13 ). Current ORS formulations do not directly address the hypersecretion, and cholera remains an important cause of mortality despite the availability of ORS ( 14 ). An improved ORS formulation with antisecretory properties, in addition to fluid and electrolyte repletion, could potentially improve clinical outcomes for patients with cholera or other secretory diarrheas. The extracellular Ca 2+ -sensing receptor (CaSR) is expressed in many tissues including parathyroid, kidney, bone, brain, and gut ( 15 ). In the intestine, the CaSR is a regulator of fluid and electrolyte transport ( 16 , 17 ). We recently showed that the FDA-approved CaSR activator cinacalcet inhibits CFTR-mediated fluid secretion and promotes NHE3-mediated fluid absorption in human colonic epithelial T84 cells and mouse intestine by promoting cAMP hydrolysis through phosphodiesterases (PDEs). Consistent with this mechanism, cinacalcet was very effective in various mouse models of secretory diarrhea ( 18 , 19 ). Although CaSR activation is a promising treatment strategy for secretory diarrheas, and short-term cinacalcet use for cholera might provide marked clinical benefits, long-term cinacalcet use for chronic diarrheas can cause systemic side effects due to CaSR activation in other organs. CaSR is physiologically activated by organic cations, including Ca 2+ , which the receptor is named after ( 20 ). We hypothesized that CaSR activation by enteral supplementation of its endogenous agonists can be used as a simple, safe, and effective treatment for secretory diarrhea. Although Ca 2+ is considered the major CaSR agonist, we found here that the extracellular Ca 2+ concentration had a minimal effect on CaSR activity in human intestinal epithelial cells and mouse intestine. Interestingly, CaSR activity and cAMP-induced Cl – secretion in human intestinal epithelial cells and mouse intestine strongly correlated with the extracellular concentration of Mg 2+ , an often-neglected CaSR agonist. Although Mg 2+ causes osmotic diarrhea at high stool concentrations (>100 mM) ( 21 ), we found here that Mg 2+ largely inhibited cyclic nucleotide–induced Cl – secretion at physiological levels (10–20 mM) seen in stool. Considering that stool Mg 2+ concentrations are essentially zero in patients with cholera ( 22 ), we postulate that oral Mg 2+ supplementation (either alone or in ORS) can be a potential treatment for cholera and other secretory diarrheas.

Methods Cell culture. T84 cells (ATCC CCL–248, human colon carcinoma cells) were cultured as previously described ( 18 , 29 ) on inserts (12 mm diameter, 0.4 μm pore size; Corning Life Sciences) and used for I sc experiments 7 days after plating. FRT cells stably expressing human WT CFTR (FRT-CFTR cells) were cultured as described previously ( 50 ) on inserts and used for I sc experiments 5 days after plating. Well-differentiated HBE cells were cultured at an air-liquid interface on inserts as described before ( 51 ). HBE cells were used for I sc experiments 21 days after plating, when they typically form a tight epithelium (TEER >1,000 Ω cm 2 ). I sc measurements. Cells were mounted in Ussing chambers containing bicarbonate-buffered Ringer’s solution at pH 7.4 (120 mM NaCl, 5 mM KCl, 1 mM CaCl 2 , 1 mM MgCl 2 , 10 mM D-glucose, 5 mM HEPES, 25 mM NaHCO 3 ). The MgCl 2 and/or CaCl 2 concentrations of both apical and basolateral solutions were altered (0–20 mM) in separate experiments as indicated in each figure. Secretagogues and ion channel inhibitors were added to both apical and basolateral bathing solutions. In some experiments, to measure apical Cl – conductance, the basolateral membrane was permeabilized with 500 μg/mL amphotericin B for 30 minutes and a 60 mM basolateral-to-apical Cl – gradient was applied. For these experiments, Ringer’s was the basolateral bathing solution (120 mM NaCl), and the apical solution contained 60 mM NaCl and 60 mM sodium gluconate. To measure basolateral membrane K + conductance, the apical membrane was permeabilized with 20 μM amphotericin B for 30 minutes, and an apical-to-basolateral potassium gradient was applied. The apical solution (pH 7.4) contained 142.5 mM K-gluconate, 1 mM CaCl 2 , 1 mM MgCl 2 , 0.43 mM KH 2 PO 4 , 0.35 mM Na 2 HPO 4 , 10 mM HEPES, and 10 mM D-glucose. In the basolateral solution (pH 7.4), 142.5 mM K-gluconate was replaced with 5.5 mM K-gluconate and 137 mM N -methylglucamine. The solutions were aerated with 95% O 2 /5% CO 2 and maintained at 37°C during the experiments. The I sc was measured using an EVC4000 multichannel voltage clamp (World Precision Instruments) via Ag/AgCl electrodes and 3 M KCl agar bridges. In parallel experiments, T84 cells were grown on permeable filters as described above and bathed with Ringer’s solution containing 0–10 mM CaCl 2 and/or MgCl 2 for 60 minutes. TEER was measured using a Millicell-ERS Resistance System with a dual-electrode volt-ohmmeter (MilliporeSigma) to test the effects of various Ca 2+ and Mg 2+ concentrations on epithelial barrier function. Net TEER (Ω/cm 2 ) was calculated by subtracting the resistance of cell-free media from the measured resistance ( 52 ). Chemicals. All chemicals were purchased from MilliporeSigma except CFTR inh -172 (MedChemExpress) and ST a toxin (Bachem Americas). CaSR activity and cAMP measurements. For intracellular Ca 2+ measurements, T84 cells were plated in 96-well, black-walled microplates (Corning). Confluent cells were loaded with the Ca 2+ indicator Fluo-4 NW (Invitrogen) according to the manufacturer’s instructions. Fluo-4 fluorescence was measured in each well continuously with a Tecan Infinite M1000 plate reader (Tecan Group) at excitation/emission wavelengths of 495 nm/516 nm after addition of 10 mM CaCl 2 or MgCl 2 . In some experiments, cells were pretreated with the PLC inhibitor U73122 (10 μM) or the sarcoplasmic/endoplasmic reticulum Ca 2+ -ATPase (SERCA) inhibitor thapsigargin (1 μM) for 10 minutes prior to addition of MgCl 2 . For the IP1 assay, T84 cells were grown in 384-well opaque plates (PerkinElmer) and treated with 0–20 mM CaCl 2 or MgCl 2 for 30 minutes. Next, cells were lysed and the IP1 concentration in each well was quantified using the IP-One Gq kit (Cisbio) according to the manufacturer’s instructions. For the cAMP assay, T84 cells were grown in clear 24-well plates and pretreated for 20 minutes with 10 mM MgCl 2 (or CaCl 2 ) with or without 500 μM IBMX or vehicle control (0.2% DMSO). After that, cells were treated with 10 μM forskolin (for 5 min) and lysed by repeated freezing/thawing and centrifuged to remove cell debris. The supernatant was assayed for cAMP using the cAMP Parameter immunoassay kit (R&D Systems) according to the manufacturer’s instructions. Animals. Casr fl/fl (strain 030647, C57BL/6 background), Vil1-Cre (strain 004586, C57BL/6 background), and WT mice (C57BL/6 or CD1 background) were obtained from The Jackson Laboratory. Intestinal epithelium–specific Casr-KO mice ( Vil1-Cre Casr fl/fl ) were generated by crossbreeding, and the genotype was confirmed by PCR. Animals were bred at the UCSF Laboratory Animal Resource Center, and experiments were conducted in adherence to the NIH Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011). Both male and female mice were used in all experiments. Intestinal I sc measurement in mice. Jejunum was excised under anesthesia and soaked in an iso-osmolar solution containing 300 mM mannitol and 10 μM indomethacin. Mucosa was stripped from the serosa/muscle layers under a dissection microscope and mounted onto Ussing chambers containing Ringer’s solution on the basolateral side. For the apical side, a similar solution was used, except 120 mM NaCl was replaced with 60 mM NaCl and 60 mM sodium gluconate, and glucose was replaced with 10 mM mannitol. I sc was measured as described above. Mouse models of cholera. For closed-loop model, 8- to 12-week-old CD1 mice were fasted overnight with access to 5% dextrose in water but no solid food. Mice were anesthetized with isoflurane, and body temperature was maintained at 36°–38°C during surgery using a heating pad. After a small abdominal incision to expose the small intestine, mid-jejunal loops (2–3 cm in length) were isolated by sutures as described previously ( 50 , 51 ). Loops were injected with 100 μL PBS (pH 7.4, 137 mM NaCl, 2.7 mM KCl, 8 mM Na 2 HPO 4 , 1.8 mM KH 2 PO 4 , 1 mM CaCl 2 , and 0.5 mM MgCl 2 ) containing 1 μg cholera toxin or PBS alone. Some loops were injected with PBS with or without cholera toxin with 20 mM Mg 2+ . After loop injections, the abdominal incision was closed with sutures, and mice were allowed to recover from anesthesia. Intestinal loops were removed 3 hours after surgery, and the loop length and weight were measured to quantify fluid secretion. Loop fluid was aspirated with a syringe, and the Mg 2+ concentration was quantified by colorimetric assay (MilliporeSigma). For the intestinal perfusion model, 10- to 14-week-old C57BL/6 mice were fasted overnight. Under isoflurane anesthesia, the proximal duodenum and terminal ileum were cannulated. The small intestine was lavaged gently with saline (at 37°C) to clear the luminal contents. After draining the intestine, the ileal catheter was clamped, and cholera toxin (10 μg in 2 mL saline) or the saline control was instilled and dispersed through the small intestine. After 2 hours, the clamps were opened, and whole-gut perfusion was initiated at 0.2 mL/min with Ringer’s solution containing 2 mM ferrocyanide [Fe(CN) 6 , nonabsorbable volume marker] and 1 or 10 mM MgCl 2 (in separate mice). After equilibration for 60 minutes, effluent samples were collected from the ileal catheter. The ferrocyanide concentration was determined in the infusate (ferrocyanide i ) and effluent (ferrocyanide e ) via absorbance, and net fluid transport (μL/min/cm) was calculated using the following previously described formula ( 53 ): perfusion rate: [perfusion rate × (ferrocyanide i )/(ferrocyanide e )]/length of gut (cm) × 1,000. In some experiments, we used the WHO ORS (TRIORAL, Trifecta Pharmaceuticals; 75 mM Na + , 65 mM Cl – , 75 mM glucose, 20 mM K + , 10 mM citrate) with and without 10 mM MgCl 2 . Statistics. Experiments with 2 groups were analyzed using a 2-tailed, unpaired Student’s t test. For 3 or more groups, analysis was performed with 1-way ANOVA and a post hoc Newman-Keuls multiple-comparison test. In all analyses, a P value of less than 0.05 was considered statistically significant. Data indicate the mean ± SEM. Study approval. The experimental protocols were approved by the IACUC of UCSF. Data availability. Values for all data points in graphs are available in the Supplemental Supporting Data Values file.

Results Extracellular Ca 2+ has minimal effect on CFTR-mediated Cl – secretion in T84 cells. To test the effect of extracellular Ca 2+ on Cl – secretion, short-circuit current (I sc ) measurements were done in human colonic T84 cells bathed with varying concentrations of Ca 2+ . Changing the extracellular Ca 2+ concentration had minimal effect on forskolin-induced I sc , as suggested by similar maximal forskolin responses in the presence of 0.1 to 20 mM Ca 2+ ( Figure 1A ). In Ca 2+ -free solution, forskolin caused slightly increased I sc changes compared with 10 mM or higher Ca 2+ concentrations ( Figure 1B ). The forskolin-induced secretory current in T84 cells was partially reversed by selective CFTR inhibitor (CFTR inh -172) treatment ( Figure 1A ). CFTR activity in T84 cells was not dependent on the extracellular Ca 2+ concentration, as indicated by similar CFTR inh -172 responses in the presence of 0.1–20 mM Ca 2+ ( Figure 1, A and C ). In the same experiments, the CaSR activator drug cinacalcet inhibited forskolin and CFTR inh -172–induced I sc changes by 80% ( Figure 1, B and C ). These results suggest that extracellular Ca 2+ had a minimal effect on CaSR activity and CFTR-mediated Cl – secretion in human intestinal epithelial cells. CFTR-mediated Cl – secretion in T84 cells is strictly dependent on the extracellular Mg 2+ concentration. Since Ca 2+ had minimal effect on CaSR activity and Cl – secretion in T84 cells, we next investigated the effects of Mg 2+ , a less-studied physiological CaSR agonist. Forskolin-induced I sc in T84 cells inversely correlated with the extracellular Mg 2+ (as MgCl 2 ) concentration ( Figure 2A ). Mg 2+ inhibited forskolin-induced maximal I sc by 70% at 10 mM or higher concentrations ( Figure 2B ), which is the physiological Mg 2+ concentration in human stool ( 21 ). The antisecretory effect of Mg 2+ was due to CFTR inhibition, as suggested by substantially lower I sc responses to CFTR inh -172 with increasing Mg 2+ concentrations ( Figure 2, A and C ). The antisecretory effects of 10 mM or higher Mg 2+ were comparable to the effect of the CaSR activator cinacalcet, as indicated by the similar forskolin and CFTR inh -172 responses. Altering extracellular Ca 2+ and Mg 2+ concentrations simultaneously had inhibitory effects on forskolin ( Supplemental Figure 1 , A and B; supplemental material available online with this article; https://doi.org/10.1172/JCI171249DS1 ) and CFTR inh -172 responses ( Supplemental Figure 1 , A and C), similar to that seen with alteration of Mg 2+ alone. Mg 2+ also had concentration-dependent inhibitory effects on forskolin and CFTR inh -172–induced I sc changes when citrate (MgC 6 H 6 O 7 , Supplemental Figure 2 , A and B) or sulfate (MgSO 4 , Supplemental Figure 2 , C and D) salts of Mg 2+ were used. Increasing solution Mg 2+ in these experiments slightly increased the solution osmolality (<10% for 10 mM MgCl 2 vs. 1 mM MgCl 2 ). Although equal osmolality increases by CaCl 2 did not have antisecretory effects ( Figure 1 ), we performed control studies to directly rule out any potential effects of increased solution osmolality on secretory currents in T84 cells. Addition of 30 or 60 mM mannitol to the solutions (equivalent to adding 10 or 20 mM MgCl 2 , respectively) did not affect forskolin or CFTR inh -172 responses, whereas 10 mM Mg 2+ had marked antisecretory effects in side-by-side studies ( Supplemental Figure 3 ). To rule out any potential effects of Mg 2+ on barrier permeability, we measured transepithelial electrical resistance (TEER) in the presence of various Mg 2+ and Ca 2+ concentrations. Altering extracellular Mg 2+ or Ca 2+ concentrations had no effect on TEER under the I sc study conditions ( Supplemental Figure 4 ). These results suggest that extracellular Mg 2+ is the major CaSR agonist and regulator of CFTR-mediated Cl – secretion in human intestinal epithelial cells. Extracellular Mg 2+ exerts its antisecretory effect by indirect inhibition of CFTR through CaSR activation. To determine whether Mg 2+ or Ca 2+ has direct CFTR inhibitory effects, we performed I sc studies in CFTR-transfected Fischer rat thyroid (FRT-CFTR) cells, which do not express the CaSR ( 23 , 24 ) and are commonly used to study CFTR modulators ( 18 , 25 – 27 ). With basolateral membrane permeabilization and a 60 mM basolateral-to-apical Cl – gradient, we found that forskolin induced a large Cl – secretory current in FRT-CFTR cells that was completely reversed by CFTR inh -172 treatment ( Figure 3A ). Increasing the extracellular Mg 2+ (or Ca 2+ ) concentration from 1 to 10 mM had no effect on forskolin or CFTR inh -172 responses ( Figure 3B ), suggesting that Mg 2+ did not have a direct CFTR inhibitory effect. The effects of Mg 2+ and Ca 2+ on CFTR-mediated Cl – secretion were also investigated in well-differentiated human bronchial epithelial (HBE) cells that express both CFTR and the CaSR, and have robust forskolin-induced Cl – secretory responses ( 18 ). Similar to intestinal epithelial cells, we found that increasing the extracellular Mg 2+ concentration from 1 to 10 mM inhibited forskolin and CFTR inh -172 responses in HBE cells by approximately 50% ( Figure 3, C and D ). However, increasing the extracellular Ca 2+ concentration from 1 to 10 mM had no effect on forskolin or CFTR inh -172 responses in HBE cells, suggesting that Mg 2+ is also the key CaSR agonist in airway epithelial cells. The antisecretory effect of Mg 2+ in T84 cells occurs through inhibition of apical membrane Cl – and basolateral membrane K + conductance. cAMP-induced secretory I sc in intestinal epithelial cells involves the coordinated action of the apical membrane CFTR Cl – channel and basolateral membrane K + channels ( 28 ). To selectively investigate the effect of Mg 2+ on apical CFTR conductance, we conducted experiments using T84 cells with selective basolateral membrane permeabilization and a basolateral-to-apical Cl – gradient ( 29 , 30 ). Under these conditions, addition of 10 mM Mg 2+ to the bathing solution inhibited forskolin and CFTR inh -172–induced I sc changes by 70% ( Figure 4, A and B ). To study the effect of Mg 2+ on basolateral membrane K + channels, we performed experiments with selective apical membrane permeabilization and an apical-to-basolateral K + gradient ( 29 , 31 ). In this setting, addition of 10 mM Mg 2+ to the bathing solution largely reduced basolateral membrane K + conductance, as shown by an 80% reduction in I sc changes in response to forskolin and BaCl 2 (cAMP-activated K + channel inhibitor) ( Figure 4, C and D ). These results suggest that Mg 2+ exerted its antisecretory effect in T84 cells via inhibition of the apical membrane CFTR Cl – channel and basolateral membrane K + channels. Mg 2+ reduces cAMP levels in T84 cells through activation of phospholipase C and PDEs. Activation of phospholipase C (PLC) and consecutive mobilization of intracellular Ca 2+ by IP3 is the key downstream pathway of CaSR activation, which leads to PDE activation and cAMP hydrolysis in human intestinal epithelial cells ( 16 , 32 ). To test the role of this mechanism in the Mg 2+ effect, we measured intracellular Ca 2+ by Fluo-4 fluorescence. Extracellular addition of 10 mM Mg 2+ resulted in a marked elevation of intracellular Ca 2+ , which was abolished by pretreatment with the PLC inhibitor U73122 ( Figure 5A ). Mg 2+ -induced intracellular Ca 2+ elevation was from intracellular stores, as suggested by the prevention of a Ca 2+ increase after endoplasmic reticulum stores were depleted by thapsigargin. Consistent with the lack of its antisecretory effects, 10 mM Ca 2+ had no effect on intracellular Ca 2+ levels in T84 cells ( Figure 5A ). IP1 is the stable downstream metabolite of IP3, and its quantification is considered the standard method to assess CaSR activity ( 33 ). Similar to the above-cited studies, we observed that extracellular Mg 2+ concentration-dependently increased intracellular IP1 levels in T84 cells, whereas Ca 2+ had minimal effect only at high concentrations ( Figure 5B ). Since cAMP is the major activator of apical CFTR and basolateral K + channels, reduced cAMP levels via CaSR activation might explain the antisecretory effect of Mg 2+ in T84 cells. Consistent with this mechanism, extracellular addition of 10 mM Mg 2+ (but not Ca 2+ ) reduced the forskolin-induced cAMP elevation in T84 cells ( Figure 5C ). The effect of Mg 2+ on cAMP levels was completely reversed with the PDE inhibitor IBMX, pointing to PDE activation as the key mechanism of the Mg 2+ effect. Collectively, these results further confirmed that Mg +2 (but not Ca 2+ ) was the major CaSR agonist in human intestinal epithelial cells and that Mg 2+ exerted its antisecretory effects via the known CaSR signaling pathways including PLC-mediated intracellular Ca 2+ mobilization and PDE activation. Mg 2+ does not affect Cl – secretion induced by Ca 2+ agonists. Although elevation of cAMP — and hence CFTR activation — is the key mechanism in cholera, in certain secretory diarrheas, elevation of intracellular Ca 2+ is the major driver of Cl – secretion via Ca 2+ -activated Cl – channels (CaCCs). To test the effects of Mg 2+ on CaCC activity, we performed I sc studies in T84 cells using the cholinergic agonist carbachol. In the presence of 1 or 10 mM Mg 2+ , carbachol induced comparable secretory currents ( Supplemental Figure 5 ), suggesting that CaSR activation by Mg 2+ did not affect CaCC activity. Mg 2+ inhibits cholera toxin, heat-stable E. coli enterotoxin, and vasoactive intestinal peptide–induced Cl – secretion in T84 cells. Cyclic nucleotide–mediated (cAMP- or cGMP-mediated) CFTR activation and consequent Cl – secretion represent the key pathology in certain secretory diarrheas including cholera, and traveler’s diarrhea and diarrhea caused by vasoactive intestinal peptide–secreting (VIP-secreting) tumors (VIPomas) ( 5 , 6 , 34 ). To test the efficacy of Mg 2+ in these settings, we conducted I sc experiments using T84 cells treated with cholera toxin, heat-stable E. coli enterotoxin (ST a toxin), and VIP as secretagogues. Increasing extracellular Mg 2+ concentration from 1 to 10 mM suppressed I sc changes induced by cholera toxin ( Figure 6, A and B ), ST a toxin ( Figure 6, C and D ), and VIP ( Figure 6, E and F ) by greater than 50%. In a similar manner, 10 mM Mg 2+ treatment resulted in reduced CFTR activity in all experiments, as suggested by markedly lower CFTR inh -172 responses compared with controls ( Figure 6, B, D, and F , right panels). Mg 2+ has antisecretory effects in mouse intestine via CaSR activation. As done for T84 cells, we tested the antisecretory effect of Mg 2+ in mouse jejunal mucosa. In WT mice, increasing extracellular Mg 2+ (but not Ca 2+ ) concentration from 1 to 10 mM reduced forskolin-induced secretory I sc by 40% ( Figure 7A ). Parallel studies were performed in intestinal epithelium–specific CaSR-KO mice ( Vil1-Cre Casr fl/fl ), in which 10 mM Mg 2+ had no antisecretory effects ( Figure 7B ). These results suggest that CaSR activation was the key mechanism for the antisecretory effect of Mg 2+ in mouse intestine. Mg 2+ also had marked antisecretory effects when applied only to the luminal side of the intestine ( Figure 7C ), which suggests its potential efficacy with oral treatment. Efficacy of Mg 2+ in mouse models of cholera. We tested the efficacy of Mg 2+ in an intestinal closed-loop mouse model of cholera ( Figure 8A , left). In this model, cholera toxin caused marked intestinal fluid accumulation at 3 hours, as suggested by an increased loop weight/length ratio. Intraluminal 20 mM Mg 2+ treatment at time zero (together with cholera toxin) inhibited the increase in the loop weight/length ratio by 40% ( Figure 8A , center and right). We measured the remaining Mg 2+ concentration in the loops at the end of the 3-hour period and found that the luminal Mg 2+ concentration dropped to 5.2 ± 1.2 mM, which suggests that the luminal Mg 2+ concentration might have become slightly subtherapeutic in some loops toward the end of this study. A potential approach to using Mg 2+ for diarrhea treatment is to fortify the ORS with Mg 2+ , which might provide sustained CaSR activation in the intestine for even higher efficacy. To test this idea, we established a mouse intestinal perfusion model of cholera ( Figure 8B , left), in which the Mg 2+ concentration of the perfusate was controlled. In this model, cholera toxin administration resulted in net intestinal fluid loss, as demonstrated by the negative fluid transport rate. Increasing the Mg 2+ concentration from 1 to 10 mM reversed net secretion into net absorption, as indicated by positive fluid transport rates ( Figure 8B , right). Mg 2+ (10 mM) also had antisecretory effects in the perfusion model when the WHO ORS solution was used, particularly at 90 minutes, when the cholera toxin effect was fully established ( Figure 8C ). These results further support our idea of developing a Mg 2+ -fortified ORS for cholera and other secretory diarrheas.

Discussion Here, we showed that CFTR-mediated Cl – secretion in human intestinal epithelial cells and mouse intestine was dependent on extracellular Mg 2+ concentration, which exerted its effect through CaSR activation (see Figure 9 for the proposed mechanisms). Interestingly, Ca 2+ , which is considered the main physiological CaSR agonist, had minimal effect on CaSR activity and Cl – secretion in intestinal epithelial cells. The antidiarrheal effect of Mg 2+ shown here might sound contradictory, since oral Mg 2+ supplements can cause osmotic diarrhea at high doses ( 21 ). The normal range of feces-soluble Mg 2+ concentration is 10–30 mM in healthy individuals, which increases to 100–150 mM in individuals with Mg 2+ -induced diarrhea ( 21 ). Although Mg 2+ can cause osmotic diarrhea at concentrations of greater than 100 mM, our findings suggest that physiological Mg 2+ concentrations in the intestinal lumen have antidiarrheal effects through CaSR activation. In patients with cholera and VIP-induced diarrhea, there is a lack of a stool osmotic gap ( 22 , 35 ), which indirectly suggests the possibility that stool Mg 2+ concentrations might be low in these conditions. Our findings suggest that increasing stool Mg 2+ concentrations to physiological levels (10–20 mM) by oral supplementation might offer a simple, safe, and effective therapy for secretory diarrheas. In addition, the stool Mg 2+ concentration can potentially be implemented as a secretory diarrhea biomarker to identify patients with low stool Mg 2+ who are likely to benefit from Mg 2+ supplementation. However, we would like to note that the earlier studies mentioned above did not directly measure stool Mg 2+ concentrations in patients, and thus the lack of a stool osmotic gap in these studies may also be explained by other factors such as potential measurement errors. Future clinical studies formally quantifying stool Mg 2+ concentrations in patients with cholera or other forms of diarrhea may be informative to validate its utility as a biomarker. The current treatment for cholera primarily relies on the ORS that was developed after the discovery of intact glucose-dependent Na + absorption in secretory diarrheas ( 11 , 12 ). However, the ORS does not have any effects on hypersecretion or stool output ( 4 , 13 ). Based on our results, Mg 2+ can be used as an adjunct therapy that can reduce fluid secretion and stool output. One potential issue with oral Mg 2+ treatment is its relatively rapid intestinal absorption as shown in our closed-loop studies, which may require frequent administration in severe diarrheas such as cholera. Additional dose/frequency determination, pharmacokinetics, and pharmacodynamics studies may be informative prior to testing the efficacy of intermittent oral Mg 2+ treatment. Alternatively, we postulate that the addition of 10 mM Mg 2+ to the ORS (Mg 2+ -fortified ORS) can provide sustained CaSR activation in the intestine and reduce stool output, in addition to repleting fluid and electrolytes. Although we present evidence for the efficacy of this approach in an animal model, further preclinical studies are required to optimize the formulation of Mg 2+ -fortified ORS that could ultimately be tested side-by-side with traditional ORS in clinical trials. A theoretical concern with the use of Mg 2+ in diarrhea treatment is potential hypermagnesemia as a side effect, since Mg 2+ salts have 50%–67% oral bioavailability ( 36 ). However, serum Mg 2+ levels are tightly regulated by the kidneys, which can rapidly decrease or increase Mg 2+ excretion according to dietary changes ( 37 , 38 ). Thus, the absorbed Mg 2+ is predicted to be rapidly excreted by the kidneys with minimal or no elevation in serum Mg 2+ levels. Another theoretical concern might be that repeated Mg 2+ treatment may result in depletion of the IP3-sensitive Ca 2+ pool in intestinal epithelia. As discussed above, the physiological Mg 2+ concentration in human stool fluid is 10–30 mM, which potentially suggests constitutive CaSR activation in the intestine under normal conditions. Considering the potentially low Mg 2+ concentrations in cholera and other secretory diarrheas discussed above, short-term Mg 2+ treatment may be effective in acute secretory diarrheas by restoring physiological Mg 2+ concentrations and CaSR activity in the intestine. Future studies testing the efficacy of long-term Mg 2+ treatment and potential tolerance development in chronic diarrheas may be informative to demonstrate the efficacy of Mg 2+ treatment in the chronic setting. Earlier studies investigating the roles of the CaSR in intestinal fluid transport solely focused on Ca 2+ and indicated Ca 2+ dependence of forskolin-induced Cl – secretion in rat colonocytes ( 32 , 39 ). On the basis of these results, the effects of Ca 2+ supplementation on diarrhea were studied in earlier clinical trials. Oral Ca 2+ supplementation was shown to have a mild antidiarrheal effect in traveler’s diarrhea, though mainly by preventing bacterial colonization ( 40 ). A large randomized, controlled trial in children compared the effects of low-calcium (50 mg/day) and regular-calcium (440 mg/day) milk on the number and duration of diarrhea episodes and found no benefits of higher calcium intake ( 41 ). Although the role of the CaSR in diarrhea has been known for decades ( 4 , 42 ), there are no large-scale clinical studies showing beneficial effects of Ca 2+ supplementation. Our findings here suggest that Mg 2+ (but not Ca 2+ ) is the key CaSR agonist in intestinal epithelia, which can potentially explain the lack of antisecretory effects of Ca 2+ in previous clinical studies. Certain compounds that elevate intracellular Ca +2 (such as cholinergic agonists) induce Cl – secretion, which is thought to be mediated by CaCCs. Although CaSR activation by Mg 2+ also increased intracellular Ca 2+ , Mg 2+ did not induce a secretory current in T84 cells, similar to what we have previously found with the CaSR-activator drug cinacalcet ( 18 ). In addition, Mg 2+ did not affect cholinergic agonist carbachol-induced secretory currents as shown here. Although cytosolic Ca +2 elevation is a shared mechanism between CaSR and cholinergic agonists, additional unshared signaling pathways, including crosstalk with EGF signaling ( 43 , 44 ), might be important determinants for the secretory effects of cholinergic agonists, but not CaSR agonists. Certain secretory diarrheas (such as those caused by rotavirus) are characterized primarily by CaCC-mediated Cl – secretion. Here, we found that CaSR activation by Mg 2+ largely inhibited cyclic nucleotide agonist–induced Cl – secretion, without any effects on Ca 2+ agonist–induced Cl – secretion. Thus, Mg 2+ may not be effective in secretory diarrheas in which CaCC activation is the major driver of intestinal fluid loss. Although our results here showed a marked inhibitory effect of Mg 2+ on CFTR-mediated Cl – secretion, CaCCs may also play a role in intestinal fluid loss in cyclic nucleotide–mediated diarrheas. In addition, there can be crosstalk between cAMP and Ca 2+ pathways, which can lead to activation of both CFTR and CaCCs in certain secretory diarrheas ( 4 ). Future studies investigating the effects of Mg 2+ on different secretory pathways and crosstalk mechanisms may be lead to a better understanding of its mechanisms of action and potential efficacy in patients with diarrhea. Although we showed here that Mg 2+ was effective in both cAMP- and cGMP-mediated diarrhea models, cyclic nucleotide elevation is not a common pathology in all diarrheas. Thus, Mg 2+ (alone or in an ORS) may not be effective as a general antidiarrheal, but it can potentially be used as a specific and targeted treatment for cyclic nucleotide–mediated diarrheas such as cholera and traveler’s diarrhea and diarrhea induced by VIPoma and GUCY2C mutations. The majority of earlier studies on CaSR agonists used bovine parathyroid cells or HEK-293 cells transfected with CaSR ( 45 ). In both settings, Ca 2+ is 2- to 3-fold more potent CaSR agonist than Mg 2+ ( 46 – 48 ). Consistent with this, the serum ionized Ca 2+ concentration is the primary determinant of parathyroid hormone (PTH) secretion in vivo ( 49 ). We show here that in human intestinal and airway epithelial cells natively expressing the CaSR, Mg 2+ was the key agonist for this receptor. These results also suggest that the term “calcium-sensing receptor” might be an oversimplification of the biological roles of this receptor. In conclusion, we have demonstrated that the extracellular Mg 2+ concentration was the major regulator of CaSR activity and cAMP-induced Cl – secretion in intestinal epithelial cells. Oral Mg 2+ supplementation, either alone or in an ORS, can offer a simple, safe, targeted and effective treatment for cyclic nucleotide–mediated secretory diarrheas such as cholera and traveler’s diarrhea or those induced by VIPoma and GUCY2C mutations.

Authorship note: LDSG and TC are co–first authors and contributed equally to this work. Cholera is a global health problem with no targeted therapies. The Ca 2+ -sensing receptor (CaSR) is a regulator of intestinal ion transport and a therapeutic target for diarrhea, and Ca 2+ is considered its main agonist. We found that increasing extracellular Ca 2+ had a minimal effect on forskolin-induced Cl – secretion in human intestinal epithelial T84 cells. However, extracellular Mg 2+ , an often-neglected CaSR agonist, suppressed forskolin-induced Cl – secretion in T84 cells by 65% at physiological levels seen in stool (10 mM). The effect of Mg 2+ occurred via the CaSR/Gq signaling that led to cAMP hydrolysis. Mg 2+ (10 mM) also suppressed Cl- secretion induced by cholera toxin, heat-stable E . coli enterotoxin, and vasoactive intestinal peptide by 50%. In mouse intestinal closed loops, luminal Mg 2+ treatment (20 mM) inhibited cholera toxin–induced fluid accumulation by 40%. In a mouse intestinal perfusion model of cholera, addition of 10 mM Mg 2+ to the perfusate reversed net fluid transport from secretion to absorption. These results suggest that Mg 2+ is the key CaSR activator in mouse and human intestinal epithelia at physiological levels in stool. Since stool Mg 2+ concentrations in patients with cholera are essentially zero, oral Mg 2+ supplementation, alone or in an oral rehydration solution, could be a potential therapy for cholera and other cyclic nucleotide–mediated secretory diarrheas. Extracellular Mg2+ regulates CaSR activity and inhibits CFTR-mediated Cl- secretion in human intestinal epithelial cells and mouse intestine.

Author contributions OC made the original discovery, conceptualized the study, and designed the experiments; LDSG, TC, RM, PDC, QG, and OC performed the experiments. LDSG, TC, and OC analyzed the data. OC obtained funding, supervised the study, and wrote the manuscript. LDSG and OC revised the manuscript. All authors read the manuscript and agreed to the submitted form. The order of the co–first authors’ names was determined alphabetically. Supplementary Material

This study was supported by grants from the NIH (DK126070 and DK072517) and the Cystic Fibrosis Foundation. 11/14/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e171249

oa_package/08/f8/PMC10786700.tar.gz

PMC10786701

38015629

Introduction Aging introduces a series of complex molecular, structural, and functional changes to blood vessels, with important consequences for organ physiology ( 1 ). In the brain, compromised vascular function can trigger a spectrum of pathologies, ranging from acute incidents like strokes to enduring and incapacitating conditions such as cerebral hypoperfusion that can ultimately result in cognitive impairment and dementia ( 2 – 4 ). These vascular dysfunctions can be accelerated and intensified by concurrent medical conditions that negatively affect blood vessels, including diabetes, atherosclerosis, hypercholesterolemia, and hypertension ( 3 ). However, it is important to note that the presence of these conditions alone does not predict development of dementia, and their absence does not necessarily assure freedom from cognitive decline during the aging process. Consequently, we must inquire: What are the specific triggers and molecular markers associated with aging that can offer more robust predictions for the occurrence of cerebral hypoperfusion and subsequent cognitive decline? Answers to this question will contribute to the understanding, prevention, and treatment of cognitive deficiencies associated with aging and offered the foundation for this study. At the core of vascular physiology resides cerebral blood flow (CBF) as the absolute readout of brain perfusion ( 5 ). The dynamic nature of CBF enables responses to various stimuli, such as increased brain activity and vasoactive challenges ( 6 ). Remarkably, cerebral autoregulation plays a pivotal role in maintaining CBF stability at a baseline level to ensure optimal brain perfusion. Notably, alterations in CBF have been observed in several neurodegenerative conditions, including Alzheimer’s disease, emphasizing the potential significance of CBF changes in the context of cognitive decline ( 7 ). The control of CBF lies within the purview of vascular smooth muscle cells (VSMCs) and pericytes, both contractile cells that finely tune blood flow and pressure ( 8 , 9 ). Their capacity to contract and relax facilitates the necessary adjustments in CBF to meet the heightened metabolic demands of the neuronal tissue ( 10 ). This intricate process, initiated early in development, is closely intertwined with the brain parenchyma itself ( 11 , 12 ). In fact, neuronal activity and blood flow are coupled and meticulously regulated, supporting the concept that the health of the brain relies on the crosstalk and integration of a neurovascular unit, rather than the parenchyma alone ( 13 , 14 ). Furthermore, regulation necessitates a high-level coordination with upper branches of the vascular tree, as, upon increased specific needs of flow in one region (hyperemia), upper branches must dilate to prevent reductions in downstream microvascular pressure ( 15 ). Consequently, within the brain, a well-coordinated flow response relies on vasodilation from distal to proximal arterial segments and myogenic mechanisms that enhance flow in response to decreased pressure. As key regulators of CBF, we focused our investigation primarily on the alterations linked to aging in VSMCs and pericytes. The initial point was the identification of Notch signaling as the predominant molecular indicator of aging-related changes in blood vessels. In particular, our findings revealed a significant reduction in both Notch3 and Jagged1 during the aging process, accompanied by decreases in downstream targets and regulators of the signaling pathway. Building on these results, we embarked on an integrated analysis of patient samples to validate these findings and additional mechanistic experiments with animal models. Using a mouse with Notch3 inactivation, we comprehensively explored how absence of this signaling pathway affects the molecular physiology, cell biology, and vascular function of brain vessels during aging. In the process, we uncovered a critical connection between Notch signaling and vascular contractility with consequences beyond regulation of CBF.

Methods Mice B6;129S1- Notch3 tm1 Grid/J ( Notch3 –/– ) (JAX:010547) and C57BL/6J (JAX:000664) mice were purchased from The Jackson Laboratory. Animals were housed at UCLA, Northwestern Medicine, and National Heart, Lung, and Blood Institute facilities. Human brain tissue Formalin-fixed frontal lobe sections were acquired from the Alzheimer’s Disease Research Core at the University of Michigan and the Northwestern Nervous System Tumor Bank. Detailed characteristics including age, sex, and NOTCH3 mutation are described in Supplemental Table 3 . Immunostaining Tissue sections. Paraffin-embedded blocks were sectioned at 5 μm, and antigen retrieval was performed in either citrate buffer (pH 6) or Tris-EDTA buffer (pH 9) followed by primary and secondary antibody incubations. Vibratomed slices. Fixed tissue samples were vibratomed at 100 μm and blocked for 48 hours, followed by incubation with primary antibodies for 48–72 hours (4°C) and secondary antibodies overnight (4°C). Fixed cells. Cells were fixed with 2% PFA or methanol for 10 minutes followed by permeabilization for 2 hours, and incubated with primary antibodies overnight and secondary antibodies for 2 hours. Information on antibodies can be found in Supplemental Methods . Images were collected with the Nikon A1R confocal system with NIS Elements acquisition software and analyzed with Imaris software (9.9.0, Bitplane). A subset of samples were imaged using a Nanozoom RS digital slide scanner (Hamamatsu) (H&E) and analyzed with the companion NDP Viewer software or the Nikon Eclipse light microscope using a ×10 air objective (PAS). A list of antibodies can be found in Supplemental Methods . Vascular casting of mouse brains Animals were injected with heparin (1 U/g, i.p.) and then euthanized. A catheter prefilled with 10 −4 M sodium nitroprusside (SNP) in PBS was introduced into the thoracic aorta and secured in place with 2 sutures. SNP/PBS was perfused to both maximally dilate the vasculature and remove all blood; then Microfil (Flowtech Inc.) was injected at an 8:1:1 ratio (polymer/diluent/curing agent) until the distal vasculature of the brain was filled ( 53 ). Brains were dissected and placed in 10% buffered formalin. Micro-CT scan and parameters and reconstruction Brain samples were scanned on a Bruker Skyscan 1276 system with voltage 50 kV, current 100 μA, and a 0.25 mm Al filter. Image capture was set with a pixel size of 6 μm. Per scan, we acquired an average of 2,100 projection images at a 0.2° rotation step over 360° total to improve the signal-to-noise ratio. Reconstruction of the data was generated with NRecon software (Bruker) for .bmp files for downstream processing by Imaris (9.7.2, Bitplane). Additional information on visualization and quantitative analysis is provided in Supplemental Methods . Retinal VSMC coverage analysis Quantification of retinal VSMC coverage of arterial vessels was performed using the NIS Elements software to calculate the relative percentage of αSMA + area. Additional details and imaging parameters are provided in Supplemental Methods . Brain vessel isolation and single-cell sequencing Notch3 –/– and control littermates were euthanized at 1, 12, and 24 months. After sacrifice, mice were perfused with versene, and brains were dissected in PBS. Using a microscope, meninges and penetrating vasculature were carefully dissected from the brain and incubated in 500 μL of digestion solution for 25 minutes at 37°C on an orbital shaker while pipetting up and down every 2–3 minutes to mix the solution and generate a single-cell suspension. Detailed information on suspension cocktail components as well as postdigestion clean-up prior to cell counting and viability assessment is given in Supplemental Methods . Libraries were prepared using 10× Genomics Chromium Single Cell 3′ Library & Gel Bead Kit v3 per the manufacturer’s protocol. For the generation of single-cell gel beads in emulsion, cells were loaded onto the Chromium single-cell instrument (10X Genomics) with an estimated targeted cell recovery of approximately 5,000 cells. Sequencing was performed on an Illumina Novaseq 6000. Libraries were processed using the Cell Ranger pipeline (10X Genomics) and analyzed using the R package Seurat ( 54 ). The following R packages were used for data visualization in combination with Seurat: ggplot2, ggraph, igraph, and tidyverse. Vascular smooth muscle culture and contraction assays A portion of the descending aorta was dissected into small fragments (1 mm 3 ). Explants were plated in DMEM + 10% FBS. Cells were allowed to grow and trypsinized only after 1 week for subculture. Differential plating (5 minutes) was performed to separate fast-attaching cells (mostly fibroblasts) from VSMCs. The purity of the VSMC culture was confirmed by immunodetection of αSMA. To assess vascular smooth muscle contractility, a neutralized solution of collagen I was mixed with VSMC cell suspension and plated into 6-well plates. Gels were allowed to polymerize for 2 hours, and then medium was added to the wells. A p200 pipette tip was used to detach the gels from the lateral aspects of the wells. Gels were returned to the incubator for 24 hours and then fixed in 2% PFA for imaging and staining. Western blotting Lysates from aortic VSMCs were prepared in 8 M urea and run under denatured conditions on an SDS-PAGE 4%–12% gradient gel, transferred onto nitrocellulose membranes, and blocked and incubated overnight at 4°C with primary antibodies. Immunocomplex detection was performed with the enhanced chemiluminescence SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Fisher Scientific) and SuperSignal West Pico Maximum Sensitivity Substrate (Thermo Fisher Scientific) using the ChemiDoc XRS+ Molecular Imager (Bio-Rad). Densitometry analysis was performed using ImageLab Software. Information on antibodies used is available in Supplemental Methods . In vivo hemodynamics and vascular reactivity Under anesthesia, a pressure catheter (1.0F, model SPR1000, Millar Instruments) was introduced into the right carotid artery and advanced to the ascending aorta of the animal. Each animal was allowed to acclimate for 5 minutes, at which time isoflurane was reduced to 1%. Pressures were recorded using Chart 8 software (ADInstruments) and analyzed from 1 to 4 minutes. For reactivity, a central venous catheter (Instech, PE-10 tubing) was placed in the jugular vein; in addition, an arterial pressure catheter (Millar Instruments) was used to measure continuous blood pressure (BP) and heart rate. Baseline BP was established, and mice were then injected with increasing bolus concentrations (10 μg/kg and 100 μg/kg) of phenylephrine or increasing bolus concentrations (10 μg/kg and 100 μg/kg) of acetylcholine diluted in PBS. BP and heart rate were monitored throughout the procedure, and the animal’s BP was allowed to return to baseline before administration of the next dose. Response to vasoactive drugs was calculated as absolute maximal deviation of systolic BP from baseline after drug administration(s) before return to baseline. MRI perfusion and angiography MRI experiments were conducted using a 7T ClinScan MRI (Bruker) and dedicated mouse brain coils. Details of image acquisitions, processing, and analysis are given in Supplemental Methods . Surgical cisterna magna cannulation and thinned-skull preparation Mice were anesthetized with ketamine/xylazine at 100 mg/kg and 10 mg/kg, respectively, and placed in a stereotactic frame under a heating pad. A midline incision was made across the scalp, and the skin and periosteum were removed to expose the skull surface. Artificial cerebral spinal fluid at 37°C was applied to the exposed skull surface, and a high-speed drill was used to thin 2 circular areas of the skull (~1–1.5 mm in diameter) lateral to the superior sagittal suture between the bregma and lambda sutures anteriorly and posteriorly, respectively, corresponding to the cortical middle cerebral artery surface bilaterally ( 55 ). Cisterna magna cannulation was performed by tilting of the animal’s head forward within the stereotactic frame to expose the neck muscles and occipital crest. The overlying skin and dorsal neck muscles were separated with forceps along the midline, exposing the cisterna magna. A 30G sterile needle was filled and inserted at 45° at the center of the cisterna magna at a depth of approximately 1 mm. The needle tip was secured in place using cyanoacrylate glue with accelerator on the adjacent dural membrane. Once secured, the tubing was cut and sealed. A thin silicone ring was affixed to the exposed skull at the margins of the skin using Vetbond tissue adhesive to create a well for artificial cerebrospinal fluid (aCSF) during intravital imaging. Thirty minutes before initiation of intravital imaging, mice underwent retro-orbital injection with non-blocking rat anti–mouse PECAM-1 antibody for blood vessel labeling conjugated to DyLight-650. Intravital imaging of perivascular flow Imaging was performed using an Olympus BX-51WI Fixed Stage illuminator with a Yokogawa CSU-X1-A1 spinning disk, a Hamamatsu EMCCD C9100-50 camera, and a Modular Laser System with solid-state diode lasers with DPPS modules for 488, 561, and 640 nm and the appropriate filters as previously described ( 56 ). Synchronization was managed by a Prosync 2 Controller. Z axis movement and objective positioning were controlled by a Piezoelectric MIPOS100 System. Fifteen minutes before intravital imaging, green fluorescent polystyrene microspheres (FluoSpheres, 1.0 μm, 505/515 nm, 0.4% solids in aCSF) were briefly sonicated and infused at 2 μL/min for 5 minutes through the cannula using an infusion pump. Images were collected with a ×20 water immersion objective using Volocity software. No UV light/excitation was used for this method. Images from the Volocity software were transferred to Imaris. The largest artery in each visual field was measured along the widest axis for diameter analysis. Aneurysms were counted per vessel from 2–6 vessels per animal with a total of 5–8 animals per genotype quantified. Image analysis and fluorescent bead tracking Images were analyzed using ImageJ software. Fluorescent microsphere tracking was performed using individually acquired 1-minute single-channel recordings obtained across multiple intravital imaging sessions. Microsphere flow speed was measured by tracking of total perivascular distance traveled over time within a given series of frames and averaging of flow speed from 3 separate observed particles per recording. Dextran clearance experiments FITC-conjugated 70 kDa dextran (5 μL at 1% in aCSF) was infused into the cisterna magna of a mixed-sex cohort of 3-month-old C57BL/6J mice. Animals were euthanized from 5 to 180 minutes after injection, and 400 μL of blood was collected using cardiac puncture. Blood was mixed with 10% citrate buffer to obtain plasma, which was evaluated using a Synergy LX plate reader. To assess changes in early clearance in Notch3 –/– mice and controls, the sample process was applied and plasma was measured 5 minutes after injection. Quantification of diameter of penetrating arteries and perivascular space One-hundred-micrometer coronal sections injected with PECAM-1–647 were scanned using a Nikon A1R confocal microscope. In Imaris, 3D renderings of the cerebral vasculature were generated and diameter measured at 100 μm from the meninges into the cortex. A total of 10–16 vessels were measured per mouse, 5 mice per genotype quantified. To measure perivascular space, transversal sections stained with aquaporin 4 antibodies along with DAPI were scanned using a ×40 objective and images transferred into Imaris. For each vessel, the smallest distance was measured from the top of DAPI + nuclei to the edge of the aquaporin 4 staining. Nine to sixteen vessels were quantified per mouse, 3 mice per genotype. Quantification of GFAP coverage in penetrating arteries Two-hundred-micrometer coronal sections were stained with PECAM-1 and GFAP antibodies and scanned using a Nikon A1R confocal microscope. Images were processed using NIS Elements. In Imaris, surface feature was used to generate 3D renderings of the cerebral vasculature. For each vessel, 2 measurements across the total length of the vessel were generated using measurement tools and then averaged to obtain percentage of vessel length covered by GFAP + astrocytes. Nine to thirteen vessels were measured per animal, 5 animals per genotype. Identification and quantification of PAS aggregates Five-micrometer paraffin-embedded coronal sections were stained with periodic acid–Schiff (PAS). Images were collected with Nikon Eclipse light microscope using ×10 air objective. Images measured 1,378.25 total μm area. Five images per field of view were captured per animal and quantified using NIS Elements software spot quantification macro. Statistics For non-transcriptomic data sets, data were analyzed using GraphPad Prism version 10.0 for Windows. Two-tailed unpaired t test was used for data with a Gaussian distribution and equal variance. Two-tailed unpaired t test with Welch’s correction was used for data with a Gaussian distribution and unequal variance. Alternatively, data with non-parametric distributions were analyzed by Mann-Whitney U test. P values less than 0.05 were regarded as statistically significant. For multiple conditions, the Kruskal-Wallis test was used with non-parametric distributions with multiple testing corrections. All data are presented as the mean ± SD, unless noted. For scRNA-Seq the R package Seurat (version 3.0.2) was used. The FindAllMarkers Seurat function was used to identify cluster markers for each cell population; the default Wilcoxon’s rank-sum test was used to calculate statistical significance, and genes were filtered using adjusted P value ≤ 0.05. Pathway enrichment analysis was performed with Metascape, which uses hypergeometric distribution in the calculation of significance for each gene enrichment category. For KEGG pathway analysis, the Database for Annotation, Visualization and Integrated Discovery (DAVID) was used. Study approval For all animal studies, protocols and experimental procedures were previously reviewed and approved by the Institutional Animal Care and Use Committees of UCLA, Northwestern University, and the NIH. These protocols were conducted in accordance with federal regulations as outlined in the Guide for the Care and Use of Laboratory Animals (National Academies Press, 2011). For human studies, samples were collected and processed in accordance with the standards for human research set by the University of Michigan and Northwestern University and in accordance with Declaration of Helsinki principles. Data availability All scRNA-Seq data sets were deposited in the NCBI’s Gene Expression Omnibus database (GEO GSE204803), and values for all data points in graphs are reported in the Supporting Data Values file.

Results Single-cell transcriptomics reveals Notch3 decline as a molecular readout of vascular aging. To evaluate the cellular changes associated with vascular aging, we examined central retinal arteries of mice from 1 month to 24 months of age ( Figure 1, A and B ). The reason to use the retina relates to the stereotypical structure of its vasculature and the ease of identifying the same artery across multiple mice. Importantly, the retina also shares the embryological origin of vessels in the central nervous system. VSMC coverage of main retinal resistance arteries was complete and indistinguishable from 1 month up to nearly 12 months of age ( Figure 1A ). Thereafter and over time, we noted some minor disorganization in the arrangement of VSMCs followed by a precipitous loss of cells ( Figure 1, A and B ). Subsequently, we applied single-cell RNA sequencing (scRNA-Seq) on medium and small brain vessels to identify aging-associated transcriptional changes focusing specifically on VSMCs and pericytes ( Figure 1C ). After data quality control ( Supplemental Figure 1 , A–C; supplemental material available online with this article; https://doi.org/10.1172/JCI166134DS1 ), we clustered cells based on their expression profile using the Pagoda 2 pipeline ( https://cran.r-project.org/web/packages/pagoda2/index.html ) and identified 12 clusters that had approximately the same number of young and aged cells ( Supplemental Figure 1 , D and E). Clusters were further evaluated and annotated based on the expression of genes for cell type identity ( Supplemental Figure 1 , F and G). Vascular cells were identified by expression of Pecam1 and Cdh5 (endothelial cells); Acta2 and Myh11 (smooth muscle cells); and Pdgfrb and Rgs5 (pericytes). In addition, we noted the presence of cortical neurons ( Pcp4 and Enpp2 ) and microglia ( Aif1 and Trem2 ) populations ( Supplemental Figure 1 , F and G). After strict quality controls, young and aged smooth muscle cells and pericytes were mined for transcriptional changes ( Supplemental Tables 1 and 2 ). A heatmap of the top 50 differentially expressed genes highlighted clear changes in genes associated with vascular smooth muscle contractile properties ( Figure 1D , green dots). Importantly, of the 72 genes identified as significantly altered between young and aged VSMCs (adjusted P value < 0.05), 2 members of the Notch signaling pathway, Notch3 and Jagged1 , were within the top downregulated transcripts in aged VSMCs compared with controls ( Figure 1, D–F ). This in combination with downregulation of the downstream Notch targets Heyl and Nrarp ( Figure 1, G and H ) as well as decreased expression of contractile markers ( Mylk and Myh11 ) ( Figure 1, I and J ) suggested that age-related loss of Notch signaling could be the driver of VSMC paucity. Upregulated genes included several heat shock and ER-stress transcripts ( Hspa8 , Hsp90aa1 , Hspb1 , Txnip ) and prostaglandin H 2 ( Ptgds ) ( Figure 1D ) as well as apolipoprotein E and Klf2 ( Apoe , Klf2 ) ( Figure 1, K and L ). Next, we examined expression of NOTCH3 in brain vessels from 27 human subjects ranging from 31 to 87 years of age who died from causes not associated with vascular dementia ( Figure 1M ). Evaluation of 115 small (1–2 VSMC layers) arteries from those patients revealed a consistent decline in nuclear and total NOTCH3 expression ( Figure 1, N and O ). These findings were consistent with the transcriptomics data from mice and pointed to Notch3 as an important molecular readout of aging in small brain vessels. The Notch signaling pathway is essential to vascular morphogenesis, including VSMC recruitment, specification, and differentiation. These functions are mostly performed by Notch1 during development ( 16 – 19 ). However, expression of Notch3 emerges later in fetal life in arterial smooth muscle cells and remains as the predominant Notch receptor throughout adulthood ( 16 ). In fact, under physiological conditions, Notch3 is a marker for arterial mural cells (VSMCs and pericytes) ( 20 ). Surprisingly, inactivation of Notch3 does not have deleterious effects on the viability of mice ( 21 ). Despite its high expression in VSMCs, large elastic arteries are relatively unaffected. In contrast, medium- and small-caliber arteries from Notch3-null mice experience loss of VSMCs in young adults ( 22 , 23 ). Together, our data showing age-dependent reduction in Notch3 and loss of VSMCs in wild-type (WT) mice, and the published findings linking this gene with VSMC coverage, suggested that Notch3 might be a key regulator of VSMC aging. Inactivation of Notch3 results in accelerated VSMC aging with progressive dedifferentiation and detachment of VSMCs that leads to vascular abnormalities first manifested in the brain. To characterize the time-dependent dynamics of VSMC loss in Notch3 –/– animals, retinal VSMC coverage in Notch3 –/– mice was quantified at multiple time points from 2 weeks to more than 104 weeks of age ( Supplemental Figure 2A ). At 2 weeks, arteries from Notch3 -null mice were indistinguishable from control ( Supplemental Figure 2B ). Nonetheless, by 4 weeks and subsequent ages until 2 years old, we found progressive disorganization and detachment of VSMCs from main retinal arteries. Essentially, a 4-week-old Notch3 –/– artery was equivalent to a 2-year-old artery from WT mice in terms of VSMC loss. Interestingly, VSMC detachment was not as pronounced in precapillary branches ( Supplemental Figure 2D ), indicating that it was the resistance arteries where the phenotype was most severely manifested. Brain resistance arteries were equally impacted by loss of Notch3 . Specifically, we found that while large arteries like the carotid showed no apparent abnormalities, the middle cerebral artery and pial arteries showed loss of VSMCs ( Supplemental Figure 3 , A and B) followed by significant (2- to 3-fold) vessel enlargement in older mice ( Supplemental Figure 3 , C and D). We also found that reduction in smooth muscle cell coverage was preceded by decline in VSMC differentiation markers; in particular, calponin ( Cnn1 ) was exquisitely sensitive to inactivation of Notch3 ( Supplemental Figure 3E ). To understand the molecular mechanisms associated with the observed vascular abnormalities in Notch3 -null mice, we performed single-cell transcriptomics on a cohort of n = 8 twelve-month-old (mature) Notch3 –/– and WT littermates ( Figure 2A ). After quality control of the data ( Supplemental Figure 4 , A–H), we focused our evaluation on VSMCs and pericytes ( Figure 2 and Supplemental Tables 1 and 2 ). As anticipated, we were able to collect more WT cells than Notch3-null cells, in keeping with their progressive loss with age; nonetheless, the recovery of over 1,000 VSMCs in the control and over 700 VSMCs in the null mouse enabled a robust analysis ( Figure 2B ). While VSMCs from both genotypes expressed α-smooth muscle actin (αSMA), levels of phospholamban ( Pln ), a gene product that regulates sarcoplasmic reticulum Ca 2+ -ATPase, were lower in Notch3 -null than in WT cells ( Figure 2C ), suggesting differences in contractile function. In fact, a heatmap of the top 50 up- and downregulated genes revealed a signature highlighting loss of contractile properties ( Figure 2D , green dots) and gain of extracellular matrix ( Figure 2D , blue arrowheads). For example, a marked reduction in transcripts for proteins associated with calcium regulation ( Pln , Mylk ), sodium/potassium transport ( Atp1b1 ), and muscle structure ( Myh11 , Sgcd , Utrn ) underscored disrupted contractility and altered VSMC identity. Similarly, upregulation of many matrix transcripts ( Col3a1 , Col8a1 , Lgals1 , Ogn , and Sulf1 ) indicated higher synthesis of matrix proteins with resulting stiffness/fibrosis of the vasculature. Analysis of Gene Ontology categories is consistent with a phenotype that is deficient in supramolecular muscle organization and committed to increase deposition of matrix proteins ( Figure 2E ). It is well established that biological sex contributes to vascular differences in the context of aging, including modulation of smooth muscle cell contractility and vascular stiffness ( 24 ). The initial 12-month scRNA-Seq data set was generated with pooled samples of equivalent female and male mice to avoid sex bias in the identification of Notch3 regulated genes. Using classical X and Y chromosome transcripts, we reidentified sex from the pooled libraries to assess whether sex was a modulator of the Notch3 effect in VSMCs ( Supplemental Figure 5A ). We identified 486 female and 407 male cells in control and 226 female and 179 male cells in Notch3 –/– VSMCs. In addition, we identified a third set of cells, labeled N/A due to lack of clear expression of Xist or any of the Y chromosome markers. N/A cells made up 278 control and 247 Notch3-null cells ( Supplemental Figure 5B ) and were not used in the analysis. Differential expression of the sex-segregated data sets identified 327 (male to male) to 356 (female to female) transcripts between Notch3-null and control VSMCs. Overlay of these 2 data sets showed a strong overlap between sexes with 225 genes as shared between comparisons ( Supplemental Figure 5 , C and D). When these sex-stratified gene signatures were then compared with the top 50 VSMC differentially expressed genes in the original data set, we found up to 99% concurrence between the sex-specific comparisons and joint data, supporting the conclusion that biological sex does not act as a modifier of Notch3 in cerebral vascular VSMCs. Pericytes were also affected by the loss of Notch3 ( Figure 2, F–I ), and their differentially expressed gene signature overlapped (by 69%–75%) with the differentially expressed gene signature from VSMCs ( Figure 2H , yellow stars). This concurrence in transcriptional profiles from 2 different cell types strongly suggested a core gene program that was most likely regulated by Notch3 ( Figure 2H ). Gene Ontology enrichment analysis of Notch3 –/– pericytes revealed alterations in proteins related to glycosaminoglycan metabolism ( Dcn , Ogn , Gpc6 , Aldoc , and Pgam1 ) and wound healing ( Figure 2I ). Given the critical decrease in transcripts associated with contractility and the increase in matrix proteins, we hypothesized that over time Notch3 –/– mice would progressively show structural abnormalities and deficiencies in contractility. To examine structural changes in the brain vasculature, we performed micro-CT on aged littermate WT and Notch3 –/– mice ( Figure 3A ). From these evaluations, a consistent micro-CT signature emerged: Notch3-null aged mice exhibited tortuosity, vascular enlargement, and microaneurysms. Remarkably, enlargement of medium-sized vessels like the middle cerebral artery was only present as the artery ascended on the lateral aspects of the brain, suggesting that it was topologically associated with areas that required increased contractile strength ( Figure 3B , overlapping vasculatures of null mice [red] and WT littermate [white]). In those areas, the vessel showed a 3- to 4-fold increase in volume and increased tortuosity ( Figure 3, C–F , and Supplemental Figure 6 , A and B). Secondary branches of the middle cerebral artery were characterized by beading (dilations and constrictions) along the course of the vessels in aged Notch3-null, but not in WT or heterozygous littermates ( Figure 4, A–G ). The microaneurysms were associated with disorganization or absence of smooth muscle cell coverage ( Figure 4B ). Beading of the vessel and development of microaneurysms were age dependent and were only observed after 6 months of age with progression over time ( Figure 4, C, F, and G ). Importantly, while reductions in VSMC coverage were also noted in other resistance arteries systemically in aged mice ( Supplemental Figure 6C ), the presence of microaneurysms and beaded vessels was confined to the brain. Notch3 is essential to maintain the contractile phenotype of VSMCs. Considering the progressive age-dependent effects in the Notch3-deficient vasculature, we performed 2 additional scRNA-Seq experiments at 1 month and 24 months of age. The objective was to potentially identify direct targets (1 month; Figure 5 ), but also to highlight the downstream compounded effect of aging in the context of Notch3 inactivation (24 months; Figure 6 ). Differential gene expression analysis at 1 month revealed that VSMCs from WT and null mice were already distinguishable ( Figure 5, A–C , Supplemental Tables 1 and 2 , and Supplemental Figure 7 , A–H). A heatmap of the top 25 upregulated and downregulated genes at 1 month uncovered a signature with loss of contractile properties (green dots) and gain of matrix synthesis (blue arrowheads; Figure 5D ). For example, a marked reduction in transcripts for proteins associated with calcium regulation ( Pln , Rcan2 ), sodium/potassium transport ( Atp1b1 , Tesc ), and Rho regulation ( Arhgap29 ) underscored disrupted contractility. Similarly, upregulation of many matrix transcripts ( Col3a1 , Sparc , Col6a1 , Col8a1 , Eln , Col5a2 , Mgp , Col1a1 , Thbs1 ) implied deposition of matrix proteins with resulting stiffness/fibrosis of the vasculature. Transcriptional increase in matrix proteins was further supported by upregulation of matrix metalloproteinase inhibitors that block matrix remodeling ( Timp3 ) and of connective tissue growth factor ( Ctgf ) and Pmepa1 , which also regulate TGF-β signaling (also identified previously; refs. 18 , 19 ). At 1 month, we also noticed the upregulation of 2 thymosins ( Tmsb4x and Tmsb10 ) that bind and sequester actin monomers, as well as an increase in tropomyosin alpha 4 ( Tpm4 ), possibly a compensatory response to rectify deficiencies in contraction. Analysis of Gene Ontology categories is consistent with a phenotype that is deficient in supramolecular fiber organization and committed to increase deposition of matrix proteins ( Figure 5E ). A similar differential gene expression analysis was performed at 24 months to compare WT and Notch3- null littermates ( Figure 6, A–C ; Supplemental Tables 1 and 2 ; and Supplemental Figure 8 , A–H) and elucidate the compounded effect of age and absence of Notch3 . The total cell recovery during isolation of VSMCs in the 2-year-old cohort of null mice was lower than that from control, an expected outcome associated with the progressive loss of VSMCs already discussed. Nonetheless, the findings from this third scRNA-Seq provided further support and rigor for the findings with a sizable number of differentially expressed transcripts overlapping between all three data sets. These findings clarified changes that were associated with aging and those that were more inherent to direct regulation by Notch3 ( Supplemental Figure 9 ). Consistent with the 1-month and 12-month data, at 24 months of age we also found a decrease in transcripts for genes that regulated calcium levels, with some of these overlapping ( Pln , Calm2 ) and others new ( Rrad ). The data also showed consistency in the reduction of sodium/potassium transport transcripts ( Atp1b1 , Tesc ), as well as changes in prostaglandin D 2 production ( Ptgds , Enpp2 ). Much like at 1 month and 12 months, transcripts for some extracellular matrix proteins were also increased in 24-month null animals, including Thbs1 , Mgp , Col3a1 , and Fn1 among the top 25 upregulated genes. In addition, we noted upregulation in proteoglycans, particularly biglycan ( Bgn ) and syndecan 4 ( Sdc4 ); a significant increase in extracellular sulfatase 1 ( Sulf1 ), which regulates sulfatioæ’n of proteoglycans extracellularly; and upregulation of transcripts associated with stress-induced apoptosis ( Sod3 , Uaca , Ndrg1 ) ( Figure 6D ). Gene Ontology enrichment analysis, similar to observations from the 1-month mice, identified deficiencies in supramolecular fiber organization and increases in extracellular matrix, but it also found deficiencies in ATP metabolism that highlight evidence for biological stress in smooth muscle cells ( Figure 6E ). To place the findings in perspective, we combined all 3 data sets and compared WT with Notch3 -null smooth muscle cells over time ( Figure 6, F–K ). In WT cells, Notch3 transcripts decreased with aging, as did the contractility transcripts Myh11 and Map3k7cl ; this decrease was more pronounced in the absence of Notch3 . In contrast, matrix proteins such as elastin, biglycan, and Sulf1 significantly increased upon loss of Notch3 . Findings were also validated at the protein level by immunocytochemistry of a cohort of up- and downregulated genes. We found consistent changes for all 8 of the targets tested in the middle cerebral artery ( Supplemental Figure 10 , A–H). Inactivation of Notch3 also affected pericytes with similar outcomes to smooth muscle cells ( Supplemental Figure 11 , A–C, and Supplemental Table 2 ). Finally, we also found alterations in brain endothelial cells, which likely relate to dysfunctional endothelial cell–mural cell interactions, as endothelial cells do not express Notch3 ( Supplemental Figure 11 , D–F, and Supplemental Table 2 ). Importantly, previous publications have documented alterations in the blood-brain barrier of Notch3 -null young adults ( 23 ). Next, given the apparent direct effect of Notch3 on the expression of key contractile genes such as Mylk , we examined the functional effects of Notch3 loss on VSMC contractility. The molecular changes identified by scRNA-Seq were physiologically tested in collagen contraction assays using smooth muscle cells isolated from 1-month WT and Notch3 –/– littermates ( Figure 7, A and B ). In fact, absence of Notch3 impaired contractility. WT cells were able to contract collagen gels by 43% in 24 hours, in contrast to only 8% contraction by Notch3- null cells for the same time period, representing a 5-fold difference in contractile function. This effect was consistently reproduced with VSMCs isolated from 4 independent mouse cohorts of both sexes ( Figure 7C ). Evaluation of VSMCs in vitro confirmed loss of the contractile phenotype and acquisition of a synthetic phenotype by cells that lacked Notch3 . In fact, while αSMA was present in both WT and null cells, the levels of phospho–myosin light chain 2 (p-MLC2) were significantly reduced in Notch3 –/– cells. Furthermore, the elongated morphology of WT cells contrasted with the polygonal aspect of null cells ( Figure 7D ). Both attributes were consistent with the synthetic phenotype that has been previously ascribed when smooth muscle cells lose contractile properties ( 14 ). Reduction of total and phosphorylated MLC2 was also confirmed on protein lysates from aortae of WT and null mice at 1, 6, and 12 months ( Figure 7E , see supplemental material for full, uncut gels) and further quantified using multiple independent lysates from control and null mice at 1 month of age ( Figure 7F and Supplemental Figure 12 , see supplemental material for full, uncut gels). Physiological assessment of the aged Notch3 –/– animals showed altered cardiovascular parameters including pulse pressure and heart rate, suggestive of attempted chronic compensation for poor vascular reactivity ( Figure 7, G and H ). In vivo hemodynamic responses to cholinergic and adrenergic stimuli were significantly affected in mutant mice ( Figure 7, I and J ). The findings were consistent with an inability of smooth muscle cells to contract and revealed inadequate vasoreactivity. Absence of Notch3 leads to chronic cerebral hypoperfusion, glymphatic dysfunction, and neurodegeneration. The limitations in VSMC contractility were further supported by MRI studies ( Figure 8A ). MRI evaluation showed a reduction in cerebral blood flow across multiple regions in Notch3 –/– mice when compared with controls ( Figure 8, B–D ). Dynamic multi-gradient echo sequence was used to acquire sequential R2* maps between air-gas exchange to assess cerebral oxygenation ( Figure 8E ). These studies confirmed impaired oxygenation of the prefrontal cortex in Notch3 –/– mice, supporting the conclusion that deficient vasoreactivity in Notch3 -null mice contributes to poor oxygenation in comparison with control littermates, leading to chronic hypoperfusion. It has been demonstrated that poor vascular contractility affects the glymphatic system, a perivascular network that subserves a pseudo-lymphatic function and promotes fluid balance and interstitial waste removal ( 25 ). To ascertain the effect of compromised vascular reactivity on the glymphatic system, we first injected fluorescent beads into the cisterna magna and observed their flow in anesthetized mice whose blood vessels had been labeled by a non-blocking anti-PECAM monoclonal antibody ( Figure 9A ). Supporting our previous findings, evaluation of PECAM-labeled arteries of live mice consistently showed dilations in the null mouse and narrow straight arteries in WT littermates ( Figure 9, B–D , and Supplemental Videos 1 and 2 ). Fluorescent beads were found to travel in the perivascular space in both groups; however, the number of beads was notably reduced in null mice ( Figure 9, B, E, and F , and Supplemental Videos 1 and 2 ). The distribution of beads in the perivascular space was also different, showing flow close to the vessel in control and farther from the vessel in the Notch3-null cohort ( Figure 9B , yellow arrows). Of the bead events identified in both cohorts, we observed no significant difference in bead velocity between null and control littermates ( Figure 9E and Supplemental Videos 1 and 2 ). To determine that relatively equal numbers of beads were injected into the cisterna magna, we evaluated the distribution of beads at the base of the brain following euthanasia and dissection ( Figure 9G ). Importantly, the lateral view revealed that in the Notch3 –/– brains, a large proportion of the fluorescent beads were retained in the perivascular space of the middle cerebral arteries, while this was not the case in controls ( Figure 9G , yellow arrows). The pattern of retention was consistent with the tortuosity noted in the middle cerebral arteries as they ascend the lateral aspects of the brain ( Figure 3, B and D ). This retention explained the low number of beads found in the pial arteries ( Figure 9G ) of the Notch3 -null mice but did not explain the distance of the beads from the vessels, which gave the impression that a blockade, perhaps from matrix proteins, was obstructing the flow. While the use of fluorescent beads enabled a crude assessment of flow, a caveat to this approach was the size of the beads (1 μm), which cannot be compared with the flow of small molecules. Thus, we injected FITC-conjugated dextran in the cisterna magna of Notch3 –/– and WT littermates and evaluated the course of drainage into the right ventricle ( Figure 9H ). Initial validation of the procedure in a large cohort ( n = 5–7 mice per time point) indicated a reproducible, time-dependent, and quantifiable detection of fluorescence ( Figure 9I ). Using this approach, we observed a statistically significant reduction in the drainage of FITC dextran in the Notch3 -null mice when compared with WT littermates ( Figure 9J ). Structurally, several consistent alterations were noted in the penetrating arteries of Notch3 -null mice. First, loss of smooth muscle cell coverage and enlargement of luminal inner diameter were regularly found by 2 years of age ( Figure 10, A and B , and Supplemental Figure 13A ). Upon injection of PECAM antibodies, penetrating arteries were easily identifiable in WT mice as daggers infiltrating the brain ( Supplemental Figure 13A ). In contrast, owing to their dilation, penetrating arteries of age-matched Notch3 –/– littermates were indistinguishable from other vessels at lower magnification ( Supplemental Figure 13 , A and B). Furthermore, we observed detachment of astrocytes, providing further support to the alterations in the perivascular space, which was also enlarged ( Figure 10, C and D , and Supplemental Figure 13 , C–F). Given the structural and functional deficiencies in the glymphatic system, we predicted that Notch3 –/– animals may show increased protein accumulation in the brain parenchyma. While we were unable to detect APP or tau, there was an accumulation of chondroitin sulfate–positive material in the brain parenchyma in close proximity to capillaries. Chondroitin sulfate was also noted in intracellularly in capillaries ( Figure 10E ). Using periodic acid–Schiff (PAS) to detect multiple negatively charged glycosaminoglycans and proteoglycans, we found an excess of PAS + granules in Notch3 -null mice when compared with littermates. These granules also contained biglycan, indicating an accumulation of proteoglycans and glycosaminoglycans in the Notch3 –/– mouse brain ( Figure 10, F and G ) consistent with the scRNA-Seq data. Thus, it seems that brain vessel dysfunction impairs the glymphatic system with accumulation of glycosaminoglycans in the parenchyma. This observation was also tested in brain tissue from 23 patients with CADASIL (cerebral autosomal dominant arteriopathy with subcortical infarcts and leukoencephalopathy) and controls ( Supplemental Table 3 ). Our data show a significant increase in the frequency of PAS + granules in CADASIL brains relative to matched controls ( Figure 10, H and I ). We also evaluated the presence of chondroitin sulfate proteoglycans of WT mice at 2 months and 24 months of age. While we did not detect chondroitin sulfate proteoglycans at 2 months, accumulation was noted in the perivascular space by 24 months ( Supplemental Figure 13G ). The levels of such accumulation were more pronounced in the absence of Notch3 by 12 months ( Supplemental Figure 13H ). Finally, scRNA-Seq evaluation of the neuronal constituency of Notch3 –/– at 12 and 24 months revealed progressive alterations in transcripts associated with neuroinflammation and neurodegeneration ( Figure 11, A–F ). Furthermore, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses highlighted multiple neurodegenerative diseases as signatures associated with Notch3 –/– aged mice ( Figure 11, C and F ). Importantly, these transcripts were not identified in young Notch3 –/– mice; instead they were preceded by alterations in metabolism ( Supplemental Figure 13 , I–K) that subsequently evolved into neurodegeneration. These findings underscore the effect of vascular insufficiency on neuronal function and offer a window on how age-dependent vascular dysfunction affects brain health.

Discussion Here, we used single-cell transcriptomics to identify age-associated molecular changes in brain vessels and found that Notch3 and Jagged1 were drastically decreased in aging, implying age-dependent reductions in Notch signaling. Indeed, we verified such reductions in human brain vessels at the protein level. Next, we used a mouse model with deletion of Notch3 to understand the resulting functional implications for the brain vasculature and the downstream consequence to the brain parenchyma. Notch3 is essential for maintaining a differentiated state in smooth muscle cells ( 23 , 26 ), which here we characterized at the single-cell transcriptome level as being associated with regulation of calcium and contractile proteins. Thus, absence of Notch3 leads to a progressive loss of contractile function. Contractility is a critical component of the functional and structural response of the vessel wall during acute or chronic changes in blood pressure. First, we found that defective contractility and loss of VSMCs resulted in structural alterations in the microvasculature, some of these unique to the brain (microaneurysms and vascular beading). Next, we showed a progressive accumulation of extracellular matrix, which altered the perivascular space and reduced astrocyte-vascular association. The combination of impaired vascular contractility and altered structure of the perivascular space manifested as reduced glymphatic flow. Importantly, glymphatic flow was further challenged by the high production of matrix proteins with notable accumulation of glycosaminoglycans in the brain parenchyma. These pathological changes ultimately translated into a neuronal transcriptional response seen in neurodegenerative diseases with decrease in chaperones (like HSP70, as shown by reduction in Hspa1a and Hspa1b transcripts) and increases in ubiquitination and oxidative stress (NDUFA4, COX6C) particularly in a subset of cortical neurons that are susceptible to chronic stress ( 27 ). Overall, our findings provide support for the temporal sequence of events and the molecular connections that link vascular dysfunction to neuronal degeneration and identify Notch3 as a critical culprit in cerebral small vessel disease that emerges with age. Cerebral small vessel disease is an age-dependent disorder that adversely affects brain health ( 28 ). The condition is associated with stroke and vascular dementia and frequently co-occurs with Alzheimer’s disease pathology, where it may result in accelerated and more severe neurodegeneration ( 29 ). It is estimated that over half of the elderly population exhibits radiological evidence of small vessel disease, yet a granular understanding of factors that contribute to this pathology particularly in aging has remained incomplete and its molecular regulation poorly characterized ( 30 ). The brain imaging signatures of sporadic, age-dependent small vessel disease include white matter hyperintensities, lacunar infarcts, microhemorrhages, and dilated perivascular spaces. These findings are also prominently observed in CADASIL, a genetic disorder linked to NOTCH3 missense mutations that result in vascular dementia ( 31 ). Interestingly, several adult patients with heterozygous Notch3 stop codon mutations leading to haploinsufficiency have been diagnosed with cerebral small vessel disease ( 32 ). These and other emerging data indicate that inactivation of a single NOTCH3 allele in humans can result in a late-onset autosomal dominant small vessel disease with incomplete penetrance. Importantly, albeit less frequent, missense biallelic null mutations in NOTCH3 were recently identified in patients and are clinically characterized by migraines, seizures, recurrent strokes starting in early childhood, and progressive cognitive impairment ( 33 – 35 ). Thus, the use of Notch3 -null mice as a proxy to understand the neurovascular consequences of vascular dysfunction is biologically relevant, and the findings presented here might shed light on disease progression. A unifying theme between the null mouse model and the human disease is impaired smooth muscle cell contractility and altered blood flow ( 26 , 36 – 39 ). Our scRNA-Seq data revealed a critical role of Notch3 in the regulation of calcium dynamics and myosin light chain kinase, both absolutely required for contractility. Importantly, we also identified significant accumulation of glycosaminoglycans specifically in the Notch3-deficient model that was also found in patients with CADASIL. Recognizing the caveats of direct extrapolation from experimental models to human disease, and the existing pathological disparity between Notch3 -transgenic mice models and CADASIL patients, these findings provide compelling evidence for neurovascular mechanisms underpinning small vessel disease–related neurodegeneration. An important aspect of this work is the link between aging and progressive reduction in Notch3. We found that in aging, VSMCs significantly and progressively show a decline in Notch3; this finding was reproduced in human brain vessels at the protein level. We then showed molecularly and physiologically that Notch3 is responsible for maintaining vascular contractility. Combined, the results imply that NOTCH3 deficiency in aging underlies impaired vasoreactivity and vascular stiffness. The data also underscore the importance of VSMCs in regulating vascular reactivity, a process fundamental for efficient cerebral autoregulation and particularly vulnerable to age-related risk factors ( 40 – 45 ). It is unclear whether additional changes, such as detachment of astrocytes, might be a direct or indirect consequence of NOTCH3 deficiencies. However, the pathway is primed to coordinate cell-cell communication and function. Importantly, loss of Notch3 has been previously associated with impaired perivascular macrophage recruitment ( 46 ). This outcome negatively affects vascular health, which relies on macrophages, and alters the immune compartment of the brain with unclear consequences during the aging process. Epidemiological and clinical data have established that age-related cerebral small vessel disease, manifested as white matter hyperintensities on brain MRI, is associated with neurovascular dysfunction as well as age-related motor and cognitive decline ( 8 , 12 , 47 – 49 ). The structural and functional changes seen in the perivascular space, which result in impaired clearance via the glymphatic system, provide intriguing evidence for one possible mechanism whereby vascular risk factors may directly result in the accumulation of aberrant proteins (i.e., proteinopathies, and here we include glycosaminoglycans) in age-related neurodegenerative diseases ( 50 – 52 ).

Vascular aging affects multiple organ systems, including the brain, where it can lead to vascular dementia. However, a concrete understanding of how aging specifically affects the brain vasculature, along with molecular readouts, remains vastly incomplete. Here, we demonstrate that aging is associated with a marked decline in Notch3 signaling in both murine and human brain vessels. To clarify the consequences of Notch3 loss in the brain vasculature, we used single-cell transcriptomics and found that Notch3 inactivation alters regulation of calcium and contractile function and promotes a notable increase in extracellular matrix. These alterations adversely impact vascular reactivity, manifesting as dilation, tortuosity, microaneurysms, and decreased cerebral blood flow, as observed by MRI. Combined, these vascular impairments hinder glymphatic flow and result in buildup of glycosaminoglycans within the brain parenchyma. Remarkably, this phenomenon mirrors a key pathological feature found in brains of patients with CADASIL, a hereditary vascular dementia associated with NOTCH3 missense mutations. Additionally, single-cell RNA sequencing of the neuronal compartment in aging Notch3- null mice unveiled patterns reminiscent of those observed in neurodegenerative diseases. These findings offer direct evidence that age-related NOTCH3 deficiencies trigger a progressive decline in vascular function, subsequently affecting glymphatic flow and culminating in neurodegeneration. <p>Notch3 protects against small vessel disease, and age-related declines in Notch3 triggers vascular deficiencies, glymphatic dysfunction, and neurodegeneration in mice.</p>

Author contributions MCR designed and performed experiments and wrote and edited the manuscript. FM performed bioinformatics analysis. RHK, AM, GEH, JS, SM, AB, DPS, DP, SB, EK, AS, JW, and CH performed experiments. KM coordinated patient sample distribution. AMS provided human patient samples and supervised experiments. MMW and FAS provided human patient samples and intellectual input. ETL, EAF, MB, WAM, and BAK provided intellectual discussion. MLIA conceived the study, designed experiments, and wrote and edited the manuscript. All authors had the opportunity to comment on the final manuscript. Supplementary Material

We thank the UCLA Broad Stem Cell Research Center for sequencing scRNA-Seq libraries and the Mouse Histology and Phenotyping Core and the Small Animal Imaging Facility at Northwestern University. A special thanks to Michelle Steel for assistance with husbandry and mouse experimentation. This work was supported by NIH grants R35HL140014 and U01151203; the Leducq Foundation (ReVAMP) (to MLIA); the Northwestern University Molecular and Translational Cardiovascular Training Program (T32HL134633; SP0040691) (to MCR); NIH grant F31HL165767 (to JS); Howard Hughes Medical Institute Gilliam Fellowship (GT11560) (to GEH); and American Heart Association 23POST1022462 (to AM). BAK was supported by NIH DIR HL006247. MMW was supported by grants from the Department of Veterans Affairs (BX003824 and NS099160). The Northwestern Nervous System Tumor Bank is supported by NIH-P50CA221747 SPORE. 11/28/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e166134

oa_package/22/30/PMC10786701.tar.gz

PMC10786702

38015640

Introduction Glycogen storage disease type III (GSDIII) is a rare inborn error of metabolism, with an incidence of 1 in 100,000, caused by mutations in the AGL gene, which encodes for the glycogen debranching enzyme (GDE; also known as amylo-α-1,6-glucosidase) ( 1 , 2 ). GDE acts along with glycogen phosphorylase to degrade glycogen in the cytosol of virtually any cell. In liver and muscle, GDE function is required to maintain glycemia and to sustain fast muscle contraction, respectively ( 1 ). The earliest manifestations of GSDIII, such as fasting hypoglycemia, hepatomegaly, elevation of liver enzymes, and growth retardation, usually appear from early childhood ( 1 , 2 ). A severe myopathy develops during adolescence, with exercise intolerance and muscle weakness followed by loss of ambulation in adult patients ( 2 – 4 ). Histological analysis of muscle biopsies from patients with GSDIII shows glycogen accumulation in large vacuoles, which disrupts the myofibril architecture ( 5 ). Most adult patients also display left ventricle hypertrophy on echocardiography, but only 15% have overt cardiomyopathy ( 2 , 4 ). Although it was long thought that liver involvement improved with age, recent studies report liver fibrosis from an early age with cirrhosis and liver tumor development in adult patients ( 3 , 6 ). To date, no curative treatment is available for patients with GSDIII, and therapeutic options mostly consist of frequent meals, the use of complex carbohydrates such as uncooked cornstarch, and a high-fat high-protein diet ( 1 , 7 ). Dietary management reduces the frequency of hypoglycemia and improves, in some cases, the cardiac phenotype, but has little effect on the myopathy ( 1 , 7 ). Gene therapy using recombinant adeno-associated virus (rAAV) vectors is a promising strategy to treat inherited diseases ( 8 , 9 ) and has already been successfully used in spinal muscular atrophy ( 10 ), hemophilia A and B ( 11 , 12 ), and congenital blindness ( 13 ), with marketing approval in the treatment of these conditions. Multiple clinical trials are currently ongoing for liver or muscle genetic diseases, with encouraging results ( 9 ). One of the major constraints in the use of rAAV for gene transfer is that its encapsidation size is limited to 5 kb, including the inverted terminal repeats (ITRs) ( 8 ). In the setting of GSDIII, the 4.6 kb GDE cDNA represents a technical challenge toward the development of a single vector strategy for GSDIII ( 14 ). Dual AAV vector strategies relying on the presence of homologous recombination sequences or on inteins peptides in each vector to achieve the expression of the full length protein, e.g the full-length GDE, in the cell ( 15 – 17 ). A few years ago, our team published a proof-of-concept of GSDIII correction using a dual overlapping rAAV gene therapy strategy in which 2 vectors were coinjected, each encoding half of the expression cassette ( 15 ). By using 2 distinct dual rAAV vector approaches, we demonstrated complete correction of the muscle and heart phenotype, but only partial correction of the liver disease ( 15 ). An improvement of this approach, based on the use of a tandem liver-muscle promoter ( 18 ) in combination with an immunosuppressive treatment, was recently reported ( 19 ). Despite ongoing optimization ( 16 , 17 ), the dual vector strategy has several limitations: it requires 2 vectors and a consequent 2-times-higher dose, the recombination efficacy could be suboptimal, and the expression of truncated proteins derived from nonrecombined genomes could trigger potential immune toxicities. Another strategy has been developed using a bacterial debranching enzyme (pullulanase), whose 2.2 kb cDNA is short enough to be encapsidated in a single rAAV vector ( 20 ). Although this approach allows initial liver and muscle correction ( 20 ), it led to immune response toward the bacterial protein and subsequent loss of correction ( 21 ). Given the limitations of the existing rAAV gene transfer technology, a strategy based on a single vector expressing a shorter and active form of GDE may represent a valid solution for a safe and efficient rAAV gene therapy for GSDIII in people. In patients with GSDIII, the liver displays liver fibrosis from early childhood ( 2 , 3 , 6 ) and severe, sometimes lethal, liver toxicities have been reported in clinical trials using high-dose rAAVs ( 22 , 23 ). Based on these considerations, we focused on the treatment of heart and muscle impairment while detargeting the liver with a recently developed muscle-tropic rAAV vector ( 24 ). Here, using Agl -KO mouse and rat models, as well as a human cellular model of GSDIII, we demonstrate the possibility of engineering a single vector encoding a mini-GDE enzyme to correct the muscular and cardiac manifestations of the disease, the major disease burden in adults with GSDIII ( 2 , 3 , 7 ).

Methods In vivo studies. The Agl knock-out mice ( Agl –/– , Agl tm1b(EUCOMM)Wtsi ) were previously described ( 15 ) and are of mixed background (C57BL/6J and Balb/c). The Agl knock-out rats ( Agl –/– ) were recently generated using the CRISPR tool and are of Sprague Dawley background. rAAV vectors encoding human GDE were either i.v. administered via the tail vein to 4-month-old male Agl –/– mice and WT littermates ( Agl +/+ ) or intramuscularly in the tibial anterior muscle of 4-month-old female Agl –/– mice and Agl +/+ WT littermates. Intramuscular injection was performed in the tibialis anterior muscle under anesthesia by ketamine and xylazine. rAAV vectors encoding GDE were also administered in the tail vein of 6-week-old male Agl –/– rats and Agl +/+ WT littermates. Finally, rAAV vectors encoding mSEAP or Luciferase were i.v. administered via the tail vein to 6-week-old WT male C57BL/6J mice. All animals were randomized to receive rAAV or PBS as controls. Mice were euthanized by cervical dislocation and rats were sacrificed by anesthetic overdose. Production of rAAV vectors. All rAAV vectors used in this study were produced using an adenovirus-free transient transfection method and purified using a chromatographic method as described earlier ( 49 ). Titers of the rAAV vector stocks were determined using a real-time quantitative PCR using primers for the codon-optimized GDE cDNA (forward: 5′-CTG AAG CTG TGG GAG TTC TT-3′ and reverse: 5′-CTC TTG GTC ACT CTT CTG TTC TC-3′) or for ITRs (forward: 5′-GGA ACC CCT AGT GAT GGA GTT-3′ and reverse: 5′-CGG CCT CAG TGA GCG A-3′), for vectors encoding mSEAP or Luciferase. The mSEAP expression cassette contains the specified promoter, the SV40 intron, the mSEAP cDNA, and an SV40 poly A signal. The Luciferase expression cassette contains the CMV promoter, the luciferase cDNA, and an SV40 poly A signal. The GDE expression cassette contains the mini CMV promoter (corresponding to the nucleotides 175050_175400 of the CMV genome, NC_006273), the human full-length or truncated codon-optimized GDE cDNA ( 15 , 50 ), and either a bGh or a pA58 poly A signal ( 43 ). All cassettes were flanked by the ITRs of AAV serotype 2 for vector packaging. The capsid used (AAV-MT) is a hybrid between AAV9 and AAVrh74, harboring a P1 peptide, as previously described ( 24 ). Analytical ultracentrifugation. Analytical ultracentrifugation measures the sedimentation coefficient of macromolecules by following over time the optical density of a sample subjected to ultracentrifugation. The difference in the sedimentation coefficient, measured by Raleigh interference or 260-nm absorbance, depends on the content of viral genome in the capsid. Analysis was performed using a Proteome Lab XL-I (Beckman Coulter). An aliquot of 400 μL rAAV vector and 400 μL formulation buffer were loaded into a 2-sector velocity cell. Sedimentation velocity centrifugation was performed at 20,000 g and 20°C. Absorbance (260 nm) and Raleigh interference optics were used to simultaneously record the radial concentration as a function of time until the lightest sedimenting component cleared the optical window (approximately 1.5 hours). Absorbance data required the use of extinction coefficients to calculate the molar concentration and the percent value of the empty and genome-containing capsids. Molar concentrations of both genome-containing and empty capsids were calculated using Beer’s law, and percentages of full genome-containing and empty capsids were calculated. Viral genome analysis on agarose gel. DNA was extracted using the High Pure Viral Nucleic Acid Kit (Roche). Purified viral DNA was then loaded on a 1% agarose gel (Eurobio Scientific) stained with SybrSafe Gel Stain (Invitrogen) to visualize the viral DNA. Western blot analysis. Mouse and rat tissues were homogenized with FastPrep lysis tubes (MP Biomedicals) in PBS with cOmplete protease inhibitor cocktail (Roche). Protein concentration was determined using the Pierce BCA Protein Assay (Thermo Fisher Scientific) according to the manufacturer’s instructions. A fraction of 50 μg of total proteins were loaded in each well for both PBS- and rAAV-injected Agl –/– mice and rats and 10 μg of total proteins per well for PBS-injected Agl +/+ mice and rats. SDS-PAGE electrophoresis was performed in a 4%–12% Bis-Tris gradient polyacrylamide gel (NuPAGE, Invitrogen). After transfer, the membrane was blocked with Intercept Blocking buffer (LI-COR Biosciences) and incubated with either an anti-GDE rabbit polyclonal antibody (16582-1-AP, Proteintech for muscles or AS09454, Agrisera for liver; 1:1,000) and an anti-vinculin mouse monoclonal antibody (V9131, Sigma-Aldrich) for tissue lysates and with an anti-Myom3 rabbit polyclonal antibody (17692-1-AP, Proteintech, 1:1,000) for plasma. The membrane was washed and incubated with the appropriate secondary antibody (LI-COR Biosciences, 1:10,000) and visualized by Odyssey imaging system (LI-COR Biosciences). For in vitro experiments, normalization was performed using an anti-actin mouse monoclonal antibody (66009-1-Ig, Proteintech, 1:1,000). When the number of samples required the use of two gels, both were processed in parallel, including running and transfer within the same tank, incubation with the same antibody solution, and visualization at the same time by the Odyssey imaging system. HEK-293T cells transfection. HEK 293T cells were transfected in 6-well plates using the Opti-MEM medium and Lipofectamine 3000 transfection reagent (Thermo Fisher Scientific) according to the manufacturer’s instructions. Cells were harvested 48 hours after transfection and lysed in PBS with 1% Triton X-100. Supernatants were collected after centrifugation at 11,000 g and 50 μg of total proteins were used for Western blots. Measurement of glycogen content in tissues. Glycogen content was measured indirectly in tissue homogenates as the glucose released after total digestion with Aspergillus niger amyloglucosidase (Sigma-Aldrich). Samples were incubated for 10 minutes at 95°C and then cooled at 4°C. After the addition of amyloglucosidase (final concentration 4 U/mL) and potassium acetate pH 5.5 (final concentration 25 mM) at 37°C for 90 minutes, the reaction was stopped by incubating samples for 10 minutes at 95°C. A control reaction without amyloglucosidase was prepared for each sample and was incubated in the same conditions. The glucose released was determined using a glucose assay kit (Sigma-Aldrich) and the resulting absorbance was acquired on an EnSpire Alpha plate reader (PerkinElmer) at a wavelength of 540 nm. Glucose released after amyloglucosidase was then normalized by the total protein concentration. Measurement of glycemia and plasma aspartate aminotransferase and alanine aminotransferase levels. Blood samples were taken from mice anesthetized with isoflurane. Glycemia was measured using a glucometer (Accu-Chek). Aspartate and alanine aminotransferase levels were measured on plasma using micro-chip DRI-CHEM SLIDE (Fujifilm, AST-P III, ALT-P III) and DRI-CHEM NX500 spectrophotometer (Fujifilm) following the manufacturer’s instructions. Measurement of cardiac troponin and NT-pro-BNP levels in rat plasma. Quantification was performed following the manufacturer’s instructions at the minimal possible plasma dilution (i.e., 1:2), using the Meso Scale Discovery (MSD) ELISA Rat Cardiac Injury Panel 2 Kit (ref K15155) and Rat NT-pro-BNP Assay Kit (ref K153JKD). Histology. Heart, triceps brachii, quadriceps femoris, soleus, EDL, and liver were snap frozen in isopentane previously chilled in liquid nitrogen. Serial 8-μm cross sections were cut in a Leica CM3050 S cryostat (Leica Biosystems). To minimize sampling error, 2 sections of each specimen were obtained and stained with HPS, PAS, and/or Sirius Red (SR) according to standard procedures. Images were digitalized using Axioscan Z1 slide scanner (Zeiss) under a Zeiss Plan-Apochromat 10X/0.45 M27 dry objective (Zeiss). Tile scan images were reconstructed with ZEN software (Zeiss). Quantification of images were processed using QuPath 0.4.3 Software ( 51 ). For the PAS staining, a first pixel classifier was trained on different types of muscle slices for detecting tissue and eliminating artefacts such as folding, bubbles, and tearings. For mice, this PAS contour pixel classifier is a Random Tree with a resolution of 14.08 μm/pixel, includes 3 channels, 5 scales (0.5, 1, 2, 4, and 8), 8 features (Gaussian, Laplacian of Gaussian, weighted deviation, gradient magnitude, structure tensor max eigenvalues, structure tensor middle eigenvalues, structure tensor min eigenvalues, and structure tensor coherence), and no local normalization. The quantification of PAS staining was performed using a pixel classifier trained on healthy tissue and impaired tissue. The output parameter is then the ratio of the surface area of impaired tissue over the surface area of the total tissue slice (obtained by the contour pixel classifier). The PAS quantification classifier is an Artificial Neural Network with a resolution of 1.76 μm/pixel (for rats) or 3.51 μm/pixel (for mice), includes 3 channels, 4 scales (0.5, 1, 2, and 4), 6 features (Gaussian, Laplacian of Gaussian, weighted deviation, gradient magnitude, structure tensor max eigenvalues, and structure tensor middle eigenvalues), and no local normalization. For the HPS staining, a pixel classifier was trained on different types of muscle slices for detecting tissue and eliminating artefacts such as folding, bubbles, and tearings. This HPS contour pixel classifier is a Random Tree with a resolution of 14.08 μm/pixel, includes 3 channels, 3 scales (2, 4, 8), 4 features (Gaussian, Laplacian of Gaussian, weighted deviation, and gradient magnitude), and no local normalization. The quantification was performed using a pixel classifier trained on healthy tissue and impaired tissue. The output parameter is then the ratio of the surface area of impaired tissue over the surface area of the total tissue slice (obtained by the contour pixel classifier). The HPS quantification classifier is an Artificial Neural Network with a resolution of 7.03 μm/pixel, includes 3 channels, 5 scales (0.5, 1, 2, 4, and 8), 5 features (Gaussian, Laplacian of Gaussian, weighted deviation, gradient magnitude, and structure tensor max eigenvalues), and no local normalization. For the quantification of fibrosis, the pixel classification feature from QuPath ( 51 ) 0.4.3 was used by creating 2 classifiers, using each time 2 images for ground truth. The first pixel classifier identifies pixels belonging to the tissue slice, excluding veins, fold, dust, bubbles, or any artifact encountered, and draw an annotation of the analyzable tissue. The second classifier was trained to identify fibrosis and healthy tissue, based on manual annotation of picrosirius red staining. The fibrosis classifier was then applied in the annotation created by the first classifier. The total surface of SR staining was divided by the total surface area of the muscle slice resulting then in a fibrosis ratio (% of fibrotic tissue) for each tissue slice. Vector genome copy number determination. Vector genome copy number was determined using a quantitative PCR assay as previously described ( 27 ). The PCR primers used in the reaction were located in the glucosyltransferase domain of the full-length and truncated codon-optimized GDE (forward: 5′-CTG AAG CTG TGG GAG TTC TT-3′ and reverse: 5′-CTC TTG GTC ACT CTT CTG TTC TC-3′) or in the ITRs (forward: 5′-GGA ACC CCT AGT GAT GGA GTT-3′ and reverse: 5′-CGG CCT CAG TGA GCG A-3′) for mSEAP expressing cassettes. As an internal control, primers located within the mouse (forward: 5′-AAA ACG AGC AGT GAC GTG AGC-3′ and reverse: 5′-TTC AGT CAT GCT GCT AGC GC-3′) or rat (forward: 5′-AAA ACG AGC GGT GAC ATG AGC-3′ and reverse: 5′-TTC AGT CAT GCT AGC GCT CC-3′D) Titin gene were used. RNA expression analysis. Total RNAs were extracted from cell lysates using Trizol (Thermo Fisher Scientific) and the RNeasy Mini Kit (Qiagen). DNA contaminants were removed using the Free DNA kit (Thermo Fisher Scientific). Total RNAs were reverse transcribed using random hexamers and the RevertAid H minus first strand cDNA synthesis kit (Thermo Fisher Scientific). Quantitative PCR was performed with oligonucleotides specific for the codon-optimized GDE transgene (forward: 5′-CTG AAG CTG TGG GAG TTC TT-3′ and reverse: 5′-CTC TTG GTC ACT CTT CTG TTC TC-3′) and normalized by the levels of expression of the P0 ribosomal protein Rplp0 mRNA (forward: 5′-CTC CAA GCA GAT GCA GCA GA-3′; reverse: 5′-ATA GCC TTG CGC ATC ATG GT-3′). mSEAP quantification. mSEAP was quantified in tissue lysates using the Phospha-Light Kit (Applied Biosystems) following the manufacturer’s instructions and was normalized by the total protein concentration measured using the Pierce BCA Protein Assay (Thermo Fisher Scientific). Luciferase quantification. Snap-frozen tissues were homogenized in PBS with FastPrep lysis tubes (MP Biomedicals), followed by centrifugation at 10,000 g for 10 minutes. Supernatants were collected and diluted in lysis buffer (1 mM DTT, 25 mM Tris/base, 1 mM EDTA, 8 mM MgCl 2 , 15% glycerol, and 0.4% Triton [Sigma Aldrich]) in a white opaque 96-well plate (PerkinElmer). Luciferase activity was measured using EnSpire (PerkinElmer) through sequential injections of assay buffer (1 mM DTT, 25 mM Tris/base, 1 mM EDTA, 8 mM MgCl 2 , 15% glycerol, and 2 mM ATP [Sigma Aldrich]) and luciferine (Interchim). Luciferase relative luminescence unit was normalized by the total protein concentration measured using the Pierce BCA Protein Assay (Thermo Fisher Scientific). Muscle function. A forelimb wire-hang test was performed as already reported ( 52 , 53 ) at baseline and each month until euthanasia. A 4-mm-thick wire was used to record the number of falls over a period of 3 minutes. The average number of falls per minute was reported for each animal. Grip strength and Rotarod tests were performed as previously reported ( 15 ). Transduction of hiPSC-derived skeletal muscle cells. The GSDIII CRISPR hiPSCs have been previously generated, using CRISPR knock down of the AGL gene ( 42 ). Control hiPSCs were the isogenic cell line (CTRL1). GSDIII CRISPR and CTRL1 hiPSCs were differentiated into skMb, as previously described ( 42 ). After expansion in 96-well plate, hiPSC-derived skMb were transduced with LK03-rAAV vectors encoding either GFP or ΔNter2-GDE under the control of the miniCMV promoter, at a multiplicity of infection (MOI) of either 75,000 or 15,000 for 72 hours. Then, hiPSC-derived skMb were differentiated into skMt, as previously described ( 42 ). Measurement of Glycogen Content in skMt. After 4 days of differentiation into skMt, hiPSC-derived skMt were starved for 3 days in a no-glucose DMEM medium with 10% FBS (Thermo Fisher Scientific) in order to induce glycogen degradation in CTRL1 skMt as previously described ( 42 ). Cells were lysed using HCl 0.3M and Tris 450 mM pH 8.0. Glycogen was then quantified using the Glycogen-Glo assay kit (Promega) and normalized using the CellTiter-Glo Luminescent Cell Viability Assay (Promega). Immunostaining assay. SkMt derived from hiPSCs were fixed with 4% paraformaldehyde (Euromedex) for 10 minutes at room temperature. After 2 washes in PBS, cells were permeabilized with 0.5% Triton X-100 for 5 minutes and blocked in PBS solution supplemented with 1% BSA (Sigma-Aldrich) for 1 hour at room temperature. SkMt were stained for specific skeletal myogenic markers overnight at 4°C using primary antibodies (Desmin, ref AF3844 R&D 1:200; MHC/MF20, ref 3ea DSHB 1:50; Titin ref T5650 US Biological 1:50). After 3 washes in PBS, staining was revealed by appropriate Alexa Fluor secondary antibodies 1:1,000 (Donkey anti-goat AF488, ref A11055 Invitrogen 1:1,000; Donkey anti-mouse AF488, ref A21202 Invitrogen 1:1,000) in the dark for 1 hour at room temperature, and nuclei were visualized with Hoechst solution 1:3,000 (Invitrogen). Cell imaging was carried out with a Zen Black software-associated LSM-800 confocal microscope (Zeiss) with a 20× objective. PAS staining on skMt. PAS staining on hiPSC-derived skMt was performed with the PAS Staining Kit (Sigma-Aldrich) following the manufacturer’s instructions. Briefly, cells were fixed with 4% paraformaldehyde for 10 minutes at room temperature. After 2 washes in PBS, cells were treated with PAS for 5 minutes at room temperature. After 3 washes in distilled water, cells were treated with Schiff’s reagent for 15 minutes at room temperature. Finally, after 4 washes in tap water, staining was visualized using an EVOS XL Core microscope (Invitrogen). Images were processed and analyzed using FIJI custom-made scripts ( 54 ). First, colors were split and only the green channel was kept as it was the most contrasted channel. Images were manually thresholded into binary images where PAS signal was black and background white. The threshold was set for maximizing the difference between genotypes, and, once calculated, the same threshold was applied to all images to quantify. The quantification of PAS staining was obtained using this formula: Area of PAS staining / Total area of image × 100, giving a percentage of PAS staining within the image. Molecular modelling. A three-dimensional model of the full-length human GDE was retrieved from the AlphaFold2-database ( 30 ). The 3D model of ΔNter2-GDE was predicted using AlphaFold2 v.2.1 ( 29 ) and the default parameters. Molecular dynamics simulations were performed using Gromacs 2021.3 ( 55 ) with CHARMM36 forcefield ( 56 ). Solvation was done using explicit TIP3 water model. A cut-off of 1.2 nm and a switch function from 1.0 to 1.2 nm were used for short-range electrostatic and van der Waals interactions, respectively. Long-range electrostatic interactions were treated with particle mesh Ewald (parameters by default) and periodic boundary conditions. The following protocol was used: (a) minimization of the system with 50,000 steps of steepest descent algorithm, (b) 100 ps of NVT equilibration, (c) 100 ps of NPT equilibration, (d) 250 ns production phase without constraints and a timestep of 2 fs. molecular dynamics simulations were run at 303 K in triplicates. To analyze molecular dynamics trajectories, we used the molecular dynamics analysis packages ( 57 ). The RMSD was computed using the Cα atoms along the simulation, RMSF per amino acid residue were calculated using the residue Cα atoms along the simulation and the inter-domain distances were measured between the center of mass of the different domains. Structural images were prepared using Pymol2.5 (Schrodinger Inc.). Statistics. All the data shown in this paper are reported as mean ± SEM. The GraphPad Prism software was used for statistical analysis. P values < 0.05 were considered significant. For all the data sets, data were analyzed by parametric tests, α = 0.05 (1-way and 2-way ANOVA with Tukey’s post hoc correction). The statistical analysis performed for each data set is indicated in figure legend. Study approval. All animal studies were performed according to the French and European legislation on animal care and experimentation (2010/63/EU) and approved by the local institutional ethical committee, Comité d’éthique de Genopole en expérimentation animale (CEGEA) (CEEA - 051 [Evry, France]). Data availability. A Supporting Data Values file with all reported data values is available as part of the supplemental material.

Results Generation of a single-vector approach for GSDIII based on a functional, truncated GDE. Our team previously demonstrated the possibility to correct the muscle and heart phenotype of a GSDIII mouse model ( Agl –/– mice) using a dual vector approach harboring the cytomegalovirus (CMV) promoter ( 15 ). A GDE expression cassette containing the human full-length GDE cDNA and the CMV promoter used in the previous study is close to 6 kb, largely beyond the packaging size limitation of a single rAAV vector. As a first step toward the optimization of the gene therapy approach for GSDIII, we carefully optimized each component of the transgene expression cassette to reduce their size at minimum. First, we compared in vivo the strength of promoters with proven expression in muscle and sizes spanning 334 to 1,050 bp such as miniCMV, SPc5-12, tMCK, eSyn, Desmin, or CK6 ( 25 , 26 ) ( Supplemental Figure 1A ; supplemental material available online with this article; https://doi.org/10.1172/JCI172018DS1 ). Mice were i.v. injected with rAAV9 encoding the reporter gene mouse secreted embryonic alkaline phosphatase (mSEAP) under the control of each promoter, and mSEAP expression was assessed in 2 skeletal muscles. MiniCMV, one of the shortest promoters (351 bp), showed the highest mSEAP expression with similar vector genome copy numbers (VGCN) among all tested promoters ( Supplemental Figure 1 , B and C) in triceps and quadriceps. Muscle targeting using rAAV usually requires high doses and may expose patients to liver overload and potential toxicities ( 22 , 23 ). Our team has already reported the generation of a potent muscle-targeting and liver-detargeting rAAV chimeric capsid referred to as rAAV-MT ( 24 ). A biodistribution study performed a month after i.v. injection in mice confirmed that rAAV-MT enabled higher expression of the reporter gene in skeletal muscles as well as a pronounced reduction of expression in liver ( Supplemental Figure 2 , A and B). Introns are known to increase the stability of mRNA, the export to the nucleus, and, ultimately, enhance protein production ( 27 ). To understand the requirement of the intron in our expression cassette, we assessed the ability of 2 similar, oversized expression cassettes with or without the SV40 intron. For this and the following experiments to comparatively evaluate the efficacy of the truncated enzymes, intramuscular injections were used ( Supplemental Figure 3A ). Intramuscular vector administration in Agl –/– mice represents a fast method to evaluate the activity of GDE toward its native substrate in a diseased context, in the absence of a reliable and robust in vitro test to evaluate enzymatic activity. Following intramuscular injection, similar vector transduction and GDE expression, as measured by VGCN and Western blot, resulted in a significant decrease of glycogen accumulation in the injected muscle of mice treated with both vectors ( Supplemental Figure 3 , B–E), suggesting that the presence of an intron is not necessary to achieve a high level of GDE expression. Without the intron, the size of the expression cassette enables the generation of oversized rAAV vectors to assess efficacy in our Agl –/– mouse model after i.v. injection. We then tested 2 single rAAV vectors, containing either a poly adenylation (poly A) signal of 58 base pairs (pA58) or of 169 (bGh) base pairs, encapsidated in rAAV-MT ( Supplemental Figure 4A ). Three months after injection, muscle transduction and GDE expression, respectively measured by VGCN and Western blot, were similar in all evaluated muscles, including heart ( Supplemental Figure 4 , B–D). The low levels of VGCN observed in muscle likely reflected the use of an oversized AAV vector cassette with suboptimal packaging. Similar but partial efficacy was observed with the 2 expression cassettes, with 50% to 60% reduction of glycogen measured in heart, triceps, quadriceps, soleus, and extensor digitorum longus (EDL) muscles ( Supplemental Figure 5A and Supplemental Table 1 ). In line with partial reduction of tissue glycogen, histological analysis revealed a limited normalization of the muscle histology ( Supplemental Figure 5B ). Taken together, these results allowed us to select the minimal regulatory sequences necessary to efficiently express GDE with a single vector. However, the optimized transgene expression cassette including the native GDE cDNA had a size of 5.3 kb and showed partial efficacy in Agl –/– mice, thus supporting efforts to reduce the size of the GDE transgene. Although no three-dimensional structure of human GDE has been experimentally determined so far, the structure of a GDE homolog from Candida glabrata (PDB id: 5D06 and 5D0F), showing 38% of sequence identity with the human GDE, was solved using X-ray diffraction ( 28 ). With the emergence of AlphaFold2, an artificial intelligence system developed by DeepMind ( 29 ), high quality models are now available, notably via the AlphaFold2 database ( 30 ), where a model of human GDE can be found (entry AF-P35573-F1, Supplemental Figure 6 , A and B). This model was retrieved from the database and compared with the crystallographic structure of C . glabrata GDE, enabling the identification in human GDE of the different domains described earlier for C . glabrata ( 28 ), namely M1, GT (corresponding to the assembly of 3 subdomains: A, B and C), M2, and GC ( Supplemental Figure 6 , C). The mapping of the missense mutations described in GSDIII patient onto the 3D model revealed the presence of most mutations in the catalytic GT and GC domains ( 2 , 5 , 6 , 28 , 31 – 38 ), while almost no mutations were reported in the M1 and C domains ( Figure 1A ). Interestingly, the crystallographic structure of inactive C . glabrata GDE in complex with maltooligosaccharides did not reveal the presence of any sugar binding site in these 2 domains (adapted from PDB id: 5D06 and 5D0F, Supplemental Figure 6B ). These considerations support the assumption that these 2 domains may not play a relevant role in catalysis or ligand binding and could be promising targets to shorten the human GDE protein while retaining activity. We first generated 4 GDE mutants with deletion in the C-domain, named ΔC1 to ΔC4, by selecting regions that were less likely to disturb formation of secondary structures, in particular the β-strands mainly composing the C-domain ( Supplemental Figure 6D and Supplemental Figure 7A ). All expression cassettes containing the mutants were around 5 kb or lower ( Figure 1B ). Transfection in HEK-293T cells showed that all 4 ΔC mutants were expressed at similar levels ( Supplemental Figure 7B ). Treatment of Agl –/– mice with rAAV vectors encoding the ΔC mutants showed truncated GDE expression and similar VGCN in heart and skeletal muscle of all injected mice ( Figure 1C and Supplemental Figure 7 , C–E). Importantly, a 50% glycogen reduction was obtained in triceps and quadriceps, suggesting residual activity of the ΔC truncated GDE enzymes, with higher correction achieved in mice treated with rAAV-ΔC1 ( Figure 1D , Supplemental Figure 7F , and Supplemental Table 2 ). Unexpectedly, no correction of glycogen accumulation was observed in the heart of injected Agl –/– mice ( Figure 1D ), despite robust tissue transduction and GDE expression. Consistently, a slight improvement of muscle strength assessed by the wire-hang test — a functional test predictive of rAAV gene therapy efficacy ( 15 ) — was observed only in mice treated with ΔC1-expressing vector ( Figure 1E ). These results suggest that, although C-domain–truncated GDE mutants retain some activity, the deletions may affect GDE function in an organ-specific manner, possibly by altering its regulation in the heart. We then generated deletion mutants in the M1 domain, named ΔNter1 to ΔNter6, by selecting regions less likely to disturb the formation of the β-strands composing this domain ( Supplemental Figure 8A ). After transfection of HEK-293T cells, we observed expression of all the truncated proteins, although ΔNter1, ΔNter2 and ΔNter4 displayed the higher expression, comparable to the ΔC1 mutant ( Supplemental Figure 8B ). We then evaluated the efficacy of the ΔNter1 to ΔNter6 truncated proteins in Agl –/– mice by intramuscular injection in the tibialis anterior muscle, compared with the ΔC1 mutant and the full-length GDE ( Figure 2A ). Western blot analysis performed in the injected muscle a month after vector injection revealed lower protein expression in all mutants compared with the full-length GDE despite a similar VGCN among the groups ( Figure 2, B and C and Supplemental Figure 8C ). Of note, the site of truncation had a striking impact on the protein expression level, since ΔNter2-GDE exhibited the highest protein expression, while, in line with our in vitro data, ΔNter5 and ΔNter6 mutants had very low expression ( Figure 2, B and C ). Importantly, all truncated enzymes displayed some degree of activity. This was reflected by the glycogen reduction in the injected muscle with ΔNter2-GDE showing the highest efficacy, comparable to that of full-length GDE ( Figure 2D ). We hypothesized that variations in protein expression could be related to differences in protein stability. Indeed, the RNA levels of GDE in the tibialis anterior was similar in mice treated with rAAV encoding either ΔNter2 or full-length GDE, suggesting that the lower ΔNter2 protein levels were likely due to reduced protein stability ( Supplemental Figure 8D ). Therefore, to further investigate the impact of the ΔNter2 truncation on protein stability, we performed molecular dynamics simulations (250 ns) for both the full-length GDE and the ΔNter2-GDE. Analysis of the conformational variations along the simulation time, monitored by the root mean square deviation (RMSD), revealed a slightly higher flexibility of the truncated protein compared with the parental protein ( Supplemental Figure 9A ). More detailed inspection of the amino acid fluctuations over the simulation time, described by the root mean square fluctuation (RMSF), showed main differences in the flexibility of the B and GC domains ( Figure 2E and Supplemental Figure 9B ). Monitoring of the distances between domains along the simulation ( Supplemental Figure 9C ) suggested that fluctuations did not appear to be directly related to conformational changes inside the domains but rather due to a relative reorientation of the domains. Indeed, distance fluctuations between B and GC domains appeared larger in ΔNter2-GDE than in the full-length protein ( Supplemental Figure 9C ), likely resulting from A and M2 domains moving apart due to the absence of M1 domain in ΔNter2-GDE. Indeed, the M1 domain is known to maintain the structural integrity of the middle region composed of A, C, and M2 domains ( 28 ). Despite of a slightly larger flexibility observed for ΔNter2-GDE compared with the full-length GDE, the structural integrity of the mutant was maintained, and catalytic sites of GC and GT were not affected, indicating that ΔNter2-GDE could be a good candidate for gene therapy. Oversized rAAV vector production has a large impact on the yields and the quality of the final product, which are critical parameters for the translation to the clinic of rAAV-based gene therapies. To evaluate the impact of the use of the truncated ΔNter2-GDE on the production yields and the quality of the vector, 3 distinct oversized expression cassettes expressing human full-length GDE were compared with the 5 kb expression cassette bearing the ΔNter2-GDE ( Supplemental Figure 10A ). The use of the ΔNter2-GDE cDNA allowed for approximately 10-fold increase of rAAV production yields ( Supplemental Figure 10B ) while allowing for the efficient encapsidation of a nontruncated genome ( Supplemental Figure 10C ). Improved yields and genome quality resulted in a dramatically improved full-to-empty particles ratios, as measured by analytical ultracentrifugation, with up to 37% of full particles measured in rAAV-ΔNter2-GDE ( Supplemental Figure 10D ). Taken together, these data indicate that, while ΔNter2-GDE has an in vivo efficacy similar to the full-size enzyme, it allows for production of rAAV vectors with higher yields and quality, thus providing a potential gene therapy candidate for the treatment of GSDIII. rAAV encoding ΔNter2-GDE rescues the cardiac and muscle phenotype of Agl –/– mice. Next, we evaluated the efficacy of an rAAV-MT vector expressing the ΔNter2-GDE in Agl –/– mice via tail vein injection at a dose of 1 × 10 14 vg/kg ( Figure 3A ). Agl –/– mice were treated at 4 months of age when they showed extensive glycogen accumulation in all muscles and functional impairment as measured by wire-hang ( 15 ). Three months after vector injection, VGCN and GDE proteins were detected in all the analyzed muscle tissues ( Figure 3B and Supplemental Figure 11 , A–C). Glycogen quantification showed almost normalized glycogen levels in the heart and in different skeletal muscles ( Figure 3, C and D , Supplemental Figure 11D , and Supplemental Table 3 ). In line with these data, histological analysis performed on the same tissues showed complete normalization of the muscle architecture on hematoxylin phloxine saffron (HPS) staining as well as an important glycogen reduction by periodic acid-schiff (PAS) staining ( Figure 3, E and F and Supplemental Figure 12 , A–C). We also evaluated by Western blot the levels of myomesin3 (Myom3) fragments, a known biomarker of muscle dystrophies ( 39 ). Although GSDIII is not a muscular dystrophy, we found that Myom3 was elevated in the plasma of Agl –/– mice at baseline. It normalized 3 months after treatment in mice injected with rAAV encoding for ΔNter2-GDE ( Figure 3G and Supplemental Figure 12 , D and E). Importantly, treatment with rAAV-ΔNter2-GDE rescued muscle strength assessed by the wire hang test. Before injection, Agl –/– animals exhibited a high frequency of falls (26.0 falls/min ± 1.4), compared with Agl +/+ mice (3.7 falls/min ± 0.8) ( Figure 3H ). Three months after injection, rAAV-ΔNter2-GDE–injected mice showed less frequency of falls compared with the PBS-injected group (9.8 falls/min ± 1.8 versus 31.4 falls/min ± 1.4), almost reaching WT levels ( Figure 3H ). To confirm these results, muscle strength was also evaluated 5 months after injection of rAAV-ΔNter2 in another cohort of Agl –/– mice, using an extended set of functional evaluations ( Supplemental Figure 13A ). Improvement of muscle strength was observed in rAAV-ΔNter2-treated animals by both the wire hang and the grip strength tests ( Supplemental Figure 13 , B and C). However, the rotarod test performances were not impaired in Agl –/– mice and remained unchanged in treated animals ( Supplemental Figure 13D ). Finally, we confirmed the strong liver detargeting by analyzing the liver of Agl –/– mice treated with rAAV-ΔNter2-GDE. Considering the dose used, low VGCN were observed in the liver of Agl –/– mice 3 months after injection, with no detectable GDE expression ( Supplemental Figure 14 , A and B). Therefore, no correction of the liver weight, accumulation of glycogen, glycemia, or of the aspartate and alanine aminotransferases levels in plasma was observed ( Supplemental Figure 14 , C-F). Liver histology was similar between untreated and treated Agl –/– mice, with similar levels of fibrosis ( Supplemental Figure 14 , G and H). These data clearly demonstrate that treatment with rAAV-ΔNter2-GDE reverses the muscle impairment in adult, symptomatic GSDIII mice at both biochemical and functional levels, in the absence of liver targeting. rAAV encoding ΔNter2-GDE corrects glycogen accumulation in an Agl –/– rat model of GSDIII. To confirm the efficacy of the optimized rAAV-ΔNter2-GDE in a larger animal model, we administered the vector to 6-week-old Agl –/– rats. Agl –/– rats, treated with 1 × 10 14 vg/kg of rAAV-ΔNter2-GDE via tail vein injection, were analyzed 3 months after injection ( Figure 4A ). Vector genomes and ΔNter2-GDE proteins were detected in all evaluated muscles ( Figure 4B and Supplemental Figure 15 , A–C). Treatment with rAAV-ΔNter2-GDE reduced glycogen content in different skeletal muscles to levels close to those measured in Agl +/+ rats. ( Figure 4C , Supplemental Figure 15 , D–F, and Supplemental Table 4 ). A significant, 50%, glycogen reduction was observed in the heart ( Figure 4D ), which was even more evident in the PAS staining ( Figure 4E ). Histological analysis also showed normalization of muscle architecture, as well as an important glycogen reduction in all skeletal muscles tested ( Figure 4F , Supplemental Figure 15 , G–I, and Supplemental Figure 16 , A and B). Analysis of the Myom3 levels in 6-week-old Agl –/– rats showed moderate elevation compared with age-matched controls ( Supplemental Figure 16 , C and D). Importantly, Agl –/– rats treated with rAAV-ΔNter2-GDE showed reduced Myom3 levels, similar to Agl +/+ rats, 3 months after vector injection ( Figure 4, G and H ). Acute cardiac toxicity has already been reported following high-dose rAAV-based gene therapy targeting the heart in animal models and in patients ( 40 , 41 ). No death, weight loss, or clinical signs of acute toxicity were reported in treated rats. Quantification of plasma cardiac troponin I and T (cTnI and cTnT) and of the N-terminal fragment of the probrain natriuretic peptide (NT-pro-BNP) levels, clinically relevant markers of cardiac damage and heart failure, respectively, did not reveal any difference between PBS-injected and rAAV-ΔNter2-GDe–injected animals 1 month and 3 months after injection ( Supplemental Figure 16E ). In conclusion, the data obtained by the treatment of Agl –/– rats with an rAAV vector encoding for ΔNter2-GDE confirm those obtained in the mouse model of the disease and further support the clinical translation of rAAV-ΔNter2-GDE to treat the muscle disease in patients with GSDIII. ΔNter2-GDE reduces glycogen accumulation in a human skeletal muscle cell model of GSDIII. To evaluate the activity of the truncated ΔNter2-GDE in a human pathological context, we took advantage of a recently reported in vitro human skeletal muscle model of GSDIII derived from human induced pluripotent stem cells (hiPSCs) edited by CRISPR/Cas9 technology (GSDIII CRISPR ) ( 42 ). Skeletal muscle cells derived from the isogenic control hiPSC line (CTRL1) were used as an unaffected control ( 42 ). We produced an rAAV vector expressing ΔNter2-GDE or GFP suitable for in vitro use and we treated GSDIII CRISPR and CTRL1 hiPSC-derived skeletal myoblasts (skMb) following a recently reported protocol ( 42 ) ( Figure 5A ). After transduction with rAAV, skMb were differentiated into skeletal myotubes (skMt). Transduction with rAAV expressing GFP or ΔNter2-GDE did not alter the differentiation of skMb, that showed similar expression of skeletal myogenic markers by immunostaining analysis ( Figure 5B and Supplemental Figure 17A ). Cell viability was reduced in GSDIII CRISPR skMt compared with CTRL1 skMt and was similar between GFP- and ΔNter2-GDetransduced cells ( Supplemental Figure 17B ). GSDIII CRISPR skMt transduced with rAAV expressing ΔNter2-GDE exhibited a significant reduction of glycogen content when compared with GSDIII CRISPR skMt transduced with the control vector ( Figure 5C ). PAS staining performed on transduced-skMt confirmed these results ( Figure 5, D and E ). These data demonstrate that the truncated ΔNter2-GDE reduced glycogen accumulation not only in mouse and rat muscles, but also in a human skeletal muscle model of GSDIII.

Discussion In GSDIII, inactivating mutations on the AGL gene result in the absence of a functional GDE and accumulation of limit dextrin in liver, heart, and skeletal muscles. The hepatic and metabolic phenotypes (hepatomegaly and hypoglycemia) appear early, in patients of around 1 year of age ( 2 ). Moreover, recent publications report that children exhibit liver fibrosis as early as 1 year of age ( 6 ). Although hypoglycemia and liver enzymes usually improve with age, the liver disease progresses and some adult patients present with cirrhosis and tumors and may require liver transplantation ( 2 , 3 ). Nonetheless, the major disease burden at adulthood is the heart and muscle phenotype, with up to 80% of patients presenting with muscle weakness and one-third with loss of ambulation ( 2 , 3 ). Whereas dietary treatment helps to control the hypoglycemic episodes, it has little effects on the peripheral myopathy ( 1 , 7 ). In the absence of a curative treatment for GSDIII, the muscle and heart manifestations remain as the most critically unmet medical needs. Gene therapy with rAAV vectors appears as an attractive option to correct the muscle and heart phenotype of GSDIII ( 14 ), but the size of the GDE cDNA (4.6 kb) along with the required regulatory sequences prevents its efficient encapsidation in a single rAAV. Two different strategies have been published, each with their own drawbacks: a dual vector approach ( 15 ) and a pullulanase-based approach ( 20 , 21 ). While the dual vector approach was efficient for muscle and heart correction ( 15 ), it required 2 vectors and might expose patients to toxicities. The second strategy relied on a single vector encoding the bacterial enzyme pullulanase and induces immune response toward the transgene ( 21 ). In the present work, we used rational engineering to generate a truncated version of the human GDE efficiently packaged in a single rAAV vector, intended for the correction of the cardiac and muscular manifestations of the disease in adult patients. Two arguments supported the rationale for liver detargeting, achieved by the use of a recently developed engineered AAV capsid ( 24 ): first, the correction of the liver disease in GSDIII at adulthood is challenging in preclinical models, likely due to hepatocyte proliferation and liver fibrosis, as previously reported ( 15 ). In particular, incomplete and heterogeneous correction was achieved after injection of a liver-optimized high-dose dual rAAV vector ( 15 ). Human patients also present with fibrosis starting from young ages ( 6 ) and might require an early treatment in childhood before marked fibrosis has developed. In this setting, rAAV would not be the first-line option, because of potential rAAV dilution after liver growth. Second, and most importantly, high-dose gene therapy clinical trials for neuromuscular disease have revealed the presence of severe, sometimes lethal, liver-associated toxicities, especially in the presence of an underlying liver disease ( 22 , 23 ). Because of the presence of liver fibrosis in GSDIII early in life, as well as cirrhosis and portal hypertension in a third of adult patients ( 2 , 3 , 6 ), we decided to detarget the liver to address the main disease manifestations during adulthood while ensuring safety. A translationally viable strategy to achieve efficient rAAV packaging and broader muscle cell transduction with large transgenes is the generation of shorter proteins ( 43 , 44 ). A similar approach has also been successfully used for Wilson disease ( 45 ). GDE is a complex enzyme that involves 2 distinct enzymatic activities in a single polypeptide chain ( 28 ). As such, even slight modifications of the sequence may induce a dramatic reduction of the activity. To reduce the size of GDE, we took advantage of the large number of reported missense variants in patients with GSDIII. We identified 2 putative regions for which we derived different truncated candidates. Intriguingly, deletion mutants within the C domain displayed activity in skeletal muscles, but not in the heart. Further characterization of the underlying mechanisms is ongoing although the working hypothesis is that the C domain participates in the regulation of GDE activity in a tissue-specific manner. Different from the C-domain mutants, the N-terminal–truncated ΔNter2-GDE was active and decreased glycogen accumulation in both heart and skeletal muscle. This truncation was predicted by molecular modelling to have limited impact on the structural integrity of the protein and its catalytic sites. However, molecular dynamics simulations indicated a higher flexibility of ΔNter2-GDE compared with the full-length GDE, which is coherent with the lower protein stability observed experimentally. The use of molecular dynamics simulations is therefore a valuable tool for the design of truncated proteins to enable rAAV gene therapy in diseases that involve large transgenes. The use of ΔNter2-GDE allowed us to generate a potent, high-quality product for GSDIII. rAAV-ΔNter2-GDE corrected muscle and heart phenotypes in both mouse and rat models of GSDIII with extensive glycogen reduction, reflecting broader cell transduction. Treatment of Agl –/– mice with rAAV-ΔNter2-GDE vector promoted the recovery of muscle strength 3 months after injection. This suggests that in GSDIII, the myopathy — which displays low fibrosis and muscle fiber renewal, in contrast to what has been observed in muscular dystrophies ( 5 , 46 ) — can be reversed, at least in rodent models. A recently developed human skeletal muscle model of GSDIII derived from hiPSCs edited by CRISPR/Cas9 technology ( 42 ) offers a unique opportunity to evaluate the efficacy and safety of the expression of the truncated ΔNter2-GDE in a human pathological context. These cells, in contrast to explanted cells, have self-renewal capacities and the possibility of differentiation in a variety of cell types ( 47 ). In the context of GSDIII, the limited access to liver and muscle biopsies makes in vitro human models of the affected tissues essential to evaluate safety and efficacy of therapeutic approaches. In our study, we demonstrated that ΔNter2-GDE clears the accumulated glycogen, suggesting that the truncation on the N-terminal region does not affect the enzyme function and regulation in a human cellular context. Furthermore, cell viability and myogenic markers were similar between rAAV-GFP- and rAAV-ΔNter2-transduced cells, suggesting the absence of specific alteration related to the expression of the truncated mutant in human muscle cells. These data are extremely valuable in the prospect of translating this gene therapy approach into humans. Beyond the correction of the muscle disease, further optimizations could improve single rAAV vector gene transfer for GSDIII. Targeting of the liver may prove complex due to the underlying liver disease ( 2 , 3 , 6 ) and the complexity of developing AAV capsids able to target the liver without overloading at the doses used for muscle gene transfer. However, the use of a muscle and liver tropic capsid combined with a short promoter able to achieve both liver and muscle transgene expression could potentially allow complete correction of GSDIII phenotype. Indeed, tandem promoters with both efficient liver and muscle expression have been described, but are too large to be efficiently encapsidated along with the ΔNter2-truncated GDE in a single rAAV vector ( 18 , 21 ). Combination of smaller promoters and novel rAAV capsids with a liver-muscle targeting tailored to GSDIII may be used in the next future to improve the safety and efficacy of the approach. Another potential approach to provide full rescue of the disease phenotype would be the combination of our approach with mRNA expressing the transgene directly in hepatocytes, as recently reported for other metabolic diseases ( 48 ). In conclusion, our work provides proof-of-concept of the use of a single rAAV vector expressing a truncated form of GDE for the rescue of muscle and heart impairment in 2 GSDIII rodent models as well as in a recently described human muscle cell model of GSDIII, which supports the clinical translation of this approach to provide a one-shot, definitive treatment for this burdensome disease.

Authorship note: AG and JR contributed equally to this work. Glycogen storage disease type III (GSDIII) is a rare inborn error of metabolism affecting liver, skeletal muscle, and heart due to mutations of the AGL gene encoding for the glycogen debranching enzyme (GDE). No curative treatment exists for GSDIII. The 4.6 kb GDE cDNA represents the major technical challenge toward the development of a single recombinant adeno-associated virus–derived (rAAV-derived) vector gene therapy strategy. Using information on GDE structure and molecular modeling, we generated multiple truncated GDEs. Among them, an N-terminal–truncated mutant, ΔNter2-GDE, had a similar efficacy in vivo compared with the full-size enzyme. A rAAV vector expressing ΔNter2-GDE allowed significant glycogen reduction in heart and muscle of Agl –/– mice 3 months after i.v. injection, as well as normalization of histology features and restoration of muscle strength. Similarly, glycogen accumulation and histological features were corrected in a recently generated Agl –/– rat model. Finally, transduction with rAAV vectors encoding ΔNter2-GDE corrected glycogen accumulation in an in vitro human skeletal muscle cellular model of GSDIII. In conclusion, our results demonstrated the ability of a single rAAV vector expressing a functional mini-GDE transgene to correct the muscle and heart phenotype in multiple models of GSDIII, supporting its clinical translation to patients with GSDIII. Single adeno-associated virus vector gene therapy corrects muscle and heart impairment in rodent and human models of glycogen storage disease type III.

Author contributions AG, JR, VMR, LR, PV, RL, MV, YKB, TLB, JC, JN, LJ, RF, LVW, JE, IA, and LH were involved in data generation. BB and GD were involved in rAAV production and purification. ND, XN, and GR supervised the experimental activities. AG, JR and, GR wrote the manuscript, and all authors provided review and editing of the manuscript. AG and JR contributed equally (order assigned alphabetically). Supplementary Material

This work was supported by Genethon, the “Association Française contre la Myopathie,” the “Association Francophone des Glycogénoses,” and the National Research Agency (ANR-17-CE18-0014 to GR and ANR 22 CE17 0031 to LH). AG is recipient of a “Poste d’Accueil Inserm” doctoral fellowship from the French National Institut of Health and Medical Research (Inserm); MV is recipient of a doctoral fellowship funded by the Ile-deFrance Region PhD program in context of the DIM Biotherapies program. YKB is funded by the France Relance Program (ANR-21-PRRD-0002-01). The authors are Genopole’s members, first french biocluster dedicated to genetic, biotechnologies and biotherapies. We are grateful to the “Imaging and Cytometry Core Facility” of Genethon for technical support, to Ile-deFrance Region, to Conseil Départemental de l’Essonne (ASTRE), to “Institut National de la Santé et de la Recherche Médicale” (Inserm) and GIP Genopole, Evry for the purchase of the equipments. We would like to thank the regional computing meso center CALMIP (Grant 2023-p22027) for granting access to High Performance Computing resources. Address correspondance to: Giuseppe Ronzitti, PhD, Genethon, 1bis rue de l’internationale, 91000 Evry, France, Université Paris-Saclay, University Evry, Inserm, Genethon, Integrare research unit UMR_S951, 1bis rue de l’internationale, 91000 Evry, France. Email: [email protected]. 11/28/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e172018

oa_package/6f/03/PMC10786702.tar.gz

PMC10786703

37943617

Introduction ER-associated degradation (ERAD) is the key cellular quality-control mechanism underlying the clearance of misfolded proteins from the ER, thereby generating a conducive environment for protein folding, maturation, and maintaining ER homeostasis. The suppressor of lin-12-like–HMG-CoA reductase degradation 1 (SEL1L-HRD1) complex, together with lectin osterosarcoma amplified 9 (OS9), ER lectin 1 (ERLEC1, also known as XTP3B), and degradation in ER (DERLIN) proteins, represents one of the most conserved branches of ERAD ( 1 – 6 ). In vivo, global or acute deletion of Sel1L or Hrd1 in germline and adult mice causes embryonic or premature lethality, respectively ( 7 – 10 ). Subsequent studies using cell type–specific KO mouse models, including those from our group, have established the vital importance of SEL1L-HRD1 ERAD in different cell types, including hematopoietic stem cells, various immune cells, pancreatic β cells, podocytes, hepatocytes, and adipocytes, in many physiological processes ( 4 – 6 , 8 , 11 – 31 ). However, its relevance and importance in humans remain unexplored. As the folding capacity within the ER varies greatly among different cell types, it has been hypothesized that cells may exhibit differential dependency on SEL1L-HRD1 ERAD ( 4 – 6 ). However, to date, the molecular evidence for this model remains circumstantial and a challenging question for the field. Intriguingly, we and others recently reported that SEL1L-HRD1 ERAD is indispensable for B cell development by targeting pre–B cell receptor (pre-BCR) for proteasomal degradation ( 11 , 13 ). However, in mature B cells, which secrete large amounts of immunoglobulin (Ig) proteins, SEL1L-HRD1 ERAD function seems to be dispensable ( 11 ). In the absence of SEL1L, B cell development is blocked at the large pre–B cell stage due to the accumulation of pre-BCR at the cell surface ( 11 , 13 ). In contrast, SEL1L deficiency in developing T cells attenuated, but failed to block, the development of αβ T cells while having no effect on that of γδ T cells ( 15 ). Hence, cell type–specific dependency of SEL1L-HRD1 ERAD in mammals remains an open, but exciting, question. In the accompanying paper, we reported 3 hypomorphic biallelic SEL1L and HRD1 variants causing a group of inherited disorders in 6 patients with ERAD-associated neurodevelopmental disorders with onset in infancy (ENDI) ( 32 ). Here, we report another SEL1L variant, p.Cys141Tyr ( SEL1L C141Y ), in 5 patients from a large family, with similar ENDI. However, unlike the other ENDI patients who are in their teens and 20s, these 5 patients exhibited B cell depletion and agammaglobulinemia and died at very early ages as a result of frequent infections. Mechanistic studies showed that this variant caused the most severe SEL1L-HRD1 ERAD dysfunction among all 4 variants by causing disulfide bond–mediated aggregation and HRD1-mediated degradation of SEL1L. These studies together demonstrate that functionality of SEL1L-HRD1 ERAD is inversely correlated with disease severity in humans.

Methods Human subjects. The patient cases were gathered through the web-based tool GeneMatcher ( 48 ) ( https://genematcher.org/statistics/ ). We present 5 developmentally delayed children with agammaglobulinemia. They were raised from birth in a big family of Roma population living in an isolated region in the southern part of Slovakia. Patients 1 and 2 (IV-1 and IV-2) were born in 2006 and 2014, respectively, from a pair of consanguineous parents in the family. Patient 1 (IV-1) presented hypotonia, developmental delay, frequent vomiting after eating, facial dysmorphisms, and agammaglobulinemia since her birth. At 15 months, she developed a 30% weight deficit, representing dystrophy of the third degree (marasmus). The physical examination revealed at the age of 3 years of life, a short stature (–3.28 SD), underweight (–5.81 SD), microcephaly (–1.89 SD), and severe developmental delay ( Supplemental Table 2 ). She was treated because of acute renal failure and pulmonary hypertension. She died due to multiorgan failure at the age of 2.9 years. Patient 2 (IV-2) presented axial hypotonia and facial dysmorphism since her birth. She got bronchopneumonia and died suddenly 2 months after birth. Her Ig levels were not tested. Patients 3–5 (IV-3, IV-5, and IV-6) were born in 2009, 2014, and 2016, respectively, from another pair of consanguineous parents in the family. All 3 patients presented with hypotonia, developmental delay, hypotrophy, facial dysmorphism, repeated vomiting after eating, and agammaglobulinemia since birth. Patient 3 (IV-3) was born with a ventricular septal defect. He was delivered at 37 weeks of gestation, birth weight of 2700 g (percentile = 25th) and length of 48 cm (percentile = 50th) with normocephaly. Because of the episode of unspecified epilepsy, a brain MRI was done, but showed normal myelinization. He was treated by valproate therapy. The genetic examination for cystic fibrosis and the hereditary agammaglobulinemia- BTK gene was negative. The child lived later in an orphanage, and his last physical examination showed at the age of 2.5 years short stature (–2.62 SD) and that he was underweight (–4.21 SD) ( Supplemental Table 2 ). He was lying in bed, exhausted from severe sepsis, and died from respiratory failure. Patient 4 (IV-5) did brain MRI at the age of 10 months, showing unspecific leukoencephalopathy frontal and occipital bilaterally, but she had no seizures. At the age of 6 years, she could turn, sit with support, and tried to stand with support. She didn’t speak. At the age of 4.5 years, her physical examination showed short stature (–3.12 SD), underweight (–3.03 SD), and microcephaly with 43 cm (–2.83 SD). Eye examination showed palpebral ptosis with bilateral partial papillary atrophy. Laboratory examination showed sideropenic anemia, increased folate level, and hypovitaminosis D. She was treated every 4 weeks with intravenous Ig, silymarin, hepatoprotective essential phospholipids, and pyridoxine, and during infection, with antibiotics. Two months before she died, she became sick with COVID-19 with a prolonged course compared with that of other children with hereditary agammaglobulinemia. Patient 5 (IV-6) was born at 39 weeks gestation with neurotrophic data (3025 g/50 cm) with microcephaly (head circumference: 31 cm, >3 pc). At the age of 2.5 years, physical examination revealed short stature (–2.25 SD), that he was underweight (–2.64 SD), microcephaly (–2.64 SD), hypotonia, and severe developmental delay ( Supplemental Table 2 ). Brain MRI was done and revealed the leukoencephalopathy frontal, parietal, and occipital bilaterally and later, at the age of 4.5 years, discrete supratentorial cortical atrophy. He did not develop seizures. The eye examination showed bilateral palpebral ptosis and bilateral papillary excavation. He had a micropenis, central hypothyroidism, hypoplastic thymus, hepatopathy, and dystrophic nails. He was intensively treated with antibiotics and oxygen and every 4 weeks with antibody supplementation therapy. At 5.1 years, he died due to severe pneumonia and respiratory failure. Patient IV-4 was a girl, born with severe hypotonia, alobar holoprosencephaly, and multiple malformations. The patient presented cheilognathopalatoschisis, 1 nostril with missing part of nasal wings, “frog” eyes, almost closed vision field, low-set deformed ears, microcephaly, short neck, pterygium colli, narrower chest, transverse groove on the left hand, 4 fingers with claw-like position on the right hand, and feet with malformed fingers with polydactyly and syndactyly of fourth and fifth fingers. The anus was malformed; multiple contractures on all joints were seen. She presented congenital heart defects with atrial ventricular septum defect and patent ductus arteriosus. The patient died at 9 days because of cardiorespiratory failure. The patient had no infection or problem with immunity as in the other siblings with agammaglobulinemia. CRISPR/Cas9-based KO and KI HEK293T cells. HEK293T cells, obtained from ATCC, were cultured at 37°C with 5% CO 2 in DMEM with 10% fetal bovine serum (Fisher Scientific). To generate SEL1L-, HRD1- and RNF5-deficient HEK293T cells, sgRNA oligonucleotides designed for human SEL1L (5′-GGCTGAACAGGGCTATGAAG-3′), human HRD1 (5′-GGACAAAGGCCTGGATGTAC-3′), or human RNF5 (5′-CACCTGTACCCCGGCGGAA-3′) were inserted into lentiCRISPR, version 2 (Addgene 52961). Cells grown in 10 cm petri dishes were transfected with indicated plasmids using 5 μl 1 mg/ml polyethylenimine (PEI) (MilliporeSigma) per 1 μg of plasmids for HEK293T cells. The cells were cultured 24 hours after transfection in medium containing 2 μg/ml puromycin for 24 hours and then in normal growth medium. SEL1L C141Y KI HEK293T cells were generated as described in the accompanying paper by Wang et al. ( 32 ). ΔFNII KI HEK293T cells were generated using the CRISPR/Cas9 approach. We designed 2 gRNAs (gRNA1 and gRNA3) flanking the SEL1L exon 4, which encodes the SEL1L FNII domain, and an additional gRNA (gRNA2) in close proximity to gRNA1 to enhance editing efficiency. These gRNAs were synthesized by Integrated DNA Technologies (IDT). These gRNAs and Cas9 protein were introduced into the cells via electroporation, followed by culturing and single-cell isolation with the desired genomic modification. SEL1L C127Y KI HEK293T cells were generated using the CRISPR/Cas9 cytosine base editing (CBE) system ( 49 ). Oligos with the gRNA sequence were annealed at 95°C for 5 minutes and cooled to room temperature for 30 minutes. The duplex was inserted to BbsI-treated (NEB) and gel-purified pYZ122-pSMART HCKan-sgRNA-Sp-BbsI plasmid, a gift from the Yan Zhang Laboratory (University of Michigan Medical School). The ligated product was transformed and amplified in DH5-α E . coli cells, and the plasmid sequence was confirmed by Sanger sequencing (rurofins). To transfect HEK293T cells, 125 ng of the plasmid containing the gRNA sequence and 375 ng of pCAG-CBE4max-SpG-P2A-EGFP (Addgene, 139998) were introduced to the cells with a confluency of 80% via Lipofectamine 3000 (Thermo Fisher) in 1 well of a 24-well plate. The CRISPR-processed cells were cultured at 37°C with 5% CO 2 . After 3 days of incubation, the genomic DNA of the cell culture was extracted with 50 mM NaOH. DNA fragments covering the target sites were amplified by PCR, using HotStart Taq 2× PCR Master (ABclonal), and analyzed by Sanger sequencing (eurofins) to estimate the percentages of mutant allele in the cell pool. In parallel, the cell culture was diluted into 8 cells per mL and cultured in 96-well plates (100 μL per well) for single-cell isolation. After 10 days, 100 single-cell colonies were transferred into 24-well plates. The SEL1L C141 or SEL1L C127 region of each colony was amplified using a 20 μL PCR reaction and sequenced. Cell colonies with homozygous SEL1L C141Y or SEL1L C127 alleles were selected and transferred into 6-well plates for further experiments. For the ΔFNII KI cell line, total RNA was extracted using TRI Reagent and BCP Phase Separation Reagent (Molecular Research Center, TR 118), followed by cDNA library generation using the High Capacity cDNA Reverse Transcription Kit (Thermo Fisher). The region encoding the FNII domain was amplified using a 20 μL PCR reaction and sequenced. Sequences were as follows: SEL1L C141Y crRNA (guide sequence): guide 1: 5′-ATGAATGTACATCAGATGGG-3′, guide 2: 5′-ATTCATCATACTCCTTATCT-3′; ΔFNII crRNA (guide sequence): guide 1: 5′-GGTAACTTCCGTGTCGTGTA-3′, guide 2: 5′-AACTTCCGTGTCGTGTACCC-3′, guide 3: 5′-ACTACAAAGCAGATGAAAAG-3′; HDR donor oligo(mutation sites are underlined): SEL1L C141Y : 5′-CACTTCCCTTTTCTTTTCCTAGATAAGGAGTATGATGAAT A TACATCAGATGGGAGGGAAGATGGCAGACTGTGGTGTGCTACAACCT-3′, SEL1L C127Y gRNA oligos (gRNA sequence is underlined): F: 5′-CACCG AGTGGCAGGGCTCCCCATG -3′, R: 5′-AAACCATGGGGAGCCCTGCCACTc-3; amplification PCR primers: SEL1L C127Y and SEL1L C141Y : F: 5′-TCAGCTAGCCATGCTCACTAAA-3′, R: 5′-TGACTTGAGTGACAGCCTGAAA-3′; ΔFNII: F: 5′-CTGCAGGCAGAGTAGTTGCT-3′, R: 5′-TGCATCTGCCGTCTCTTAGC-3′. Plasmids. The following plasmids were used in the study (h denotes human genes; m denotes mouse genes): mSel1L cDNA was cloned from mouse liver cDNA and inserted into the pcDNA3 to generate pcDNA3-mSEL1L(WT)-FLAG. pcDNA3-h-proAVP(G57S)-HA were described previously ( 24 ). SEL1L C141Y mutations in this study were generated using site-directed mutagenesis with pcDNA3-mSEL1L(WT)-FLAG as the template. All plasmids were validated by DNA-Seq. The cloning primers were as follows: mSEL1L-FLAG-F: 5′-CGCGGATCCACCATGCAGGTCCGCGTCAGGCTGTCG-3′; R: 5′-CGCTCTAGACTATTTATCATCATCATCTTTATAATCTCCGCCCTGTGGTGGCTGCTGCTCTGG-3′. C141Y-FLAG-F: 5′-GTATGATGAGTACACCTCAGACG-3′; R: 5′-CGTCTGAGGTGTACTCATCATAC-3′. Western blot and antibodies. Cells were harvested and snap-frozen in liquid nitrogen. The proteins were extracted by sonication in NP-40 lysis buffer (50 mM Tris-HCl at pH7.5, 150 mM NaCl, 1% NP-40, 1 mM EDTA) with protease inhibitor (MilliporeSigma), DTT (MilliporeSigma, 1 mM), and phosphatase inhibitor cocktail (MilliporeSigma). Lysates were incubated on ice for 30 minutes and centrifuged at 16,000 g for 10 minutes. Supernatants were collected and analyzed for protein concentration using Bio-Rad Protein Assay Dye (Bio-Rad). From 10 to 30 μg of protein was denatured at 95°C for 5 minutes in 5× SDS sample buffer (250 mM Tris-HCl pH 6.8, 10% sodium dodecyl sulfate, 0.05% bromophenol blue, 50% glycerol, and 1.44 M β-mercaptoethanol). Protein was separated using SDS-PAGE, followed by electrophoretic transfer to PVDF (Fisher Scientific) membrane. The blots were incubated in 2% BSA/TBST with the following primary antibodies overnight at 4°C: anti-HSP90 (Santa Cruz Biotechnology Inc., sc-13119, 1:5,000), anti-SEL1L (homemade, against SEL1L, 23–205 aa, ref. 50 , 1 :10,000), anti-SEL1L (Abcam, ab78298, against SEL1L, 330–400 aa, 1:1000), anti-HRD1 (Proteintech, 13473-1, 1:2,000), anti-OS9 (Abcam, ab109510, 1:5,000), anti-ERLEC1 (Abcam, ab181166, 1:5,000), anti-CD147 (Proteintech, 11989-1, 1:3,000), anti-IRE1α (Cell Signaling Technology, 3294, 1:2,000), anti-UBE2J1 (Santa Cruz Biotechnology Inc., sc-377002, 1:3,000), anti-ubiquitin (Santa Cruz Biotechnology Inc., P4D1, 1:1000), anti-LC3 (Cell Signaling Technology, 2775), anti-RNF5 (Bethyl, A303-594A, 1:2000), anti-FLAG (MilliporeSigma, F1804, 1:1,000), anti-HA (MilliporeSigma, H3663, 1:5,000), anti-PERK (Cell Signaling Technology, 3192, 1:5000), anti-–p-PERK (Cell Signaling Technology, 3179, 1:1,000), anti-eIF2α (Cell Signaling Technology, 9722, 1:5000), anti-p-eIF2α (Cell Signaling Technology, 9721, 1:1000), anti-BiP (Abcam, ab21685, 1:5,000), and anti-PDI (Enzo, ADI-SPA-890, 1:5,000). Membranes were washed with TBST and incubated with HRP-conjugated secondary antibodies (Bio-Rad, 1:10,000) at room temperature for 1 hour for ECL Chemiluminescence Detection System (Bio-Rad) development. Band intensity was determined using Image Lab (Bio-Rad) software, version 6.1. For additional information, see Supplemental Methods . Statistics. Statistics tests were performed using GraphPad Prism, version 8.0 (GraphPad Software). Unless indicated otherwise, values are represented as means ± SEM. All experiments were repeated at least 2 to 3 times and/or performed with multiple independent biological samples from which representative data are shown. All data sets passed normality and equal variance tests. Statistical differences between the groups were compared using unpaired 2-tailed Student’s t test for 2 groups or 1-way ANOVA or 2-way ANOVA for multiple groups. P < 0.05 was considered statistically significant. Study approval. Study protocols and protocols for written, informed consent were approved by the Johannes Kepler University Ethics Committee (JKU-EC, approval no. 1253/2021), the Institutional Review Boards of the University of Michigan Medical School (IRBMED, HUM00227482), and the Institutional Review Board for Health Sciences Research (IRB-HSR, University of Virginia, HSR230351). Patients and parents provided written, informed consent prior to participation in the study. Written, informed consent was received for use of the photographs. Data availability. Materials and reagents used are either commercially available or available upon request. All materials used for the manuscript are included in Methods. Values for all data points in graphs are reported in the Supporting Data Values file.

Results Identification of a biallelic SEL1L C141Y variant in humans. Five siblings from 2 consanguineous families in a large Slovakian family presented with developmental delay, neurological disorders, and agammaglobulinemia in childhood ( Figure 1 ) and were suspected of inherited genetic disorder. Array comparative genomic hybridization (aCGH) analysis performed in patients 4 and 5 did not reveal the presence of any larger deletions or amplifications within the genome ( Supplemental Figure 1 ; supplemental material available online with this article; https://doi.org/10.1172/JCI170882DS1 ). Analyses of whole-exome sequencing of DNA samples from patients 3 and 5 and their parents (III-3 and III-4) did not identify any known monogenic inborn errors of neurological disorders and agammaglobulinemia. Since both patients were born to the same consanguineous parents, variants were queried from the database according to the assumption of a recessive inheritance with 100% penetrance. Annotated variants were then filtered against their allele frequency (smaller than 1% or unknown) and predicted deleteriousness ( Figure 1B ). Two variants were identified in individual patients, SEL1L p.Cys141Tyr (NM_005065.6: exon 4: c.422G>A) and fatty acyl-CoA reductase 2 ( FAR2 ) p.Arg490Trp (NM_001271783.2: exon 12: c.1468C>T) ( Figure 1B and Supplemental Table 1 ). As Sanger sequencing confirmed the segregation of the FAR2 variant with symptoms in patient 2 ( Supplemental Figure 2 ) and as loss of FAR2 function is not linked to neurological disorder or agammaglobulinemia ( 33 ), we excluded it from being causal for these patients. Moreover, Sanger sequencing further confirmed the biallelic SEL1L C141Y variant in all 5 patients, but not in parents or unaffected siblings ( Figure 1, C and D ). Hence, on the basis of the known function of SEL1L protein, and after applying stringent filters of the exome sequencing data ( Supplemental Table 1 ) according to the American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP) 2015 guidelines for clinical interpretation of genetic variants ( 34 ), we determined that SEL1L C141Y is a potential candidate. SEL1L C141Y variant in patients with ENDI-agammaglobulinemia. The clinical presentation was uniform among the 5 patients with ENDI-agammaglobulinemia (ENDI-A). All patients started to have problems with food intake soon after birth, as they repeatedly vomited after eating ( Table 1 and Supplemental Table 2 ). They were unable to gain weight, had pale skin color, gradually developed cachexia, and had similar facial dysmorphisms, including triangle faces, big ears, etc. ( Figure 1A ). All patients showed severe axial hypotonia and general developmental delay with short stature and microcephaly ( Table 1 ). Three patients (patients 1, 3, and 5) could not sit up, hold their heads, or raise their heads while supporting themselves on their elbows. Only patient 4 was able to sit without support at the age of 4.5 years. All patients exhibited intellectual disability and were unable to speak words and sentences ( Table 1 and Supplemental Table 2 ). Of note, the IV-4 individual, carrying 1 allele SEL1L C141Y , was born with multiple malformations (alobar holoprosencephaly, “frog” eyes, anus malformation, congenital heart defects, etc.) and died from cardiorespiratory failure at 9 days of age ( Figure 1A ). Hence, these patients exhibited typical ENDI symptoms as described in the accompanying paper ( Table 1 ) ( 32 ). However, unlike the other ENDI patients, these patients were frequently admitted to the hospital due to recurrent severe lower respiratory infection almost every month, starting from otitis media, sinusitis (bilateral maxillary and ethmoidal), to bronchitis and pneumonia, as diagnosed by doctors. During pneumonia, children developed respiratory insufficiency often caused by food aspiration. As a result, they were hospitalized nearly monthly and treated with antibiotics and sufficient oxygen supply and were regularly supplemented with Igs. In patients 3 and 4, hearing impairment was diagnosed at 1 year of age, which was likely secondary to multiple otitis media. Immunological tests were performed with peripheral blood from patients 3, 4, and 5, which revealed no detectable mature CD19 + memory B cells in the circulation or circulating Igs at the age of 12 months and beyond ( Table 2 ). Consequently, IgG was given as a replacement therapy every 4 weeks in these patients. Indeed, after replacement therapy was initiated, a substantial improvement in acute infections was noted. However, chronic respiratory symptoms, phlegm, and cough in some forms continued to recur. The early treatment strategy by IgG replacement therapy and antiinfectious prophylaxis very likely postponed infectious complications in patients 3, 4, and 5. However, all 3 were observed to have a sudden dramatic deterioration of the clinical condition due to sepsis with multiorgan failure. A gut biopsy of patient 3 showed subtotal villous atrophy of the duodenum as Marsh IIIb enteropathy with no CD20 + B cells and moderate to high amounts of CD3 + , CD4 + , and CD8 + intraepithelial cells compared with noncarriers ( Supplemental Figure 3A and Supplemental Table 3 ). Although the absolute numbers of CD4 + helper and CD8 + cytotoxic T cells were largely in the normal ranges, the ratio of CD4 + to CD8 + T cells was reduced ( Table 2 ), pointing to an impaired T cell development with defective SEL1L-HRD1 ERAD. Moreover, patient 4 had a COVID-19 infection with respiratory failure at the age of 7.7 years. After the patient was administered with therapeutic anti–SARS-Cov-2 monoclonal antibodies, the viral load was effectively decreased. However, the course of treatment was prolonged and differed from that of healthy children or children with X-linked agammaglobulinemia (XLA). Indeed, no specific cellular response measured by COVID antigen–specific memory T cells was detected in peripheral blood ( Supplemental Figure 3B ), indicative of defects in COVID antigen–induced T cell activation. Hence, these patients lacked mature B cells and exhibited agammaglobulinemia, with impaired T cell development and/or function. Sequence and structural analyses of SEL1L C141Y variant. This variant affects a conserved residue in the luminal N-terminal fibronectin type II (FNII) domain of SEL1L, a domain with unknown function ( Figure 2A ). Interestingly, unlike other domains of SEL1L, the FNII domain is not conserved in invertebrates ( Figure 2B ). Position-specific scoring matrix (PSSM) analysis ( 35 ) showed that Cys at this position was evolutionarily selected and that the Cys-to-Tyr mutation may be detrimental to SEL1L function ( Figure 2C ). Structural modeling of the human SEL1L-HRD1-OS9-DERLIN1 protein complex (SEL1L, 107–723 aa; HRD1, 1–334 aa; OS9, 33–655 aa; DERLIN1, 1–213 aa) using the AI-based AlphaFold2 prediction network ( 36 ) showed 2 short antiparallel β sheets connected by 2 long random coils at the FNII domain with 2 disulfide bridges in close, quasiorthogonal juxtaposition ( Figure 2, D and E ): Cys141-Cys168 and Cys127-Cys153. SEL1L C141Y variant causes ERAD complex instability and dysfunction. We next tested to determine whether and how SEL1L C141Y affects ERAD function using skin fibroblasts derived from patients and those from noncarrier individuals as WT controls. Strikingly, SEL1L protein levels were significantly reduced and largely undetectable in the patient fibroblasts compared with in WT cells ( Figure 3, A and B ). In keeping with our previous findings that SEL1L is required for HRD1 protein stability ( 37 ), HRD1 protein levels were also significantly reduced, by over 90%, in patient cells ( Figure 3, A and B ). The reduction of SEL1L and HRD1 protein levels was confirmed using immunofluorescent staining in patient skin fibroblasts ( Supplemental Figure 4 , A and B) as well as immunohistochemical staining of the duodenal biopsies from the patients ( Figure 3C ). This reduction in SEL1L-HRD1 protein levels was not due to gene transcription, as their mRNA levels were unchanged compared with those in healthy cells ( Supplemental Figure 4C ). SEL1L and HRD1 proteins became unstable in SEL1L C141Y patient cells treated with a translation inhibitor, cycloheximide ( Figure 3, D and E ), while 2 known ERAD substrates, inositol-requiring enzyme 1α (IRE1α) ( 37 ) and CD147 ( 38 ), were accumulated and stabilized in SEL1L C141Y patient cells ( Figure 3, A, B, D, and E ). Much to our surprise, 2 lectins that help recruit substrates to the SEL1L-HRD1 complex, OS9 and ERLEC1, were significantly decreased and destabilized in SEL1L C141Y patient cells ( Figure 3, A–E ), uncoupled from their gene transcription ( Supplemental Figure 4C ). Furthermore, a model ERAD substrate proarginine vasopressin (proAVP) mutant, Gly57Ser (proAVP G57S) ( 24 ), formed significantly more high–molecular weight (HMW) aggregates in SEL1L C141Y knockin (KI) HEK293T cells compared with WT cells, to levels similar to those in ERAD-KO HEK293T cells ( Supplemental Figure 5 , A and B). ER staining showed an increase in ER volume in SEL1L C141Y KI HEK293T cells ( Supplemental Figure 5C ). SEL1L C141Y causes the most severe ERAD dysfunction among all 4 variants. We next compared the SEL1L C141Y variant to the other hypomorphic SEL1L and HRD1 variants (SEL1L M528R , SEL1L G585D , and HRD1 P398L ) described in the accompanying paper by Wang et al. ( 32 ) in terms of ERAD function by generating KI HEK293T cells using the CRISPR/Cas9 system expressing individual variants ( Supplemental Figure 6 , A–E). Indeed, compared with cells expressing other hypomorphic variants, SEL1L C141Y cells had the lowest SEL1L and HRD1 protein levels, but the highest protein levels of the ERAD substrates CD147 and IRE1α ( Figure 4, A and B ). Hence, we conclude that SEL1L C141Y -expressing cells exhibit the most severe ERAD dysfunction among all the hypomorphic variants. Lack of an overt unfolded protein response in SEL1L C141Y cells. We next asked whether ERAD dysfunction in SEL1L C141Y cells induces an overt unfolded protein response (UPR). UPR was measured using standard protocols as previously described ( 39 ). ER chaperones Ig heavy chain–binding protein (BiP) and protein disulfide isomerase (PDI) were significantly elevated ( Figure 4C ). While IRE1α protein levels were elevated in SEL1L C141Y cells, IRE1α was not phosphorylated based on the phos-tag system ( 39 , 40 ) ( Figure 4D ). Consistently, X-box–binding protein 1 ( XBP1 ) mRNA splicing was not detected in SEL1L C141Y cells ( Figure 4E ). Although the absolute levels of phosphorylation of protein kinase R-like ER kinase (PERK) and eukaryotic initiation factor 2α (eIF2α) were elevated in patient fibroblasts when normalized to the loading control HSP90, there was no difference in the percentage of phosphorylated PERK and eIF2α upon normalization to total PERK and eIF2α proteins ( Figure 4, F and G ). Changes in these markers were not affected by the treatment of MG132 ( Figure 4, D and G ). Hence, we conclude that SEL1L C141Y causes severe ERAD dysfunction, but is not associated with an overt UPR. This scenario likely resulted from the upregulation of ER chaperones and the expansion of the ER volume. The disulfide bonds in the FNII domain are required for ERAD complex stability and function. We next asked mechanistically how the SEL1L C141Y variant affects ERAD complex stability. To this end, we first performed immunoprecipitation to examine the complex formation. Upon overexpression, SEL1L C141Y had no effect on the interactions between SEL1L and other ERAD components, such as OS9, ERLEC1, HRD1, and ubiquitin-conjugating E2 enzyme J1 (UBE2J1) ( Figure 5A ), thus excluding the possibility that the SEL1L C141Y variant interferes with the complex formation. Next, as SEL1L has 2 disulfide bond pairs, C127-C153 and C141-C168, in the FNII domain ( Figure 2A ), we asked whether each disulfide bond had a similar impact on ERAD complex stability. We disrupted another disulfide bond by generating SEL1L C127Y KI HEK293T cells ( Supplemental Figure 6 , A–E). Indeed, similarly to SEL1L C141Y , SEL1L C127Y reduced the protein levels of the ERAD complex and stabilized the known ERAD substrates ( Figure 5, B and C ). Importantly, the effects of both variant/mutant on protein levels of the ERAD complex and substrates were relatively milder compared with those of SEL1L –/– HEK293T cells ( Figure 5, B and C ), suggesting that SEL1L C141Y is not a complete loss-of-function variant, which may explain why the patients could survive for months or even years. Moreover, cycloheximide experiments showed that both variants rendered the ERAD complex unstable while increasing the stability of ERAD substrates such as IRE1α and CD147 ( Figure 5, D and E ). It is worth noting that, unlike in ERAD-deficient cells, where both lectins are stabilized, OS9 and ERLEC1 were unstable in both SEL1L C141Y and SEL1L C127Y cells ( Supplemental Figure 7 and Figure 5E ), pointing to an additional impact of free Cys in SEL1L on other ERAD components. Hence, disulfide bonds in the FNII domain of SEL1L are indispensable for ERAD complex stability and function. SEL1L FNII domain is dispensable for ERAD function. We next explored the importance of the FNII domain in ERAD function. Interestingly, the FNII domain (aa 122–170) is not conserved and is absent in fly or yeast SEL1L homolog Hrd3 ( Figure 6A ). We generated FNII-less SEL1L HEK293T cells using CRISPR/Cas9-mediated deletion of the entire exon 4 encoding residues 115 to 170 ( Supplemental Figure 8 , A–C). Initial experiments using a homemade N-terminus–specific (which includes the FNII domain) antibody showed an approximately 90% reduction of SEL1L protein levels in FNII-less SEL1L KI HEK293T cells ( Figure 6B ). However, using a C-terminus–specific antibody (from Abcam), we noted that loss of the FNII domain caused an approximately 60% reduction of SEL1L protein levels compared with that in WT cells, which was much higher than those in cells expressing Cys variants ( Figure 6, B and D ). The difference between these 2 antibodies was also further confirmed in SEL1L WT/ΔFNII cells (ΔFNII HET) ( Figure 6B ). These findings not only confirmed the deletion of the FNII domain in FNII-less SEL1L, but also showed that the FNII domain is important for SEL1L protein stability. By comparison, both antibodies detected very little, if any, SEL1L protein in SEL1L C141Y KI HEK293T cells ( Figure 6, B and D ), hence excluding the possibility that the failure to detect SEL1L protein was due to antibody recognition affected by Cys mutations. Moreover, HRD1 protein levels in FNII-less SEL1L KI HEK293T cells were reduced by 30% compared with those in WT cells, but doubled compared with those expressing the Cys variants ( Figure 6, C and D ). Further examination of ERAD substrates such as IRE1α and CD147 showed mild, if any, changes in their protein levels and stability in FNII-less SEL1L KI HEK293T cells compared with WT HEK293T cells ( Figure 6, C–F ), pointing to largely normal ERAD function associated with FNII-less SEL1L. Hence, we conclude that the FNII domain of SEL1L is dispensable for SEL1L-HRD1 ERAD function and that the SEL1L C141Y variant affects ERAD complex stability, likely through the unpaired Cys. SEL1L C141Y variant causes proteasome-mediated self-destruction. We next further explored mechanistically how the SEL1L C141Y variant causes the instability of the SEL1L-HRD1 ERAD complex. We first asked whether proteasomes are required in this process. Treatment with the proteasomal inhibitor MG132 elevated the protein levels of both SEL1L and HRD1 ( Figure 7A and Supplemental Figure 9A ), pointing to the involvement of the proteasomes in the reduction of the ERAD complex. Next, as previous studies have implicated HRD1 ( 41 , 42 ) or RING finger protein 5 (RNF5/RMA1) E3 ligase ( 43 ) in HRD1 turnover, we asked which the E3 ligase is involved in the degradation of the SEL1L-HRD1 complex in the presence of the variants. We generated HRD1 –/– or RNF5 –/– HEK293T cells expressing (via KI) SEL1L C127Y and SEL1L C141Y . Strikingly, deletion of HRD1, but not RNF5, significantly rescued the protein levels and stability of SEL1L in SEL1L C127Y and SEL1L C141Y KI HEK293T cells ( Figure 7, B, C, E, and F ). This was consistent using 2 different SEL1L antibodies recognizing different regions of SEL1L protein ( Figure 7D and Supplemental Figure 9B ). The accumulation of SEL1L C141Y protein in the absence of HRD1 protein led to the formation of HMW aggregates in HEK293T cells ( Figure 7G ). Similarly, OS9 and ERLEC1 proteins accumulated ( Figure 7B and Supplemental Figure 9C ) and formed HMW complexes ( Supplemental Figure 9D ) in HRD1 –/– ;SEL1L C127Y and HRD1 –/– ;SEL1L C141Y HEK293T cells. Additionally, both IRE1α and CD147 were also further increased upon the deletion of HRD1 compared with the parental SEL1L C127Y and SEL1L C141Y KI cells, pointing to residual HRD1 function in these KI cells ( Figure 7B and Supplemental Figure 9E ). In conclusion, SEL1L C141Y causes proteasome-mediated self-destruction of the SEL1L-HRD1 ERAD complex.

Discussion Here we report a group of patients expressing a new biallelic SEL1L C141Y variant with clinical features of ENDI-A ( Figure 8A ). They resemble the ENDI patients described in the accompanying paper ( 32 ) in terms of neurodevelopmental disorders characterized by infantile-onset developmental delay, intellectual disability, microcephaly, hypotonia, and facial dysmorphisms, i.e., ENDI. However, unlike those ENDI patients, they exhibited severe B cell immunodeficiency, suffered frequent infections that required Ig replacement therapy, and died due to respiratory insufficiency. These differences likely reflect different degrees of ERAD dysfunction among these variants; while SEL1L (p.Gly585Asp, p.Met528Arg) and HRD1 (p.Pro398Leu) variants are hypomorphic with moderate ERAD dysfunction, the SEL1L p.Cys141Tyr variant is much more severe, with a significant loss of the ERAD complex and much more severe ERAD dysfunction ( Figure 8, A and B ). Agammaglobulinemias are congenital diseases characterized by a lack of functional B cells and antibodies ( 44 ). Here, we show that in humans, SEL1L-HRD1 ERAD dysfunction is likely associated with agammaglobulinemia. This finding is supported by previous reports that SEL1L-HRD1 ERAD plays a key role in B cell development in mice as a checkpoint to control the degradation and hence abundance of pre-BCR ( 11 , 13 ). SEL1L-HRD1 deficiency in B cell lineage causes B cell developmental blockade at the large pre–B cell stage, leading to a significant reduction of mature B cells ( 11 , 13 ). In comparison, the effect of SEL1L p.Cys141Tyr variant on T cells is more moderate with largely normal absolute number of T cells in the periphery. Nonetheless, the ratio of CD4 + to CD8 + T cells is altered, in line with a known role of SEL1L-HRD1 ERAD in αβ T cell development ( 15 ). Hence, given the severity of ERAD dysfunction, we believe that the SEL1L C141Y variant likely causes agammaglobulinemia by blocking B cell development in humans. Definitive evidence will come from further studies with mouse models carrying the variant, which will also be useful to delineate how T cell development and function are affected. From combining evidence from genetics and in silico and in vitro analysis, SEL1L C141Y has been considered to be a pathogenic variant with a total score of 14 points based on ACMG criteria ( 34 ). At the same time, we also considered the possibility of a second biallelic variant causing immune deficiency phenotypes in ENDI-A. Other variants associated with neurodevelopmental disorders or hypogammaglobulinemia were examined independently with chromosome analysis, SNParray, and next-generation sequencing (NGS); however, these tests did not show any other genomic imbalances or other associated genetic variants. Genetically, the chance of 5 affected children showing similar symptoms of 2 different diseases with a genetic linkage of 2 different biallelic variants is extremely low. Therefore, we consider the SEL1L C141Y variant as the only possible causative variant in these patients. With that being said, we acknowledge that we cannot firmly establish disease causality without a KI mouse model carrying the variant. SEL1L has 2 disulfide bonds, both of which are in the FNII domain. Here our data reveal the importance of disulfide bonds in the little-known FNII domain of SEL1L, while the FNII domain itself is dispensable in SEL1L-HRD1 function. The FNII domain probably formed during evolution via exon shuffling ( 45 ) and is also present in other proteins, such as coagulation factor XII ( 46 ) and the cation-independent mannose-6-phosphate/insulin-like growth factor-II receptor (IGF2R) ( 47 ). Disrupting either disulfide bond causes HRD1-mediated self-destruction of the complex, presumably due to the formation of aberrant disulfide bonds ( Figure 8A ). This effect of the disease variant is distinct from that of a simple loss of function of SEL1L, where there is no free cysteine. In the latter case, loss of SEL1L causes HRD1 self-degradation, while leading to the stabilization and accumulation of lectins such as OS9/ERLEC1 ( Supplemental Figure 10 ). While much more in disease pathogenesis associated with this variant awaits further investigation, the identification of this variant not only provides exciting opportunities for studying ERAD biology, but also further establishes the (patho-)physiological importance of SEL1L-HRD1 ERAD. Together with the findings reported in the accompanying paper ( 32 ), our data have uncovered an inverse correlation between SEL1L-HRD1 ERAD and disease severity in humans. It paves the foundation for future efforts to therapeutically target this important protein complex in the treatment of human diseases.

Authorship note: DW, LLL, and HHW contributed equally to this work. RGF, JAM, and LQ are co–senior authors. Suppressor of lin-12-like–HMG-CoA reductase degradation 1 (SEL1L-HRD1) ER-associated degradation (ERAD) plays a critical role in many physiological processes in mice, including immunity, water homeostasis, and energy metabolism; however, its relevance and importance in humans remain unclear, as no disease variant has been identified. Here, we report a biallelic SEL1L variant (p. Cys141Tyr) in 5 patients from a consanguineous Slovakian family. These patients presented with not only ERAD-associated neurodevelopmental disorders with onset in infancy (ENDI) syndromes, but infantile-onset agammaglobulinemia with no mature B cells, resulting in frequent infections and early death. This variant disrupted the formation of a disulfide bond in the luminal fibronectin II domain of SEL1L, largely abolishing the function of the SEL1L-HRD1 ERAD complex in part via proteasomal-mediated self destruction by HRD1. This study reports a disease entity termed ENDI-agammaglobulinemia (ENDI-A) syndrome and establishes an inverse correlation between SEL1L-HRD1 ERAD functionality and disease severity in humans. A biallelic SEL1L C141Y variant causes premature death in 5 patients with early-onset neurodevelopmental disorders and agammaglobulinemia (ENDI-A) due to severe SEL1L-HRD1 ER-associated degradation dysfunction.

Author contributions DW, HMW, ALP, SW, KS,GH, AS, LD, and JAM obtained clinical, molecular, and biochemical data. LLL, HHW, ZJL, ZW, XW, and TTY designed and performed biochemical experiments. KK and RGF assisted with some experiments and analysis. PC acquired clinical and immune profile data. DW, RGF, JAM, and LQ directed the study. DW, PC, HHW, LLL, and LQ wrote the manuscript. All authors commented on and approved the manuscript. Supplementary Material

We are deeply grateful for all patients and their families for their consent and willingness to participate in this study. We thank members in the Qi, Mayr, Wolf, Ciznar, Danisovic, Soltysova, Skalicka, and Weis laboratories for technical assistance and insightful discussions; Shengyi Sun, Fowzan S. Alkuraya, Nicola Brunetti-Pierri, Claude Besmond, Gabriela Hrckova, Frantisek Valacsai, and Margit Burmeister for their constructive comments; Yan Zhang for CRISPR plasmids and suggestions on generating KI cells; László Kovács from the children’s clinic of the Faculty of Medicine of the Comenius University in Bratislava and his wife for acquiring the photo of patient IV-1 (patient 1). This work was supported by 1R35GM130292 and the Michigan Protein Folding Disease Initiative (to LQ) and the Austrian Science Fund (FWF; I4695-B, GENOMIT to JAM). LLL and ZJL are supported in part by National Ataxia Foundation Post- and Predoctoral Fellowships (NAF 918037 and 1036307). 11/09/2023 In-Press Preview 01/16/2024 Electronic publication

CC BY

no

2024-01-16 23:40:17

J Clin Invest.; 134(2):e170882

oa_package/43/e7/PMC10786703.tar.gz

PMC10788183

38226110

Introduction Liver surgery is a primary treatment modality for hepatocellular carcinoma (HCC) and colorectal liver metastasis, markedly reducing mortality and morbidity over the last decade [ 1 - 4 ]. Advances in diagnostic imaging such as computed tomography (CT) and magnetic resonance imaging (MRI) have led to earlier identification of liver malignancies such as HCC, with an increase in the incidence of small tumors. These tumors had previously not been considered surgical targets but are now regarded as an indication for hepatectomy [ 5 ]. In the fourth Japan Society of Hepatology-HCC guidelines, surgical resection is recommended as the first-line therapy for solitary HCC, regardless of size, with Child-Pugh A/B liver function and without extrahepatic metastasis or vascular invasion [ 6 ]. In addition, an increasing number of cases in which liver tumors that have shrunk due to successful chemotherapy, such as liver metastasis from colorectal cancer, are being resected [ 7 ]. Gastrointestinal tumors may recur after completion of chemotherapy. Particularly, complete response (CR) on imaging and pathological diagnosis could be inaccurate among patients with primary or metastatic liver cancer. Therefore, several patients who achieve CR on imaging still require exploratory surgery. Moreover, despite improvements in the treatment and prognosis of HCC, recurrence remains a clinical challenge [ 8 ]. The number of patients with micro-recurrent HCC who will undergo resection is predicted to increase further [ 9 ]. Liver tumors are usually identified using intraoperative ultrasound (IOUS), and a hepatectomy is accordingly performed to include the liver tumor. However, in our experience, small liver tumors are often unidentifiable on IOUS. In such cases, excision is usually performed based on the positional relationship between the vessel running in the liver and the liver tumor determined from preoperative imaging tests. Hence, there is always a possibility of a wider extent of hepatectomy, and, in rare cases, the tumor may not even be resected. This is a major disadvantage for patients and a major challenge in current hepatectomy techniques. To date, only one clinical study on percutaneous CT-guided marking for metastatic liver tumors has been reported [ 10 ], and, to our knowledge, no reports of transcatheter marking in liver resection have been published. Preoperative marking is a standard practice for deep-seated tumors in other solid organs, such as the lung [ 11 ]. Furthermore, the usefulness of coil marking for lung tumors has been reported [ 12 , 13 ]. Preoperative marking via endoscopic instillation or clipping for gastrointestinal tumors is also common. Originally, we would have liked to conduct a Phase II study to confirm the usefulness of the procedure; however, as mentioned above, only a few case series for percutaneous marking have been reported [ 10 ], and, to our knowledge, there are no reports of trials on transcatheter marking or a study combing the two procedures. Accordingly, it is necessary to first evaluate the safety and reliability of our preoperative marking method, which was developed according to a previous report [ 14 ], in a Phase I trial.

Materials and methods Trial design and participants This exploratory prospective study included patients between August 2017 and March 2020 based on the criteria shown in Table 1 . Patients were excluded according to the criteria shown in Table 2 . Patient data were collected from one month before marking to one month after surgery. During data collection, we de-identified the data and accessed only the anonymized information during analysis to protect patient privacy. Sample size estimation Discussions with the protocol committee at the facility led to the decision that because this was a new protocol that was scientifically valid in other surgical areas but lacked evidence in liver resection and an exploratory study with no proof of safety, including a large number of patients was not desirable; as such, the next phase of the study would be planned after the safety of the protocol was confirmed in about 20 patients. Treatment protocol Marking was performed in the vicinity of the tumor before hepatectomy (Figure 1 ) using the percutaneous or transcatheter method. Given that the benefit of this procedure to the patient is yet to be proven, marking was performed simultaneously with the examinations for liver tumors. Therefore, additional invasive procedures were avoided by selecting the percutaneous method for patients in whom CT and ultrasound-guided biopsies were performed. The transcatheter method was also performed simultaneously with angiography. If the tumor could be confirmed by plain CT or ultrasound and punctured from the body surface, the percutaneous method was selected; if it could not be confirmed by such methods or was difficult to puncture, the transcatheter method was selected. We avoided predetermining the method and attempted to select the most appropriate method for each case. The marking procedure was performed by an interventional radiology specialist certified by the Japanese Society of Interventional Radiology using either method. The distance between the tumor and the marker was measured using horizontal or coronal CT sections. The presence of marker residue was determined using postoperative CT. Preoperative chemotherapy In patients with liver metastasis, chemotherapy was administered after marking [ 17 ]. Preoperative chemotherapy was administered according to Japanese guidelines. The regimen consisted of eight courses of the CapeOX + Bmab regimen (capecitabine + oxaliplatin + bevacizumab) for colorectal cancer and four courses of the SOX regimen (tegafur-gimeracil-oteracil potassium + oxaliplatin) for gastric cancer. Chemotherapy response was assessed by CT and/or MRI every four courses according to the Response Evaluation Criteria in Solid Tumors [ 18 ]. Percutaneous marking Percutaneous marking was performed using local anesthesia under CT/ultrasound guidance. The same site was punctured again following a tumor biopsy for marking. The puncture was performed using a Chiba needle (Cook Medical, Japan). The needle tip was inserted near the tumor (within the range that could be resected at the same time by partial resection), and one microcoil was placed via the needle. The needle was not advanced into the tumor to avoid the risk of dissemination due to direct puncture. The puncture needle was then removed after confirming the absence of internal bleeding. The position of the marker was confirmed by CT before and after its placement (Figure 2 ). Cases in which tumors can be confirmed by preoperative ultrasound should also be confirmed by IOUS, and thus, they should not be marked; however, clinically, they were selected for tumors that were expected to be eliminated with preoperative chemotherapy. Transcatheter marking Transcatheter marking was performed following the primary angiography. No new blood vessel puncture was performed for the marking. A 4 Fr catheter was introduced through the femoral artery using the Seldinger technique, and abdominal vessel angiography was performed to identify the feeding artery of the tumor. A microcatheter was inserted superselectively into the vicinity of the tumor (within the range that could be resected at the same time by partial resection). After the positional relationship between the tumor and microcatheter was confirmed via CT, a microcoil was placed. The marker position was then confirmed again via CT (Figure 3 ). Liver resection An IOUS was used to confirm the marker placed preoperatively, and the tumor was resected based on the marker. Per oncological principles, the route of the percutaneous puncture was included in the excision range. Hepatectomy procedures were performed as previously described by Itamoto et al. [ 19 ] and Kuroda et al. [ 20 ]. In addition, the Pringle maneuver was used during hepatectomy, as necessary. After hepatectomy, the required number of drains was placed where they were required. The drains were removed within one week; however, the duration of drainage was extended if a bile fistula or intra-abdominal abscess occurred. Tumors and markers were searched in all cases using Sonazoid® (common name: perfluorobutane; GE Healthcare Pharma, Tokyo, Japan) contrast ultrasonography. Outcome measures The primary outcome measure was the safety of the tumor marking protocol. This was assessed according to the incidence of adverse events and major postoperative complications. The secondary outcome measures were the success rate of tumor resection and changes in blood biochemical parameters. Ethics approval and consent to participate The protocol for this research project was approved by a suitably constituted ethics committee at Hiroshima University Hospital (approval number: C-198) and conformed to the provisions of the Declaration of Helsinki. Informed consent was obtained from all study participants.

Results A total of 19 patients with liver tumors measuring ≤20 mm and requiring resection were treated between August 2017 and March 2020 (Figure 1 ). Of these patients, nine (47.3%) had primary liver cancer, whereas 10 (52.7%) had metastatic tumors. Percutaneous marking was performed in six patients, all of whom had metastatic tumors. Transcatheter marking was performed in 13 patients. One patient in the percutaneous group had an unresectable tumor due to tumor growth during post-marking chemotherapy; thus, five patients underwent post-marking resection. In the transcatheter group, marking could not be completed in two patients because of the long distance from the hepatic hilum to the tumor and because of tortuous blood vessels; therefore, the catheter could not be advanced to the tumor. However, all patients underwent resection. The characteristics of patients in the percutaneous and transcatheter groups are summarized in Table 3 and Table 4 , respectively. Additionally, the details of the surgery are presented in Table 5 . While the marker should be removed maximally, we intentionally left the remaining marker, as we decided not to remove excess liver tissue (larger than that needed for hepatectomy) to remove the marker because of the distance between the marker and the tumor. The rates of postoperative complications and postoperative liver failure were comparable to those of normal hepatectomy.

Discussion Small liver nodules are difficult to distinguish intraoperatively. Tumors that cannot be identified by ultrasonography are challenging to treat locally, with procedures such as radiofrequency ablation; and when resection is performed, it can only be conducted in approximation based on anatomic landmarks. This carries the risk of an unnecessarily large hepatic volume resection or leaving parts of the tumor undetected. Furthermore, although a report of successful percutaneous preoperative marking has been published [ 10 ], most tumors cannot be marked percutaneously in clinical practice. It is hypothesized that an adaptable choice of percutaneous marking or transcatheter marking would prove to be useful. To our knowledge, the current study is the first to perform preoperative marking of primary HCC using an intravascular catheter. As a preliminary step to a Phase II trial to demonstrate the usefulness of this method, we were able to establish the safety of this method in a Phase I trial. Marking metastatic liver cancer using the percutaneous method has been reported previously [ 21 ]. In the present study, preoperative tumor marking was successfully performed via the percutaneous and transcatheter approaches. Percutaneous marking was possible in all six patients with metastatic liver tumors. Meanwhile, in the transcatheter group, marking failure only occurred in 2/13 patients. Four patients in this group had metastatic liver tumors and nine had HCC. Both cases with marking failure were of HCC. Small HCCs may be difficult to identify without angiography [ 22 ]. In addition, factors such as tumor location (e.g., under the dome, deep in the dome, or near the main vessel) make marking more challenging in the percutaneous method, which could explain the higher rate of marking failure in the transcatheter group, as marking of some HCC tumors could not be performed using the percutaneous method. This suggests that in marking liver tumors, the modalities and approaches that can be used to mark the tumor differ depending on the location and size of the tumor. Therefore, our new method, which allows flexibility in choosing between percutaneous and transcatheter approaches is practical. Various methods such as indocyanine green (ICG) fluorescence [ 23 ] and intraoperative contrast-enhanced ultrasound [ 24 ] have been developed for navigation during hepatectomy. However, ICG fluorescence can only be performed on the surface of the liver. Moreover, detecting a small tumor in liver cirrhosis is often difficult using intraoperative contrast-enhanced ultrasound. Marking navigation is a useful technique addressing the limitations of these methods, allowing surgical navigation even beyond the liver surface and in liver cirrhosis. Nonetheless, the accuracy of marking and the safety of the marker being placed near the tumor are yet to be established. In this study, the percutaneous markers could be placed within 10.5 mm from the tumor, and transcatheter markers could be placed within 12 mm. The farthest distance was 20 mm. This distance between the tumor and the marker is acceptable for effective navigation in actual surgery. Markers were identified by IOUS in all patients, indicating their effectiveness in identifying the excision site. Marking was also highly effective for one patient in whom the tumor disappeared unexpectedly following chemotherapy. The tumor could be resected in all patients with the marker as a guide. For two patients in the transcatheter group, the catheter could not reach the tumor site because of the distance from the hepatic hilum to the tumor and because of tortuous blood vessels; thus, the catheter could not be advanced to the tumor. One of the two patients had an anatomic variation in which the right hepatic artery diverged from the superior mesenteric artery, which may have contributed to the difficulty in marking. Future studies should investigate alternative methods, such as switching to the percutaneous method, for marking in such cases. With respect to safety, no marking-related complications occurred in either the percutaneous or transcatheter groups. One patient developed a postoperative complication, but it was unrelated to the preoperative marking. Meanwhile, six marker remnants were found. These showed two patterns. The first pattern was where chemotherapy was not successful after marking and hepatectomy was not indicated due to tumor growth. This pattern was observed in one patient who underwent percutaneous marking. In the second pattern, the markers were not within the resection range because they were far from the tumor. This was found in one patient in the percutaneous group and four patients in the transcatheter group. In the transcatheter method, the coil is left on the central side of the transected part of the vessel because of the straightening of the coil. To avoid such a problem, a shorter coil could be considered in the future. However, the coil used in this method is the one used for embolizing blood vessels and was originally designed on the assumption that it will remain in the body permanently. As such, the remnants may have no adverse impact on organ function. No adverse events related to the residual coil were observed in this case. In addition, even if the marker remains, as it is not too large, only slight artifacts are visible on postoperative CT, and this did not interfere with the diagnosis. This study has some limitations. As this was an exploratory study and the safety of the protocol could not be confirmed, the number of cases was purposely kept small. Given that safety was not confirmed, marking was performed concurrently with other preoperative examinations such as angiography and liver biopsy. The optimal patients for marking are yet to be determined; hence, patient enrollment based on optimum benefit could not be performed. Although preoperative marking has originally been considered most effective against tumors that are difficult to detect, the evaluation of its efficacy is limited because the study was conducted on detectable tumors. Large-scale Phase II trials are required to further verify the safety and effectiveness of preoperative marking in liver tumors. In fact, given the results of this study, which have provided evidence of the safety of this method, we have currently initiated a Phase II trial to demonstrate its efficacy.

Conclusions This study demonstrated the safety and feasibility of preoperative marking for small liver tumors. Preoperative marking for the intraoperative identification of liver tumors does not have any adverse effects and enables surgical navigation, making it a safe and effective modality. These findings provide a basis for improving the complete resection rates in these tumors. To show the efficacy of preoperative marking, a future prospective study will focus on liver tumor cases that are difficult to confirm preoperatively.

Background Small tumors in liver cirrhosis are difficult to distinguish using intraoperative ultrasonography. In addition, preoperative chemotherapy for metastatic liver cancer may diminish tumor size, thus making tumors difficult to identify intraoperatively. To address such difficulties, we devised a method to mark liver tumors preoperatively to facilitate intraoperative identification. This study aimed to investigate the safety of a preoperative liver tumor marking method. Methodology This exploratory prospective clinical trial included patients with liver tumors measuring ≤20 mm requiring resection. Preoperative marking was performed by placing a coil for embolization of blood vessels near the tumor using either the transcatheter or percutaneous approach. The tumor was identified and resected by intraoperative ultrasonography based on the marker. The study was registered in the University Hospital Medical Information Network Clinical Trials Registry (UMIN000028608). Results Overall, 19 patients (9 with primary liver cancer and 10 with metastatic tumors) were recruited. The transcatheter and percutaneous methods were used in 13 and 6 patients, respectively. Marking was not possible in two patients in the transcatheter group because the catheter could not be guided to the vicinity of the tumor. There were no marking-related complications. Hepatectomy was performed in all but one patient who was not fit for hepatectomy owing to the development of a metastatic liver tumor. The markers were adequately identified during hepatectomy. Additionally, there were no difficulties in the surgical procedure or postoperative complications. Conclusions Preoperative marking with embolization coils can be performed safely for intraoperative identification of liver nodules.

The authors thank Editage (www.editage.jp) for English language editing.

CC BY

no

2024-01-16 23:40:17

Cureus.; 15(12):e50603

oa_package/db/96/PMC10788183.tar.gz

PMC10788185

38226314

Introduction In 1906, carotene, lutein and chlorophyll were successfully isolated from green leaves [1] . Carotenoids are isoprene metabolites derived from isoprene pyrophosphate (IPP) and its isomer, dimethylallyl diphosphate (DMAPP) [2] . The carotenoid synthesis pathway involves nine catalytic enzymes, including octahydro lycopene synthase (PSY), lycopene ε-cyclase (LCY-ε), lycopene β-cyclase (LCY-β), and 9-cis-epoxy carotenoid dioxygenase (NCED), etc. These fat-soluble pigments act as antioxidant stress protectors in plants [3] . For instance, when green plants are overexposed to sunlight, the photosynthetic system is damaged, and photosynthetic free radicals are quenched with carotenoids in the plant to protect the plant from strong light damage [4] . Carotenoids also play a crucial role in the drought resistance of higher plants. Studies have shown that radish ( Raphanus sativus L) has a significant increase in the content of carotenoids and malondialdehyde (MDA) in the roots when subjected to drought stress [5] . Additionally, after treating the carrot taproot with PEG, it was found that the expression of 9-cis-epoxycarotenoid dioxygenase (DcNCED1) and DcNCED2 increased significantly under drought stress, and lutein levels exhibited varying degrees of increase. This indicated that NCED genes could respond to drought stress [6] . In plants, carotenoids are mainly divided into carotene and lutein, of which lutein accounts for 59% of the total carotenoids [7] . Lutein, a crucial branching product in the carotenoid synthesis pathway, is a yellow pigment closely related to the photosynthetic reaction center and plays a crucial role in plant removal of oxidative damage and response to drought stress [8] . lycopene cyclase, as a central enzyme for lutein synthesis, is of great significance for regulating lutein production. The lycopene cyclase family (LCYs) was initially identified in bacteria, exhibiting high conservation from bacterial to plant domains [9] . Subsequent experiments confirmed that the amino acid sequences of lycopene cyclase of plants and cyanobacteria are highly similar, and they are both composed of 400 amino acid residues and a polypeptide with a molecular weight of 43kD [10] . LCYs have been demonstrated in various plants, classified into two subfamilies, lycopene ε-cyclase and lycopene β-cyclase [11] . In 1996, the function of β and ε lycopene cyclases was analyzed, revealing that it has the function of regulating the proportion of downstream carotenoids produced [12] . The close relationship between lycopene cyclase and plant stress resistance has been demonstrated, with the overexpression of LCY-β enhancing tolerance to abiotic stresses in tomatoes [13] . Cotton, an important cash crop, is susceptible to various environmental stresses during its growth, and drought stress is one of the main abiotic stress factors affecting cotton yield. Drought stress not only affects the growth and development of cotton, but also severely weakens its productivity, and this impact is likely to expand further as global warming intensifies. Therefore, it is of great significance to the study cotton drought tolerance. Drought stress negatively affects plant morphological, physiological and biochemical processes [14] . Drought stress induces various responses in plants, including the regulation of osmotic substances to mitigate damage. At the cellular level, drought stress triggers Ca 2+ signaling, reactive oxygen species (ROS), and abscisic acid (ABA) hormone-mediated signaling. Carotenoids act as signaling molecules, reducing oxidative damage and enabling plants to adapt to various stresses [15] . Carotenoids, as synthetic precursors of ABA, improve the drought resistance of cotton by regulating the production of ABA. Carotenoids are essential for survival in clearing reactive oxygen species ROS in the body [16] . Lutein has been observed to affect the functional chlorophyll antenna size of photosystem II, preventing ROS formation by inhibiting harmful substances by binding at site I of the plant's major LHCII complex [17] . Some carotenoid genes are involved in regulating plant responses to various biotic and abiotic stresses. Overexpression of the lycopene β-cyclase gene in tobacco enhanced the salt and drought tolerance of transgenic plants [18] . Despite extensive studies on lycopene cyclase in various species, its function in cotton has not been reported, making the study of lycopene cyclase in cotton necessary. In this research, 12 drought-tolerant materials were selected to determine the content of lutein in cotton seedlings after the treatment with 15% polyethylene glycol (PEG 6000) and lycopene cyclase (LCYs) were mined in order to explore the relationship between lutein and drought resistance. The obtaied gene GhLCYε-3 was subjected to a VIGS experiment, from LCYs, revealed increased wilting under drought stress, accumulated more ROS, faster water loss on leaves, increased malondialdehyde and proline, and decreased chlorophyll and lutein levels. The results provide a reference for the functional study of LCYs in other species and the breeding of drought-tolerant varieties of cotton.

Materials and methods Experimental materials 12 materials were chosen to determination of the lutein content in cotton seedlings ( Table 1 ), with 30 capsules per material and three replicates. PEG6000 (Solarbio,Beijing) served as the standard for simulating drought conditions [19] . The treatment concentration was set at 15% PEG6000 and the processing time was 3 days, based on our previous research results [15] . The 12 materials are drought-tolerant and are preserved by the Institute of Cotton Research of Chinese Academy of Agricultural Sciences. The seeds were placed vertically and evenly on cut filter paper in a tray, with the filter paper moistened either with ddH2O or a 15% PEG6000 solution. The filter paper is rolled into straight tubes, and the rolled filter paper (growth point down) is transferred vertically to a beaker containing 15% PEG6000 (adding PEG solution to 1/5 of the total beaker range) and ddH 2 O. The beaker is transferred to a biochemical incubator at 25 °C/24 h for dark germination, removed after 72 h of treatment, observed seed germination and recorded. Germination potential was calculated as the number of germinations divided by the total number, multiplied by 100% (germination was considered when the germ length reached half of the seed length). Uniform-sized cotton seedlings were preserved in liquid nitrogen. Due to the fat solubility of carotenoids, acetone extraction methods were used in this study [20] . The whole seedlings are placed in a pulverizer and the sample is ground thoroughly. Lutein was extracted from seedlings using 50% acetone, left to stand protected from light for 24 h and centrifuged for analysis. Absorbance photometric values at 470, 485, 642, 665 nm were measured using a microplate reader (BioTek Synergy HT, USA). Chlorophyll a (mgL -1 )= 9.99A 665 -0.0872A 642 , Chlorophyll b(mgL -1 )= 17.7A 642 -3.04A 665 , Lutein= (mgL -1 ) = 10.2A 470 -11.5A 485 -0.0036[a]− 0.652[b], Lutein content (mgL -1 ) = Lutein (mg·L -1 ) × Total volume of extract / Sample fresh weight (g). Identification of LCYs gene family members Genome sequences of four Gossypium species including G. hirsutum , ZJU; G. barbadense , HAU; G. arboreum , CRI; and G. raimondii, JGI were used to identify the gene family [21] . Protein sequences of tetraploid ( G. barbadense and G. hirsutum ) and diploid ( G. arboreum and G. raimondii ) cotton varieties were downloaded from Cotton FGD ( https://cottonfgd.org ) [22] . Additionally, genome datas of six other plants, A. thaliana, Vitis vinifera (V. vinifera), Populus trichocarpa (P. trichocarpa), Theobroma cacao (T. cacao), Oryza sativa (O. sativa) and Zea mays (Z. mays) were downloaded from the Phytozome V12.1 online site ( https://phytozome-next.jgi.doe.gov/ ) [23] . The Hidden Markov models (HMMs) (version 3.0) profiles of (Pfam05834) were downloaded from online website Pfam ( https://pfam.xfam.org/ ) to obtain the LCY proteins with default parameters, then we further screened the LCY proteins by using. In order to identify potential genes for the LCYs in the terrestrial cotton genome, genes with the conserved domain of the lycopene cyclase protein (PF05834) were screened using the hidden Markov model as a query file. The Pfam database was used for additional analysis (PF05834) to manually exclude genes with incomplete domains based on the conserved domain of the LCYs protein. Phylogenetic evolutionary analysis of GhLCYs and prediction of cis-acting elements The obtained LCYs sequences were submitted to CottonFGD ( https://cottonfgd . net) to get physicochemical parameters like protein length, molecular weight (MWs), isoelectric points (pIs) and subcellular localization. Subsequently, 49 LCYs protein sequences were entered into MEGA7.0 ( https://www.megasoftware.net/ ) in order to examine the evolutionary relationships between LCYs. Phylogenetic trees for LCY proteins from four cotton species and six other species were constructed using the Neighbor Joining (NJ) method with default parameters in MEGA 7 software. The construction was based on the LG+G model, and 1000 bootstrap repeat sequences were utilized for robustness analysis [24] . The 2000 bp DNA sequence in the upstream region of the GhLCYs gene was obtained from the Cotton FGD website ( https://www.cottonfgd.org/ ). The Plant CARE website ( http://bioinformatics.psb.ugent.be/webtools/plantcare/html/ ) was used to predict the cis-acting elements in the promoter region of LCYs genes. The cis-acting elements schematic was constructed by TBtools software. Four cotton motif composition analysis and chromosome location The protein sequences were entered into the online MEME software to predicted the LCYs conserved motif sequence, The analysis was set to allow a maximum of 15 motifs, and other parameters were kept at default settings ( http://meme-suite.org/tools/meme ) [22] . Genome-wide annotation files and General Feature Format (GFF) files for four different cotton cultivars were acquired from CottonFGD ( http://www.cottonfgd.net/ ). The TBtools (TBtools-ll Toolbox for Bioologists v1.108, parameter default https://github.com/CJ-Chen/TBtools/releases ) was used by default to locate and analyze the LCYs gene chromosomes of fou r cottons [25] . FPKMs of LCYs are collected from Gossypium Resource and Network Database online database ( https://grand.cricaas.com.cn/page/tools/expressionVisualization ). The expression of diferent GhLCYs and GbLCYsin cotton response to salt (400 mM NaCl), drought (20% PEG), cold (4 °C) and heat (37 °C) for diferent time points (1, 3, 6 and 12 h) were analyzed. LCYs protein interaction prediction GhLCYε-3 protein sequences were downloaded from CottonFGD ( https://cottonfgd.net ). GhLCYε-3 protein sequences were entered into online websites ( https://cn.string-db.org/ ) where prediction protein interactions, and finally the most similar basic sequences were selected for plotting [26] . I-TASSER was used to identify 8 proteins of GhLCYs, enter protein sequences into online website prediction ( https://zhanglab.ccmb.med.umich.edu/I-TASSER/ ), and selected the most similar to plot [27] . 8 GhLCYs protein sequences were entered into the SOPMA ( https://npsa.lyon.inserm.fr/cgi-bin/npsa_automat.pl?page=/NPSA/npsa_sopma.html ) online website to predict protein structures, with default parameters. Plant treatment and fluorescence quantitative detection of GhLCYs expression analysis Genome-wide annotation files and General Feature Format (GFF) files for four different cotton cultivars were acquired from CottonFGD [28] . The seeds were sown in sand and incubated at 26 °C for 16 h during the day and 8 h at night. To investigate the effect of PEG on GhLCYs expression, 15% PEG6000 treatment was used. 0% PEG6000 treatment was used as a control and samples were taken separately until the phenotype of the treatment appeared. True leaf samples were frozen at − 80 °C, and RNA extraction was carried out using the Adlai EASYspin Plus kit, employing the lysate milling method. As a template for qRT-PCR, RNA was extracted and reverse-transcribed into cDNA. To determine the response of genes to drought, 8 genes were selected under pressure. GhLCYs qRT-PCR primer sequences were created by Premier V6 ( http://www.premierbiosoft.com/ ) design ( Supplementary Table S1 ). The PCR production size range for primers is set to 50–100, and all other parameters are default. Three biological replicates were used in qRT-PCR experiments on the Bio-Rad 7500 Fast Photo-PCR platform in accordance with the manufacturer's instructions for TransStart Top Green qPCR Supermix (TransGene Biotech Co., LTD, Beijing, China). The reaction conditions were predenatured 95 °C, 300 s, denatured 95 °C for 20 s, annealed 55 °C for 20 s, extension 72 °C for 20 s, and the total cycle was set to 40 times. The 2- ΔΔCt standardized method was used to determine the relative expression level of genes [29] . SPSS software (IBM SPSS Statistics 21, https://www.ibm.com/spss ) is used for statistical analysis of data confidence testing. GhUBQ7 served as the internal control and was stably expressed in cotton plants, remaining unaffected by treatment or genotype [30] . Preliminary functional verification of GhLCYs Previous studies have shown that VIGS technology is widely used in cotton [31] . To explore the function of Lycopene ε cyclase ( GhLCYε-3 ), a 300 bp silent fragment ( Supplementary Table S1 ) was designed ( https://vigs.solgenomics.net/ ) and Ten genes downstream of GhLCYε-3 were fluorescently assayed for changes, and its primers were added in ( Supplementary Table 1 ). The constructed VIGS vector was transformed into Escherichia coli (DH5-α competent cells, weidi), and the plasmid was extracted after the bacterial solution was sequenced correctly. The pYL156 vector was maintained by the laboratory for a long time. The silenced fragments were attached to the pYL156 vector preserved in this experiment, and three treatments were designed with a negative control (CK), a positive control (pYL156), and a silencing plant pYL156- GhLCYε-3 . The bacterial fluid was injected into the leaves of Zhong H177, followed by incubation in a dark chamber at 25 °C for 24 h, transitioning to 25 °C with a 16 h light/8 h darkness cycle. Both VIGS plants and pYL156 plants were treated using PEG (PEG treatment for 3 h), and samples were collected upon leaf wilting and stored at − 80 °C. Proline (Pro) and malondialdehyde (MDA) content detection The leaf to be measured is placed on the instrument, and the chlorophyll II content of the leaf is determined using a content analyzer (Konica Minolta China Investment Co., Ltd.). The leaves of the pYL156 plant and the pYL156- GhLCYε-3 plant were used to determine chlorophyll with three biological replicates per sample. The determine of chlorophyll II content is performed by a content analyzer (Konica Minolta (China) Investment Co., Ltd.). Leaves of pYL156 and pYL- 156-GhLCYε-3 plants were used to determine PRO and MDA. Take 0.1 g of the sample to be tested, grind the leaves separately, and then determine the content of Pro and MDA [32] with proline (Pro) and malondialdehyde (MDA) detection kits (Nanjing Institute of Bioengineering) according to the instructions [32] . Hydrogen peroxide (H 2 O 2 ) was detected through staining with diaminobenzidine (DAB) after reaction, and ROS accumulation under drought stress was measured in leaves of plants pYL156 and pYL156-GhLCYε-3 (DAB) [33] . The leaves are placed in 1 g/L DAB staining solution, allowed to stand for 12 h, and decolorized with absolute ethanol [34] . The staining of the leaves was observed and recorded, and the black particles on the leaves were ROS [34] . Statistical analysis A minimum of three biological replicates was considered necessary for each dataset. The data was analyzed using a single factor completely randomized trial, with the least significant difference (LSD) test employed to determine significance, denoted as *P < 0.05 or * *P < 0.01.

Results and analysis Germination test and PEG tolerance evaluation of 12 cotton germplasm Existing studies have shown that lutein is closely related to stress resistance. As shown in Table 1 , the content of Zhong H177 lutein in 12 materials was high, and the germination potential, fresh weight root length were also at a high level. Therefore, Zhong H177 was selected for the next step of research. Fig. 1 illustrates that, during normal cotton germination, there exists a positive correlation between germination potential, fresh weight, and lutein, while root length shows a negative correlation with lutein ( Fig. 1 ). To pinpoint the key gene responsible for regulating lutein, an analysis was conducted on the lycopene cyclase family, which governs lutein synthesis. Identification and characterization of lycopene cyclase genes in four cotton species The target genes were put into the CottonFGD ( https://cottonfgd.net// ) website, and the domain screen finally obtained 24 LCYs genes, 8 Gossypium hirsutum (GhLCYs), 8 Gossypium barbadense (GbLCYs), 4 Gossypium arboreum (GaLCYs), and 4 Gossypium Raimondii (GrLCYs), LCYs are highly conserved in the process of evolution. Screening in Arabidopsis thaliana using the domain, it was found that there were 3 genes LCYs in Arabidopsis, AT2G32640 (ATLCYβ-1), AT3G10230 (LCYβ-2), AT5G57030 (ATLCYε-1) , after comparing the Arabidopsis sequence with LCYs in cotton, the 24 LCYs identified in four cotton species were divided into two groups of LCY-β and LCY-ε. The LCY-β group exhibited a length ranged from of 487 to 497 amino acids ( Table 2 ), while the length of LCY-ε group ranged from 497 to 769. Based on the gene's chromosomal position, we renamed it in the four cotton genomes. The isoelectric point, protein molecular weight, protein length and subcellular location of these genes were predicted and analyzed. The isoelectric points of the proteins encoded by LCYs ranged from 5.732–7.492 in the LCY-ε group, 7.864–8.526 in the LCY-β group, 55.469kD∼86.48kD in the LCY-ε group, and 54.639kD∼56.441kD in the LCY-β group. The protein sequences encoded by LCYs ranged from 497kD to 769kD amino acids. Subcellular localization prediction showed that LCY-ε was distributed in chloroplasts, and LCY-β group were distributed in chloroplasts and chromosomes. Conservative motif analysis of LCYs proteins Conserved motifs are essential for gene function studies [35] . Four important presumed conserved motifs were identified for LCY-β and LCY-ε based on the conserved motifs in the four cotton species. The conservative motifs of the LCY-β and LCY-ε groups in the four cotton varieties were delineated. Despite these differences, numerous highly similar motifs were identified among LCYs in the four cotton species ( Fig. 2 ). The LCY-ε group exhibited 36–50 similar conserved motifs, while the LCY-β group displayed 50 similar conserved motifs, suggesting a higher level of conservation in the LCY-β group in these species. It was clearly shown that LCYs in Gossypium barbadense and Gossypium hirsutum had higher basic similarity, possibly due to more complete gene copies of cotton in tetraploid. Our results showed that LCYs in four cotton species were highly conserved. Phylogenetic analysis of LCYs Four intra-cotton phylogenetic trees were constructed to study LCY kinship ( Fig. 3 ). Total of 49 protein sequences were derived from G. hirsutum, G. barbadense, G. arboreum, and G. raimondii, A. thaliana, O. sativa, V. vinifera, P. trichocarpa, T. cacao, O. sativa, and Z. mays were used to construct the interspecific tree. The phylogenetic tree is divided into LCY-ε and LCY-β groups, each with 12 genes in four cotton species, which is also divided into two groups, with 31 genes in group 1 and 18 genes in group 2 in 10 species. Among them, 8 LCYs genes were identified in Gossypium barbadense, 8 in Gossypium hirsutum, 4 in Gossypium raimondii, 4 in Gossypium arboreum, 3 in Arabidopsis, 3 in Oryza sativa, 10 in Populus trichocarpa, 3 in grapes, 3 in Theobroma cacao and 3 in Vitis vinifera . Notably, the phylogenetic trees depicted a distinct separation between monocots and dicots, emphasizing the highly conserved characteristics of LCYs across the 10 species. Chromosome location of LCYs The chromosomal arrangement of genes plays a pivotal role in the development and evolution of an organism's features [36] . The LCY-β group was distributed on chromosomes 09 and 13, and the LCY-ε group was distributed on chromosomes 05 and 07 except for Gossypium raimondii . In both Gossypium barbadense and Gossypium hirsutum , 4 genes were evenly distributed in subgroup A and subgroup D, and demonstrating a uniform arrangement. Similarly, in Gossypium arboreum and Gossypium raimondii , exhibited an even distribution of 4 genes across four chromosomes. In particular, except for Gossypium raimondii , the genes of the other three cotton varieties are distributed on chromosomes 05 and 07. As shown in Fig. 4 the distribution of genes in the cotton species, Gossypium hirsutum , Gossypium arboreum , Gossypium arboreum are similar in chromosomal location, suggesting that LCYs were conserved during evolution. However, the LCYs gene of Gossypium raimondii displayed a different pattern, with distribution across chromosomes 01, 06, 09, and 13. Notably, the gene distribution on chromosomes 06 and 09 in Gossypium raimondii differed from that in other cotton species, indicating a phenomenon of chromosome reversal. Structural analysis of motif and cis-acting elements of LCYs Exon, intron and encoded protein distribution patterns are reportedly employed for evolutionary study at various taxonomic levels [37] . In our investigation of the potential structural development of LCYs, we constructed a phylogenetic matrix and conducted a motif association analysis ( Fig. 5 ). The results reveal that class I (LCY-ε) contains 9–13motifs and all have the same motif except GbLCY ε-2. Meanwhile, the class II (LCY-β) contains 11–12 motifs, but they have up to ten similar motifs. The LCY-ε and LCY-β groups were found to have 6–10 similar motifs, indicating the conservation of LCYs during evolution. Similar exon and intron patterns were identified in the LCY-ε group. Interestingly, the LCY-β group displayed almost no introns, suggesting its ability to respond more swiftly to stress. To further understand the response of LCYs genes to stress, PlantCARE was used to predict the LCYs promoter region. It was observed that most of the genes in LCYs contain photoreactivity, salicylic acid reactive MYB, drought, low temperature, some elements of defense stress response. Fig. 5 illustrated that the LCYε group contains more photosensing elements, accounting for about 8.0% of the total element, while the LCY-β group has more elements involved in defense and stress response, accounting for 7.6% of the total element. Analysis of expression patterns of LCYs To investigate the response mechanism of LCYs to abiotic stressors, FPKM (Replace with log 2 ) of LCYs were gathered from an online database. The Gossypium Resource and Network Database were used to assess the expression profiles of these genes under various stresses (cold, heat, salt, and PEG), build a phylogenetic matrix, and correlate the expression heat map ( Fig. 6 ). The findings indicated that multiple LCYs genes corresponded to various abiotic stresses, including GhLCYε-1, GhLCYε-2, GhLCYε-3, GbLCYβ-2, GhLCYβ-2, GhLCYβ-4 and GbLCYβ-4 . In particular, the expression level of GhLCYε-3 gene showed significant differences. Color indicates Gene expression levels, red indicates high gene load, significantly differential expression. Yellow represents medium expression, blue expression is low expression or no expression. Interaction network of LCYs proteins Protein-protein interaction analysis can reveal unknown functions of proteins [38] . Based on cotton protein sequence search, the interaction network of LCYs in cotton was analyzed ( Fig. 7 ). 11 protein (Contains GhLCYε-3) interactions were identified, such as: Zeta-carotene desaturase (GH_A11G2866, GH_A13G1988, GH_D13G1948, GH_D11G2894)、Prolycopene isomerase (GH_D01G1424,GH_A01G1349), GhLCYβ-4 (GH_D13G0022), GhLCYβ-2 (GH_A13G0026), Phytoene synthase (GH_A07G0815) and Violaxanthin de-epoxidase (GH_A11G2696). This suggests that GhLCYε-3 may regulate the synthesis of lutein and ABA by interacting with GhLCYβ-2 and GhLCYβ-4 to complete the response to drought stress. Three-dimensional (3D) structure prediction of GhLCYs proteins I-TASSER was used to obtain the 8 GhLCYs proteins structures ( Fig. 8 ). Interestingly, the structure of GhLCYs proteins is mostly similar, while GhLCYε-1 has a special protein conformation. The structures of GhLCYs proteins consist of α-helices, β-strands and random coils. Specifically, the LCY-ε group exhibits α-helices ranging from 40.69% to 42.21%, random coils from 36.02% to 38.88%, and β-strands from 15.51% to 17.64%. On the other hand, the LCY-β group features α-helices ranging from 38.23% to 40.64%, random coils from 37.22% to 42.25%, and β-strands from 15.90% to 17.30%. Expression patterns of s under drought stress The fluorescence quantitative results showed that the gene expression of GhLCYs was higher in leaves, and under normal growth conditions, such as GhLCYβ-3 and GhLCYβ-4 expression tended to stabilize in stems. Under PEG stress, there were significant differences in expression such as GhLCYε-1、 GhLCYε-3、GhLCYβ-1 and GhLCYβ-4 . Fig. 9 illustrates that GhLCYε-3 and GhLCYε-1 are expressed much higher after treatment than before treatment. In particular, the expression of GhLCYε-3 gene increased by about 2 times compared with before treatment, and the expression level in the stem was significantly increased, so we selected GhLCYε-3 gene for further study. Cotton plants with the GhLCYε-3 gene silenced by VIGS were sensitive to PEG stress VIGS was used to verify the effect of GhLCYε-3 under drought stress. In Fig. 10 A VIGS plants withered more severely than control group (CK). Fig. 10 B shows a 63% decrease in the VIGS plant pYL156: GhLCYε-3 . After silencing GhLCYε-3 , we found that chlorophyll and lutein content were significantly reduced in silent plants. A decrease of about 27.2% in chlorophyll and 6.7% in lutein was observed, a ratio of about 4 multiples ( Fig. 10 C and 10D). Biochemical indexes MDA activity and Pro activity were measured for CK and VIGS plants, revealing a significant increase in the content of MDA and Pro in VIGS plants. Additionally, the speed of leaf water loss was observed to be faster in VIGS plants ( Fig. 10 E and 10 F). Furthermore, the expression of lutein-related genes downstream of VIGS plants is down-regulated, such as:PP2C8, AIP1, CYP97C1, CYP97A3, ABF2, PP2C, AIP1, ABF2, BHY, ABI5in ( Fig. 10 I). Fig. 10 I showed that the reactive oxygen species of GhLCYε-3 VIGS plants increased. The results showed that the reactive oxygen species scavenging system of GhLCYε-3 VIGS plants was weakened, and the drought tolerance of cotton was reduced.

Discussion Abiotic stress significantly impacts on the productivity of cotton. Lutein, known for its ability to remove ROS, plays a crucial role in plants' response to abiotic stress. The identification of LCYs, key genes in the biosynthetic pathway of lutein, is of great significance for improving the stress tolerance of crops. In order to explore the relationship between lutein and stress resistance, LCYs that control lutein synthesis were selected for further research. The research analyzed the expression of LCYs in four cotton species with conserved motifs, evolutionary relationships, acting elements and various abiotic stresses. Studies have shown that the GhLCYε-3 gene affects the drought resistance of cotton by regulating the content of lutein MDA and PRO content. The results of this study further elucidate the relationship between LCYs and lutein and enhanced stress resistance. In order to explore the relationship between lutein with germination, fresh and root length, this experiment was designed. Lutein content analysis indicated a positive correlation between germination potential and fresh weight with lutein, and a negative correlation with root length. Lycopene cyclases (LCYs), crucial enzymes regulating lutein production, have been shown to increase lutein accumulation upon overexpression [39] . Based on previous research, we analyzed LCYs in cotton to explore the relationship between LCYs regulating lutein. The research shows that the length of this gene protein is 487–769 kD, which is consistent with previous studies [40] . Subcellular prediction showed localization on chloroplasts, possibly lycopene cyclase controlling lutein production, while lutein and chlorophyll belong to the photosynthetic system, and analysis of LCYs cis-acting elements found that the LCYε group had more cis-acting elements [41] . The lycopene cyclase family is divided into two branches, LCY-ε and LCY-β, which is consistent with previous studies, and interestingly, in algae plants, LCYs is divided into three groups, CrtL-b, CrtL-e, and CrtL, speculating that CrtL may have other important roles [42] LCYs genes were found to be closely related to four types of cotton, and it may be that these four cotton species evolved from a common ancestor and had more homologous genes [43] . For instance, GhLCYε-3 and GhLCYε-1 are a pair of homologous genes. We found that the four cotton species of LCYs protein were closely related to cocoa, but were distantly related to maize. While the LCYs protein of the four cotton species closely relates to cocoa, it is distantly related to maize, possibly due to their common ancestry in the mallow family. LCYs was found to be more conserved during evolution, and it is speculated that such a highly conserved LCYs is important for maintaining the normal function of organisms, so the gene is not easily mutated and is highly preserved [44] . Chromosome mapping analysis further supports the evolutionary link of LCYs,with genes evenly distributed on four chromosomes. The similarity in gene positions strengthens the notion that cotton's heterotetraploid state resulted from doubling chromosomes after A-genome diploid hybridization [45] . LCY-ε and LCY-β groups, despite a large number of similar motifs, showcase an interesting case where (GhLCYε-2, GbLCYε-2) possesses 3–4 motifs less, hinting at potential lost functions during evolution. Notably, the LCYε group containing introns contrasts with the intron absence in the LCYβ group, indicating potential rapid development through reverse transcription or replication [46] . Introns are the parts that are cut off during messenger RNA processing and are not present in mature messenger RNA. Non-coding regions in GrLCYε-1, GrLCYβ-2, GaLCYβ-1, and GrLCYβ-2 reveal potential locations for LCY transcription factors [47] . Since promoters are typically located in non-coding regions, non-coding regions have regulatory functions for gene expression. It has been observed in non-coding regions of GrLCYε-1, GrLCYβ-2, GaLCYβ-1, and GrLCYβ-2 , and we speculate that transcription factors of LCY may be located in these regions [48] . Protein prediction showed that the GhLCYε-3 pit regulated the content of lutein and ABA by interacting with 10 proteins, thereby completing the response to drought stress. Quantitative fluorescence analysis showed that the expression level of LCYs was observed to be quite different from that of the control after PEG treatment of cotton, and it was speculated that LCYs gene played an important role in responding to drought stress in cotton. Two enzymes, lycopene β cyclase and lycopene ε cyclase, are found in most species and play an important role in regulating the abundance of carotenoids [12] , [49] . It is hypothesized that lycopene cyclase ε in cotton can also regulate the content of carotenoids, thereby improving the stress resistance of plants [42] . By comparing LCYs sequences in Arabidopsis , we identified a putative GhLCYε-3 gene for VIGS validation. Lutein serves as an effective antioxidant to protect cells from oxidative damage caused by ROS [50] . Our research establishes a molecular mechanistic model was developed, unveiling GhLCY-3 's role in controlling the synthesis of ABA and lutein ( Fig. 11 ). Drought stress induces an increase in ROS, leading to observed lutein cycle disorders in silent plants, resulting in ROS accumulation and wilted leaves [51] . The inverse relationship between lutein and MDA levels, coupled with elevated MDA and Pro levels in VIGS plants, indicates increased stress and severe damage to cell membranes [52] . The expression of lutein and ABA-related genes may be impacted by the interaction between lycopene β - cyclase and lycopene ε - cyclase [53] . GhLCYε-3 is a key gene in lutein synthesis. exhibits up-regulated expression under drought stress, increasing lutein content and enhancing drought resistance by reducing oxidative damage ( Fig. 11 ). The signaling mechanism during plant growth and development is activated by adversity stress, and this adaptation mechanism of plants to adversity leads to a variety of molecular and physiological changes. In conclusion, the regulation of lutein, MDA and PRO by GhLCYε-3 is an important response mechanism of cotton to drought stress, which provides valuable insights for improving cotton stress resistance. In this research, from lutein and stress resistance, material screening, bioinformatics analysis of LCYs family, gene mining and functional verification, it was observed that the GhLCYε-3 gene responded to cotton drought stress, and its function was verified by VIGS. However, this experiment also has certain limitations, the research is limited to one material and one treatment. Further investigations are warranted to explore the interaction and regulation of GhLCYε-3 with other genes involved in stress response pathways under different abiotic stresses.

Conclusion In this research, aimed at unraveling the regulatory role of lutein in enhancing cotton's drought resistance, conducted a comprehensive evaluation of LCYs gene structure, phylogeny, and expression patterns. Fluorescence quantitative experiments showed that GhLCYs genes responded to various abiotic stresses, and the function of GhLCYε-3 was preliminarily verified by VIGS. In sum, this research provides new insights into the molecular mechanism of gene function and regulation of lycopene cyclase in cotton. The findings not only aid in enhancing the resistance and quality of cotton under drought stress but also offer valuable guidance for cotton cultivation in arid regions.

These authors contributed equally to this work. Drought stress significantly affects crop productivity. Carotenoids are essential photosynthetic pigment for plants, bacteria, and algae, with signaling and antioxidant functions. Lutein is a crucial branch product in the carotenoid synthesis pathway, which effectively improves the stress tolerance of higher plants. lycopene cyclase, a central enzyme for lutein synthesis, holds great significance in regulating lutein production. This research establishes a correlation between lutein content and stress resistance by measuring the drought resistance and lutein content of various cotton materials. To identify which crucial genes are associated with lutein, the lycopene cyclase family (LCYs) was analyzed. The research found that LCYs form a highly conserved family divided into two subfamilies, LCY-ε (lycopene ε-cyclase) and LCY-β (lycopene β-cyclase). Most members of the LCY family contain photoresponsive elements and abscisic acid elements. qRT-PCR demonstrates showed that most genes responded positively to drought stress, and GhLCYε-3 was expressed significantly differently under drought stress. Virus-induced gene silencing (VIGS) assay showed that the content of GhLCYε-3 was significantly increased with MDA and PRO, and the contents of chlorophyll and lutein were significantly decreased in pYL156 plants. The decrease in GhLCYε-3 expression is speculated to lead to reduced lutein content in vivo, resulting in the accumulation of reactive oxygen species (ROS) and decreased drought tolerance. This research enriched the understanding of LCY gene family and lutein function, and provided a new reference for cotton planting in arid areas. Synopsis Lycopene cyclase plays an important role in enhancing the ability of scavenging ROS and drought resistance of plants. Graphical Abstract Keywords Abbreviations Abscisic acid Lycopene cyclase family phytoene synthase phytoene dehydrogenase lycopene ε-cyclase carotene ε –monooxygenase lycopene β-cyclase 9-cis-epoxycarotenoid dioxygenase Protein lutein deficiency Virus-induced Gene Silencing Gossypium arboreum Gossypium hirsutum Gossypium raimondii Gossypium barbadense Quantitative real-time polymerase chain reaction Fragments Per Kilobase of exon model per Million mapped fragments Malondialdehyde Proline

Ethics approval and consent to participate Not applicable. Consent to publication All authors consent this version for publication. Funding This work was supported by The 10.13039/501100012166 National Key Research and Development Program of China (2023YFD1200300). CRediT authorship contribution statement YW conceived the experiments. NK, LX, LS, LF, ZY, and HY performed the experiments. NK, ZY, FY, CR and HH analyzed the data. NK wrote the initial draft of the manuscript. CX, WJ, WS, GL, ZL and WY supervised the project. All the authors read and revised the final manuscript. Acknowledgements The authors acknowledge technical assistance of Cotton Research Institute of the Chinese Academy of Agricultural Sciences. Declaration of Competing Interest The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Supplementary material . . . Data Availability The datasets generated and/or analyzed during the current study are available in the CottonFGD ( https://cottonfgd.org/ ). Additional: Supplementary Table S1 .

CC BY

no

2024-01-16 23:40:17

Comput Struct Biotechnol J. 2023 Dec 22; 23:384-395

oa_package/b2/f0/PMC10788185.tar.gz

PMC10788186

38225983

Introduction Patients with acute ischemic stroke can receive intravenous thrombolysis up to 4.5 h after symptom onset ( 1 ). However, up to 25% of all stroke patients have an unknown time of symptom onset because stroke occurs while sleeping or the time of onset cannot be communicated due to aphasia or a disturbed level of consciousness ( 2 ). Patients with unknown time of symptom onset who are suitable for thrombolysis can be identified by penumbral imaging, i.e., the identification of hypoperfused but potentially salvageable brain tissue ( 3 ). Such evaluation of perfusion imaging requires the application of dedicated software with limited availability ( 4 ). An additional method for the identification of patients with unknown time of symptom onset suitable for thrombolysis is a so-called “tissue clock” approach, whereby a visible lesion mismatch between diffusion-weighted imaging (DWI) and fluid-attenuated inversion recovery (FLAIR) sequences indicates a stroke within the time window of thrombolysis ( 5 , 6 ). This approach is limited by the restricted availability of MRI for acute stroke triage. Recently, in the multicenter MissPerfeCT study, we have established the new and simple CT-based concept of hypoperfusion-hypodensity mismatch ( 7 ). The method is based on evaluating a low net water uptake in native cranial CT. This low net water uptake is considered a marker for the overestimation of the ischemic core in corresponding perfusion imaging ( 8 , 9 ). We showed that this hypoperfusion-hypodensity mismatch, i.e., the absence of a hypodensity on native CT within the hypoperfused core lesion on perfusion CT, identifies patients within the 4.5-h time window with high accuracy ( 7 , 10 ). This approach is easily and rapidly applicable with standard radiological software worldwide and without requiring additional software tools ( 3 , 7 , 11–13 ). We, therefore, performed a further analysis of the MissPerfeCT study and compared the evidence-based method of computed tomography (CT) automated perfusion imaging with the new CT hypoperfusion-hypodensity mismatch approach regarding the accuracy of identifying patients suitable for thrombolysis.

Methods Study design and patients This analysis of the retrospective multicenter MissPerfeCT observational cohort study (08/2009–11/2017) ( 7 ) includes consecutive patients with known onset of symptoms from seven tertiary stroke centers who were clinically diagnosed with acute ischemic stroke and received multimodal CT on admission, including standard native cerebral CT (NCCT), CT angiography (CTA), and perfusion CT (CTP). Consecutive patients were included to reduce the risk of bias. Patients from the following university medical centers were included: Bochum (10/2016–08/2017), Goettingen (10/2016–11/2017), Dresden (05/2015–12/2016), Greifswald (09/2015–10/2017), Luebeck (03/2015–12/2016), LMU Munich (08/2009–06/2012), and Muenster (05/2016–11/2017). Inclusion criteria were as follows: 1. evidence of acute intracranial vessel occlusion (any supratentorial proximal or peripheral artery of the ACA, MCA, or PCA territory) by ischemic perfusion deficit and/or CT hyperdense thrombus and/or CTA vessel occlusion, 2. acute symptoms attributable to the ischemic CTP lesion, and 3. sufficient NCCT quality for judgment of early ischemic hypodensity (potential limitations were old infarcts, severe white matter disease, and movement artifacts); sufficient CTP quality for judgment of the ischemic core lesion (potential limitations were insufficient contrast bolus or movement artifacts), and processing by RAPID automated perfusion software. All datasets of patients fulfilling the inclusion criteria were additionally processed with automated perfusion software (RAPID, RapidAI, Ca, United States), and mismatch criteria were defined according to the Extending the Time for Thrombolysis in Emergency Neurological Deficits (EXTEND) trial criteria (ratio of mismatch >1.2 between CBF < 30% and Tmax >6 s or absolute mismatch >10 mL if total core volume < 70 mL) ( 3 , 11 ). The study was approved by the institutional ethics committee ( Ethikkommission der Ärztekammer Westfalen Lippe ; reference number 2017-233-f-S), which waived informed consent because all identifying information was removed before the retrospective analysis. The local ethics committees of all participating centers gave approval according to their local protocol for sharing retrospective and anonymized data. All study protocols and procedures were conducted in accordance with the Declaration of Helsinki. The MissPerfeCT study is a retrospective observational multicenter cohort study and is registered with clinicaltrials.gov (NCT04277728). This study followed the standards for reporting of diagnostic accuracy (STARD) reporting guidelines. Imaging protocol Patients received NCCT, CTA, and whole-brain CTP performed on 64 or 128 dual slice scanners (Siemens Definition AS+; Siemens Definition Flash, Siemens Healthcare, Forchheim, Germany; Philips Brilliance 64, Philips Medical Systems, Eindhoven, Netherlands). CT: 120 kV, 280 to 320 mA, 5.0 mm slice reconstruction; CTA: 100 to 120 kV, 260 to 300 mA, 1.0 mm slice reconstruction, 5 mm MIP reconstruction with 1 mm increment; and CTP: 80 kV, 200 to 250 mA, 5 mm slice reconstruction (max. 10 mm), slice sampling rate 1.50 s (min. 1.33 s), scan time 45 s (max. 60 s), biphasic injection with 30 mL (max. 40 mL) of highly iodinated contrast medium with 350 mg iodine/mL (max. 400 mg/mL) injected with at least 4 mL/s (max. 6 mL/s) followed by 30 mL of NaCl chaser bolus. All perfusion parameter maps were calculated on a dedicated workstation (Syngo VE52A with VPCT-Neuro; Siemens Healthcare, Forchheim, Germany) based on a deconvolution model by least mean squares fitting, including cerebral blood volume (CBV), cerebral blood flow (CBF), mean transit time (MTT), and time to drain (TTD) ( 14 , 15 ). Image analysis The algorithm to identify a hypoperfusion-hypodensity mismatch was defined as previously published ( Figure 1 ) ( 7 ). It is an easy-to-use method in which two raters, blinded to clinical information, conduct a side-by-side panel comparison. In brief, it consisted of the following steps: First, the total ischemic area was identified by the visual inspection of sensitive MTT or TTD maps. Within this ischemic area of bolus delay, the core lesion, defined as a lesion of high infarct probability, was identified in CBV parameter maps, showing significantly reduced perfusion values of less than 2 mL/ 100 mL or less than 30% relative to the normal side. If no CBV lesion was present, a CBF lesion was used to define the core lesion with significantly reduced perfusion values of less than 30 mL/100 mL/min or less than 60% relative to the normal side. Then, the corresponding region in the NCCT was identified and judged for the presence of a hypodense lesion with respect to the healthy side, consistent with early acute infarct. For this purpose, the NCCT and perfusion CT maps displaying the ischemic core were presented slice by slice because a hypoperfusion-hypodensity match should encompass all slices. The goal was to rate NCCT with high specificity for definite early infarct: in case of doubt as to whether there was a clear hypodensity present, images were rated as an absence of hypodensity. Overall, the judgment about the presence of a CT hypoperfusion-hypodensity mismatch requires 1–2 min. Statistical analysis Patients were classified as eligible or not eligible for thrombolysis. Eligibility was given when patients presented within 4.5 h or for patients beyond 4.5 h when a perfusion mismatch was detected with automated perfusion software according to the EXTEND criteria. We compared patients eligible for thrombolysis with those not eligible by using Pearson’s chi-square test for categorical variables and Student’s t-test or the Mann–Whitney U-test for continuous variables. We calculated the area under the curve (AUC), sensitivity, specificity, positive predictive value, and negative predictive value (with Clopper–Pearson 95% confidence intervals) for the identification of patients eligible for iv-tPA. Statistical analyses were carried out using SPSS (version 26). Data availability statement The data that support the findings of the study are available from the corresponding author upon request.

Results Patient baseline characteristics A total of 247 patients were included in our study, of whom 219 (88.7%) were suitable for thrombolysis and 28 (11.3%) were not suitable for thrombolysis (comprising patients with an onset <4.5 h and those with an onset >4.5 h fulfilling automated mismatch criteria). The median time from onset to CT was 2 h and 42 min (standard deviation 2 h 18 min) in patients suitable and 8 h 16 min (SD 6 h 6 min) not suitable for thrombolysis. Table 1 shows the patients’ baseline characteristics. Patients suitable for thrombolysis more often had hypertension than those not suitable for thrombolysis. Groups were comparable regarding age, sex, other comorbid conditions, and NIHSS score on admission. Presence of hypoperfusion-hypodensity mismatch or automated perfusion mismatch Of 219 patients suitable for thrombolysis, 197 (90.0%) were presented within 4.5 h and 22 (10.0%) beyond 4.5 h ( Table 2 ). Of the 197 patients within 4.5 h of symptom onset, 190 (96.4%) were identified by the presence of hypoperfusion-hypodensity mismatch and 88 (44.7%) by automated perfusion mismatch. Of the 22 patients presenting beyond 4.5 h classified as suitable for thrombolysis by automated perfusion mismatch, 5 (22.7%) were also identified by hypoperfusion-hypodensity mismatch. There were 28 patients who presented beyond 4.5 h and were classified as not suitable for thrombolysis by automated perfusion analysis. Within this group, eight patients (28.6%) were likewise classified as not suitable for thrombolysis by hypoperfusion-hypodensity mismatch. Using hypoperfusion-hypodensity mismatch to classify patients suitable for thrombolysis yielded a sensitivity of 89.0% (95% CI 84.1–92.9), a specificity of 71.4% (95%CI 51.3–86.8), a positive predictive value of 96.1% (95%CI 92.3–98.3), and a negative predictive value of 45.5% (95%CI 30.4–61.2) ( Table 3 ). Automated perfusion mismatch quantification identified patients suitable for thrombolysis with 50.2% (95%CI 43.4–57.9) sensitivity, 100.0% (95%CI 57.7–100.0) specificity, 100.0% (95%CI 96.7–100.0) positive predictive value, and 20.4% (95%CI 14.0–28.2) negative predictive value.

Discussion This multicenter study shows that the assessment of hypoperfusion-hypodensity mismatch identifies patients who are eligible for thrombolysis with higher sensitivity (89.0%) compared with automated perfusion analysis (50.2%) and might thus increase the number of patients treated with thrombolysis among those with unknown time of symptom onset. Of note, the positive predictive value was also very high with the assessment of hypoperfusion-hypodensity mismatch (96.1%), even though it was naturally lower than with automated analysis, where patients with a favorable imaging profile were defined as “ground truth” for eligibility for thrombolysis in this study, inevitably leading to specificity and positive predictive value of 100%. For hypoperfusion-hypodensity mismatch evaluation, the false negative rate was 54.4%, and for automated mismatch analysis, it was 79.6%. This implies that patients eligible for thrombolysis may occasionally be missed by both methods, but less frequently with hypoperfusion-hypodensity mismatch. Thus, in a real-world setting, our approach is likely to perform even better than in this study. The concept of hypoperfusion-hypodensity mismatch is based on the pathophysiology of cerebral ischemia, i.e., the uptake of water into the ischemic brain ( 10 ). This tissue water uptake after cerebral artery occlusion follows a characteristic course that is visualized by a decreasing CT density within the hypoperfused brain region ( 16 , 17 ). We have recently shown that the hypoperfused region and the region of hypodensity usually match after 4.5 h and vice versa, patients without such a match, i.e., a hypoperfusion-hypodensity mismatch, are suitable for thrombolysis ( 7 ). However, from our previous study, it was not clear how this approach performs compared to the automated assessment of perfusion mismatch that was validated in the randomized EXTEND trial ( 3 , 13 ). The present study shows that hypoperfusion-hypodensity mismatch increases the proportion of patients suitable for thrombolysis among those with an unknown time of symptom onset compared to automated perfusion mismatch quantification. This finding is in line with a recent MRI study that found that the yield of a tissue clock imaging approach to select patients eligible for thrombolysis in an unknown time window was double that of ischemic core–perfusion mismatch-based patient selection ( 13 ). One might argue that a tissue clock approach, such as the hypoperfusion-hypodensity mismatch method, might miss patients beyond 4.5 h who are suitable for thrombolysis. However, previous studies suggest that the majority of patients with unknown time of stroke onset have had their symptom onset within 4.5 h ( 16 , 17 ). Thus, the number of patients identified by hypoperfusion-hypodensity mismatch presumably outnumbers patients identified by automated perfusion mismatch quantification. In addition, the present study showed that of the 22 patients with symptom onset beyond 4.5 h who were suitable for thrombolysis, as identified by automated perfusion mismatch, 5 patients were also identified by hypoperfusion-hypodensity mismatch. The low proportion of patients with symptom onset beyond 4.5 h in our study might have contributed to the high positive predictive value. The imbalance between patients with symptom onset within 4.5 h and beyond should therefore be acknowledged as a limitation. Nevertheless, the high proportion of patients with symptom onset within 4.5 h reflects the real-world situation, as previous studies showed that a majority of wake-up stroke patients are very likely to be in the 4.5 h time window ( 18 ). A further potential weakness of our method is that it detects only a small proportion of patients with symptom onset of more than 4.5 h who are suitable for thrombolysis. However, the majority of patients with wake-up stroke (i.e., patients with unknown time of symptom onset) are very likely to be within the 4.5-h time window ( 18 ). It is worth noting that patients with hypertension and coronary artery disease are more likely to be eligible for thrombolysis. Although observed in previous studies, the reason remains unclear ( 7 ). One possible explanation could be that patients with a history of cardiovascular disease may be more vigilant regarding acute cardiovascular events. The multicenter, randomized, double-blind WAKE-UP trial confirmed the rationale of the “tissue clock” approach for the identification of stroke patients with unknown time of symptom onset eligible for thrombolysis ( 6 ). Following this concept, patients with visible changes on DWI but normal FLAIR are likely within the time window of thrombolysis. The WAKE-UP trial showed that thrombolysis in patients with unknown time of symptom onset guided by MRI DWI-FLAIR mismatch results in a significantly better functional outcome ( 6 ). However, compared with MRI, CT is less affected by contraindications (such as pacemakers) and is more generally available in the acute setting in most hospitals that treat acute ischemic stroke patients, and thus is the primary imaging modality used globally. Overall, the hypoperfusion-hypodensity mismatch method has several advantages over existing imaging-based methods for the identification of patients with unknown time of symptom onset who are eligible for thrombolysis, including the speedy accessibility of computed tomography worldwide compared to MRI and the dispensability of automated software tools. A limitation of our study is its retrospective design. However, all images were assessed by readers blinded to clinical information.

Conclusion Eligibility for thrombolysis among stroke patients with unknown time of symptom onset can be determined by the detection of a hypoperfusion-hypodensity mismatch with higher sensitivity compared to automated perfusion mismatch quantification, the gold standard that was established in a randomized trial. Thus, the simple method of hypoperfusion-hypodensity mismatch can potentially increase the proportion of patients with unknown time of stroke onset who are treated with thrombolysis.

Edited by: Wen-Jun Tu, Chinese Academy of Medical Sciences and Peking Union Medical College, China Reviewed by: Gabriel Broocks, University of Hamburg, Germany; Jin Lv, Pla Rocket Force Characteristic Medical Center, China † These authors have contributed equally to this work and share first authorship ‡ These authors have contributed equally to this work and share last authorship Background and purpose Automated perfusion imaging can detect stroke patients with unknown time of symptom onset who are eligible for thrombolysis. However, the availability of this technique is limited. We, therefore, established the novel concept of computed tomography (CT) hypoperfusion-hypodensity mismatch, i.e., an ischemic core lesion visible on cerebral perfusion CT without visible hypodensity in the corresponding native cerebral CT. We compared both methods regarding their accuracy in identifying patients suitable for thrombolysis. Methods In a retrospective analysis of the MissPerfeCT observational cohort study, patients were classified as suitable or not for thrombolysis based on established time window and imaging criteria. We calculated predictive values for hypoperfusion-hypodensity mismatch and automated perfusion imaging to compare accuracy in the identification of patients suitable for thrombolysis. Results Of 247 patients, 219 (88.7%) were eligible for thrombolysis and 28 (11.3%) were not eligible for thrombolysis. Of 197 patients who were within 4.5 h of symptom onset, 190 (96.4%) were identified by hypoperfusion-hypodensity mismatch and 88 (44.7%) by automated perfusion mismatch ( p < 0.001). Of 22 patients who were beyond 4.5 h of symptom onset but were eligible for thrombolysis, 5 patients (22.7%) were identified by hypoperfusion-hypodensity mismatch. Predictive values for the hypoperfusion-hypodensity mismatch vs. automated perfusion mismatch were as follows: sensitivity, 89.0% vs. 50.2%; specificity, 71.4% vs. 100.0%; positive predictive value, 96.1% vs. 100.0%; and negative predictive value, 45.5% vs. 20.4%. Conclusion The novel method of hypoperfusion-hypodensity mismatch can identify patients suitable for thrombolysis with higher sensitivity and lower specificity than established techniques. Using this simple method might therefore increase the proportion of patients treated with thrombolysis without the use of special automated software. The MissPerfeCT study is a retrospective observational multicenter cohort study and is registered with clinicaltrials.gov (NCT04277728).

Data availability statement The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation. Ethics statement The studies involving humans were approved by Ethikkommission der Ärztekammer Westfalen Lippe (reference number: 2017-233-f-S). The studies were conducted in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required from the participants or the participants' legal guardians/next of kin in accordance with the national legislation and institutional requirements. Author contributions PSp: Formal analysis, Visualization, Writing – review & editing, Conceptualization, Investigation, Writing – original draft, Methodology, Project administration. AK: Formal analysis, Visualization, Conceptualization, Investigation, Methodology, Writing – review & editing. LM: Visualization, Writing – review & editing, Formal analysis, Investigation, Methodology, Data curation. CK: Formal analysis, Investigation, Methodology, Writing – review & editing. VP: Formal analysis, Investigation, Methodology, Writing – review & editing. KT: Formal analysis, Investigation, Methodology, Writing – review & editing. MD: Formal analysis, Investigation, Methodology, Writing – review & editing. CL: Formal analysis, Investigation, Methodology, Writing – review & editing. DK: Formal analysis, Investigation, Methodology, Writing – review & editing. SL: Formal analysis, Investigation, Methodology, Writing – review & editing. AB: Investigation, Methodology, Writing – review & editing, Formal analysis. LR: Formal analysis, Investigation, Methodology, Writing – review & editing. WK: Formal analysis, Investigation, Methodology, Writing – review & editing. CB: Formal analysis, Investigation, Methodology, Writing – review & editing. WH: Formal analysis, Investigation, Methodology, Writing – review & editing. JF: Formal analysis, Investigation, Methodology, Writing – review & editing. PSc: Formal analysis, Investigation, Methodology, Writing – review & editing. HW: Formal analysis, Investigation, Methodology, Writing – review & editing. HM: Formal analysis, Investigation, Methodology, Writing – review & editing. MP: Investigation, Methodology, Writing – review & editing, Formal analysis. JM: Conceptualization, Investigation, Methodology, Writing – review & editing, Formal analysis.

Conflict of interest JM has received grants from Deutsche Forschungsgemeinschaft, Bundesministerium für Bildung und Forschung (BMBF), Else KrönerFresenius-Stiftung, EVER Pharma Jena GmbH, and Ferrer International, travel grants from Boehringer Ingelheim, and speaking fees from Bayer Vital and Chugai Pharma. MD has received honoraria for lectures from Bayer Vital and Sanofi Genzyme. Consultant for Hovid Berhad and Roche Pharma. CL received consulting and speaker’s honoraria from Biogen Idec, Bayer Schering, Bristol-Myers Squibb, Daiichi Sanykyo, Merck Serono, Novartis, Sanofi, Genzyme and TEVA. JF has received grants from German Ministry of Science and Education (BMBF), German Ministry of Economy and Innovation (BMWi), German Research Foundation (DFG), European Union (EU), Hamburgische Investitions- und Förderbank (IFB), Medtronic, Microvention, Route92, Stryker. Consultant for: Acandis, Bayer, Boehringer Ingelheim, Cerenovus, Evasc Neurovascular, MD Clinicals, Medtronic, Microvention, Penumbra, Phenox, Stryker, Transverse Medical. Stock holder: Tegus Medical. CK has received honoraria and travel grants from Bayer Vital and Daiichi-Sankyo. DK receives a grant from Else Kröner-Fresenius-Center for Digital Health. PS has received grants from Siemens and Penumbra. Consultant for: Penumbra, Phenox, Stryker, Cerus endovascular. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The author(s) declared that they were an editorial board member of Frontiers, at the time of submission. This had no impact on the peer review process and the final decision. Publisher’s note All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

CC BY

no

2024-01-16 23:40:17

Front Neurol. 2023 Dec 29; 14:1320620

oa_package/e8/02/PMC10788186.tar.gz

PMC10788187

38220147

Background Extreme weather and climate change continue to cause concern among researchers worldwide, and the increase in the frequency and intensity of extreme weather events poses a threat to human health [ 1 – 3 ]. Numerous studies in the past have confirmed the adverse effects of extreme temperatures on human health [ 4 – 8 ]. Recent studies have found the significant associations between non-optimal temperatures and increased risk of mortality of cardiovascular (CVD), nervous system and respiratory diseases [ 9 – 12 ]. Additionally, a study in China reported that non-optimal temperatures could rapidly increase the risk of acute heart attack [ 13 ]. More than five million people worldwide die each year from extreme heat or cold under a changing global climate, accounting for 9.43% of all deaths according to a 20 year cohort study [ 14 ]. Non-optimal temperature is a very important health hazard, and further research is needed to reveal its adverse effects on human health. A wide range of multicenter studies is available in developed countries. Gasparrini et al. found that the burden of non-accidental deaths attributable to non-optimal temperatures was 5.86% and 6.52% in 135 cities in the United States and 51 cities in Spain, respectively [ 15 ]. In recent years, there has also been a growing number of multicenter studies on the effects of temperature on human health in China. For example, Chen et al. found that the attributable burden of cardiovascular disease mortality caused by non-optimal temperatures in 272 cities in China was 17.48% [ 16 ]. Effects of non-optimal temperatures on nervous system diseases were assessed in five Chinese cities [ 9 ]. The effects of non-optimal temperatures on human health were measured in 31 cities [ 17 ], 43 counties [ 18 ], and 66 communities [ 19 ] in China. The Sichuan Basin is located in the south-central part of the Asian continent, in the heart of China. A study found that since 1960, the annual average temperature of the Sichuan Basin has increased at a rate of 0.17 °C per decade [ 20 ]. Some studies predict that by 2030, the Sichuan Basin will be about 1 °C warmer than the decadal average temperature observed in 2000. This warming will continue beyond 2060, and by the end of the century, the annual average temperature will have likely exceeded 20 °C [ 21 ]. In the past, most studies have analyzed the effects of single-city non-optimal temperatures on human health in the Sichuan Basin. For example, Cui et al. found that the attributable risks of respiratory and cardiovascular diseases caused by non-optimal temperatures in Chengdu in the Sichuan Basin were 19.69% and 11.40%, respectively, with cold responsible for a higher proportion of deaths than heat [ 22 ]. Yin et al. found an association between temperature and the incidence of HFMD. High temperatures have acute and short-term effects, while the effects of low temperatures will persist for longer periods of time. Males and children under one year of age were more vulnerable to temperature changes [ 23 ]. Therefore, a multicenter study based on previous single-city studies was conducted to more accurately reflect the effect of temperature on population health in southwest China. We collected data from four cities in the Sichuan Basin, southwest China. We also investigated the impact of both high and low temperatures on the mortality risk of patients with non-accidental diseases, and examined the vulnerable populations by age, gender, education level, and marital status in a stratified analysis. The results of the study will provide valuable information for developing effective intervention measures and public health policies in the Sichuan Basin, southwest China.

Methods Study area The Sichuan Basin, located in southwest China, is one of the four major basins in China. The climate of the basin belongs to the humid sub-tropical monsoon type, with high temperatures in the east and low temperatures in the north and west. In addition, it is one of the most populous regions in China and in the world. The four study cities were Chengdu, Zigong, Guangyuan, and Panzhihua. Chengdu (east longitude: 102°54′–104°53′ and north latitude: 30°05′–31°26′) is located in the central part of the Sichuan Basin and is the provincial capital city of the Sichuan Basin. Zigong (east longitude: 102°02′–105°16′ and north latitude: 28°55′–29°38′) is located in the southern part of the Sichuan Basin. Guangyuan (east longitude: 104°36′–106°45′ and north latitude: 31°31′–32°56′) is located in the northern part of the Sichuan Basin. Panzhihua (east longitude: 101°08′–102°15′ and north latitude: 26°05′–27°21′) is located in the southern part of the Sichuan Basin (Figure 1 ). Mortality data Daily non-accidental deaths for the four cities were obtained from the Population Death Information Registration Management System (PDIRMS), which covers all the mortality information of residents in the four cities. Deaths of residents were confirmed by hospitals or doctors in the residents’ homes, and the data were recorded in the system. Due to availability, the data covered different periods of time in the four study cities. Chengdu and Zigong had records from January 1, 2016, to December 31, 2021; Guangyuan and Panzhihua had records from January 1, 2018, to December 31, 2021. According to the International Classification of Diseases, Tenth Revision (ICD-10), A00-R99 is classified as non-accidental deaths (referred to as “total deaths” in this study; ICD-10: A00-R99). Finally, a stratified analysis was performed by gender, age group (0–64 years old and ≥65 years old), education level (low education or high education), and marital status (married or alternative marriage statuses). Meteorological and air pollution data Daily meteorological data were provided by the Meteorological Administration of each study city. These data included daily maximum, mean, and minimum temperatures (°C) and mean relative humidity (RH%). Daily air pollution data including particulate matter <2.5 μm in aerodynamic diameter (PM 2.5 , 24-h mean μg/m3) and ozone (O 3 , 8-h mean μg/m3) were obtained from municipal environmental monitoring sites in Chengdu, Zigong, Guangyuan, and Panzhihua. Statistical analysis First stage analysis The distributed lag non-linear models (DLNM) was developed by Gasparrini in 2010, a modelling framework that can simultaneously represent non-linear exposure-response dependencies and delayed effects [ 24 ]. A Poisson-distributed distributed lag non-linear model was used to evaluate the association between the extreme temperature and non-accidental mortality for each city in our study. In this study, the minimum mortality temperature (MMT) was used as the reference temperature, which corresponds to a minimum mortality percentile between the first and the 99th percentiles, was derived from the best linear unbiased prediction of the overall cumulative exposure-response association in each location [ 15 ]. We referred to the MMT as the optimum temperature, and used it as the reference for calculating the attributable risk.The model was as follows: In this model, Yit represents the number of non-accidental deaths in city i on day t ; cb(Tempit) is the cross-base matrix generated by DLNM with a maximum lag time of 25 days, which included a quadratic B spline with three internal knots placed at the 25th, 50th, and 75th percentiles of location specific temperature distributions, and the lag response curve with a natural cubic B spline with an intercept and three internal knots placed at equally spaced values in the log scale [ 15 , 25 ]; ns(RH,4), ns(PM 2.5 ,4), and ns(O 3 ,4) represents natural cubic spline functions with four degrees of freedom; RH is the daily average relative humidity; ns(Time i , df*year) means spline functions with eight degrees of freedom per year to control for seasonal and long-term trends; DOW means the day of the week effect. Second stage analysis In the second stage, we used a multivariate meta-analysis to obtain summary estimates for the four cities [ 26 ]. A BLUP approach involved a trade-off between city-specific associations and second-stage pooled estimation, providing more precise estimates, especially in cities with small numbers of deaths [ 16 ]. We then calculated the number of attributable deaths and the proportion of attributable deaths during the present day and 25 lagged days according to previous studies [ 27 ]. Attribution fractions for cold (below MMT) and heat (above MMT) were calculated by deeming the MMT as the baseline reference. Based on the cutoff values of the 1% and 99% temperature percentiles and the MMT, we divided the temperature of each city into four levels, i.e., extreme cold (lower than the 1st percentile of the daily mean temperature), moderate cold (from the 1st percentile of the daily mean temperature to the MMT), moderate heat (from the MMT to the 99th percentile of the daily mean temperature), and extreme heat (greater than the 99th percentile of the daily mean temperature). Finally, the 95% empirical confidence intervals (eCIs) of attributable mortality were computed using Monte Carlo simulation [ 27 ]. Additionally, we also stratified the analysis by gender, age group, education level, and marital status through the above steps. Sensitivity analysis To test the stability of the model, we observed the exposure-response relationship by adjusting the maximum lag days (21/25 days), the df of the time trend (7–9), and we also controlled for ozone and fine particulate matter (PM 2.5 ) in the sensitivity analysis. We used R software 4.1.2 for data analysis. Specifically, the “dlnm” package [ 25 ] was used to estimate city-specific temperature-mortality associations, and the “mvmeta” package [ 28 ] was used for the meta-analysis.

Results Table 1 shows descriptive data for total non-accidental deaths, mean temperature, minimum temperature, maximum temperature, mean relative humidity, and air pollutants for four cities in the Sichuan Basin during the study period. The total number of deaths recorded in the four cities was 751,930. The total numbers of deaths in Chengdu, Zigong, Guangyuan, and Panzhihua were 530,228, 127,446, 68,061, and 26,195, respectively. The number of daily non-accidental deaths in each city varied widely, from 18 in Panzhihua to 242 in Chengdu. The daily mean temperature and daily mean relative humidity in the four cities were 18.3 °C (−1.6 °C–34.6 °C) and 72.4% (11.8%–100%), respectively. The mean concentrations of PM 2.5 and O 3 were 41.6 μg/m3 (3.8 μg/m3–300.8 μg/m3) and 85.4 μg/m3 (5.0 μg/m3–278.0 μg/m3), respectively. Table 2 shows the overall and heat and cold estimated attribution fractions for the four cities. Overall, the MMP was 80% (25.1 °C), and 10.16% of non-accidental mortality was attributed to cold and heat. The MMP distribution was around 80% in all cities, and the MMT distribution was between 20 °C and 30 °C. The majority of temperature-related non-accidental deaths were attributed to cold, accounting for 9.10% (95% eCI: 5.50%, 12.19%), and heat effects accounted for only 1.06% (95% eCI: 0.76%, 1.33%). This difference was caused by a higher minimum mortality percentile, with most daily mean temperatures being below the MMT. Figure 2 shows the best linear unbiased predicted (BLUP) estimates of the exposure–response relationships between temperature and non-accidental mortality in the four cities, with corresponding MMT and the cutoffs to define extreme and moderate temperatures. The temperature-mortality association curve was described by a U or L shape. The RRs of extreme heat and cold were higher; the cold effect lasts longer, and most of the daily mean temperatures were distributed below the MMT. Figure 3 shows the effects of the extreme heat (31.8 °C) and extreme cold (1.1 °C) on non-accidental mortality at lags of 0–25 days. The effect of extreme heat is immediate and disappears after 3–4 days. The effect of extreme cold lasted for a long lag period. Figure 4 presents the attributable fractions of low and high temperatures for non-accidental mortality in subgroups by sex, age, education level, and marital status. Overall, AF was significantly higher for moderate temperatures than for extreme temperatures. The greatest proportion of non-accidental deaths were attributable to moderate cold (from 6.23% to 12.30%), with only a small proportion of non-accidental mortality attributable to extreme cold (0.24%–0.48%) or extreme heat (0.10%–0.37%). Those over 65 years, females, and those with a low education level or alternative marriage status were more sensitive to extreme temperatures. The mortality burden attributable to non-optimal temperatures was higher among those under 65 years old, females, and those with low education level or alternative marriage status, with overall AF of 12.42% (95% eCI: −1.85%, 21.90%), 12.37% (95% eCI: 6.71%, 17.43%), 10.54% (95% eCI: 6.56%, 11.11%), and 14.74% (95% eCI: 9.56%, 18.97%), respectively.

Discussion Our study assessed the associations between low and high temperatures and non-accidental mortality using more than 0.7 million deaths from four cities in the Sichuan Basin of southwest China, and we further estimated the mortality burden attributable to heat and cold. The results showed that both heat and cold significantly increased the mortality risk. Overall, the MMP was 80%, and 10.16% of non-accidental deaths could be attributed to non-optimal temperatures. The majority of temperature-related non-accidental deaths were caused by cold temperature at 9.10%, with heat effect accounting for only 1.06%. We observed that low temperature had a greater effect on mortality and lasted longer, while the heat effect was immediate, consistent with previous studies [ 29 , 30 ]. Stratified analysis by age, sex, education level, and marital status showed that those over 65 years of age, females, and those with low education level or alternative marriage status were more sensitive to extreme temperatures, while the mortality burden attributable to non-optimal temperatures was higher among those under the 65 years old, females, and those with a low education level or alternative marriage status. Many studies have found that the effects of temperature on human health were generally characterized by “U,” “V,” and “J” shapes [ 31 , 32 ]. Our study found that temperature was associated with mortality in a “U” or “J” shape, with both hot and cold increasing the risk of death, and low temperature having a higher effect than high temperature. This is consistent with previous findings [ 15 , 16 , 33 ]. For example, Chen et al. found that the attributable percentages of total non-accidental mortality, cardiovascular disease, and respiratory disease mortality caused by non-optimal temperature in 272 cities in China were 14.33%, 17.48%, and 10.57%, respectively. The mortality risk of extreme cold temperature lasted for more than 14 days, whereas the risk of extreme hot temperature appeared immediately [ 16 ]. Gasparrini et al. also found that the total proportion of deaths caused by heat and cold was 7.71% in 384 locations, and more temperature-attributable deaths were caused by cold (7.29%) than by heat (0.42%) [ 15 ]. A study by Scovronick et al. found that 3.4% of deaths (∼290,000) in South Africa were attributable to non-optimum temperatures during the study period, with heat effects occurred immediately after exposure and diminished rapidly, while cold effects were delayed and persistent [ 33 ]. Temperatures in the Sichuan Basin are higher than in other regions at the same latitude, with extreme high temperatures in the southeastern part of the basin often exceeding 40 °C. Our results showed that 10.16% of non-accidental deaths were attributable to non-optimal temperatures, similar to the findings of Gasparrini [ 15 ] and Ma [ 10 ], lower than those of Chen [ 16 ] and Zhang [ 34 ], and higher than those of Scovronick [ 33 ] and Cao et al. [ 35 ] The different results may be related to the different study locations, socioeconomic characteristics, life patterns, medical conditions and demographic characteristics. According to Bannister’s study, under the CMIP5 high-emission future climate change scenario, the temperature in the Sichuan Basin is expected to increase by about 4 °C by 2100, and more frequent extreme temperature events could have adverse effects on population health [ 21 ]. Therefore, it is important to develop adaptation plans to reduce the adverse effects of climate change. Previous studies have confirmed that individual social factors (e.g., age, sex, and marital status) can modify the effect of temperature on mortality [ 36 , 37 ]. Our study found that the elderly over 65 years, females, and those with a low education level or alternative marriage status were more sensitive to extreme temperatures. This is consistent with some previous studies [ 33 , 38 , 39 ], but inconsistent results were observed in other studies [ 9 ]. A study in South Africa found that people over the age of 65 were more sensitive to extreme temperatures [ 33 ]. Physiological studies have shown that most elderly individuals have comorbidities (co-morbidities) that make them more sensitive to extreme temperatures or temperature changes than younger adults. Moreover, elderly have weaker neurothermoregulatory mechanisms and reduced thermoregulatory capacity and may not be able to maintain their core temperature at safe levels, and prolonged exposure to temperature extremes may lead to associated diseases or other fatal events [ 40 – 42 ]. We found that women were more susceptible to the effects of heat and cold than men. Previous studies have yielded mixed results, with some studies finding a higher risk of temperature-related mortality in women than in men [ 16 , 36 , 43 ]. However, some studies also reported a greater sensitivity to extreme temperatures in men [ 9 ]. Previous studies found that people with low or no education were at higher risk of heat stroke [ 37 , 44 ], consistent with our results. Education may be an indicator of socioeconomic status, and this may be associated with poor baseline status, limited health care coverage, and associated housing conditions [ 38 ]. Only a few studies have analyzed the modifying effect of marital status on temperature in the past. Son et al. found a higher risk of death due to heat and cold in widows [ 37 ]. Our results showed that individuals with alternative marriage status were more susceptible to extreme temperatures. In cold conditions, body core temperature drops, resulting in the depletion of physical reserves of the heart, liver, muscles, breathing, and heartbeat, which may manifest as shivering, respiratory depression, cardiac arrhythmias, and impaired mental functions, causing vessel spasm development and inefficient circulation, even progressing to cardiac arrest or coma [ 45 – 47 ]. One study found that hypothermia deaths were twice as frequent as hyperthermia deaths [ 48 ]. Under conditions of high temperature, the thermoregulatory capacity of the body may be exceeded, leading to illness due to overheating. If the resultant body water deficit is not adequately replenished, it can lead to dehydration. This may lead to heat stroke and increased cardiovascular strain, and even death [ 49 , 50 ]. In addition, high internal body temperature (39–40 °C), increased ischemia, and oxidative stress after blood redistribution can lead to cell, tissue, or organ damage, with organs such as the brain, heart, kidneys and lungs being at greater risk [ 4 ]. Our study suggest that the local government should pay more attention to vulnerable people and take measures to reduce adverse effects in extreme climates. On the one hand, various forms of targeted activities for publicity and education ought to be carried out. Provide scientific knowledge on the correct response and self-rescue in the event of extreme weather to enhance the public’s awareness of self-protection. On the other hand, local government should actively carry out assessment and investigation of the impact of extreme weather on the health of the population, improve and optimize the early warning system of heat and cold waves health risks. In order to ensure the stability of the study results, we conducted sensitivity analyses. The lag days were chosen to be 21 and 25 days; the time trend degrees of freedom was from 7 to 9, and the mean relative humidity and pollutants ranged from 3 to 5 (Table S1 ). The AF calculated according to different df were similar. Therefore, the results calculated by the model were reliable. To our knowledge, this is the first study to explore the association between non-appropriate temperatures and mortality in a multi-city area of the Sichuan Basin in southwest China. Second, the mortality data were derived from the PDIRMS and were therefore authentic and reliable. In addition, we conducted subgroup analyses, which were more comprehensive and included age, sex, education level, and marital status, to better reveal how susceptible populations are affected by non-optimal temperatures. However, our study also has some limitations. Since ecological studies have inherent limitations, we could not obtain individual exposure data. In addition, our study included four cities in the central, northern, and southern parts of the Sichuan Basin in southwest China, while other cities were not included. Therefore, the results may not generalize to other regions.

Conclusions Our study indicates that a significant association between non-optimal temperature and non-accidental mortality. Most temperature-related deaths were caused by low temperatures, with moderately high and low temperatures representing the majority of the mortality burden. Those under 65 years old, females, and those with a low educational level or alternative marriage status had the highest attributable burden. Our findings may be helpful to policymakers at local levels in developing adaptation plans.

Yizhang Xia, Chunli Shi and Yang Li made the same contributions to this manuscriptspondence Background There are few multi-city studies on the association between temperature and mortality in basin climates. This study was based on the Sichuan Basin in southwest China to assess the association of basin temperature with non-accidental mortality in the population and with the temperature-related mortality burden. Methods Daily mortality data, meteorological and air pollution data were collected for four cities in the Sichuan Basin of southwest China. We used a two-stage time-series analysis to quantify the association between temperature and non-accidental mortality in each city, and a multivariate meta-analysis was performed to obtain the overall cumulative risk. The attributable fractions (AFs) were calculated to access the mortality burden attributable to non-optimal temperature. Additionally, we performed a stratified analyses by gender, age group, education level, and marital status. Results A total of 751,930 non-accidental deaths were collected in our study. Overall, 10.16% of non-accidental deaths could be attributed to non-optimal temperatures. A majority of temperature-related non-accidental deaths were caused by low temperature, accounting for 9.10% (95% eCI: 5.50%, 12.19%), and heat effects accounted for only 1.06% (95% eCI: 0.76%, 1.33%). The mortality burden attributable to non-optimal temperatures was higher among those under 65 years old, females, those with a low education level, and those with an alternative marriage status. Conclusions Our study suggested that a significant association between non-optimal temperature and non-accidental mortality. Those under 65 years old, females, and those with a low educational level or alternative marriage status had the highest attributable burden. Supplementary information The online version contains supplementary material available at https://doi.org/10.1265/ehpm.23-00118 . Keywords:

Abbreviations particulate matter <2.5 μm in aerodynamic diameter daily 8-hour mean concentrations of ozone relative humidity minimum mortality temperature minimum mortality percentile attributable fractions distributed-lag non-linear models standard deviation relative risk confidence interval Supplementary information Declarations Ethics approval and consent to participate Not applicable. This study does not involve experimental animals or individual information on human subjects. Consent for publication Not applicable. Availability of data and material The datasets used in this study are available from the corresponding author upon reasonable request. Competing interests The authors declare that they have no competing interests. Funding This study was funded by the Sichuan Provincial Cadre Health Care Research Project (No. 2021-1801), and the Sichuan Provincial Cadre Health Care Research Project (No. ZH2018-1801). Author contributions XYZ, SCL and LY coordinated the study, performed data analysis, and drafted the manuscript; RSJ, JXY and CY contributed to the statistical analyses; HW, GXF, XR, LMJ, SHY, PXJ and XRQ assisted in obtaining data; CJY coordinated the study and edited the manuscript; and ZL organized and coordinated the study and edited the manuscript. All authors have read and approved the final manuscript. Acknowledgments We thank the Center for Disease Control and Prevention of Chenhdu, Zigong, Guangyuan and Panzhihua for providing data.

CC BY

no

2024-01-16 23:40:17

Environ Health Prev Med.; 29:1

oa_package/90/71/PMC10788187.tar.gz

PMC10788188

38226039

Background Pistacia chinensis and weinmannifolia have long been used as important medicinal plants and are considered the best potential resources for developing new natural products for pharmaceutical purposes. However, genomic and genetic information are very limited for P. chinensis and not yet available for P. weinmannifolia . RNA-seq analysis was performed to provide genomic information of the two Pistacia species and also to characterize and compare differences in important genes, metabolic pathways, and networks associated with the production of secondary metabolites between the two species.

Experimental Design, Materials and Methods Sample collection Fresh leaves were collected from fully-grown P. chinensis and P. weinmannifolia in June 2015 at Yunnan, China. Leaf samples were submerged into liquid nitrogen, transferred into RNAlater solution for long-term transportation (Ambion Ins, USA), and then stored in −20 °C freezer until used for RNA extraction. Transcriptome sequencing Leaf samples removed from RNAlater solution were ground with a pestle and mortar in liquid nitrogen to extract total RNAs using TRIzol reagent (Thermo Fisher Scientific, Korea). Total RNAs were checked for quantity and purity using an RNA Pico Chip on the Agilent 2100 Bioanalyzer (Agilent Technologies, USA). A 10 μg of the total RNA was used for mRNA isolation using oligo-dT beads, the isolated mRNA was randomly sheared, and the adaptor was added by ligating to 3′ A overhang of the sheared mRNA for cDNA synthesis. The mRNA isolation and cDNA library construction was conducted by following the procedure of the SureSelect strand-specific RNA reagent kit (Agilent, USA). An equal quantity of mRNA from three different leaf samples was pooled and used for cDNA library construction. Quality check of the cDNA library was examined by Agilent DNA 1000 chip (Agilent Technologies, USA), and the Illumina Hiseq 2500 (Illumina, USA) was used for sequencing the cDNA library. De novo assembly and annotation The raw reads were trimmed and filtered to remove adaptor sequences, empty reads, and low quality reads with ≤20 of a phred quality score and ≤50 bp in length using NGS tool kits and Trimmomatic tool [2] . The filtered reads were assembled using three different assemblers, CLC Genomics Workbench (ver. 3.7.1), Trinity (release 20110519), and Velvet-Oases (ver. 1.1.04-ver. 0.1.21). A default k-mer value (25-mer) was used for the assembly with the CLC Genomics Workbench. For the assembly by Trinity and Velvet-Oases, different k-mer values (25–33 for Trinity; 21–79 for Velvet-Oases) were applied to obtain the best results. All contigs from each assembler were merged separately for further processing. As Oases does not cluster assembled contigs, CD-HIT-EST was used to cluster the contigs with an identity of more than 90 % and coverage of 100 % [3] . All data sets from each assembler were finally combined into a single dataset by collapsing identical contigs into a single contig using CD-HIT-EST with the same criteria described above. Contigs longer than 300 bp were annotated by running local BLAST with a cutoff E-value of 10 −6 against the NCBI non-redundant (NR) protein database ( http://www.ncbi.nlm.nih.gov ).

Pistacia chinensis and Pistacia weinmannifolia are small trees and are distributed in East Asia, in particular China. The data on P. chinensis presented in this article is associated with the research article, “DOI: 10.5010/JPB.2019.46.4.274 ” [1] . Both P. chinensis and P. weinmannifolia have long been used as ethnobotanical plants to treat various illnesses, including dysentery, inflammatory swelling, rheumatism, liver diseases, influenza, lung cancer, etc. Many studies have been carried out to delve into the pharmaceutical properties of these Pistacia species using plant extracts, but genomic studies are very rarely performed to date. To enrich the genetic information of these two species, RNA sequencing was conducted using a pair-end Illumina HiSeq2500 sequencing system, resulting in 2.6 G of raw data from P. chinensis (Accession no: SRR10136265) and 2.7 G bases from P. weinmannifolia (Accession no: SRR10136264). Transcriptome shotgun assembly using three different assembly tools generated a total of 18,524 non-redundant contigs (N50, 1104 bp) from P. chinensis and 18,956 from P. weinmannifolia (N50, 1137 bp). The data is accessible at NCBI BioProject: PRJNA566127. These data would be crucial for the identification of genes associated with the compounds exerting pharmaceutical properties and also for molecular marker development. Keywords

Specifications Table Value of the Data • The data provides transcripts as the first transcriptome reference for P. weinmmanifolia , and as for P. chinensis it will enrich genomic information since only a couple of transcriptome analyses have been reported. • These data will be useful for studying genes differentially expressed in Pistacia species and useful for plant taxonomists to classify Pistacia species more accurately. • These data can be used to identify compounds with pharmaceutical properties and to identify genes associated with synthetic pathways of the compounds of interest. • Also, these datasets can be used to understand whether differentially expressed genes are related to distinctive therapeutic effects of each Pistacia species and related genera. Data Description The RNA sequencing data presented in this article are obtained from leaf samples of two different Pistacia species, P. chinensis and P. weinmannifolia . The raw RNA-seq data were deposited at NCBI Sequence Read Archive (SRA) database under the accession SRR10136265 for P. chinensis and SRR10136264 for P. weinmannifolia . This Transcriptome Shotgun Assembly (TSA) projects have been deposited at DDBJ/EMBL/GenBank under the accession GKQH00000000 for P. chinensis and GKQG00000000 for P. weinmannifolia . The TSA version described in this paper is the first version, GKQH01000000 and GKQG01000000. The raw and assembled RNA sequencing data are summarized in Table 1 . The assembled contigs were annotated by running local BLAST against the NCBI non-redundant (NR) protein database. The assembled contigs showed that many of them contain only partial coding regions as shown in Fig. 1 . Comparison of two transcriptomes showed that at least 8825 contigs are highly similar in these two transcriptomes ( Fig. 2 ). Limitations None. Ethics Statement The current work does not involve human subjects, animal experiments, or any data collected from social media platforms. CRediT authorship contribution statement Azeez Bimpe Suliyat: Data curation, Writing – original draft. Dewi Komang Anggita: Data curation. Hee Soo Yang: Data curation. Sang-Woo Lee: Data curation, Resources. Wan Yi Li: Resources. Sang-Ho Choi: Data curation, Conceptualization. Ki-Young Choi: Writing – review & editing. Jong-Kuk Na: Conceptualization, Supervision.

Data Availability Transcriptome sequecing of two Pistacia species (Original data) (NCBI Bioproject) Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Acknowledgements This research was supported in part by the KRIBB Initiative Program (KGM4582322) of the Republic of Korea.

CC BY

no

2024-01-16 23:40:17

Data Brief. 2023 Dec 22; 52:110002

oa_package/1e/49/PMC10788188.tar.gz

PMC10788189

38226037

Background Biochar and wood vinegar production using the pyrolysis process has become popular due to the usefulness of the product for various applications, particularly in the agricultural sector. However, a design is needed that is applicable for smallholder farmers, and a study is required to characterise the data about the product made using this material. This study was conducted to develop a small-scale field pyrolyser that could be used to produce biochar from agricultural residues such as coconut shells, mango pruning, and carambola pruning. This was followed by a characterisation study to evaluate each of the products obtained using this process. The information can be used as references to guide stakeholders on the different qualities of the biochar produced using the system. The design and data could both be useful to farmers seeking to emulate the system for field biochar and wood vinegar production, as well as in selecting the raw materials best suited to eventual use in related agricultural activities.

Experimental Design, Materials and Methods Pyrolyser design and development The pyrolysis system used in this study comprised a two-stage production chamber with 10 kg of capacity for the initial raw materials ( Fig. 1 , Fig. 2 , Fig. 3 , Fig. 4 ). The main chamber produced biochar, while the second chamber, also known as the condensation chamber, produced pyroligneous liquid or wood vinegar through the condensation of white smoke, a by-product of the pyrolysis process. The biochar production chamber design was based on a previous study involving a bottom-up self-heating pyrolysis method practised by farmers [4 , 5] . Meanwhile, the condensation chamber design was based on the cyclonic concept and aimed to optimise the amount of pyroligneous liquid collected. According to [6] , the cyclone technique used to be the predominant way to remove particles from gas streams. Data about the utilisation of pyroligneous liquid or wood vinegar was reported in [7] . The pyrolyser was designed and drawn using SolidWorks software version 2020. Having finalised the design, fabrication was outsourced to a third party. The main production chamber was built using stainless steel, while the condensation chamber was built using a mixed-steel material. The design and drawing were introduced in [8] as a means of disseminating information among local farmers and stakeholders. It was registered as copyrighted work for public goods (ref no: CR2022/40/105) and made free for public access and reuse. Collection of raw materials and pyrolysis set-up The three biomasses used in this study - coconut shells, carambola pruning, and mango pruning - were obtained from local sources. The coconut shells were obtained from a local wet market in Serdang, Selangor, Malaysia, while the carambola and mango pruning were obtained from research plots at MARDI headquarters, Serdang, Selangor, Malaysia. The pyrolysis was undertaken using a self-heating process, whereby a small amount of biomass (e.g., coconut shells) was ignited before a chain of charring process occurred. The system underwent pyrolysation without depending on external heat sources [4] . For a batch of 10 kg capacity input in the biochar chamber, pyrolysis occurred slowly for between one and two hours. Carbonisation occurred under an uncontrolled temperature, whereby the carbonised biomass itself would provide the heat required for a pyrolysis chain to continue. To measure the temperature inside the pyrolysis chamber, a portable thermocouple was locally fabricated and assembled (TH Muda Engineering) before being used. The recorded temperatures ranged from 195 °C at the beginning to reach a maximum of 667 °C. These data were previously reported in [5] . Since the previous development, the design of the condensation chamber has been improved to enhance the collection of pyroligneous liquid or wood vinegar ( Fig. 1 , Fig. 2 , Fig. 3 ). Additionally, low-temperature pyrolysis was defined as temperatures within this range of pyrolysis temperatures [9] , [10] , [11] . The upper part of the condensation chamber had a small outlet whose function was to release excess white smoke ( Fig. 4 ). In conventional farmers’ practice (although this has yet to be documented), the white smoke has the beneficial function of repelling pests on their farms. By observation, the visible smoke normally disappeared within a three-metre radius of the chamber. Data collection and analysis for elemental composition, physicochemical, calorific value, and surface area characterisation The powdered raw materials and biochar samples (triplicate) were taken for nutrient analysis to determine the percentages of the carbon, hydrogen, nitrogen, and sulphur elements. For each sample, 30 mg of the substrate was mixed with 30 mg of tungsten and wrapped in metal foil. The samples were then analysed using an elemental analyser (vario MACRO cube). Meanwhile, the pH and electrical conductivity (EC) were analysed using the Eutech P700 (Eutech Instruments), while the ash content (percentage) was determined using a standard ashing method through incineration in a muffle furnace at 500 °C. Calorific content data were obtained by outsourcing the analysis to an accredited laboratory, MARDI Laboratory (MARDILab) at Serdang, Selangor, Malaysia. Additionally, the BET surface characteristics were determined using Micromeritics® Tristar II Plus. This analysis was carried out at the Universiti Putra Malaysia (UPM) laboratory. The chemical structure was analysed using SEM (FEI Quanta 400). The SEM analysis was carried out at the Universiti Kebangsaan Malaysia (UKM) laboratory. The SEM data presented in this study are shown at one hundred (100X) and five hundred times magnification (500X). Adsorption characteristics of biochar using fentin acetate and cypermethrin The adsorption study was conducted using a standard jar test method, whereby initial concentrations of 10 mg/l of each pesticide compound solution (from fentin acetate and cypermethrin) were prepared separately. Into the jars containing a pesticide concentration solution were added 0.1 g of biochar sample (in triplicate) from either the coconut shells, carambola pruning, or mango pruning. Following this, the mixed samples and pesticide solutions were shaken vigorously in an incubator shaker. The mixed solutions were sampled at the intervals of 0, 0.5, 1, 2, 4, and 24 h. For each sampling process, the concentrations of fentin acetate and cypermethrin were determined using liquid chromatography–mass spectrometry (LC-MS/MS) (Sciex 5500). Data analysis The elemental composition, physicochemical, and calorific values data were analysed using descriptive statistical analyses, whereby the mean and standard deviation (SD) values were calculated using Microsoft Office Excel (MS Excel). Meanwhile, the statistical model for the adsorption study was determined using the statistical analysis system (SAS) version 9.3 (SAS Institute, Inc., USA). An analysis of variance (ANOVA) was conducted, whereby different sources of biochar (treatment) were compared using a Duncan's test.

Biochar production is an effective approach to managing abundant agricultural wastes. Pruning wastes from trimming the branches of trees such as carambola and mango, as well as coconut shells, are among the agricultural wastes that have reutilisation potential, which would simultaneously reduce the space required for disposal. In this study, the potential use of these wastes by converting them into biochar was investigated. The data presented in this study highlight the design of a pyrolysis system for a low-temperature slow pyrolysis process, as well as the characterisation data of the biochar produced using this system. The data collected included the elemental composition, porosity, as well as surface and adsorption characteristics of the biochar. These data indicate that the biochar produced had certain qualities that would enable its use for specific agricultural and industrial purposes. Meanwhile, the design indicated that it could facilitate small farms with specific outputs. In brief, these data can be used as references for developing a small-scale system for agricultural waste management using different types of crops. Keywords

Specifications Table Value of The Data • The data describe the characteristics of biochar produced from coconut shells, carambola pruning, and mango pruning using low-temperature slow pyrolysis with an upward pyrolysation process. • Upward pyrolysation with self-heating is a common technique for producing biochar, especially among farmers in Malaysia. Therefore, the data could be compared with other available sources in the market. • The data also provide valuable information about biochar qualities that would enable it to be used for specific agricultural uses. Users and farmers can benefit from the application of specific types of biochar. • In addition to the widely used coconut shell biochar, biochar obtained from carambola and mango pruning demonstrated their potential use as added-value products due to the characteristics identified in this study. Data Description The data included in this study incorporate a design of a small-scale pyrolyser to accommodate the management of farm waste and the characterisation of the biochar produced. The study was conducted at the Malaysian Agricultural Research and Development Institute (MARDI) headquarters in Serdang, Selangor, Malaysia. The data have been organised into seven figures and three tables. The two main products used in the pyrolysis were biochar and pyroligneous liquid (wood vinegar). The raw materials used were coconut shells, carambola pruning, and mango pruning. Fig. 1 , Fig. 2 , Fig. 3 , Fig. 4 show the design of the pyrolyser used in this study, with four (4) perspectives showing the top view, front view, side view, and rear view, as well as the design measurements and the actual view of the pyrolyser after completion. The data can be accessed from the data files entitled “All views.png”, “Front views.png”, “Design measurements.png”, and “Actual view.png” in the Mendeley repository. Meanwhile, Table 1 , Table 2 , Table 3 show the elemental composition, physicochemcial properties, calorific values, and porosity analysis of the biochar from the three biomasses used in the study: (a) coconut shells, (b) carambola pruning, and (c) mango pruning. The data for these analyses can be accessed from the data files entitled “CHNS Analyses.xlsx”, “Physicochemical and calorific value.xlsx”, and “Porosity BET.xslx” in the Mendeley repository. The data indicated that the elemental composition of the raw materials was comparable with that of other materials [1 , 2] . Meanwhile, the carbon content of the biochar was higher than that of the raw materials and surpassed 50% elemental composition, indicating that the product materials (biochar) conserved high percentages of carbon. Table 3 shows that the biochar from the carambola pruning had the highest surface area characteristics. Fig. 5 shows the SEM analysis of the biochar produced from these three sources. This figure also highlights uniform and visibly structured hollow pores within the biochar obtained from the carambola pruning. For further reference, the SEM figures can be accessed in the file “SEM Biochar.png”, which is in the Mendeley repository. Further characterisation studies were undertaken. Fig. 6 , Fig. 7 show the results from the study of the adsorption characteristics of the biochar, which were obtained using two different agricultural pesticides, a) fentin acetate and b) cypermethrin. The purpose of the study was to evaluate how effectively the biochar would remove the recalcitrant pesticide from the effects of agricultural run-off [3] . The results show that all three sources of biochar demonstrated effective adsorption capacity in a 24-hour adsorption study, based on an initial pesticide concentration of 10 mg/L. A significant adsorption result was obtained for the biochar made from mango pruning. The data for these analyses were kept in the file entitled “Adsorption study.xlsx” in the Mendeley repository. Limitations This design could only accommodate small-scale biochar production at 10 kg per batch of initial raw materials. Therefore, in order to evaluate the quality of the biochar produced for validation purposes, additional analysis is required of the data characteristics obtained from a larger design that can process over 10 kg of initial raw materials. In addition, the current study focused on only one location as a source of raw materials. In the future, further studies should be conducted using raw materials from different location sources so that more datasets can be obtained. Overall, the data from this study can be used as a benchmark for future characterisation studies. Ethics Statement This work involved no human or animal studies, so an ethics statement is not applicable. CRediT authorship contribution statement Mohammad Shahid Shahrun: Conceptualization, Methodology, Formal analysis. Mohammad Hariz Abdul Rahman: Project administration, Methodology, Formal analysis, Writing – original draft, Writing – review & editing. Nur Adliza Baharom: Methodology, Formal analysis. Fauzi Jumat: Methodology, Formal analysis. Mohamad Jani Saad: Visualization, Investigation. Mohd Fazly Mail: Methodology, Formal analysis. Norziana Zin Zawawi: Methodology, Funding acquisition. Farah Huda Sjafni Suherman: Project administration, Funding acquisition. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper

Data Availability Design of a pyrolysis system and the characteristics of biochar produced from selected agricultural wastes (Original data) (Mendeley Data) Acknowledgements The authors would like to thank Mr Mohd Firdaus A.Wahab @ Othman, Mr Mohd Ghazali Rusli, Mr Mohamad Shahir Supian, and Mr Ashraaf Siddiqi for their technical assistance in the study. A special acknowledgement is extended to the Malaysian Agricultural Research and Development Institute (MARDI) for the research funds granted under the RMK-11 Developmental Fund (PRH-403) and RMK-12 Developmental Fund (PRP-515).

CC BY

no

2024-01-16 23:40:17

Data Brief. 2023 Dec 21; 52:109997

oa_package/e1/b7/PMC10788189.tar.gz

PMC10788192

38226029

Experimental Design, Materials and Methods Questionnaire Structure • The data were collected using an anonymously administered questionnaire. • The questions were designed to be simple and unambiguous, such as ‘How old are you (years)?’ or ‘What Nigerian ethnic group do you belong to?’ • The questionnaire had three question types - open-ended, closed-ended, and 5 point likert scale questions. 10 of the 43 questions were open-ended, for which text fields were provided into which the respondents could type in their responses. One such question was regarding the symptoms being experienced by the respondents. Responses provided here could be passed into artificial intelligence tools for automatic symptom elicitation and inferences. 25 of the 43 questions were closed-ended questions, with respondents provided dropdown options to choose from. For instance, a question such as ‘What is your marital status?’, had four possible response options (‘Single’, ‘Married’, ‘Divorced’, ‘Widow/Widower’) from which the respondent could choose. The remaining 8 questions were 5 point likert scale, ranging from 1 (strongly disagree) to 5 (strongly agree). Data Collection Method • All data were collected via askNivi, a conversational health tool accessible via WhatsApp. The primary purpose of askNivi is to provide health education and referrals for healthcare systems [1] . • Participants (respondents) were patients or visitors to healthcare centres in villages and towns in northern Nigeria. • Respondents were offered incentives to complete the questionnaire in the form of mobile airtime credit (as most mobile lines are prepaid in Nigeria). During the pilot study, the 1093 respondents were split into 3 groups of 364, 364, and 365, respectively. They were then offered NGN 500, NGN 1500, and NGN 2500 (USD 0.50, USD 1.50, USD 2.50) as incentives to fill the questionnaire. However, no significant difference in response rate was observed across the three groups. At NGN 500, only 84 of the 364 participants (23%) responded; at NGN 1500 there were only 77 (21%) responses, and only 70 (19%) responded at NGN 2500. Based on this, respondents were offered an incentive of NGN 500 during the main data collection phase. • During the main data collection phase, an askNivi link was sent to the respondents mobile phones via whatsapp. Literate respondents filled out the questionnaires by themselves while non-literate respondents (who chose to participate), were assisted by literate family members or administrative staff of the healthcare centres. • Data were collected between September 2022 and March 2023.

Gender equity, particularly in healthcare, has been gaining increasing attention in recent years. The goal is to ensure that everyone has equal access to quality healthcare services irrespective of age, gender, or socio-economic status. However, most countries in sub-Saharan Africa struggle to meet this goal, due to several challenges, including poverty, poor infrastructure, and gender-bias. Using Nigeria as a case-study, it is common knowledge that gender inequality and discrimination is predominant in the northern region of the country. This work sought to gather data to assess the level of healthcare accessibility from a gender-based perspective in northern Nigeria. Data were sourced anonymously from residents in about 500 locations across the northern region of Nigeria, using WhatsApp-based questionnaires, in two phases and two languages - English and Hausa. About 4700 participants took part in the survey and each had to answer 43 questions, split into demographic, socio-economic, wellness check, and diversity, equity, and inclusion (DEI) in health care services obtained. Keywords

Specifications Table Value of the Data • This data is useful for assessing gender-based healthcare accessibility in northern Nigeria, a region where gender discrimination is prevalent. • Data analysis can be carried out on the data to assess the impact of various socio-economic factors, such as marital status, level of education and income levels, on access to healthcare. This can help answer questions such as “does being well educated or wealthy improve the chances of receiving prompt services from health care professionals in rural areas?” • The data can be used to train artificial intelligence tools, such as Natural Language Processing (NLP) and/or Named Entity Recognition (NER), to automatically identify prevalent medical symptoms in the region of study. • Using advanced NLPs, ChatBots could be built to serve as triage solutions for first-line medical respondents. Data Description This article describes the datasets collected from respondents in northern Nigeria regarding equitable healthcare in the region. Data was collected in English and Hausa languages. The curated data files, saved in Microsoft Excel (XLSX) format, and associated questionnaire (in PDF format) are available in Ref. [ 2 ]. There are a total of 6 data files in the repository, the first MS Excel file contains responses from an initial pilot study. The second two and third files contain responses from the main study in English and Hausa languages respectively. The fourth MS Excel file contains English interpretations of Hausa words in the Hausa data file, while the fifth file is a codebook describing the variables in the other four data files. The sixth file is the questionnaire in PDF format. The collected data were from several towns and villages across the northern region of Nigeria as shown in Fig. 1 . Data were collected using questionnaires administered using WhatsApp. The questionnaires had 43 questions, split into four sections, (i) demographic information, (ii) socio-economic information, (iii) wellness check, (iv) healthcare DEI; and a mix of three questions types, viz.: (i) open-ended questions, (ii) closed-ended questions with dropdown options (e.g., “yes/no”, “employed/unemployed/self-employed”), (iii) scaled questions (5 point likert scale). Table 1 provides a concise summary of the 43 questions types and corresponding variables in the datasets. 8 additional variables are included, named as ‘variableName_Translated’, which are related to the fourth dataset (English translation of the Hausa dataset). These additional variables are the English translations of their corresponding variables in Hausa language. Table 2 summarises the datasets and gives information of the collection period. Table 3 shows a summary of the distribution of the respondents per dataset and sheds light on the sample population considered. Regarding the response rates, of the number of respondents who participated in the data collection process, 21% completed the survey and answered all questions, while 53% provided partial responses. Partial responses means that certain questions were left unanswered or skipped. Finally, data collection was a one-off process for each respondent, with no deadlines set or reminders sent to respondents. Though the data provided here are intended as inputs or precursors to more detailed analyses, some quick insights can be drawn from them. For instance, from the distribution of respondents in the data, as shown on Table 3 , it can be seen that despite the widespread stereotyping and repression of women in northern Nigeria, more women participated in the survey than any other genders. This perhaps suggests that providing a safe space and an enabling environment might be instrumental in tackling the repression against women. Further, the distribution of the respondents also reveals that a large percentage of the population are youths aged 35 years and younger, who live in Towns and have at least secondary school level education. Despite this youthful population, the high level of insecurity and constant insurrections in the region are perhaps responsible for limiting the economic power of residents, as most earn less than USD 500 annually. Limitations • In northern Nigeria, women are often not allowed to go out on their own or interact with the general public without a male chaperone (in the person of her father, husband, or brother). This severely limits womens' freedom of expression and ability to provide unbiased / uninfluenced responses to the questions. • Non-binary gender is not legally recognized in Nigeria, hence the small number of respondents who identified as non-binary. • In the region, men are often not as open to discussing their personal / medical conditions as women, hence why there were more female respondents than other genders. • During the pilot study, the data collection process was not well moderated hence there were lots of incorrect or invalid responses in the data. • For the open-ended questions, there were several typographical errors in the provided responses which made automatic preprocessing a challenge. To use the data for further analysis, significant manual preprocessing would be required. For example, omission of spaces between words or misspelt words, especially those related to the medical symptoms, can be challenging. Ethics Statement 1. Data collected were from respondents who gave their consent to participate. All respondents were presented with a first page on the Nivi app, which clearly stated that the data being collected was for research and analytic purposes, and would remain completely anonymous without any form of tracking. All respondents were required to accept these terms and conditions before being allowed to participate in the data collection exercise. A copy of this consent form has been submitted with this article. 2. This study is institutional review board exempt, as the data collection process relied exclusively on surveys and data was collected anonymously, such that the identity of the respondents cannot be ascertained [ 3 ]. 3. Data were collected using askNivi, an app specifically designed for healthcare related data collection. Privacy policy, consent, and related information about askNivi can be found at www.nivi.io/privacy-policy 4. In the region of interest, northern Nigeria, almost 50 % of girls are married and are parents to multiple kids by the age of 15 [ 4 , 5 ]. This is also reflected in our collected data. For this study, the investigators ensured that all respondents were at least 15 years old. CRediT authorship contribution statement Chika Yinka-Banjo: Conceptualization, Methodology, Supervision. Mary Akinyemi: Validation, Methodology, Writing – review & editing. Olasupo Ajayi: Investigation, Software, Writing – original draft. David Tresner-Kirsch: Data curation, Software, Methodology.

Data Availability Data on Gender-Equitable Healthcare Accessibility in Northern Nigeria (Original data) (Mendeley Data). Acknowledgments This work was supported by USAID and DAI Global - “Nivi and the University of Lagos (UNILAG): Partnering to Create a Gender-Aware Auditing Tool.” Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:40:17

Data Brief. 2023 Dec 18; 52:109979

oa_package/9d/ac/PMC10788192.tar.gz

PMC10788193

38226161

Introduction There is growing interest in the use of aquatic eDNA for studying population genetics of focal species. 1 , 2 , 3 To date much of this research has focused on variation in target regions of the mitochondrial genome, and it has been possible to resolve spatial genetic structure in a range of vertebrate species. 2 , 3 , 4 , 5 Mitochondrial DNA (mtDNA) has been employed for population genetic inference in this context as it is reliably amplified from eDNA and sequenced, and the relatively high evolutionary rate allows populations to be distinguished. Moreover, since variable sections of mtDNA are often flanked by phylogenetically conserved regions, useful primers can be readily identified, and eDNA-derived sequences can be assigned to focal species. 6 However, mtDNA may be suboptimal for resolving genetic structure among very recently diverged populations; it is typically maternally inherited and non-recombining, so the whole mtDNA genome acts as a single locus and primarily provides only information on female ancestry. 7 Furthermore, nuclear insertions of mitochondrial origin (numts), or selection on the mitochondrial genome, can also confound population genetic inference. 7 , 8 Thus, there has been interest in utilizing nuclear eDNA for population genetic analysis. 3 , 9 , 10 , 11 , 12 To date, support for nuclear eDNA as a reliable tool for population genetic inference is limited, although it has been shown that microsatellite allele frequencies of round goby ( Neogobius melanostomus ) derived from eDNA have a strong correspondence to allele frequencies of source fish, in both mesocosms and the natural environment. 11 Importantly, it has also been shown that eDNA-derived microsatellite allele frequencies of round goby can be used to describe structure among geographically separated populations, and that this pattern closely resembles that derived from microsatellite-based analyses of tissue samples. 12 Here we build on this research by investigating if population genetic structure can be determined over a small spatial scale within a single lake using single nucleotide polymorphism (SNP) variants from the nuclear genome.

STAR★Methods Key resources table Resource availability Lead contact Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Martin Genner ( [email protected] ). Materials availability The study did not generate new unique reagents. Data and code availability • Sequence data have been deposited at the Sequence Read Archive and are publicly available as of the date of publication. Accession numbers are listed in the key resources table . • All original code is publicly available as of the date of publication. DOIs are listed in the key resources table . • Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Experimental model and study participant details The study was based on analyses of environmental DNA samples and pre-existing genotyping data, it did not use experimental models or study human participants. Method details Sampling eDNA in Lake Masoko Water samples were collected in sterile containers using SCUBA from five depths in Lake Masoko (3, 7, 12, 18 and 22 m below the water surface) on the 3 September 2019 ( Tables S1 and S3 ). In total we collected nine samples from 3 m depth, nine samples from 22 m depth, and four from each of 7 m, 12 m and 18 m depths. We used 25 of those 30 samples for the work reported here ( Tables S1 and S3 ). All samples were collected from one location on the southeast of the lake (9.3362°S, 33.7566°E). A thermocline was present at the time of sampling, although temperatures were not recorded. Water samples were taken to the shore for processing. Two samples of bottled drinking water were processed as field controls. Research permission was issued by the Tanzania Commission for Science and Technology. Each sample was filtered through a 0.22 μm Sterivex-GP polyethersulfone (PES) filter (Merck Millipore) using repeated loadings of a 60 ml syringe. We aimed to maximise the volume of water sampled in each filter, with the final volume reached when the filter became clogged. The total volume of water sampled in each filter was recorded. Water was expelled, and the outlet of the filter was sealed by melting the plastic. Environmental DNA within the filters was preserved by adding 0.37 ml of ATL buffer to the filter cartridge (Qiagen) using a 1 ml syringe, before a combistopper was used to seal the inlet of the Sterivex. Each sample was then individually placed in a WhirlPak bag, labelled, and stored in shade and away from heat sources. For long-term storage, filters were placed in -20°C freezer three days after collection. A detailed field collection protocol is at https://zenodo.org/record/4687985 . DNA extraction Environmental DNA was extracted from Sterivex filters using a modified DNeasy Blood and Tissue kit (Qiagen) protocol ( https://doi.org/10.5281/zenodo.4741283 ), in a dedicated trace DNA extraction laboratory. Each aliquot of extracted eDNA was passed through a OneStep PCR inhibitor removal column (Zymo Research). The quantity of eDNA in each sample was then quantified using spectrophotometry (NanoPhotometer N60-Touch, Implen). ONT sequencing of microbial communities in eDNA In total nine samples were used for ONT (Oxford Nanopore Technologies) sequencing of the bacterial community. These covered the sampled depth range, and included three samples from 3 m depth, three samples from 22 m depth, and one from each of 7 m, 12 m and 18 m depths ( Table S1 ). Using a starting volume of 500 ng of DNA in 10 μl, we added 15μl of nuclease-free water, and used DNA repair (NEBNext FFPE DNA Repair Mix, NEB) and end-prep (NEBNext Ultra II End repair/dA-tailing Module, NEB) following the manufacturer’s protocol. The samples were then subject to a bead-clean up (30μl AMPure XP SPRI reagent, Beckman Coulter: 30μl of sample) using 75% fresh ethanol. After clean-up, 1μl was quantified using a Qubit 3.0 fluorometer with a Broad Range Assay (Invitrogen). The DNA was then adaptor-ligated (NEBNext Quick Ligation Module, NEB) and cleaned again using AMPure XP beads (as above). An ONT buffer (either SFB or LFB) was used to wash beads and enrich fragments. Samples were stored in LoBind tubes (Eppendorf) and 1μl from each sample was quantified using the Qubit fluorometer, ensuring that they contained between 3 and 20 fmol. Samples were sequenced on Flongle flow cells for MinION (Oxford Nanopore Technologies) following the manufacturers protocol, ensuring a minimum active pore number of 60 prior to loading. Illumina-based genotyping of fish eDNA In total, 16 samples were used for analyses of fish SNP allele counts in the eDNA. These covered the sampled depth range, and included five samples from 3 m depth, five samples from 22 m depth, and two from each of 7 m, 12 m and 18 m depths ( Table S3 ). We used a set of 100 pairs of PCR primers flanking regions containing SNPs, as identified by, 14 to amplify target sequences ( Table S2 ). The primers were assigned to 26 different groups according to annealing temperatures, and PCR reactions were performed in multiplex for each group. Three replicate PCRs were performed on each eDNA template with each primer group. Each PCR was conducted in a 10 μl volume comprising: 5 μl AmpliTaq Gold 360 Master Mix (Applied Biosystems); 0.5 μl forward primer from each group (5 μM); 0.5 μl reverse primer from each group (5 μM); 3 μl molecular-grade water; and 1 μl eDNA template. Thermocycling initially comprised a polymerase activation step at 95°C for 10 min. This was followed by 40 cycles of: denaturation at 95°C for 30 s, annealing (estimated annealing temperature plus 3°C–4°C) for 30 s, extension at 72°C for 60 s. The final extension was at 72°C for 10 min. The eDNA extractions, pre-PCR preparations, and post-PCR procedures were carried out in separate rooms. We tested the following controls for amplification: two field negatives (the bottled water filtered in the field), two laboratory extraction negatives (new Sterivex filters that underwent the extraction protocol), and one PCR negative consisting of molecular biology grade ultrapure water. There was no indication of an amplification product in any negative control visible on agarose gel. PCR products from all 78 amplifications of each sample were pooled. These were loaded on agarose gels (2%), and amplicons excised and purified using the QIAquick Gel Extraction Kit (Qiagen) and the Oligo Clean & Concentrator kit (Zymo Research) using a modified version of the manufacturer’s protocol ( https://doi.org/10.5281/zenodo.10083982 ). Illumina library preparation was conducted using xGen UDI-UMI adapters (IDT), ligated to the amplicons using the PCR-free KAPA HyperPrep Kit (Roche) following the manufacturer’s protocol. A total of 16 libraries using unique indexes were created. Libraries were then quantified individually using a NEBNext Library Quant Kit for Illumina (NEB) qPCR assay, standardized, and sequenced on an Illumina NextSeq 500 using v2.5 (75 bp paired-end reads) high output chemistry and 10% phiX spike-in. Quantification and statistical analysis Bioinformatic analyses of microbial community ONT sequences Base-calling and quality filtering of ONT data were performed using Guppy software v4.09 ( https://nanoporetech.com ). Sequencing adaptors were removed using porechop v0.2.4 ( https://github.com/rrwick/Porechop ). For assignment of reads to microbial taxa, we used KrakenUniq v0.7.3 36 employing the minikraken_20171019_8GB microbial database ( https://ccb.jhu.edu/software/kraken/ ). Reads were filtered to only include only those assigned to bacteria at the order level, which provided sufficient resolution for a broad interpretation of ecological preferences of constituent taxa in samples. To quantify community structure across the depth gradient we used a canonical correspondence analysis using the R package vegan v2.5-7, 37 testing the association of the primary axes of variation with depth using the `anova.cca` function, with 10,000 permutations. Bioinformatic analyses of genomic variants in eDNA Demultiplexed Illumina sequencing reads were trimmed to remove adaptors using cutadapt v4.1. 38 They were then mapped to an indexed Astatotilapia calliptera reference genome (fAstCal1.2; GCA_900246225.3) using bwa v0.7.17-r1188. 39 Using SAMtools v1.9, 40 SAM files were converted to BAM format using the view function, sorted using the sort function, read group information added using the addreplacerg function, and indexed using the index function. Mapping rates were determined using the flagstat function. Ninety-eight primer pairs amplified genomic fragments located in the fAstCal1.2 genome based on a standard nucleotide BLAST (blastn; https://blast.ncbi.nlm.nih.gov/Blast.cgi ) search ( Table S2 ). We extracted variants from a VCF file of 648 Lake Masoko A. calliptera genomes (Sequence Read Archive bioprojects PRJEB1254, PRJEB10014 and PRJEB27804) aligned to the same ref. 18 using a BED file containing the genomic coordinates of the 98 targeted fragments and VCFtools v0.1.16. 41 This process yielded a set of 120 biallelic SNPs within the 98 loci. Across this set of 648 fish and across all chromosomes there was only weak linkage between SNPs ( r 2 = 0.071), but linkage within chromosomes was substantially greater ( r 2 = 0.672). We counted the number of reads assigned to each allele at SNPs within the eDNA BAM files using the ASEReadCounter function in gatk v4.3.0. 42 These allele count data were then manually curated into a combined file, including all samples, including total depth of the relevant alleles for each SNP ( Table S4 ). We retained 71 focal SNPs that were represented in at least 75% of 16 eDNA samples. We tested if reference allele frequencies of the 71 SNPs in eDNA were significantly associated with reference allele frequencies of SNPs in the fish for each of five depth strata ( Table S3 ). We mapped our eDNA samples to the same depth strata used for fish collections [3 m eDNA collection = 0-5 m fish collection; 7 m eDNA collection = 5-10 m fish collection; 12 m eDNA collection = 10-15 m fish collection; 18 m eDNA collection = 15-20 m fish collection; 22 m eDNA collection = 20-25 m fish collection]. We calculated the average reference allele frequency of each SNP in samples from each depth within the eDNA data. We extracted the 71 focal SNPs present in the eDNA data from the VCF file of 648 aligned Lake Masoko A. calliptera individuals. Next, we separated this SNP-subsetted VCF file into five separate VCF files consisting of subsets of individuals sampled at different depths using the view function of BCFtools v1.8. 40 We calculated the allele frequency among individuals from each depth range using the freq2 function in VCFtools v0.1.16. 41 We compared allele frequencies within each depth stratum using the Pearson correlation coefficient (cor.test function) in R v4.2.3. 43 We determined whether these 71 “eDNA SNPs” were able to resolve genetic structure over the Lake Masoko depth gradient, using genetic information of fish collected from known depths (530 of the 648 individuals). First, we pruned SNPs in linkage disequilibrium from the WGS data (3,107,901 of 3,881,258 variants removed) using plink v1.90b6.2, 44 with a window size of 50 kb, a step size of 10 bp, and pruning sites above an r 2 threshold of 0.1. We used Admixture 1.3.0 45 to calculate ancestry proportions of individuals based on this set of 773,357 unlinked SNPs, assuming two ancestry components (K = 2). We compared these ancestry proportions to those of the eDNA samples calculated using the 71 focal SNPs. We summarized variation across the 71 SNPs in the 16 eDNA samples ( Table S3 ) with a Principal Component Analysis (PCA) using the R package pcaMethods v1.90.0. 46 Finally, we compared shifts in allele frequencies from shallow (3 m eDNA; 0-5 m fish) to deep (22 m eDNA, 20-25 m fish) habitats between the eDNA and WGS datasets using the Pearson correlation coefficient (cor.test function in R).

Results Our study system is Lake Masoko, a crater (maar) lake in southern Tanzania that has no surface connections with nearby rivers 13 , 14 ( Figures 1 A and 1B). The lake is ∼700 m in diameter, has a maximum depth of ∼35 m with rapid attenuation of light with increasing depth ( Figure 1 E). The lake exhibits seasonal stratification, with a thermo-oxycline located at approximately 15 m ( Figures 1 C and 1D). It has been proposed that there is stronger stratification during the warmer wetter season (approximately September–March) compared to the cooler and drier season (approximately April–August), 15 although the consistency of this pattern across years is unknown. We collected environmental DNA using SCUBA from five depths in September 2019 (3, 7, 12, 18 and 22 m). To confirm two ecologically distinct water masses are present, we used Oxford Nanopore Technologies (ONT) sequencing on nine samples of bulk environmental DNA ( Table S1 ). We mapped the derived reads to a microbial database (minikraken_20171019_8GB; https://ccb.jhu.edu/software/kraken ) and quantified the number of reads assigned to each microbial order. We found a significant association between the composition of the microbial community and increasing depth (Canonical Correspondence Analysis, Anova F 1,7 = 4.582, p = 0.009), with the major axis of variation (CCA1) capturing a switch in community composition between 10 and 20 m ( Figure 2 A), coincidental with the known position of the thermo-oxycline at approximately 15 m ( Figures 1 C–1E). This shift in the functional composition of the microbial community reflected the change in environment, with shallow, well-oxygenated, warmer and more brightly lit waters possessing a higher proportion of photosynthetic cyanobacteria (e.g., Synechococcales), and deeper, poorly oxygenated, cooler and darker waters possessing a greater dominance of anaerobic bacteria (e.g., Enterobacterales) ( Figure 2 B). The lake contains a pair of ecomorphs of the cichlid fish Astatotilapia calliptera . A littoral ecomorph with yellow males dominates the shallow waters (<5 m) and feeds primarily on benthic invertebrates, while a deep ecomorph with blue males dominates the deeper waters (>20 m) and feeds primarily on zooplankton. 14 , 16 , 17 The ecomorphs show strong differences in ecologically relevant morphologies, including body shape and lower pharyngeal jaws – which are used principally for processing prey. 14 , 16 Using whole genome sequence (WGS) data of individuals collected from known depth strata, the ecomorphs have been shown to be genetically divergent, 18 and these genomic data show the prominent shift in the relative abundance of ecomorphs over the depth gradient, 18 although the total abundance of each ecomorph at each depth is not known. Using our eDNA from the five depths (3, 7, 12, 18, and 22 m), we tested if we could identify genomic DNA from the focal species in the environmental DNA. We also tested whether it was possible to determine differences in allele frequencies at SNP loci that had previously been inferred to be segregating between ecomorphs along the depth gradient. 14 We focused our analyses on a subset of 120 SNPs which had previously been identified within a set of 98 loci across the A. calliptera genome. 14 To amplify loci containing our target SNPs, we used PCR on 16 eDNA samples ( Table S2 ), with an average amplicon length of 84.86 bp (range: 49–109 bp). The resulting amplicons were sequenced using an Illumina NextSeq 550. The generated sequences were mapped to the A. calliptera genome (fAstCal1.2), with an average mapping success of 52.56% of reads (range 14.70–97.89%; Table S3 ). The mapped reads were filtered to include only target regions, and within them we located 114 of the 120 SNP positions in the eDNA read data. We used read counts to quantify variation (allele frequencies) in each of the focal SNPs. Of these 114 SNPs, 102 were variable in the eDNA samples. We further filtered the SNPs to only include those present in 12 or more of the 16 samples. This resulted in a set of 71 SNPs, that we refer to here as “eSNPs”. We tested if these eSNPs can resolve population genetic structure over the depth gradient in fish samples collected from known depths between 2013 and 2018, that have had their whole genomes sequenced (n = 530 individuals 18 ; 3,881,258 biallelic SNPs). Individual ecomorph ancestry estimates using 773,357 unlinked SNPs showed a clear break in population genetic structure between 12 m and 18 m, coincident with the location of the thermo-oxycline ( Figure 3 A). Using the 71 eSNPs, it was possible to resolve a similar pattern across fish from the five depth strata, although with a less polarized assignment of individuals to the two population genetic clusters ( Figure 3 B). If allelic compositions of known SNPs in fish were reflected in eDNA, we expected a positive association between the frequency of the reference allele (i.e., the allele matching the reference genome) in the fish from the whole genome data, and the frequency of the same allele in the eDNA, within each depth band. Focusing on the eSNPs, at each depth, there was a significant positive association between the reference allele frequencies between the eDNA and WGS datasets ( Figure 3 C; Table 1 ), confirming the non-random nature of the allelic composition of eDNA samples. A Principal Component Analysis (PCA) of the eDNA genetic variation across the eSNPs revealed clear differences between the deepest and shallowest sample groupings ( Figure 3 D). To determine whether genetic variation among the ecomorphs corresponded to the variation among the eDNA eSNPs, we compared the change in WGS and eDNA eSNP reference allele frequencies for fish sampled from 0 to 5 m to those sampled at 20-25 m. Overall, we found a significant positive association between changes in the allele frequencies in the eDNA, and the changes in allele frequencies of the fish (Pearson’s correlation, n = 71, r 2 = 0.268, p = 0.0237; Figure 3 E).

Discussion Our study demonstrates that the clear depth differences in genetic structure of the Lake Masoko A. calliptera ecomorphs are reflected in eDNA present in their immediate environment. This pattern is not consistent with complete homogenization of DNA in the water column, although we consider it plausible that there is transport of cichlid fish eDNA between the depth zones via passive mechanisms (for example, fecal particles) or active processes (for example the movement of their predators). Such processes may partially explain the variation in the association between allele frequencies observed in sampled fish (WGS) and the allele frequencies observed in the eDNA. It seems most likely that the differences in eDNA-based allele frequencies are due to strong associations of A. calliptera individuals to specific shallow or deep-water habitats, with little migration between depths – at least during the sampling period when there was stratification of the water column. Although the movements of these fish have not been individually tracked, there is clear support for habitat-specific adaptation in adults, from phenotypic, 14 , 16 , 17 genomic 14 , 16 , 18 and transcriptomic 16 , 17 evidence. Environmental regimes, including light, oxygen and food resources have plausibly led selection to favor philopatry, with strong negative effects on fitness for individuals with inappropriate phenotypes for specific habitats. 19 Notably, work to date showing strong genetic and phenotypic differences between the ecomorphs has been focused on adult individuals, but the divergence in eDNA allele frequencies between the depth zones either side of the thermo-oxycline is consistent with philopatry at all life history stages. The focal species A. calliptera is a maternal mouthbrooder, 20 and also exhibits a short period of parental care after first release of free-swimming fry, so there is scope for habitat imprinting that may contribute to strong life-long habitat affiliation. Whether the eDNA we detected is disproportionately contributed by adults, subadults or juveniles is, however, unclear. While larger individual adults can generate eDNA at a greater rate than individual juveniles, 21 contributions to total eDNA by each life stage will depend on their relative abundance, biomass and activity levels. 22 Evidence of spatial structuring of allele frequencies within eDNA across the small spatial scale of less than 20 m clearly confirms the potential for eDNA to be used to describe fine scale spatial population structure in aquatic species. Studies in both freshwater and marine systems have also revealed the potential for fine scale differences in environmental DNA composition over narrow depth gradients, but those studies have focused on community structure. For example, in shallow Canadian lakes (13–30 m deep), there was clear differentiation in the fish species resolved between depth intervals when the lakes were thermally stratified – matching the distribution of the fish species – however, the differences in community structure among depth intervals were not present when the lakes were mixed. 23 Similarly, in a New Zealand fiord, stratification in a halocline led to differentiation in fish, crustacean and echinoderm community composition resolved using eDNA over a depth only 4 m. 24 Even in the absence of a clear water mass boundary differences fish community structure have been resolved over a depth gradient of only 10 m (Californian kelp forests 25 ). It is becoming clear therefore that eDNA may be useful to describe not only gradients in diversity across larger spatial scales as is commonly recognised, 26 but also over fine spatial scales in some systems, such as Lake Masoko, that are characterized by limited water movement and/or stratification. Our study shows that sufficient information can be gained from nuclear SNP allelic frequencies from environmental DNA to infer population genetic structure. Studies that have explored the capacity for population genetics using eDNA data until now have focused on mtDNA, 4 sequence capture by hybridization 9 or nuclear microsatellites. 11 , 12 We were able to benefit from a-priori knowledge of segregating SNPs in our focal species within a well characterized environment. Targeting SNPs dispersed throughout the nuclear genome in this manner has an advantage over shotgun sequencing eDNA as it enables taxonomic specificity – but may suffer from PCR biases. Use of DNA hybridization-capture techniques to select target regions of interest from eDNA, as has been applied to amplifying species-specific mitochondrial 27 , 28 and nuclear 9 regions from eDNA, may enable us to overcome implicit PCR biases. Reference and re-sequenced genomes enable identification of SNPs for target species suitable for eDNA-based population genetics. 29 More widespread availability of genomes from non-target species in communities would help to confirm the source species of sequenced fragments. In Lake Masoko, the dominant fish species is Astatotilapia calliptera , but three other fish species are present, 30 including the cichlid Coptodon rendalli (estimated to have diverged from A. calliptera 18 Mya 31 ), the cichlid Oreochromis squamipinnis (estimated to have diverged from A. calliptera 23 Mya 31 ), the catfish Clarias gariepinus (estimated to have diverged from A. calliptera 224 Mya 31 ). In addition, there are many domestic and wild terrestrial vertebrates in the Lake Masoko crater generating environmental DNA that can enter the water. It is possible that primers would also amplify fragments of eDNA from these or other sources, which may vary spatially in the lake, and this may explain the high variation in mapping success (samples ranged from 14.7 to 89.5%; Table S3 ). We focused on known biallelic SNPs with previously characterized allelic variation, and this will have reduced the influence of heterospecifics on our results. Implementation of data filtering steps that remove reads that map more closely to heterospecifics in the environment would be additionally beneficial, and this process would ideally benefit from a bespoke array of heterospecific reference sequences for the study site. In principle, population-genomic methods using aquatic environmental DNA could become important for studying species that are challenging to sample due to rarity, catchability or habitat occupancy, 1 or because their capture and direct sampling may be ethically questionable. 32 This study has contributed to this field of research by confirming that nuclear loci can be reliably amplified from environmental DNA, and that it is possible to generate allele frequencies that can be used for population genetic inference. To conclude, our results strengthen indications that future research aimed at refining eDNA-based population genomic methods may improve our understanding of population structure of many species, including those of commercial, ecological and conservation importance, and within marine, estuarine and freshwater systems. Limitations of the study Certainly, the results will be influenced by a multitude of factors, including the eDNA collection, preservation, extraction, PCR, sequencing and bioinformatic methods. We may expect PCR to distort allele frequencies, and the approach we used did not enable us to check for consistency across PCR replicates. There are also likely to be seasonal differences in the ability to resolve spatial population genetic structure, related to the strength and distribution of stratification, and breeding seasonality may also influence the density of eDNA in the water 33 , 34 (observational evidence indicates our focal species A. calliptera breeds from March to May in Lake Masoko 30 ). There are also questions surrounding the efficacy of eDNA-based methods for informing on population level genetic processes, such as selection and mutation. Notably, the approach we used is similar to a pooled-sample sequencing approach (pool-seq), albeit with an unknown number of contributing individuals, and there may be potential to refine approaches for eDNA-based allele frequency analyses using those developed for pooled sequencing. 35

Lead contact Summary There is considerable potential for nuclear genomic material in environmental DNA (eDNA) to inform us of population genetic structure within aquatic species. We tested if nuclear allelic composition data sourced from eDNA can resolve fine scale spatial genetic structure of the cichlid fish Astatotilapia calliptera in Lake Masoko, Tanzania. In this ∼35 m deep crater lake the species is diverging into two genetically distinguishable ecomorphs, separated by a thermo-oxycline at ∼15 m that divides biologically distinct water masses. We quantified population genetic structure along a depth transect using single nucleotide polymorphisms (SNPs) derived from genome sequencing of 530 individuals. This population genetic structure was reflected in a focal set of SNPs that were also reliably amplified from eDNA — with allele frequencies derived from eDNA reflecting those of fish within each depth zone. Thus, by targeting known genetic variation between populations within aquatic eDNA, we measured genetic structure within the focal species. Graphical abstract Highlights • Intraspecific genetic variation can be identified within aquatic environmental DNA • Allele frequencies in eDNA match those in local source populations • Environmental DNA can be used to quantify population genetic divergence Environmental science; Genetics; Techniques in genetics; Evolutionary biology Subject areas Published: December 7, 2023

Supplemental information Acknowledgments We thank staff of the Tanzania Fisheries Research Institute for support during fieldwork. We thank the researchers who contributed to the collection of fish and generation and analysis of whole genome data from Lake Masoko Astatotilapia calliptera , including Emília Santos, Grégoire Vernaz, Richard Durbin, Eric Miska, and Hannah Munby. We thank Julie Johnson for fish paintings. M.J.G. and G.F.T. were supported by 10.13039/501100000270 NERC award NE/S001794/1. Z.L. was supported by a studentship from the 10.13039/501100004543 China Scholarship Council . We thank Christy Waterfall and Jane Coghill from the 10.13039/501100000883 University of Bristol Genomics Facility for sequencing support. Author contributions Conceptualization, R.A.C. and M.J.G. Data curation Z.L., R.A.C., and M.J.G. Formal Analysis, Z.L., R.A.C., and M.J.G. Funding Acquisition, Z.L., G.F.T., and M.J.G. Investigation, Z.L., M.A.K., P.N.G., K.S., A.D.S., R.A.C., A.G.H., and M.J.G. Methodology, Z.L., R.A.C., and M.J.G. Project administration, M.A.K., A.H.S., and M.J.G. Resources, A.H., T.L., G.F.T., and M.J.G. Supervision, R.A.C. and M.J.G. Visualization, Z.L., R.A.C., and M.J.G. Writing – original draft, Z.L., R.A.C., and M.J.G. Writing – review and editing, All authors. Declaration of interests The authors have no competing interests to declare.

CC BY

no

2024-01-16 23:40:17

iScience. 2023 Dec 7; 27(1):108669

oa_package/86/87/PMC10788193.tar.gz

PMC10788196

38226031

Background In the context of Bangladesh, there exists a pressing requirement to ascertain sweet potato genotypes that exhibit the high yield potential and demonstrate stability in varying environmental conditions. The original article highlights the yield potential and correlation of traits of the examined varieties. Nevertheless, the incorporation of the Multi Trait Stability Index (MTSI) in the dataset will introduce a novel aspect, encompassing stability ranking of studied varieties based on a multi-dimensional trait index.

Experimental Design, Materials and Methods Description of the study area The study took place in the 2021–22 growing season across five districts in Bangladesh: Gazipur, Bogura, Jamalpur, Chattogram, and Jashore. The research sites are located in Bangladesh, spanning latitudes 23.6850° N and longitudes 90.3563° E, covering elevations ranging from 10 m (Coastal South) to 105 m (North) above sea level. Experimental design and materials Five renowned varieties of sweet potatoes, namely BARI Mistialu-9, BARI Mistialu-10, BARI Mistialu-12, BARI Mistialu-15, and BARI Mistialu-17, were utilized for the conducted experiment at the TCRC, BARI, Gazipur. Detailed information regarding these varieties can be found in the supplementary Table 1. The study employed a randomized complete block design (RCBD) with three replications. Experimental techniques The management practices for sweet potato were done following a standard procedure of Alam et al. [3] . The harvesting took place after 130 days after vine planting, wherein each genotype consisted of 10 plants. In order to conduct a thorough quality analysis, the laboratory of Postharvest Technology Division at BARI collected and examined the storage roots of each genotype, each weighing 500 g. The following data was gathered in accordance with the methodology outlined by Alam et al. [3] : mean vine length (cm) (VL), storage root length (cm) (ARL), storage root diameter (cm) (ARD), average storage roots per plant (ARN), average storage roots weight per plant (kg) (ARW), marketable storage root number per plant (MRN), marketable storage root yield (t/ha) (MRY), non-marketable storage root yield (t/ha) (NMRW) and dry matter content (%) (DW). Statistical analysis Correlation coefficient of nine sweet potato traits was analysed using ‘metan’ package of R software. MTSI was computed to calculate the multi trait stability index using following formula [4] . The stability index for the i th genotype, labeled as MTSI i , is determined based on multiple traits. The score of the i th genotype in the j th factor is represented as , while the score of the ideal genotype in the same factor is denoted as . Calculations for genotype and trait scores were conducted through factor analysis. The length of nine variables' vector, where these variables were utilized for ideotype planning. Among nine variables, all the variable were planned with a higher desired sense except for NMRY, which was set for a low sense in analysing the MTSI index.

A study was conducted in five regions of Bangladesh, specifically Gazipur, Bogura, Jamalpur, Jashore, and Chattogram, each characterized by suitable agro-ecologies for sweet potato cultivation. The purpose of this data article was to demonstrate the correlations between traits and the selection of stable varieties based on the multi-trait stability index (MTSI). The data indicated a direct link between multiple characteristics and both the yield and factors contributing to yield. This implies that enhancing these traits might result in a higher overall production of sweet potato storage roots. Furthermore, the factor analysis for MTSI demonstrated that the desired goal for selection was achieved for all traits, except for mean vine length (VL) and storage root dry weight (DW). The broad sense heritability ranged from 0 to 0.97, and the selection gain percentage ranged from 0 to 42.8. The MTSI analysis identified the sweet potato variety BARI Mistialu-15 as the most stable among the other studied varieties. Keywords

Specifications Table Value of the Data • This dataset presents information on the relationship between sweet potato traits and their impact on storage root yield. • The dataset in this article supplies data to researchers, farmers, and industry users on the multi trait stability of sweet potato genotypes grown in different locations of Bangladesh. • The provided data has a potential for assisting in crop improvement programs and genetic studies, particularly in analyzing the stability of sweet potato based on specific traits of importance. • The dataset demonstrates the usefulness of the MTSI index in speeding up the selection of stable genotypes in crop breeding programs. Data Description The challenge of selecting genotypes for breeders is posed by the need to adapt to various environmental conditions, as the specific environment influences and interconnects the phenotypic traits of plants. The concept of a plant ideotype encompasses a set of desirable traits that can be independently measured in the field. By utilizing this approach, the selection process becomes more streamlined, systematic, and dynamic, facilitating a better understanding of the genotype-environment interaction in multi-environment trials [1] . This study conducted in Bangladesh involved the cultivation of five genotypes of sweet potato varieties across five different locations. The dataset presented in this article comprises two figures and one table. Fig. 1 presents the correlation coefficients utilized to evaluate the associations among eight yield-contributing traits and one quality trait, employing data from five different locations and five distinct varieties. The findings demonstrated noteworthy positive correlations ( p ≤ 0.05) among all the examined traits, with the exception of non-marketable root yield. Consequently, the acquired data from the correlation analysis disclosed that positively correlated traits serve as pivotal elements that impact the root yield in sweet potatoes. Table 1 displays the results of the factor analysis, which reveal the broad sense heritability, selection gain percentage, desired selection sense, and the achievement status of the goal based on factor analysis of selected sweet potato genotypes using nine traits. The factor analysis revealed that the eigenvalues of three factors were greater than 1, leading to the inclusion of these three factors in Table 1 . These factors formed three distinct groups consisting of nine traits under study. The desired selection sense was to increase for all traits, with the exception of non-marketable storage root yield (NMRY). The goal was successfully achieved for all traits, with the exception of mean vine length (VL) and the dry weight of storage root (DW). Among the studied traits, the broad sense heritability was highest for DW (0.97), followed by average root diameter (ARD) (0.93), average root length (ARL) (0.88), NMRY (0.87), VL (0.54), and marketable root yield (MRY) (0.41). Furthermore, the selection gain percentage for these traits was also higher, indicating that the selected genotype shows great promise in terms of the heritability of a few agronomic traits. The selected genotype, however, possesses a significant weakness in DW. Fig. 2 presents the representation of the order of the most stable genotype determined by the multi-trait stability indexing (MTSI). Within this context, the red circle symbolizes the threshold for selecting the stable genotype [2] . The sweet potato genotype called BARI Mistialu-15 stands out as the most stable compared to other genotypes. Limitations The dataset only covers one growing season so it may not capture the variability across different years. Factors like climate, soil, and pests can affect sweet potato yield and change every year. Ethics Statement All authors have read and follow the ethical requirements for publication in Data in Brief and our work meets these requirements. Our work does not involve studies with animals and humans. CRediT authorship contribution statement Zakaria Alam: Conceptualization, Methodology, Software, Writing – original draft. Sanjida Akter: Data curation, Software. Md. Anwar Hossain Khan: Visualization, Investigation, Supervision. Md. Harunor Rashid: Investigation, Supervision. Md. Iqbal Hossain: Software, Validation. Abul Bashar: Software. Umakanta Sarker: Writing – review & editing.

Supplementary materials Data Availability Multi trait stability indexing and trait correlation from a dataset of sweet potato (Ipomoea batatas L.) (Original data) (Mendeley Data) Acknowledgements The authors are thankful to Bangladesh Agricultural Research Institute for providing planting materials for the study. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships which have, or could be perceived to have, influenced the work reported in this article.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 21; 52:109995

oa_package/56/4d/PMC10788196.tar.gz

PMC10788197

38226040

Experimental Design, Materials and Methods Location The data collection was conducted on two different locations. The first location is a freshwater fish farm in Tasikmalaya Region, West Java Indonesia (7°20′53.6"S 108°11′14.2"E). The second location is a large open-air fishpond in a food court area in Bandung City, West Java, Indonesia (6°54′51.9"S 107°35′33.4"E). Data collection in both locations was conducted with written permission from the location owner. Both locations contain similar edible fresh-water fish, mainly the Oreochromis niloticus . The fish in the first location are bred for consumption purposes, while in the second location only grow the fish for decoration purposes. Both locations have different water treatment methods and water sources. This condition ensures that the acquired data will have condition variance. Data collection process For each individual fishpond, the water quality data was taken on several points. These points are located on the side and the corner of the pond. In each data collection point, temperature, pH level, and total dissolved solid level were measured using digital meter and recorded. This process is repeated several different times to ensure the variance on the water quality data. The image of the fishponds on data collection time is recorded from a certain height using drone at 90° angle. There are a total of 105 images taken from all the locations. The water quality data consists of temperature data, pH level, and total dissolved solid. These three parameters were chosen based on the research conducted by Siswanto et al. [2] . The research emphasizes these three parameters as the important parameters in supporting optimal fish growth. Each water quality data entry is given a status label that corresponds to the optimal growing conditions for Oreochromis niloticus . For data entry in which all parameters are inside the optimal range, the data is labeled as OPTIMUM. While for parameters that are outside the optimal range is labeled as LOW or HIGH. Image data processing The aerial image acquired from the drone has large resolution and includes any other objects found in the fishpond. The image needs to be clean so that only water image is used in the data. However, the raw image of each pond has also been provided in the dataset. An example of raw image from the drone can be seen in Fig. 1 . The image was overlaid with gridlines (see Fig. 2 ). Each gridline was positioned so that the distance between gridlines (vertically and horizontally) is 100 pixels. So, each box formed by the gridlines has size of 100×100 pixels. On the grid, 10 boxes were chosen where it only contains the water image. Each box was then cropped to create a new image. The new image is used for the dataset. This process needs to be repeated for every raw image available in the data. The process to layout and crop the raw image is conducted by using Python. A program is written in Python script to automatically conduct the grid layout and image cropping. The 10 data points are selected manually, to avoid any non-water image is selected. The selection was also conducted based on the proximity to the water quality data collection point. The condition of each cropped image is acquired from the result of the nearest data collection point.

This dataset is part of fundamental research to produce IoT monitoring in fishponds. The data consists of the results of measurements of pH, total dissolved solids (TDS), and water temperature obtained through manual sensor devices in several locations at different times. Additionally, this data also includes images taken by drones at consistent heights. These images are linked to the sensor data that has been collected. In this research, data will be used to monitor the health of fishponds through visual data. This data can be used for correlation analysis between visual data and sensor data. The hypothesis is the visual appearance of the pond (the colour) is affected by the number of mixed solid (mud and other organic material) in the water, which reflected in the TDS level of the water. In addition, the data can also be used for initial investigations into the development of machine learning models for pool condition recognition through image analysis. Keywords

Specifications Table Value of the Data • The dataset contains information on fishpond conditions, visually and numerically. The data are useful in investigation on method to determine fishpond water quality through visual means. • Researchers may benefit from this dataset through the dataset's linkage on visual data with water quality data. The researcher may study or develop means of inferring the fishpond's condition using visual data using this dataset. It is expected that the visual condition is affected by the number of solid mixed in the water, in which reflected in the water quality parameter. • These data can be processed using correlation analysis or clustering method, or any image processing method. The status label can be used to enhance the result of correlation analysis and clustering methods. Data Description The dataset comprises 1023 data entries, with each entry representing the condition and a 100×100 pixels visual image of a fishpond captured at a specific time and location. The original image file of the fishpond has also been provided as raw data. The dataset is comprised of a tabular data (in csv file), a folder of raw images (in JPG format), and a folder of cropped images (in JPG format). The csv file encompasses tabular information from 21 distinct ponds, each corresponding to different locations and collection times. Additionally, the filename for the visual image (cropped and raw images) is provided for each data entry. Table 1 shows the attributes of the tabular information within the dataset. The explanation of each attribute is as follows: a. Pond_label. Provides the label that informs where the data entry is taken. Each pond number corresponds to a different time and location of data collection. b. Temp. Provide the data about the temperature of water. The temperature is given in degree Celsius. c. TDS. Provide the data on the level of total dissolved solid on the water. It is given as mg/L. d. pH. Provide the data on the level of acidity of the water. e. State. Provide the classification of water condition in relation to the optimal growing conditions of Oreochromis niloticus. Each entry may have several labels attached, depending on the parameters in each entry. For entry in which all parameters are within the optimal range, it only be given a label of OPTIMUM. While other conditions can be a combination of LOW_PH, HIGH_PH, LOW_TEMP, or HIGH_TEMP. The TDS in this dataset is always inside the optimal range. f. Raw_images. Provide the filename of the raw image of the pond where the data entry is located. g. Images. Provide the filename of the cropped image in which the data entry is located. In relation to the state of water quality, the distribution of the label in the data entry is explained in Table 2 . Each data entry may have more than one status label, so the total number of state classification in Table 2 is greater than the total numbers of data in this dataset. The summarization of the dataset number of images and locations can be found in Table 3 . Limitations The status label data only applies to Oreochromis niloticus growing conditions. The label may not be applicable to different species of fishes. Ethics Statement The authors state to have read and follow the ethical requirements for publication in Data in Brief and confirm that the current work does not involve human subjects, animal experiments, or any data collected from social media platforms. CRediT authorship contribution statement Dany Eka Saputra: Conceptualization, Methodology, Investigation, Supervision, Writing – review & editing. Dani Suandi: Data curation, Investigation, Formal analysis, Writing – original draft. Joshua Wenata Sunarto: Data curation, Investigation. Petra Michael: Data curation, Investigation.

Data Availability Fishpond Visual Condition Dataset v2.0 (Original data) (Mendeley Data) Acknowledgements This work is supported and funded by the Ministry of Education, Culture, Research, and Technology Republic of Indonesia through Fundamental Research Grant Program with contract number: 179/E5/PG.02.00.PL/2023. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 23; 52:110009

oa_package/ec/92/PMC10788197.tar.gz

PMC10788198

38226156

Introduction The prevalence of CRPC continues to be the leading cause of mortality among male patients worldwide. 1 , 2 The rapid progression, ease of transfer, and development of late castration resistance of CRPC pose challenges to treatment. 3 , 4 The existence of androgen-repressed genes has been reported in a few studies, with some suggesting their involvement in the progression of CRPC through indirect mechanisms. 5 , 6 , 7 , 8 , 9 The identification of targets is imperative for the advancement of CRPC development. 10 Traf2- and Nck-interacting kinase was initially identified as germinal center kinases (GCKs) and plays a crucial role in regulating various fundamental cellular processes through phosphorylation of its downstream substrates. 11 , 12 The proto-oncoprotein TNIK exhibits overexpression in various malignancies, including prostate cancer (PCa), multiple myeloma, pancreatic cancer, hepatocellular carcinoma, and gastric cancer. 13 , 14 , 15 , 16 , 17 , 18 For example, TNIK protein phosphorylates AKT and promotes the proliferation of gastric cancer cells. 19 Furthermore, TNIK-deficient mice showed reduced expression of Myc and Cd44. 20 TNIK was also shown to be involved into the NF-κB and SMAD signaling pathways. 11 , 17 Emerging evidence has demonstrated the crucial involvement of epidermal growth factor receptor (EGFR) signaling pathways in prostate regulatory mechanisms during prostate tumorigenesis. Activation of the EGFR signaling pathway plays a pivotal role in promoting cancer cell survival under androgen-depleted conditions. Upregulation of EGFR signaling may be a mechanism by which PC cells escape castration-induced cell death. 21 The association between TNIK and EGFR, however, remains unestablished in existing literature. In this study, we employed GeneArray analysis to elucidate the mechanism underlying the development of CRPC. Our findings revealed TNIK as a potential driver gene for CRPC that interacts with EGFR. Notably, our evidence demonstrated that AR directly repressed TNIK transcription through the AR-H3K27me3 complex. Upon si-AR or MDV3100 treatment, TNIK gene expression was activated, leading to its binding to the extracellular domain (ECD) of EGFR and subsequent promotion of phosphorylation. Previous studies have implicated the ECD domain of EGFR in cellular drug resistance, suggesting that the interaction between TNIK and ECD may play a crucial role in CRPC emergence. 22 The activated EGFR is mainly located in the nucleus and regulates the transcription of target genes, thereby promoting the progression of CRPC. Pharmacological inhibitors of TNIK (NCB-0846) inhibited the growth of CRPC cell xenografts. Furthermore, we observed that silencing TNIK also enhances EGFR-mediated, erastin-induced ferroptosis in CRPC cells. This kinase represents a promising candidate for targeted drug therapy in CRPC and provides valuable insights for potential combination strategies involving ferroptosis-based therapies.

STAR★Methods Key resources table Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Ning Jiang ( [email protected] ). Materials availability The plasmids used in this study are available from the lead contact . Data and code availability • All data reported in this paper will be shared by the lead contact upon request. • This paper does not report original code. • Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request Experimental model and study participant details Animal Four-week-old male BALB/c mice (HFK Bio-Technology Co. Ltd, Beijing) were subcutaneously injected with 2×106 C4-2 cells suspended in 0.1 mL of Matrigel (BD Biosciences) and implanted into the dorsal flank on both sides of the mice. Once the tumors reached a size of approximately 100 mm3, the mice were orally administered either vehicle (10% DMSO in PBS) or NCB-0846 (80 mg/kg of body weight) daily for 10 days by oral gavage (n = 4 mice for each treatment). Tumor volume was measured using digital calipers and estimated using the formula LW2/2, where L represents length of tumor and W represents width. At the end of the study, mice were euthanized, and tumors were extracted and weighed. All procedures involving mice were approved by the University Committee on Use and Care of Animals at Tianjin Medical University and complied with all regulatory standards. The permit number for mouse experiments is SYXK (Jin) 2019-0004. Ethic CRPC and HSPC tumor tissues as well as paracancerous samples were obtained from patients undergoing radical prostate cancer surgery and examined by certified pathologists. All procedures performed in studies involving human participants were in accordance with the ethical standards of the Research Ethics Committee of The Second Hospital of Tianjin Medical University and with the 1964 Helsinki declaration and its later amendments. ALL written informed consent to participate in the study was obtained from prostate cancer patients for samples to be collected from them (number:KY2021K192). Cell lines and cell culture The PC cell lines LNCaP, C4-2, 22RV1, and PC3 were obtained from ATCC. The cells were cultured in RPMI 1640 medium supplemented with 10% fetal bovine serum (FBS), 1% penicillin/streptomycin, and 1% glutamine. Dihydrotestosterone (DHT) was acquired from Amersham (Braunschweig, Germany). At the beginning and end of the experiment, the above cell lines were verified through the STR genotyping-based cell line identification service provided by Gene Create Company, and no abnormalities were found. The cells were tested negative for mycoplasma by the mycoplasma detection kit (Gendx, KS201). Method details SiRNA The validated siRNAs were obtained from GenePharma (Shanghai, China). Transfections were conducted using lipo2000 (Thermo Fisher) in accordance with the manufacturer's instructions. Co-immunoprecipitation and western blotting LNCaP and C4-2 cells were harvested and lysed in lysis buffer containing 150 mM KCl, 75 mM Hepes (pH 7.5), 1.5 mM EGTA, 1.5 mM MgCl2, 10% glycerol, and 0.075% NP-40 supplemented with a protease inhibitor cocktail from Roche (USA). The extracted proteins were precleared using a mixture of protein A–Sepharose (CL-4B; GE Healthcare) and antibody overnight at 4°C. Immunoprecipitates were washed with lysis buffer, resuspended in sample buffer, boiled, and analyzed by SDS-PAGE. Individual samples (40 μg of protein) were separated on an 8% SDS polyacrylamide gel and transferred to PVDF membranes from Millipore (Billerica, MA). The membranes were blocked in a PBS-Tween20 solution with 5% fat-free milk for one hour at room temperature before being incubated overnight at 4°C with appropriate dilutions of specific primary AR or TNIK antibodies. After washing, the blots were incubated with HRP-conjugated anti-rabbit or anti-mouse IgG for one hour. Finally, the blots were developed using an ECL mixture from Vector Laboratories (Burlingame, CA) and visualized by Imager. Chromatin immunoprecipitation LNCaP cells were cultured in 1640 medium (Invitrogen) with the addition of 10% charcoal-stripped fetal bovine serum (CSF, HyClone, USA) for a duration of 12 hours. DNA cross-linking was performed by adding 1% formaldehyde to the cell cultures at room temperature for a period of 10 minutes, followed by the addition of glycine (final concentration: 0.125 M) for an additional 5 minutes to halt the cross-linking reaction. Cells were lysed using a lysis buffer containing protease inhibitors and sonicated to fragment genomic DNA into sizes ranging from 200 to 1000 base pairs. One-tenth of the cell lysate was used as an input control, while the remaining portion underwent immunoprecipitation using AR or H3K27me3 antibodies. After collecting the immunoprecipitates using protein G-agarose columns, protein-DNA complexes were eluted and heated at a temperature of 65°C to reverse the cross-linking process. Following digestion with proteinase K, DNA fragments were purified using spin columns and analyzed via PCR for a total of 35 cycles under conditions consisting of denaturation at a temperature of94°C for a durationof30 seconds, annealing at55 °C for another thirty seconds and extension at 72°C for one minute each cycle. The specific primer sets targeting a sequence within the human TNIK promoter (as listed in Table S3 ) were designed accordingly. The resulting PCR products were subsequently electrophoresed ona1.5% agarose gel stained with ethidium bromide and visualized under ultraviolet light. Whole-transcript expression array and microarray image processing The starting material for generating total RNA/Poly-A RNA controls was 1 μg/μl of total RNA, which was mixed with the GeneChip® Eukaryotic Poly-A Control Kit (Affymetrix, Inc., CA, USA). To enhance sensitivity, the majority of rRNA was eliminated from the total RNA samples prior to target labeling using the RiboMinusTM Human Transcriptome Isolation Kit (Invitrogen, CA, USA). Subsequently, cDNA synthesis was performed according to the manufacturer's instructions using the GeneChip® WT Sense Target Labeling and Control Reagents Kit (Affymetrix, Inc., CA, USA). The resulting sense cDNA was fragmented by UDG (uracil DNA glycosylase) and APE 1 (apurinic/apyrimidinic endonuclease 1), followed by biotin labeling with TdT (terminal deoxynucleotidyl transferase) using a GeneChip® WT Terminal Labeling Kit (Affymetrix, Inc., CA, USA). Once the biotin-labeled sense target DNA was prepared, it underwent hybridization to a gene chip known as The GeneChip® Human Exon 1.0 ST array. Hybridization involved incubating 5 μg of biotinylated target with a GeneChip® Hybridization Wash and Stain Kit and a GeneChip® Fluidics Station 450 (Affymetrix, Inc., CA, USA). Finally, the arrays were scanned using a GeneChip® Scanner 3000 7G(Affymetrix,Inc., CA,USA), and raw data were extracted from scanned images for analysis utilizing Agilent Technologies'GeneSpring GX software version11.5. Immunohistochemistry The tissue sections were deparaffinized in xylene and rehydrated with graded alcohol. Antigen retrieval was performed under pressure for 5 minutes in citrate buffer (pH adjusted to 6.0). Endogenous peroxidase activity was blocked with 0.3% hydrogen peroxide for 10 minutes, followed by blocking with 1.5% horse serum. Primary antibody incubation was carried out overnight at 4°C in a humidified chamber. After applying Poly-HRP anti-rabbit IgG (30 min), secondary antibody detection was performed using the Ultraview DAB detection kit (Zhongshan Co., China). All immunostained sections were evaluated under a Zeiss microscope (×200). At least ten high-power fields around each malignant gland were assessed and scored. MTT assay After 48 hours of transfection, a total of 2.0 × 103 cells per well were seeded in 96-well plates and subjected to treatment with either DMSO or erastin at a temperature of 37°C for durations of 24, 48, 72, and 96 hours. Subsequently, each well was supplemented with 30 μl of MTT solution and incubated at a temperature of 37°C for a period specified by the experiment protocol (indicated time). Following this incubation period, the MTT solution was aspirated and replaced with an addition of dimethyl sulfoxide (DMSO) measuring up to a volume of150 μl per well in order to dissolve the formazan crystals. Finally, absorbance measurements were taken at a wavelength of 490 nm using a microplate reader. Determination of lipid peroxidation After 48 hours of transfection, a total of 5.0 × 106 cells were collected and transferred into a centrifuge tube. Then, 1 ml of extract solution was added to the tube followed by cell lysis and subsequent centrifugation at 4°C for 10 minutes. The resulting supernatant was carefully collected and mixed with the specified reagent (Jining Shiye, JN24889) according to the provided instructions. After thorough mixing, the mixture was heated at 90°C for 40 minutes and then subjected to another round of centrifugation to obtain a final volume of 1 ml supernatant. Finally, the absorbance at 490 nm was measured using a microplate reader. Quantification and statistical analysis The data were presented as mean ± standard deviation. Student's t-test was used to compare two samples, and p values of 0.05 or less were considered statistically significant. Tumor weight was analyzed using GraphPad Prism software. All experiments in our study were repeated at least three times. Statistical significance is indicated with ∗P <.05, ∗∗P < .01, ∗∗∗P < .005.

Results TNIK was upregulated in castration-resistant prostate cancer Previously, LNCaP cells were cultured under androgen deprivation conditions using 10% Certified FBS Charcoal Stripped and RPMI Medium 1640 culture medium. After about four months of culture and multiple passages, the CR-LNCaP cell line was obtained. Subsequently, a model of castration-resistant LNCaP tumors (CR-LNCaP) and androgen-sensitive tumors (HS-LNCaP) was established with LNCaP xenografts. 8 In order to identify unrecognized molecular mechanisms of CRPC between intact and castrated mice, we quantified changes in mRNA levels of human genes in HS/CR-LNCAP tumors. We found that 1,884 genes were upregulated and 588 genes were repressed in CR-LNCaP compared with HS-LNCaP tumors (log 2 FC > 1.5, p < 0.05) ( Figure 1 A; Table S1 ). GO analysis revealed a significant enrichment of upregulated differentially expressed genes involved in mRNA processing and nuclear transport signaling pathway ( Figure 1 B). Among these genes, we focused on TNIK serine/threonine kinase due to its upregulated expression in CRPC associated with the EGFR signaling pathway. To validate the microarray results, we performed qPCR analysis on xenograft tumor samples from CR-LNCaP (4 castrations) and HS-LNCaP (4 uncastrations), confirming higher mRNA expression of TNIK in CR-LNCaP compared with HS-LNCaP ( Figure 1 C). Furthermore, we investigated the potential role of TNIK by analyzing prostatectomy samples obtained from patients with hormone-sensitive prostate cancer (HSPC) (26 cases), CRPC (29 cases), and their respective paracancerous tissues using immunohistochemistry staining for TNIK expression patterns ( Figure 1 D). The results showed that both cytoplasmic and nuclear compartments exhibited circumscribed expression of TNIK within the epithelial cells, with a progressively significant increase observed in CRPC samples. These findings are consistent with the data obtained from microarray analysis shown in Figures 1 C and 1D, indicating increased mRNA and protein expression of TNIK in CRPC cells. Knockdown TNIK inhibited CRPC cell proliferation To investigate the impact of TNIK on cell proliferation in CRPC, siRNA targeting TNIK was transfected into CRPC cells (C4-2, 22RV1, and PC3). Efficient knockdowns of TNIK were successfully achieved in these cell lines ( Figure 2 A). Subsequently, we observed a clear inhibition of proliferation in C4-2, 22RV1, and PC3 cells upon downregulation of TNIK ( Figure 2 B). Furthermore, the reduction of TNIK expression significantly attenuated the invasive C4-2, 22RV1, and PC3 cells as demonstrated by Matrigel assays ( Figure 2 C). These findings strongly support that TNIK plays a crucial role in promoting cell proliferation within CRPC. In summary, our results highlight the independent functions exerted by TNIK in androgen-independent PC cells while emphasizing its pivotal contribution to the growth and metastasis of castration-resistant tumors. Androgen-AR signaling suppressed TNIK gene expression The upregulated expression of TNIK in CR-LNCaP compared with HS-LNCaP tumors suggests a potential regulation of TNIK by androgens. Therefore, we investigated the impact of dihydrotestosterone (DHT) treatment (10 nM) on TNIK levels. A time-course study of DHT incubation in LNCaP cells revealed that androgen treatment suppressed the expression and phosphorylation of TNIK ( Figure 3 A). We observed increased TNIK protein expression and phosphorylation levels upon AR knockdown in PCa cells, whereas treatment of LNCaP cells with the AR antagonist MDV3100 also significantly upregulated TNIK mRNA and protein levels, as well as phosphorylation levels of TNIK ( Figures 3 B–3D). Consistent with the upregulation of TNIK detected in CR-LNCaP, we also observed that inhibition of androgen exposure led to increased nuclear abundance of TNIK through immunofluorescence imaging ( Figure 3 E). In our previous study, we reported a cooperative role between EZH2 and AR for YAP1 transcriptional repression. 23 The sole identified methyltransferase with activity toward H3K27, EZH2, is solely responsible for all methylation of H3K27. 24 Therefore, we performed co-immunoprecipitation (Co-IP) experiments to verify whether there is an interaction between AR and H3K27me3 in PCa cells. As depicted in Figure 3 F, AR formed a stable complex with H3K27me3 in PCa cells. The specificity of these protein interactions was confirmed, as no visible interaction was observed in the IgG control. Furthermore, chromatin immunoprecipitation experiments demonstrated that both AR and H3K27me3 were recruited to the TNIK gene promoter; however, treatment with MDV3100 abolished the ability of AR to form a complex ( Figure 3 G). Treatment of PCa cells with DZNeP (EZH2 inhibitor) resulted in an increase in TNIK protein expression as well as phosphorylation levels, while interestingly downregulating AR levels as well ( Figure 3 H). In order to validate the role of H3K27me3 in transcriptional repression by AR, we utilized GSK-J1 (H3K27 demethylase inhibitor) to elevate the level of H3K27me3. We observed a decrease in TNIK expression when H3K27me3 levels increased, suggesting the inhibitory effect of H3K27me3 on TNIK, whereas the non-significant change in AR expression levels confirmed that EZH2-mediated regulation of AR levels is independent of its methylation function, consistent with previous studies ( Figure 3 I). 25 In addition, the elevation of H3K27me3 effectively counteracted the promoting effect of AR downregulation on TNIK expression ( Figure 3 J). Overall, these experiments demonstrated that the AR and H3K27me3 complex mediated the androgen-driven epigenetic repression of TNIK. TNIK interacts directly with EGFR Gene set enrichment analysis (GSEA) was conducted to investigate the involvement of activating EGFR pathways in LNCaP-CR vs. LNCaP-HS, where the differential expression profile of genes was analyzed ( Table S2 ). Although the role of TNIK in certain cancers has been extensively documented, its role in CRPC remains less understood. To validate the regulatory function of TNIK on EGFR expression, we utilized C4-2 cell line with an active EGFR pathway. The interaction between TNIK and EGFR was examined in C4-2 cells by immunoprecipitating TNIK from lysates and probing for interaction with EGFR through western blotting ( Figures 4 A and 4B). As expected, direct binding between TNIK and EGFR was observed. Furthermore, using immunofluorescence imaging, we found that overexpression of TNIK led to increased abundance of nuclear-localized EGFR, whereas decreased levels of TNIK resulted in reduced nuclear import of EGFR ( Figure 4 C). In order to determine the binding domain between TNIK and EGFR more precisely, we generated several plasmids expressing truncated versions of Myc-tagged-EGFR (Myc-EGFR). Immunoprecipitation experiments were performed in C4-2 cells coexpressing these Myc-tag mutants along with TNIK. Results indicated that when the ECD domain was deleted from these mutants, they were not detected in the immunoprecipitates pulled down by anti-Myc immunomagnetic beads ( Figure 4 D). Next, we investigated the impact of TNIK overexpression or siRNA-mediated knockdown on EGFR phosphorylation and its downstream target genes. Our findings revealed that TNIK overexpression activated EGFR phosphorylation and its associated signaling pathways, including the ferroptosis-related gene NRF2. Conversely, knockdown of TNIK had the opposite effect ( Figures 4 E and 4F). Mutants with attenuated catalytic activity at S171 exhibited reduced EGFR phosphorylation and NRF2 upregulation compared with mutant controls ( Figure 4 G). To further validate these results, we suppressed the activity of the ECD domain in EGFR using a plasmid. We observed that deletion of the ECD domain reversed the inhibitory effect of TNIK knockdown on EGFR phosphorylation and the changes in EGFR-associated NRF2 levels ( Figure 4 H). This further illustrates the importance of the ECD domain for TNIK to bind and function with EGFR. Notably, knockdown of TNIK decreased C4-2 cell resistance to erastin (a ferroptosis inducer). Similar to previous results, the ferroptosis sensitivity caused by TNIK knockdown was reversed when the ECD domain was deleted. ( Figure 4 I). In conclusion, our study highlights the essential role of TNIK in optimal EGFR phosphorylation and transcriptional activation while emphasizing its regulatory effects on ferroptosis as a potential avenue for combination ferroptosis therapies. TNIK inhibitor inhibited proliferation and invasion of CRPC cell The efficacy of the small molecule TNIK inhibitor NCB-0846 on tumor cells has been confirmed. 13 , 26 To test the efficacy of NCB-0846 during prostate cancer treatment and in CRPC, we initially assessed the efficacy of NCB-0846 in suppressing TNIK protein expression in C4-2 and PC3 cells. Treatment with 1 μm or 10 μm of NCB-0846 significantly reduced the levels of TNIK protein and phosphorylation after 24 h ( Figure 5 A). Furthermore, western blot analysis revealed a concurrent decrease in p -EGFR activity upon TNIK downregulation following inhibitor treatment in both C4-2 and PC3 cells. Notably, C4-2 and PC3 cells exhibited sensitivity to NCB-0846 treatment at a concentration of 10 μm ( Figure 5 B). Additionally, NCB-0846 treatment effectively inhibited cell invasion in both C4-2 and PC3 cells at a concentration of 10 μm ( Figure 5 C). Collectively, these findings highlight the potential therapeutic targeting of TNIK as an innovative approach for treating CRPC by modulating the EGFR signaling pathway. Targeting TNIK suppressed CRPC tumor progression in vivo To investigate the impact of the TNIK inhibitor on the growth of CRPC xenograft tumors, 2 × 10 6 C4-2 and PC3 cells were subcutaneously implanted in BALB/c mice. Once the tumors reached approximately 100 mm 3 in size, the mice were randomly assigned to receive either vehicle (10% DMSO in PBS) or NCB0846 (80 mg/kg of body weight) daily via oral gavage for a duration of 10 days (n = 4 mice per treatment group). While robust subcutaneous CRPC tumors formed in the DMSO-treated mice, tumor growth was noticeably smaller in the NCB-0846-treated group ( Figures 6 A and S1 A). A significant inhibition of tumor growth was observed in the NCB-0846-treated mice compared with those treated with DMSO alone ( Figures 6 B, 6 C, S1 B, and S1 C). Importantly, administration of NCB-0846 did not result in any apparent toxicity, as it had no effect on body weight changes ( Figures 6 D and S1 D). Furthermore, we found that tumors treated with NCB-0846 exhibited reduced TNIK expression levels and decreased Ki67 and p -EGFR expression ( Figures 6 E and S1 E). Because previous articles have demonstrated that EGFR is critical for the regulation of EMT in PCa, we speculate that TNIK can also regulate the EMT process through EGFR. 27 Therefore, we tested three EMT markers (β-catenin, vimentin, and E-cadherin) and found that inhibiting TINK seemed to block EMT in PCa. More importantly, in order to further explore the therapeutic potential of TNIK-targeted therapy in PCa, we detected two more bone metastasis markers (BMP6, BMP7) in PCa. Surprisingly, NCB-0846 also had inhibitory effects on PCa bone metastasis. These findings underscored the potential therapeutic value of targeting TNIK signaling to enhance sensitivity toward CRPC therapy and inhibit its progression and metastasis.

Discussion In the present study, we have elucidated crucial components of the interplay between the AR and EGFR signaling pathways via TNIK. By selectively inhibiting TNIK to impede EGFR phosphorylation, we effectively suppressed the proliferation of CRPC. The previous study reported that TNIK was frequently upregulated in high-grade ovarian cancer tumors and serous hepatocellular carcinoma. 18 Moreover, TNIK hyperactivity contributed to human lung adenocarcinoma cell metastasis. 17 In this study, we initially identified TNIK as a potential biomarker of CRPC by analyzing gene array files from mouse models. We observed elevated expression levels of TNIK not only in CR-LNCaP tumors in mice but also in CRPC patients compared with those with localized PCa and benign prostatic hyperplasia (BPH) ( Figure 1 D). These findings suggest a correlation between TNIK and aggressive behavior in cancer. The principal findings of our study revealed that AR forms a repressive complex with H3K27me3 at the TNIK promoter, leading to the suppression of TNIK transcription ( Figure 3 ). Consequently, ADT induces TNIK mRNA expression, which in turn regulates EGFR phosphorylation to activate the EGFR signaling pathway and contribute to CRPC growth. The transcriptional activity of AR is regulated by interacting coactivators that positively modulate receptor function. Conversely, AR inhibits target gene expression through "corepressors" such as Alien, SMRT, and NCoR. 28 , 29 , 30 A few reports have studied AR inhibition of transcription. Some hinted to indirect mechanisms—DNA methylation or protein phosphorylation 23 , 31 —whereas others hinted to direct mechanisms in involving the epigenetic silencing complex. 24 , 32 Using TNIK as a model, we discovered that AR can also suppress gene expression through hormone-induced recruitment of H3K27me3 to the AR/H3K27me3 complex, thereby directly inhibiting transcription. Consequently, ADT restores TNIK expression by disrupting the association between AR and H3K27me3 at the TNIK promoter. EGFR signaling pathway also plays a crucial role in CRPC. Substantial evidence has accumulated, indicating that the expression of EGFR is associated with an increased risk of high-grade, advanced disease, as well as prostate-specific antigen recurrence. 32 Castration of mature animals can increase the expression of EGFR protein in the prostate in a time-dependent manner. 33 Our findings also suggest that castration resistance may be induced by reciprocal interaction between TNIK and the EGFR signaling pathway, with TNIK playing a major role in promoting nuclear translocation and activation of EGFR ( Figure 4 ). Phosphorylation of EGFR is crucial for its intracellular distribution and transcriptional activity. ECD domain can recognize and bind to specific ligands, thereby facilitating the activation of EGFR. In this study, we discovered that TNIK can interact with the ECD domain on EGFR to induce phosphorylation and enhance nuclear translocation of EGFR. When the ECD domain is deleted, TNIK loses its regulatory effect on EGFR and EGFR-associated ferroptosis sensitivity. It has been reported that inhibition of TNIK reduces cell viability in ERG-positive PCa cells DU145 and 22RV1. 24 Our current data further demonstrate that TNIK has the ability to phosphorylate EGFR and promote transcriptional activation of its target genes. The use of TNIK inhibitors has shown promising inhibitory effects in CRPC cells and tumors ( Figures 5 and 6 ). Interestingly, TNIK also plays a regulatory role in ferroptosis. Ferroptosis is primarily caused by an imbalance between the generation and degradation of intracellular lipid reactive oxygen species (ROS), which is closely associated with cellular metabolic activity. In comparison to normal cells, tumor cells exhibit heightened metabolic rates and consequently face an increased susceptibility to ferroptosis. Previous studies have demonstrated that TNIK regulates glucose and lipid homeostasis in both Drosophila and mice. 34 The knockout of TNIK has been shown to enhance glucose and lipid metabolism; however, there is currently no research linking TNIK to ferroptosis. Given its role as a crucial regulatory factor in CRPC, the upregulation of TNIK expression may serve as a cellular self-protective mechanism against ferroptosis. Similarly, AR also plays a significant protective role in PCa cells by inhibiting lipid peroxidation and preventing ferroptosis. 35 When AR is present, in order to sustain the normal metabolic activities necessary for cell growth, AR exerts an inhibitory influence on TNIK. During treatment, the reduction in AR levels triggers the activation of TNIK as a secondary defense mechanism against ferroptosis while concurrently promoting CRPC progression. The collaborative action of AR and TNIK ensures cellular metabolism homeostasis in PCa. NCB-0846, as a TNIK inhibitor, has shown potential therapeutic effects in a variety of tumors. As a regulatory component of the transcription complex composed of β-catenin and T cell factor 4 (TCF4), TNIK can regulate the Wnt signaling pathway related to stem cell activity. In colorectal cancer, where the Wnt signaling pathway is the most prominent, NCB-0846 has been shown to bind to TNIK in an inactive structure, thereby inhibiting the activation of the Wnt pathway and blocking intestinal tumorigenesis and tumorigenic activity. 20 In addition, the combination of NCB-0846 and ABT-263 (BCL-X) inhibitor also showed synergistic inhibition of colorectal tumors in the KRAS mutant xenograft model. 36 The inhibitory effect of NCB-0846 on Wnt target genes also shows good tumor suppressor effects in synovial sarcoma, inducing rapid apoptosis of synovial sarcoma cells. 37 Similar to colorectal cancer and synovial sarcoma, the Wnt pathway is very important for the occurrence and development of PCa, and KRAS mutations also occur in PCa. The use or combination treatment of NCB-0846 in colorectal cancer and synovial sarcoma can provide a reference for the treatment of PCa. More importantly, NCB-0846 exhibits an inhibitory effect on lung cancer cell metastasis, which is mediated through the epithelial to mesenchymal transition (EMT) mechanism. NCB-0846 blocks the activation of Smad signaling and the induction of EMT by downregulating the expression of transforming growth factor beta (TGF-β) receptor type I. 38 In PCa, the metastasis and drug resistance caused by EMT have always been problems that trouble people. Interestingly, NCB-0846 can disrupt the transport and secretion of type I procollagen and inhibit the production of matrix proteins to regulate liver fibrosis. 39 In PCa, matrix proteins can induce tumor cell metastasis and adhesion to bone cells. 39 Although previous studies have demonstrated that NCB-0846 has a good therapeutic effect in ERG-positive PCa, whether it has the effect of regulating PCa EMT and bone metastasis has not been verified. 13 Here, our findings demonstrate that AR functions as a transcriptional repressor in TNIK by binding H3K27me3 complex. When AR levels decrease, TNIK phosphorylated EGFR and activating EGFR pathway to escape from castration treatment and promoted CRPC progression. By inhibiting the TNIK/EGFR axis, NCB-0846 showed inhibitory effects on CRPC tumor cells both in vivo and in vitro and blocked EMT of tumor cells as well as bone metastasis. Through current research, we have reason to believe that NCB-0846 has great potential for the treatment of PCa. These results provide the possibility for the combined treatment of PCa with AR and TNIK, which may be the key to inhibiting the progression of CRPC and provide a reference for the launch of ferroptosis treatment. Limitations of the study In this study, we found that the expression levels of markers of PCa EMT and bone metastasis also changed after inhibiting TNIK using NCB-0846, demonstrating the therapeutic potential of TNIK-targeted therapy in PCa. However, the mechanism by which TNIK regulates the EMT process of PCa and bone metastasis remains unclear, and more phenotypic validation is needed. This is crucial for the development of future PCa therapies. Future research needs to be more specific and targeted to explore the regulation of EMT and bone metastasis by TNIK in PCa.

These authors contributed equally Lead contact Summary The development of castration-resistant prostate cancer (CRPC) is driven by intricate genetic and epigenetic mechanisms. Traf2- and Nck-interacting kinase (TNIK) has been reported as a serine/threonine kinase associated with tumor cell proliferation or unfavorable cancer behavior. The microarray approach revealed a substantial upregulation of TNIK expression levels, enabling us to investigate the functional behaviors of the TNIK gene in CRPC. Specifically, we discovered that AR suppresses TNIK gene transcription in LNCaP and C4-2 cells by forming a complex with H3K27me3. Following the reduction of AR levels induced by androgen deprivation therapy (ADT), TNIK is recruited to activate EGFR signaling through phosphorylation in C4-2 cells, thereby promoting CRPC progression. Our findings unveil a regulatory role of AR as a repressor for TNIK while also highlighting how TNIK activates the EGFR pathway via phosphorylation to drive CRPC progression. Consequently, targeting TNIK may represent an appealing therapeutic strategy for CRPC. Graphical abstract Highlights • Inhibition of AR by ADT therapy leads to an increase in TNIK • TNIK activates transcription of EGFR by acting on its ECD domain • NCB-0846 can regulate ferroptosis, EMT, and bone metastasis in prostate cancer Biochemistry; Molecular biology; Cancer Subject areas Published: December 12, 2023

Supplemental information Acknowledgments We thank Hans Clevers, Professor of Hubrecht Institute, for kind assistance with TNIK vectors. This work was supported by grants from the 10.13039/501100001809 National Natural Science Foundation of China (81872079, 81572538) and the Science Foundation of Tianjin (No.: 16JCZDJC34400, 21JCZDJC01260, ZC20131, 2022ZD070). Author contributions J.N. conceived of the study, and G.J.N. carried out its design. L.J.M. and W.Y.Z. collected experiment samples. G.T. and L.Y.H. analyzed the data and wrote the paper. Z.B.Q., G.S.Y., C.Q., and L.J.B. revised the paper. M.F.A. and N.Y.J. supplemented the article. All authors read and approved the final manuscript. Declaration of interests None of the authors have any relevant conflicts of interest pertaining to the studies and data in this manuscript. Inclusion and diversity We support inclusive, diverse, and equitable conduct of research.

CC BY

no

2024-01-16 23:40:18

iScience. 2023 Dec 12; 27(1):108713

oa_package/b3/48/PMC10788198.tar.gz

PMC10788203

38226041

Background The obtained data serves as a comparison material with other case data for the projection of the Indonesian Presidential election in 2024. For example, The obtained data could be compared with the results of obtained data during the determination of presidential candidates and throughout the campaign period, as well as the actual election results for the Indonesian Presidential election in 2024. Therefore, through a sentiment analysis approach, this research can provide reliable recommendations for projecting similar Presidential election outcomes in Indonesia or other countries.

Experimental Design, Materials and Methods Data collection, labeling & preprocessing Data collection was conducted in April 2023 using the Twitter API. Before collecting data, users must have a developer account. The obtained data is in CSV and Excel formats. The original data is stored in the available repository (Folder: original data). The next step involves the labeling process with the assistance of language experts. This stage is crucial, especially in research on sentiment analysis. Since the tweets collected are in the Indonesian language, the labeling process is also conducted in Indonesian. However, the repository folder is labeled in English (Folder: labeled data). Each tweet is categorized into positive or negative sentiment classes by language experts. Translation into English is performed after the labeling process using machine translation and validation. After translation, the data undergoes preprocessing steps such as case folding, tokenizing, stop word removal, normalization, and stemming to clean the raw data using Jupyter Anaconda tools and the Python programming language (Folder: cleaned data). Data is then filtered to remove repeated (spam) tweets and captured tweets containing empty cells. After completing all the steps, the accumulated data amounts to 23,446, as indicated in Fig. 2 . Experiment The experiment was conducted as a model for the data that had been processed in the study. This study determined the model using two classification methods, namely Naïve Bayes and Support Vector Machine (SVM) with four kernels. The Naïve Bayes method has been widely used in sentiment analysis [5] because it has the highest accuracy [6] and is applied to many applications [7] , [8] , [9] . The SVM method is also superior to other classification methods such as research [ 10 , 11 ] in the sentiment analysis approach [12] . The determination of the model was used using three data ratios, namely 70:30, 80:20, and 90:10. Data that has been labeled will go through the extraction stage of the TF-IDF (Term Frequency - Inverse Document Frequency) feature to calculate the amount of weight in a document, the more relevant important words indicate that the greater the weight owned. Furthermore, classification is carried out using two methods as shown in Table 3 . It is shown that these two methods used have an average of above 75 %. SVM has a higher accuracy average than Naïve Bayes in the data used as shown in Fig. 3 . Discussion The data available in the repository is dehydrated by removing the attribute column for the username. This is intended to follow the ethics available on the Twitter platform. In addition to the deletion of the username attribute column, the data available in the repository is no longer deleted because it is useful for analysis in other studies that use this data. Initial testing of the data was obtained using two text classification methods, namely Support Vector Machine with four kernels and Naive Bayes. Both of these methods are used because they have a good degree of accuracy when faced with text classification. Text data obtained on the Twitter platform is then carried out a classification model using both methods. It was found that the SVM method had better performance calculated based on average accuracy than the Naïve Bayes method. However, the SVM method has a longer computational time than Naive Bayes which does it quickly. Naive Bayes have lower accuracy because they require only small amounts of data [13] , this is in contrast to SVM. This proves that the SVM method can be more suitable for data models related to the Presidential election and social media-based text classification. Meanwhile, based on the dataset used, the average dataset of each candidate has fairly good accuracy in the data used today. So the data obtained today can be a recommendation for future research on the Indonesian Presidential election based on user Tweets.

Discussion The data available in the repository is dehydrated by removing the attribute column for the username. This is intended to follow the ethics available on the Twitter platform. In addition to the deletion of the username attribute column, the data available in the repository is no longer deleted because it is useful for analysis in other studies that use this data. Initial testing of the data was obtained using two text classification methods, namely Support Vector Machine with four kernels and Naive Bayes. Both of these methods are used because they have a good degree of accuracy when faced with text classification. Text data obtained on the Twitter platform is then carried out a classification model using both methods. It was found that the SVM method had better performance calculated based on average accuracy than the Naïve Bayes method. However, the SVM method has a longer computational time than Naive Bayes which does it quickly. Naive Bayes have lower accuracy because they require only small amounts of data [13] , this is in contrast to SVM. This proves that the SVM method can be more suitable for data models related to the Presidential election and social media-based text classification. Meanwhile, based on the dataset used, the average dataset of each candidate has fairly good accuracy in the data used today. So the data obtained today can be a recommendation for future research on the Indonesian Presidential election based on user Tweets.

Indonesia is one of the countries that is currently entering the political year for the election of President, Regional Heads, and Members of the Legislative in 2024. This has become a hot topic on social media, especially about the Presidential Election. Twitter is one of the platforms with the largest users in Indonesia. It is interesting to see the alignment of Twitter users towards presidential candidates who already have a carrying party, namely Ganjar Pranowo, Prabowo Subianto, and Anies Baswedan based on a sentiment analysis approach. User feedback data about Indonesian Presidential candidates are obtained from the Twitter platform using Twitter API with Python programming language. The data obtained was 30,000 data with each candidate as many as 10,000 data. Data is pulled in April 2023 with specific keywords. The time for data withdrawal is chosen based on the announcement of Presidential Candidates carried by political parties before the schedule for determining or campaigning for Presidential candidates. Current data can potentially be used again as a comparison of analysis of presidential candidates on campaign time spans and after campaigns or actual calculation results. The data that can be accessed is in CSV format and has gone through several stages such as labelling using Language experts, removing spam Tweets & empty cells and preprocessing. Keywords

Specifications Table Value of the Data • This dataset is useful for policymakers, pollsters, political and government academics, candidates and winning teams, and political parties. • This dataset provides insight into the fact that the Indonesian's involvement in freedom of speech on social media platforms over political issues, such as Indonesian presidential candidates will be able to validate the predictions of presidential election results through academic studies. • This dataset can be served as a reference for future research when a candidate's campaign time starts and after the actual results of the 2024 Indonesian Presidential Election (Example: [2] ). So this approach model can be proposed as another option for predicting presidential election results because it is considered better than traditional polling [3] . Data Description This dataset contains Tweet information about the discussion of the 2024 Indonesian Presidential candidate. The candidate dataset used was Ganjar Pranowo, Prabowo Subianto, and Anies Baswedan from October 2022 to April 2023. The withdrawal of data this time was taken because the three candidates were already carried by one party, such as Ganjar Pranowo is carried by the PDI-P party (Indonesian Democratic Party of Struggle), Prabowo Subianto is carried by the Gerindra Party (Great Indonesia Movement), and Anies Baswedan is carried by the Nasdem Party (National Democratic). However, only the PDI-P party is qualified to carry a presidential candidate because it is enough to win seats in the national parliament. However, this Indonesian presidential election is important because President Jokowi is going to end his two terms. PDI-P as Jokowi's party again carries a new candidate in the 2024 election, namely Ganjar Pranowo who is currently the Governor of Central Java. In addition, Prabowo Subianto, who was also Jokowi's minister until now, is running as a presidential candidate in 2024. This selection is his fourth candidacy in the Indonesian Presidential Election contestation, namely in 2009 as a Vice Presidential candidate, 2014 as a Presidential candidate, and 2019 as a Presidential candidate. Another candidate, Anies Baswedan had also been a minister in 2014-2016 during Jokowi's era as President. The data for each candidate is in a folder such as original data, labeled data, and cleaned data with each folder containing files specific to the candidate's name, such as “Anies Baswedan,” “Ganjar Pranowo,” and “Prabowo Subianto,”. a. original data (Data Folder: original data) The data used is in the form of user Tweets based on keywords at the time of data search. The data search used is the name of potential candidates and is followed by “President of Indonesia 2024”, such as “Ganjar Presiden Indonesia 2024”, “Prabowo Presiden Indonesia 2024”, and “Anies Presiden Indonesia 2024” in language. The use of these keywords is intended so that related data can be collected. The time frame for data collection was done in April 2023. Each candidate has a different number of Tweets to reach 10,000 data in each candidate as shown in Fig. 1 . Fig. 1 shows the number of Tweets discussing Indonesian Presidential candidates, Ganjar Pranowo from October 2022 to April 2023 reached 10,000 Tweets, Prabowo Subianto from December 2022 to April 2023 reached 10,000 Tweets, and Anies Baswedan from January to April 2023 reached 10,000 Tweets. Although Tweets consist of several attributes that contain different information, the data retrieval used in this study is only for a few attributes as shown in Table 1 . These attributes are common, but proving some significant research on this subject requires additional attributes beyond those already acquired. In the utilized data, the obtained information also includes the attribute “username”. However, the attribute is removed before publication to protect the privacy of Twitter users. Social media-based research using a sentiment analysis approach can use the attributes that have been provided as analysis material. This original data folder contains Tweets as they were originally obtained, using Indonesian. There are no other changes available in this folder from the original data obtained other than the removal of the username attribute. b. labeled data (Data Folder: labeled data) Many labeling techniques can be used in Python, but labeling can also be done manually. Manual labeling proves to have better accuracy compared to Python libraries because the annotation technique which was given by humans takes deeper terms [4] . The labeling used in this study is manual with the help of experts so that its accuracy is validated. Indonesian linguists are more trusted to analyze texts for sentiment labeling using Indonesian. Therefore, this study tried to use file labeling in Indonesian when given labeling by linguists. However, in the repository file provided, we have changed it to English to make it easier for readers. c. cleaned data (Data Folder: cleaned data) The data goes through preprocessing stages to clean the data, such as case folding, tokenizing, stop word removal, normalization, stemming, and finally, dropping duplicates to eliminate identical Tweets (spam) and empty Tweets. Empty tweets occur after going through preprocessing stages such as URL deletion, tags/symbols, hashtags, and so on. The initial data for each candidate is 10,000, but after these stages, the number of Tweets for each candidate decreases as shown in Table 2 . Limitations Several bullets describe the limitations of the data acquisition in question. • Data retrieval is limited to a limited number of attribute items on the Twitter entity through the specific keywords used. • The collected data is then analyzed in the form of labeling positive or negative sentiment classes. This is obtained from linguists/linguists as partners in research. Ethics Statement Data is collected using Twitter's APIs and filtered to conform to policies established by the Twitter platform [14] . In discussions, we process data and do not mention privacy information to protect Twitter users and ensure anonymity. CRediT Author Statement Asno Azzawagama Firdaus : Conceptualization, Methodology, Writing – Original draft, Writing - review & editing, Data curation, Formal analysis; Anton Yudhana : Funding acquisition, Project administration, Writing – review & editing, Conceptualization; Imam Riadi : Writing – review & editing, Visualization, Conceptualization; Mahsun : Resource, Validation, Data curation.

Data Availability Indonesia Presidential Candidate's Dataset, 2024 (Original data) (Mendeley Data) Acknowledgements This work was supported by the Indonesian Ministry of Education, Culture, Research, and Technology under Contract Number 0557/E5.5/AL.04/2023. The author also gratefully for the cooperation as a partner of the Nusantara Research Institute, the Center for Human Development Studies Foundation. Funding This work was supported by Indonesian Ministry of Education, Culture, Research, and Technology grant numbers with an institutional contract number between DRTPM and LLDIKTI V: 181/E5/PG.02.00.PL/2023. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 19; 52:109993

oa_package/35/e8/PMC10788203.tar.gz

PMC10788206

38226042

Experimental Design, Materials and Methods Field data collection The datasets were collected from small and medium-scale farms in significant coffee and cashew growing regions of Uganda, including in the southern, central, eastern, and northern parts of the country. However, most of the coffee images were collected from two demonstration farms operated by the Uganda National Coffee Research Institute (NaCORI) in Kituza village, Mukono district, central Uganda. NaCORI is a governmental agency responsible for researching and developing new coffee varieties. Fig. 3 shows one of the demonstration farms (i.e., Block 13) where most of the coffee images were collected. The choice of farms and purposefully sampled plants from which data was collected was advised by agricultural experts who were part of the field data collection team. The coffee and cashew nut image data was collected in three phases. In the first phase, coffee image data was collected from Bukomansimbi, Kyotera, Mukono, Buikwe, and Masaka districts in Central Uganda between November and December 2022. In the second data collection phase, coffee image data collection took place in June 2023 in the Eastern Uganda districts of Luuka, Jinja, Mbale, and Sironko. In the third data collection phase, cashew nut image data was collected from Lira, Abim, and Nakasongola districts in March 2023. Data collection was carried out during the peak of the harvest season(s) for each crop. The details of images collected per region are shown in Table 3 .

Conventional methods of crop yield estimation are costly, inefficient, and prone to error resulting in poor yield estimates. This affects the ability of farmers to appropriately plan and manage their crop production pipelines and market processes. There is therefore a need to develop automated methods of crop yield estimation. However, the development of accurate machine-learning methods for crop yield estimation depends on the availability of appropriate datasets. There is a lack of such datasets, especially in sub-Saharan Africa. We present curated image datasets of coffee and cashew nuts acquired in Uganda during two crop harvest seasons. The datasets were collected over nine months, from September 2022 to May 2023. The data was collected using a high-resolution camera mounted on an Unmanned Aerial Vehicle . The datasets contain 3000 coffee and 3086 cashew nut images, constituting 6086 images. Annotated objects of interest in the coffee dataset consist of five classes namely: unripe, ripening, ripe, spoilt, and coffee_tree. Annotated objects of interest in the cashew nut dataset consist of six classes namely: tree, flower, premature, unripe, ripe, and spoilt. The datasets may be used for various machine-learning tasks including flowering intensity estimation, fruit maturity stage analysis, disease diagnosis, crop variety identification, and yield estimation. Keywords

Specifications Table Value of the Data • Flowering intensity estimation. Flowering represents an important stage in coffee and cashew farming since it affects crop yield. It has a significant impact on yield in that flowering intensity is positively correlated with the amount of crop yield. Therefore, flowering intensity could be an important predictor of crop yield [1] . Our dataset contains a flowering class, which can be used to train machine learning models to estimate the flowering intensity of cashew crops. • Fruit detection. In crop yield estimation using computer vision techniques, accurately detecting objects of interest, e.g., fruit pods is critical. Our dataset can be used to train machine learning models for the automated detection of coffee cherries and cashew apples. The dataset may also facilitate the development of new algorithms for small object detection, currently an open research problem in computer vision, since coffee cherries and cashew apples are relatively small. • Fruit maturity stage analysis. Our datasets contain coffee cherries and cashew apples at various stages of growth or maturity. It is these maturity stages that constitute object classes in the datasets. The dataset can be used to build machine learning models for automated analysis of fruit maturity stages for various purposes, including harvest scheduling. • Coffee crop variety identification. Our coffee dataset features the two main varieties grown in Uganda, namely Robusta ( Coffea canephora ) and Arabica ( Coffea arabica ) [2] . Within the Robusta variety, there are at least ten Coffee Wilt Disease resistant (CWD-r) clones, also known as KR lines. These clones are also resistant to leaf rust, tolerant to Red Blister Disease (RBD), have larger coffee bean sizes, are higher yielding, and have better cup quality. This means that our dataset can potentially be used for building models for the automated identification of Robusta coffee varieties. • Crop yield estimation. Our coffee and cashew datasets may also be used for yield estimation using machine learning methods, similar to work in [3] , [4] , [5] . Various machine-learning approaches may be used for yield estimation with this dataset, including object detection, image-based regression, and vegetation index-based methods. • Fruit disease diagnosis. Coffee cherries and cashew apples belong to three and five classes, namely unripe, ripe, spoiled and flower, immature, unripe, ripe, and spoiled, respectively. Images belonging to the spoiled class in each dataset may be used for coffee and cashew fruit disease diagnosis. Data Description The datasets presented in this work consist of high-resolution images of coffee and cashew plants acquired using Unmanned Aerial Equipment (UAV) equipment from small and large-scale farms across Uganda. Images range approximately between 10 MB and 12 MB in size, approx. 4000 by 3200 pixels in dimension and 72 pixels/in in Dots per inch (DPI). Each image is annotated with multiple bounding boxes, each enclosing an object of interest. Each image is accompanied by metadata, including the date (timestamp) and the geographic location (latitude and longitude) where it was captured. The majority of the images for coffee capture the full height and breadth of the tree (or plant) from two opposite lateral sides. A few of the images involved imaging the same tree from an overhead position that covers the entire canopy. For cashew trees, images were captured from different lateral sides (no top-view images). The image data for coffee and cashew nuts have been meticulously annotated. These annotated datasets, stored in the YOLO (You Only Look Once) format [ 6 ], are now readily accessible on the Hugging Face platform. Tables 1 and 2 provide the number of annotated object instances per class in the coffee and cashew datasets, while Fig. 1 , Fig. 2 show sample images for the two crops. The five classes in the coffee image dataset, as shown in Fig. 1 , have the following class IDs and labels: 0: Unripe, 1: Ripening, 2: Ripe, 3: Spoilt, and 4: Coffee_tree. The cashew dataset has six class IDs and labels, as shown in Fig. 2 . The classes include 0: Tree, 1: Flower, 2: Premature, 3: Unripe, 4: Ripe and 5: Spoilt. Our coffee and cashew nut datasets for machine learning yield estimation is the first of its kind and we did not come across any similar publicly available dataset. Existing datasets such as [ [7] , [8] , [9] ] consist of coffee leaf images designed for nutritional deficiency and/or plant disease detection and classification. Our dataset was collected using UAV equipment while the studies cited above used smartphone cameras for data collection. Materials and Methods Preparatory activities were carried out before field data collection. These included obtaining authorization letters, designing the data collection guidelines, training data collectors, and pilot fieldwork. This was done to prepare the field data collection team, to test equipment and data collection instruments, and to evaluate sample images for quality assurance. The imaging equipment consisted of a UAV, commonly referred to as a drone. Specifically, we used a DJI Mini 3 Pro drone equipped with a high-quality camera that had a 48 MP 1/1.3 in CMOS sensor, lens with aperture of f/1.7 and focus range of 1 m - ∞, shutter speed of 2-1/8000s and an ISO range of 100–6400 (Auto and Manual). A custom drone flight strategy was developed and used. This included using manual flight plans, flying at low altitudes and at close distances of about 1 m from coffee and cashew trees, adjusting camera orientation for optimal exposure and visibility of objects of interest, optimal spatial resolution, and flight speed. Images were primarily collected under optimal weather conditions for flying a UAV for farm-based data acquisition, including natural illumination (sunshine), precipitation, temperature, cloud cover, wind speed, and humidity. This was done to ensure that the resulting images were of high quality. Multiple images (at least three) were acquired per each purposively sampled coffee and cashew tree, taken from different viewpoints including from the top and opposite lateral sides. Full tree height and breadth and close range (approx. 1 m) images focused on coffee cherries and cashew apples were acquired ( Fig. 4 ). Data preprocessing and annotation We conducted thorough data cleaning, eliminating blurry and overexposed images while resolving any inconsistencies. The cashew data was labelled using an online annotation tool called Makesense AI 1 . The annotated data was saved in YOLO format [ 7 ] with six class IDs representing the cashew labels: 0: Tree, 1: Flower, 2: Premature, 3:Unripe, 4: Ripe, and 5: Spoilt based on a categorization in [ 10 ]. Fig. 5 shows an example of cashew nut image annotation. For the coffee image data, the coffee specialist from NaCORI expertly handled the annotation process using an offline tool called VGG Image Annotator (VIA) [ [8] , [11] ] to annotate the images. The coffee annotated data was saved in YOLO format with 5 class IDs representing the coffee labels: 0:Unripe, 1:Ripening, 2:Ripe, 3:Spoilt, and 4:Coffee_tree based on categorisation in [12] . Fig. 6 shows an example of coffee image annotation. CRediT authorship contribution statement Rahman Sanya: Conceptualization, Data curation, Project administration, Writing – original draft. Ann Lisa Nabiryo: Data curation, Validation, Writing – original draft. Jeremy Francis Tusubira: Data curation, Validation, Writing – original draft. Sudi Murindanyi: Data curation, Writing – original draft. Andrew Katumba: Funding acquisition, Supervision. Joyce Nakatumba-Nabende: Conceptualization, Supervision, Funding acquisition, Writing – review & editing.

Data Availability Coffee and Cashew Nut Dataset (Original data) (Mendeley Data) Acknowledgements We would like to thank Dr. Sammy Olal from the National Coffee Research Institute and Mr. Joseph Okilan for their support in the collection and curation of these datasets. The work is funded by a sub-grant from Lacuna Fund No. 19497.51 between the Makerere Artificial Intelligence Lab and KaraAgroAI from Ghana. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have influenced the work reported in this paper.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 14; 52:109952

oa_package/af/3d/PMC10788206.tar.gz

PMC10788208

0

Introduction Knee arthroplasty alleviates pain and restores function for nearly 900,000 Americans every year, and by 2030, this number is expected to more than double because the aging population desires to remain active [ 1 ]. As the number of expected primary knee arthroplasties continues to grow, so will revision procedures. The most recent report from the American Joint Replacement Registry (2022) has indicated infection, aseptic loosening, mechanical complications, and instability comprise over 80% of the causes of knee revision surgery in the United States [ 2 ]. The sum of technical errors during the index procedure can significantly affect postoperative function. For instance, the removal of additional bone from the distal femur, over-resection of the posterior condyles, and addition of slope to the tibia result in gap asymmetry which contributes to prosthesis instability. In response, implant companies have attempted to standardize primary total knee arthroplasty (TKA) via robotic-assisted surgery, but the application of robotic-assisted surgery in revision scenarios is still largely uncharted. It can be difficult to adapt current technology to revision scenarios because of bone loss, ligament attenuation, and a lack of appropriate computer software. Current implant systems rely on intramedullary (IM) fixation to bypass areas of deficient bone during revision knee arthroplasty. Smaller fully cemented stems can be used without much difficulty, but longer diaphyseal engaging IM stems have been found to influence the final position of the revision implants [ [3] , [4] , [5] ]. Anatomically speaking, it can be difficult to precisely position a long-stem revision tibial component because the center of the tibial epiphysis is not colinear with the center of the diaphysis. Similarly, when a long stem of the revision femoral component contacts the sagittal bow of the femur, disruption of gap balance may occur [ 6 , 7 ]. Therefore, implant augmentation and offset couplers were developed to improve bone contact and fine tune implant positioning. Unfortunately, offset components only permit a few millimeters of fine tuning within the plane of resection yet can continue to inhibit ideal placement of revision implants. Frequently, a constrained polyethylene component is employed during revision TKAs to supplement for residual instability after revision implants are placed. Revision components can be positioned independently of IM stems during revision TKA by using a previously described technique [ 8 , 9 ]. By manipulation of a current robotic-assisted workflow, the authors completed a series of robotic-assisted revision TKAs to address a variety of patients presenting with infection, aseptic loosening, mechanical complications, and instability. The purpose of this study was to compare the 3-dimensional changes in the preoperative and postoperative positions of the knee arthroplasty components necessary to achieve balanced gaps within a patient cohort. It is hypothesized that posterior femoral condylar offset is more important to gap stability than distal femoral measurements in robotic revision TKAs.

Material and methods Institutional review board approval was obtained prior to the initiation of this study. A total of 25 consecutive cases by 2 fellowship-trained arthroplasty surgeons at a single intuition from July 2021 through March 2023 were reviewed. All subjects had undergone a prior TKA using conventional jig-based instrumentation. Patients with a prior unicompartmental knee arthroplasty or peri-prosthetic fracture were excluded. Patient demographics along with indications for revision can be found in Table 1 . All surgeries were performed using the MAKO robotic-arm system (Stryker Orthopedics, Mahwah, NJ) equipped with Version 1.0 software according to our previously described surgical technique [ 8 , 9 ]. In most cases, several attempts were necessary to ensure an overall registration error of ≤0.5 mm due to a metal artifact from the existing implants on the preoperative computed tomography (CT) scan. Anterior referencing was used for all cases due to posterior femoral bone loss and to avoid anterior femoral notching. The revision femoral and tibial components were manipulated within the coronal, sagittal, and axial planes to obtain balanced medial and lateral flexion and extension gaps ( Table 2 ). After careful implant removal, all bone cuts were performed with the robotic arm, and implant augmentation was utilized in each case depending on the requirements for gap balancing. Preparation for the IM stem and metaphyseal cone(s) was performed for the femur and tibia as necessary. For most cases, the tibial array was positioned within the tibial diaphysis while the femoral array was located at the distal femur metadiaphyseal region to ensure room for a revision construct with short IM stems. For each subject, the preoperative coronal, sagittal, and axial positions of the primary implants (PI) were compared to the final coronal, sagittal, and axial positions of the robotic revision implants (RRIs). The current MAKO computer software (Version 1.0) cannot directly measure existing implant position nor directly calculate the positional difference between PI and RRI; therefore, an indirect method was utilized for this study. After a specific CT scan of the operative extremity was obtained, the PI was outlined during the implant planning phase. Next, virtual trial implants were superimposed over the existent PI to determine preoperative implant position within the cardinal planes ( Fig. 1 ). The best-fit virtual implants were centered within areas of metal artifact to obtain the most accurate measurement possible as indicated by prior studies [ 10 , 11 ]. The following values were recorded in relation to the defined mechanical axis or transepicondylar axis (TEA) of the extremity: PI femoral component flexion/extension, varus/valgus, internal/external rotation, posterior condylar axis, and TEA and PI tibial component varus/valgus, slope, and internal/external rotation. After the robotic assisted revision surgery was completed, the final planned virtual positions of the RRI were analyzed, and the following values were recorded in relation to the defined mechanical axis or TEA of the extremity: RRI femoral component flexion/extension, varus/valgus, internal/external rotation positions, posterior condylar axis, and TEA and RRI tibial component varus/valgus, slope, and internal/external rotation. In order to obtain data on the resultant change in component positioning, screenshots of the implant planning page that depicted the final RRI that overlayed the preoperative CT scan with the PI were then transferred into TraumaCad (Brainlab Munich, Germany). Measurements from the medial and lateral epicondyles to the distal most portion of the medial and lateral condyles for the PI and RRI were recorded, respectively. The difference between these numbers yielded the net change between PI and RRI positions for the distal medial and distal lateral condyles ( Tables 3 and 4 ). Measurements for posterior condylar offset (PCO) for the medial and lateral condyles were also obtained by measurement of the distance from a line tangential to the posterior femoral shaft and apex of the posterior medial and posterior lateral condyles for the PI and RRI. The difference between these values demonstrated the net change between PI and RRI positions for the posterior medial and distal lateral condyles, respectively ( Fig. 1 ). Calibration of the screenshots was standardized from a known constant value of either the inner width of the femoral implant box (16.2 mm) or the medial-lateral width of the tibial baseplate (which varies by size). All data were secured within a password-protected spreadsheet. Repeated-measures analysis of variance (ANOVA) using the difference (in millimeters and degrees preoperatively and postoperatively) were used to determine if there were statistical differences between the final positions of the RRI and the PI. Following that determination, multiple ANOVAs between the PI and RRI absolute values in millimeters and degrees were conducted to compare differences between groups before and after the robotic revision TKA surgeries. Differences were considered statistically significant at P < .05.

Results Statistical determinations There were statistically significant differences between the placement of the RRI and the PI. The overall repeated-measures ANOVA comparing the difference between primary and revision implants (in millimeters) demonstrated that the RRI values were statistically significantly different from the PI ( P < .001). There were also statistically significant differences in the absolute values (in millimeters) for both the posteromedial and posterolateral condyles from PI to RRI ( P values). The average increase in posterior medial and lateral condyle offset was 4.2 mm and 3.3 mm, respectively. There was no statistically significant difference between the distal medial or distal lateral values between PI and RRI. The posterolateral and posteromedial offset difference between implants (in mm) was statistically significantly different from the PI ( P < .001 and P < .03 respectively). While the difference between the distal-lateral and distal-medial positioning of the implants (in mm) was not significantly different from the PI ( P = .82 and P = .87, respectively). Preoperative and postoperative measured absolute values (degrees) in the tibial component positioning, but not femoral component positioning, were different between the RRI and the PI. The tibial slope and varus/valgus values were statistically different from the PI; however, the femoral values flexion, varus/valgus, and posterior condylar axis were not significantly different from one another. The difference of tibial varus/valgus between implants (in degrees) was statistically significantly different from the PI ( P < .003). The posterior-medial distance to implant (in mm) was statistically significantly different from the PI ( P < .03). The RRI tibial slope postoperative vs preoperative (degrees) was also statistically significantly different from the PI ( P < .01). The femoral posterior condylar axis in the RRI (in degrees) was not statistically significantly different from the PI placement although it approached significance with P = .0988. Femoral flexion in the RRI procedure (in degrees) was not statistically significantly different from the PI ( P = .51). Femoral varus/valgus in the RRI procedure (in degrees) was not statistically significantly different from the PI placement ( P = .68). The RRI surgical TEA (degrees) was not statistically significantly different from the PI placement ( P = .285). A subcategory ANOVA comparing the PI and RRI implant positions among subjects with aseptic loosening, infection, and instability did not reach statistical significance although this study was not powered for subgroup analysis given the current sample size. Implant results There were no reported complications during the operative procedure (ie, fracture, neurovascular injury, ligamentous injury, or anesthesia concerns) or during the postoperative hospitalization. For all cases, the selected tibial component was within one size over/under the selected femoral component per manufacturer recommendations. Medial and lateral flexion and extension gaps were found to be within 1 mm of each other in most cases. Tibial metaphyseal cones were used more often than femoral cones. Twenty-three of the revision tibias utilized a short 12 × 50-mm IM stem. A revision femoral component was utilized in 23 of the cases. The diameter of the femoral IM stem varied by case, with the average length being 100 mm. Offset couplers were not utilized for any of the cases. The average polyethylene thickness employed during revision cases was 11 mm. Twenty cases utilized a constrained posterior-stabilized component while 5 cases utilized a regular posterior stabilized component. Medial and lateral tibial augments were used 60% of the time. One case required a step cut on the tibia, and therefore only used a single 5-mm medial tibial augment. During femoral augmentation, 5-mm augments were more commonly used distally while 10-mm augments were more commonly used posteriorly. The existing femoral component preoperative coronal plane alignment varied between 4.7° valgus and 12.5° varus while the existing tibial component preoperative alignment was found between 3.5° valgus to 14° varus. The average postoperative coronal implant positioning averaged 0.20° varus and 0.04° varus for the femoral and tibial components, respectively. The postoperative posterior condylar and trans-epicondylar axes averaged 0.24° internal rotation and 1.45° external rotation, respectively, which was slightly less than the preoperative average values of 1.26° external rotation and 2.47° external rotation, respectively. The preoperative femoral component flexion ranged from 2.6° of extension to 16.5° of flexion. Final femoral implant flexion averaged 5.9°, and this was closely monitored to balance the flexion gap during implant planning. Postoperative posterior tibial slope averaged 1.6° in this cohort.

Discussion Restoration of femoral PCO and near-neutral alignment of the tibial component are statistically significant parameters necessary to achieve more symmetric gaps in this revision TKA cohort. Review of the PI positions from the current cohort frequently revealed a larger distal femoral cut and overresection of the posterior femoral condyles during the index surgeries. The resultant instability arises from the subtle elevation of the joint line and decreased PCO of the femoral component leading to decreased ligamentous balance [ 5 , [12] , [13] , [14] ]. The ensuing flexion instability was further amplified when excess posterior slope was added to the primary tibial component. Robotic-assisted revision TKA simplifies several challenging concepts frequently encountered during revision scenarios. Knowledge of each patient’s mechanical axis can provide the surgeon with a frame of reference when orienting the femoral and tibial components to help address gap symmetry. Conventional revision TKA relies on an IM alignment rod to indirectly reference the femoral mechanical axis. This traditional method is based on “average” measurements that can vary based on patient sex, age, height, and body mass index [ 15 , 16 ]. During our study, the mechanical axis of the operative extremity was directly obtained from a preoperative CT scan, which represented ideal neutral alignment during surgery. Incorporating ligamentous gap values during virtual implant templating was preferred to standard preoperative radiographic templating. Results of this study indicated one-third of the radiographically templated femoral components were oversized while another one-third were found to be undersized when compared to the final implanted femoral component ( Table 2 ). This can be partially attributed to magnification error seen with radiograph templating. Alternatively, a different femoral size can be selected to alter the ligamentous tension to achieve gap symmetry. To the authors’ knowledge, there are no “defined” ligamentous tension values for revision TKA representing a void within TKA literature. The referenced “gap” values within our results serve as a proxy for the gap measurements during robotic revision TKA. Incorporating ligamentous tension values throughout the range of motion may enhance the surgeon’s ability to optimally position the implants, which has the potential to improve gap symmetry in revision TKA, but further research is necessary. Symmetric gaps both in flexion and extension as well as medially and laterally were the goals of the operative procedure, but it was difficult to achieve equal numbers in all cases. If this occurred, the priority was focused on a symmetric medial flexion and extension gap. This rationale is driven by the dynamic interplay of bony and soft-tissue balance within the medial compartment during normal knee mechanics [ 17 ]. The MAKO robotic arm system establishes boundaries for restricted kinematic alignment during the procedure where the implants are positioned within 3° to 5° of the mechanical axis within the coronal, sagittal, and axial planes [ 18 ]. Re-establishment of the joint line within 5 mm of the native location has been correlated with improved knee-joint function and satisfaction [ 12 ]. Joint line measurements were not obtained during this study because current robotic software (Version 1.0) cannot directly determine the joint line in a revision scenario. The authors conceptualize the joint line as a dynamic entity that is established by proper ligamentous tension. Indirect determination of a more native joint line may be possible by relying on kinematic collateral ligamentous tension throughout the knee range of motion [ 19 ]. Sagittal alignment is an underappreciated factor for stability and function during revision TKA [ 3 , 20 , 21 ]. Evaluation of the prosthesis sagittal implant position and ligamentous balance is difficult with manual instrumentation because of bone loss and soft-tissue laxity that are commonly encountered during revision scenarios [ 22 ]. PCO, a component of sagittal alignment, is an important measurement that has been studied for both primary and conventional revision TKAs. The PCO has been reported up to 40% greater than native measurements because of bone loss and soft-tissue attenuation [ 23 ]. PCO directly affects the flexion gap, but it can also indirectly affect the extension gap as theorized by Mitsuyasu et al [ 24 ]. The enlarged femoral condyles that result from increased PCO may stretch the remaining posterior knee structures when the knee assumes an extended position, thus contributing to extension gap stability [ 24 ]. A recent comparative study by Sultan et al [ 25 ] demonstrated robotic arm–assisted surgery more accurately restored PCO in primary TKA. Meanwhile Clement et al [ 26 ] reported PCO is an independent predictor of functional outcome after revision TKA. In addition, PCO directly correlates with patient satisfaction scores after revision TKA [ 26 , 27 ]. PCO along with a near-neutral tibial slope provides flexion stability and helps ensure proper ligamentous tension during mid and deep flexion [ 22 , 23 , 26 ]. For our revision TKA cohort, 3D virtual planning helped illustrate the associated changes in the sagittal alignment of the femoral and tibial components. In most cases, the revision femoral component was upsized, and the flexion orientation was optimized at the distal femur to help restore ligamentous tension and flexion gap symmetry during our revision procedures. The PCO was increased in 76% of the revision TKA in this study cohort. The medial PCO increased on average 4.2 mm while lateral PCO increased on average 3.3 mm. These medial and lateral values were not expected to be equal because of the resultant kinematic alignment for this revision TKA cohort. Within the present study, the posterior femoral condyles required larger augments than the distal femur to again address the larger flexion gap and maximize PCO ( Figure 2 , Figure 3 ). With PCO being such a critical factor in revision TKA, the accuracy and precision of robotic-assisted surgery is of particular interest for future research. The accuracy of preoperative bone resections and final coronal limb alignment using the MAKO system has previously been reported to be within 1 mm of the plan and 0.78°, respectively [ 28 ]. Unfortunately, the accuracy of the robotic system during revision scenarios is still unknown. Kang et al [ 29 ] reported that up to 7° of femoral component flexion increased the mechanical advantage of the quadriceps muscles and decreased the patellofemoral contact stress. On the contrary, excessive femoral component flexion can result in impingement between the femoral box and the polyethylene post during knee extension [ 21 , 29 ]. Between 1° and 2° of component flexion decreases the flexion gap by an average of 1 mm depending on implant geometry according to several studies [ 3 , 7 , 20 , 21 ]. For the current cohort, posterior translation of the femoral component occurred with femoral flexion beyond 5° in an attempt to keep the anterior flange flush with the anterior femoral cortex to prevent overstuffing the patellofemoral joint and disruption of extensor mechanism kinematics. The average femoral component flexion measured 5.9° for the current cohort, which was slightly larger than the 3° to 5° of flexion suggested during primary TKA [ 14 ]. The degree of femoral component flexion appears to be variable and dependent on patient sex and other anthropometric measurements including body height and weight in addition to gap balance [ 30 , 31 ]. Recent studies have shown that revision constructs that consist of cones and short stems can assist with improved implant alignment and reduced micromotion at the bone-implant interface [ [32] , [33] , [34] ]. In addition, it would be difficult to attain adequate femoral component flexion with an attached metaphyseal sleeve since they are directly coupled to the IM stem [ 4 , 33 ]. Furthermore, utilization of longer IM stems may cause the proximal tip to engage the anterior femoral cortex and inadvertently cause the femoral component to assume an extended position and disrupt gap balance [ 7 , 21 , 35 ]. Traditionally with conventional manual revision TKA, a series of trial-and-error scenarios are attempted to achieve “an acceptable” soft-tissue “feel” of a balanced knee. This method has been shown to be difficult to reproduce among surgeons because of a lack of objective data [ 36 ]. A dynamic interplay exists between the coronal, sagittal, and axial orientations of a knee prosthesis [ 37 , 38 ]. Adjustment of position within one plane can directly affect the other 2 planes due to implant geometry as calculated by the computer. Surgeons may now be able to further quantify these small incremental changes when using robotic assistance during revision procedures. Robotic revision TKA can provide the surgeon with immediate visual feedback regarding implant sizes and the need for augmentation. This information has the potential to improve operating efficiency and may limit the amount of time spent trialing during revision surgery. This article demonstrates re-establishing the PCO contributes to intraoperative gap balance. Although beyond the scope of this study, the overwhelming majority of patients reported a more stable knee after their robotic revision surgery. Early in the process, the authors utilized fully constrained components for robotic revision cases due to the learning curve associated with the off-label use of the robotic arm system. Nonetheless, as the operative technique and experience evolved, the tendency now is to use a standard posterior stabilized component because of the ability to restore soft-tissue tension throughout knee range of motion as long as the collateral ligaments have not been compromised. Finally, the ability to attain specific targets or achieve “optimal alignment” with robotic assistance does not always translate into clinical success and/or improved patient satisfaction because of the countless variables encountered during revision surgeries. Radiographs can depict ideal implant orientation and correlate with an impeccable physical examination, but in the end, the patient can still remain dissatisfied. Preoperative planning continues to be the gold standard of revision TKA surgery, but in the future, robotic assistance may permit intraoperative fine tuning of this plan, which may someday contribute to improved patient outcomes.

Conclusions This current series demonstrates the authors’ up-to-date experience using the MAKO robotic system for revision TKA. Results of this study indicate that sagittal alignment of the revision implants, specifically the femoral PCO and tibial component slope, are statistically significant considerations for a stable revision TKA. By utilization of robotic assistance, the surgeon can become more cognizant of how appropriate implant size and alignment directly affects the ligamentous tension that is important for a functional revision TKA. The MAKO robotic arm system is not yet approved for revision scenarios, so the utility of this technique and outcomes of this study are still considered investigational. Future research and software iterations will be needed to determine the feasibility of robotic-assisted revision TKA.

Background The application of robotic-assisted arthroplasty in revision knee scenarios continues to evolve. This study compares the pre- and post-revision implant positions in series of revision total knee arthroplasties (TKA) using a robotic arm system. Methods Twenty-five consecutive off-label robotic-assisted revision TKA were performed. After virtual revision femoral and tibial components were positioned to achieve “balanced” medial and lateral flexion and extension gaps, the existing primary implants (PI) were removed, and bone cuts were executed with the robotic arm system. Preoperative coronal, sagittal, and axial position of the PI was compared to the final planned positions of the robotic revision implants (RRI) for each subject. A repeated measures ANOVA using the absolute difference in millimeters and degrees between the PI and RRI orientation was completed. Results Intra-operatively, the virtual gaps were balanced within the planning software followed by successful execution of the plan. There was a statistically significant difference between posterior condylar offset and tibial component positioning for RRI compared to PI. There was no difference between the distal femoral component values between PI and RRI. Conclusions The sagittal alignment of the revision implants, specifically the femoral posterior condylar offset and tibial component slope, are statistically significant considerations for a stable revision TKA with off-label use of a robotic-arm system. Other potential benefits may include appropriate implant sizing which can affect the resultant ligamentous tension important for a functional revision TKA. Future research and software iterations will be needed to determine the overall accuracy and utility of robotic-assisted revision TKA. Keywords

Limitations This study includes a cohort of 25 patients from a single institution. A prospective randomized controlled trial comparing manual to robotic revision TKA will be necessary to evaluate long-term outcomes of this technique. Future studies will need to include more patients across several institutions after a standardized technique is established. The accuracy of this robotic system in primary TKA has been reported, but application of the system during revision scenarios is still unknown [ 28 ]. Future studies may focus on quantifying changes in gap balance with the associated change in implant positions. This work is based on basic principles of restricted kinematic alignment referenced to a lower-limb mechanical axis, but other alignment philosophies exist. Currently there are no defined ligamentous tension values for knee arthroplasty; hence, attainment of a particular gap value may not fully represent physiologic performance that is applicable for all patients. Finally, the cost of robotic revision surgery is currently unknown. Forthcoming analyses incorporating costs from a preoperative CT scan, additional operating time, robotic system maintenance, software updates, disposable equipment, and correlation with patient outcomes are in order to establish the feasibility of robotic assistance for revision TKA. Conflicts of interest The authors declare there are no conflicts of interest. For full disclosure statements refer to https://doi.org/10.1016/j.artd.2023.101310 . Author contributions M.W.B., A.C., J.L., M.L.M., and R.P. contributed to writing—review & editing and writing—original draft. M.W.B. and A.C. contributed to supervision and framed the study methodology. M.W.B., A.C., and M.L.M. contributed to investigation and study conceptualization. T.E.H. performed formal analysis and data curation. CRediT authorship contribution statement Micah MacAskill: Writing – review & editing, Writing – original draft, Investigation, Conceptualization. Richard Peluso: Writing – review & editing, Writing – original draft. Jonathan Lash: Writing – review & editing, Writing – original draft. Timothy E. Hewett: Formal analysis, Data curation. Matthew Bullock: Writing – review & editing, Writing – original draft, Supervision, Methodology, Investigation, Conceptualization. Alexander Caughran: Writing – review & editing, Writing – original draft, Supervision, Methodology, Investigation, Conceptualization.

Supplementary data Acknowledgments The authors would like to acknowledge and thank our MAKO product Specialist for their assistance with this study. Dr. Bullock Serves on the Editorial board for Arthroplasty Today. He will be recused from any peer review and editorial decisions.

CC BY

no

2024-01-16 23:40:18

Arthroplast Today. 2023 Dec 27; 25:101310

oa_package/1e/96/PMC10788208.tar.gz

PMC10788210

38226034

Experimental Design, Materials and Methods The serial process through four main steps was conducted to produce a non-commercial tourism dataset in supporting valuable data for research purposes in a smart tourism industry through a tourism recommender systems development as presented in Fig. 2 . Through the WebHarvy crawler module, the unstructured single table was produced from the TripAdvisor website. The following steps were conducted manually using Microsoft Excel incorporating Google Maps for user location search. Previous studies in dataset creation by [14] , [15] , [16] have been reviewed to build comprehensive knowledge in methodology and experimental design. The TripAdvisor website provides essential information for recommender systems regarding the available information of users, items, ratings, and other transaction attributes in tourism activities. The next step involves crawling the data resulting in a single table in Excel format for tourism activities in Bali, Malang, and Yogyakarta regions. The crawled data, rich in attributes, requires thorough pre-processing to develop an adequate dataset suitable for recommender systems. This study conducted the following steps for the crawled raw data that has been stored in a single unnormal table: • Selecting Data: selecting only the textual data from tourism transactions as candidates for the datasets. • Removing Sparsity: Selecting only known data as candidates for the dataset, excluding missing values like unknown users, unknown items (attractions), and unknown transaction attributes such as user location, item location, visit time (month, year), and visit mode. • Normalizing Data Values: Manually correcting unnormalized data in cells. For example, “Australind” and “Australian” were revised to “Australia” for country names, “zÃ1⁄4rich” was corrected to “Zurich” for city names, and unknown attraction types were identified by finding similar types. • Splitting Columns: Separating columns with multiple values into unique attributes, such as dividing a date column into VisitMonth, VisitYear, and VisitMode. • Adding Columns: Enhancing data completeness by adding new columns like region and continent for user profile completeness. The values for these new columns were derived from Google Maps searches based on existing user address data, which may include city and country. • Splitting the Table: Dividing the table into seven normalized tables: Transaction, User, Item, Type, Mode, City, and Country. • Adding New Tables: Inserting additional tables for Region, Continent. • Encoding Data Values: Converting data values into numeric types to facilitate smooth computations in various Machine Learning techniques. Fig. 3 visualizes the rating recap of the dataset through two perspectives which are the mean ratings and number of ratings according to the result of the dataset development process. A high-level interface for drawing informative statistical graphs for the dataset is provided through Seaborn, a Python-based data visualization library. This data visualization offers an overview of the average distribution of ratings for tourism products, based on the 52,930 transactions included in this dataset. It allows for diverse assessments of tourism user experiences, as reflected by the given average ratings. The dataset is ready to be used for studies in recommender systems through some techniques in machine learning.

The tourism industry has currently grown in various aspects, including the types of attractions, their quantity, and the number of tourist visits in various regions, contributing positively to both regional and global economies. Historical transactions are essential for developing recommender systems, utilizing techniques such as Collaborative Filtering and Demographic Filtering. TripAdvisor is a reputable website providing a wide range of accessible tourism information, including attractions, user profiles, and ratings. However, this unstructured raw data requires processing to create an adequate dataset for recommender systems. This study conducted a series of data processing steps on the raw data, including data restructuring, validation, content addition, integration with Google Maps, normalization, and modeling. This study successfully produced an original dataset comprising User Transaction, Item or Attraction, Attraction Type, Continent, Region, Country, City, and Visiting Mode. It also includes an entity relational model for tourism in Indonesia, particularly in Bali, Malang, and Yogyakarta regions, based on various global user experiences. This dataset is adequate and essential for developing various models of tourism recommender systems such as using Collaborative Filtering. Keywords

Specifications Table Value of the Data • The data values encompass various functionalities and aspects of tourism, including users, items (tourism attractions), ratings, and visiting attributes, all of which are essential inputs for studies in recommender systems. • To facilitate research on recommender systems using Machine Learning techniques, the dataset was normalized and selectively includes data from the last 10 years. Additionally, it features only the top 10 most popular items from each region in Bali, Malang, and Yogyakarta. • This dataset is valuable for researchers in the field of Artificial Intelligence (AI), specifically for conducting research in tourism recommender systems utilizing certain techniques in Machine Learning (ML). • The data values are logically interconnected, building a comprehensive perspective of tourism visits in Indonesia's Bali, Malang, and Yogyakarta regions. The relational keys within the dataset facilitate researchers' understanding and utilization. Data Description The tourism industry has currently grown in various aspects, including the types of attractions, their quantity, and the number of tourist visits in various regions, contributing positively to both regional and global economies [3] . There is an urgent need for studies in this sector to enhance service quality through smart solutions like recommender systems, which offer more personalized tourism experiences [4] . The tourism dataset offers crucial data for experimental studies in tourism recommender systems. The source of the dataset was crawled from a reputable website namely TripAdvisor website in providing tourism historical transactions in Indonesia. This data source draws inspiration from previous studies [ 5 , 6 ] in tourism recommender systems. The raw dataset that was crawled from this website consists of some unnormalized attributes for the descriptions of users, items, and transactions (ratings), unstructured data format, and consist of sparse data values, so it couldn't be used for some techniques in recommender systems. Through some data processing for the raw dataset incorporated with the Google Maps search engine, this study has succeeded in providing the original tourism dataset for tourism recommender systems that functionally consists of 52,930 transactions, 33,530 users, and 30 items. The relationship between the entities in the dataset is presented in Fig. 1 . Fig. 1 facilitates the description of each entity, providing detailed information about every data piece in the dataset. The relationships within the dataset serve as a guideline for studies focused on implementing recommender system applications. Definitions for each column name in the normalized dataset are detailed from Table 1 , Table 2 , Table 3 , Table 4 , Table 5 , Table 6 , Table 7 , Table 8 , Table 9 . Each user transaction is listed in separate rows, with every column displaying encoded values. Research in recommender systems utilizing this dataset can be executed using various machine learning techniques such as Collaborative Filtering [ 1 , [7] , [8] , [9] ], Content-Based Filtering [10] , Demographic Filtering [1] , Centex-Aware [ 2 , 11 ], and Hybrid Technique [ 1 , 5 ] for smart tourism solution [12] . Table 1 presents user profiles regarding their geographic location produced by combining data sources from TripAdvisor and search results from Google Maps. The other attributes are collected from Google Maps manually through the user's country and the user's city from TripAdvisor. Table 2 presents the past tourism activities of each user along with the time of occurrence, accompanied by the impressions of each user while visiting each attraction. The attributes offering geographic location information are generated using the same process as that for the User table. This data facilitates the analysis of user experience and interests, presenting opportunities for further research in the realm of recommender systems. The dataset encompasses a total of 52,930 transactions. Table 3 describes the user's continent in terms of the user's geographic location. There are 5 continents provided in this dataset. Table 4 describes the user's region in terms of the user's geographic location. There are 21 regions provided in this dataset. Table 5 describes the user's country in terms of the user's geographic location. There are 164 countries provided in this dataset. Table 6 describes the user's city in terms of the user's geographic location. There are 9,142 cities provided in this dataset. Table 7 describes the user's mode of visit when using tourism products or attractions. Based on the results of data processing, five modes of visits were obtained: Business, Couples, Family, Friends, and Solo. Table 8 describes the tourism products or attractions that have been visited by each tourist, which includes the attraction code, attraction name, type of attraction, and location of the attraction. There are 30 attractions provided in this dataset that are distributed in three areas in Indonesia: Bali, Malang, and Yogyakarta. Table 9 describes the type of attractions that were visited by each tourist. The attraction types were obtained from the website of TripAdvisor as a data source. There are 17 attraction types provided in this dataset. The transaction table, serving as a coordinating table in the relational data schema with other tables (entities), contains foreign keys that facilitate the retrieval of further information through join or merge processes with other entities. Utilizing this dataset allows for smoother soft computation compared to direct processing of the original, crawled data. The dataset consists of normalized data that has many perspectives for analysis. Table 1 provides some opportunities for study in recommender systems through some techniques in Machine Learning. Table 9 presents the aggregated number of visiting users by user's continents and visiting months. This dataset can also be utilized for analysis in the tourism industry, particularly in areas requiring greater attention, like product and service development, tailored to user demographics and visit timings. To ensure a pleasant tourism experience, it is vital to cater to various preferences, ensuring the provision of quality facilities and service [13] . Limitations While the dataset meets the requirements for use in research and development of a tourism recommendation system, its scope is limited to transactions from only three popular places in Indonesia: Bali, Malang, and Yogyakarta. This dataset still presents the ten most popular tourist attractions for each of these regions. In future developments, we propose to add several transactions for more regions and attractions. Ethics Statement The purpose of this dataset is to provide useful data for research purposes, not for commercial use. WebHarvy crawling module was also conducted for collecting public tourism information in the previous study [1] .The users' identity information from the dataset's source has been substituted with new and unique identifiers ensuring that the original identities remain inaccessible to the public . This article's dataset has been fully anonymized, and the platform's data redistribution policy has been complied with. CRediT authorship contribution statement Choirul Huda: Conceptualization, Methodology, Data curation, Software. Yaya Heryadi: Writing – original draft, Supervision. Lukas: Validation, Writing – review & editing. Widodo Budiharto: Visualization, Investigation, Supervision.

Data Availability Tourism Dataset (Original data) (Mendeley Data) Acknowledgements This research did not receive any specific grant from public, commercial, or not-for-profit funding agencies. Declaration of Competing Interest The authors state that they have no competing financial interests or personal relationships that could influence the work reported in this paper.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 19; 52:109990

oa_package/0a/9c/PMC10788210.tar.gz

PMC10788214

38226030

Experimental Design, Materials and Methods For the calculation, we have created four atom-based face-centered cubic unit cell and calculated its structural, electrical, and magnetic properties in both regular and inverse Heusler phases. It's worth mentioning here that inverse Heusler phase is a disordered structure of the full Heusler phase and can be obtained by swapping one of the X -site atoms with the Y -site. The density functional equations were solved using the plane-wave pseudopotential and projector-augmented-wave (PAW) approaches. The unknown exchange-correlation energy functional within the Khon-Sham equation was parametrized within the Perdew-Bruke-Ernzerhof (PBE) version of the generalized gradient approximation GGA. A Monkhorst-Pack special k-point mesh of 20 × 20 × 20, covering the irreducible part of the Brillouin zone was used for k-space integration. In our calculation, we didn't optimize cutoff energies, instead, we used very high cutoff to minimize error. The wave-function cutoff was set to 250 Ry for all alloy and element calculations and the kinetic energy cutoff for charge density for all cases were four times the wave-function cutoff i.e., 1000 Ry. (1 Ry = 13.6057 eV). The convergence of the total energy to a minimum value of 10–9 Ry was set as a criterion for the convergence of self-consistency loop and force convergence cutoff 10 −5 Ry was used for all alloys and element calculations. The equilibrium lattice constant of these compounds was obtained by fitting the quadratic energy-volume graph, utilizing the Birch-Murnaghan equation of state. Where E T is the total energy for given volume V, B 0 is the bulk modulus at the equilibrium volume V 0 and B′ 0 is the pressure derivative of the bulk modulus at the equilibrium volume V 0 . The ground state lattice structure was obtained by relaxing the system using conjugate-gradient method which allows the cell shape and volume change freely until it finds the ground-state. All alloys were considered spin-polarized and a small non-zero starting magnetization value was set for all atomic types in the alloy. Pseudopotentials from Quantum Espresso PSLibrary (generated by A. Dal Corso) [13] were used in our calculations and are included in the Supplementary Material. The formation energy of the alloys is calculated by subtracting the energy corresponding to individual element from the calculated total energy ( E total ) of the alloy. The value of E total was obtained from the ground state self-consistent (SCF) calculations. For elemental calculations, the calculation steps and convergence parameters were the same as what we used for alloy calculations. The known ground state structure was used for all elements, e.g., simple hexagonal structure for Scandium, Hafnium, Zirconium, bcc structure for Chromium, Iron, fcc structure for Copper, Silicon, Aluminum, etc. A non-SCF calculation was performed after the self-consistent field (SCF) calculation to get the density of states (DOS) and projected density of states (pDOS) of each of these alloys. Spin Polarization ( was calculated by using the density of states (DOS) at the Fermi energy in spin-up and -down states. The calculations were performed without considering the effect of spin-orbit coupling.

This paper contains data and results from Density Functional Theory (DFT) investigation of 423 distinct X 2 YZ ternary full Heusler alloys, where X and Y represent elements from the D -block of the periodic table and Z signifies element from main group. The study encompasses both “regular” and “inverse” Heusler phases of these alloys for a total of 846 potential materials. For each specific alloy and each phase, a range of information is provided including total energy, formation energy, lattice constant, total and site-specific magnetic moments, spin polarization as well as total and projected density of electronic states. The aim of creating this dataset is to provide fundamental theoretical insights into ternary X 2 YZ Heusler alloys for further theoretical and experimental analysis. Keywords

Specification Table Value of the Data • This dataset will provide fundamental information about 423 ternary Heusler alloys in their regular ( L2 1 ) and inverse ( XA ) Heusler phases, for a total of 846 distinct alloys calculated. Structure, magnetic properties, and spin polarization values are provided as a launching point for future studies. • This wide-scale database will serve as a valuable screening tool for identifying promising candidates and conducting thorough and comprehensive studies on ternary Heusler alloys. • The dataset is also useful for machine learning studies of structure, phase stability, and electronic and magnetic properties of alloy systems. Objective Heusler alloys have attracted substantial attention within the research community due to their potential applications in fields such as spintronics [2] and thermo-electric devices [3] . Considerable focus is now being paid to identify novel candidates with advanced properties and their possible application in various fields. The enclosed comprehensive dataset of Heusler alloys, including structure properties and stability, magnetic structure and spin polarization, and electronic structure via density of states calculations, is valuable for researchers seeking to identify novel candidates and more quickly progress to subsequent in-depth theoretical and experimental exploration. Data Description Heusler compounds first attracted interest when Cu 2 MnAl was discovered by German scientist Fritz Heusler in 1903 [4] . This remarkable face-centered cubic compound exhibited ferromagnetic properties although none of its constituent elements possessed inherent ferromagnetism. Motivated by this discovery, researchers worldwide have identified over a thousand Heusler alloys and their relatives within the past century and the quest for more continues to this day. To facilitate this quest, we have generated collections of X 2 YZ Heusler alloys. This dataset stands out as one of the few comprehensive databases accessible for researchers in this field [5] , offering a complementary source of data alongside the well-known Open Quantum Materials Datasets (OQMD) [ 6 , 7 ], Material Projects (MP) [8] and Automatic-FLOW for Materials Discovery (AFlow) [9] . Data on Materials Project (MP) are collected from two sources: (1) high-throughput calculations on supercomputing clusters; (2) academic community using MPContribs [10] . Data collected with (1) came from calculations performed using VASP [11] , a well-reputed package that uses a full-potential method. However, self-consistent calculations (SCF) of many Heusler compounds and non-self-consistent calculations (NSCF) prior to DOS calculations were performed with a low k-point grid. Data for MP, OQMD and AFLOW were uploaded via Application Programming Interface (API) [12] from a supportive academic community comprise of many experienced researchers. However, the lack of a consistency in computation methods and parameters across over a thousand Heusler compounds makes it very difficult to create models of generalized behavior of Heuslers. Moreover, MP, OQMD and AFLOW websites contain only a subset of Heusler compounds, many of which are included in our dataset. For example, the X 2 VAl Heusler series featured in this manuscript, with X-sites choices Sc, Cu, Y, Nb, Mo, Hf, Ta and W, are not included in the MP website. For our calculation, we have selected 18 D -block (including 2 rare earth) elements for X-site, most of these elements have also been used as Y-site and 6 main group elements as Z-site. Fig. 1 shows the orthogonal element choices for this work, sorted outward by atomic radius. Our calculations are based on a four atom-based face-centered cubic unit cell with 2 X-sites atoms and 1 Y and 1 Z site atoms, which represents one formula unit. Full Heusler structures consist of four interpenetrating face-centered cubic (fcc) sublattices positioned at (0, 0, 0), (1/4, 1/4, 1/4), (1/2, 1/2, 1/2), and (3/4, 3/4, 3/4). The sublattice sites (0, 0, 0) and (1/2, 1/2,1/2) are occupied by X atoms in a regular Heusler. The remaining sublattice sites (1/4, 1/4, 1/4) and (3/4,3/4, 3/4) are occupied by Y and Z elements respectively. 423 out of more than a thousand combinations have been calculated and presented in this dataset. Table 1 summarizes the data collected on a series of X 2 VAl ( X = Sc, Ti, Cr, Mn, Fe, Co, Ni, Cu, Y, Zr, Nb. Mo, Hf, TA, W) full (L2 1 ) and inverse (XA) Heusler alloys. The total energy of the system can be conceptualized as the energy required to construct the system at absolute zero temperature by assembling single atoms of constituting elements. The calculations of density of states show that Y 2 VAl, Zr 2 VAl and Hf 2 VAl are half metals in their inverse Heusler phases. Fig. 2 represents the total and atom-resolved density of states of X 2 VAl alloy series in their L2 1 and XA phases. The total density of states (DOS) and site-specific projected density of states (pDOS) of X 2 VAl where X is one of the fifteen 3d elements presented in Table 1 are shown in Fig. 2 . The L2 1 phases are displayed on the left and the XA phases are on the right side of the figure. The X -pDOS is the sum of two X atoms in the alloy. The Fermi energy ( E F ) is shifted to zero and is marked by a vertical line. List of all included alloys, organized by Y-Z combination. X 2 FeAl: Sc 2 FeAl, Ti 2 FeAl, V 2 FeAl, Cr 2 FeAl, Mn 2 FeAl, Co 2 FeAl, Ni 2 FeAl, Cu 2 FeAl, Y 2 FeAl, Zr 2 FeAl, Nb 2 FeAl, Mo 2 FeAl, Hf 2 FeAl, Ta 2 FeAl, W 2 FeAl, Dy 2 FeAl, Gd 2 FeAl X 2 FeGe: Sc 2 FeGe, Ti 2 FeGe, V 2 FeGe, Cr 2 FeGe, Mn 2 FeGe, Co 2 FeGe, Ni 2 FeGe, Cu 2 FeGe, Y 2 FeGe, Zr 2 FeGe, Nb 2 FeGe, Hf 2 FeGe, Ta 2 FeGe, W 2 FeGe, Dy 2 FeGe, Gd 2 FeGe X 2 FeSi: Sc 2 FeSi, Ti 2 FeSi, V 2 FeSi, Cr 2 FeSi, Mn 2 FeSi, Co 2 FeSi, Ni 2 FeSi, Cu 2 FeSi, Y 2 FeSi, Zr 2 FeSi, Hf 2 FeSi, Ta 2 FeSi, W 2 FeSi, Gd 2 FeSi, Dy 2 FeSi X 2 FeGa: Sc 2 FeGa, Ti 2 FeGa, V 2 FeGa, Cr 2 FeGa, Mn 2 FeGa, Ni 2 FeGa, Cu 2 FeGa, Y 2 FeGa, Zr 2 FeGa, Ta 2 FeGa, Gd 2 FeGa X 2 FeBi: Cr 2 FeBi X 2 FeSn: Cr 2 FeSn X 2 TiAl: Sc 2 TiAl, V 2 TiAl, Cr 2 TiAl, Mn 2 TiAl, Fe 2 TiAl, Co 2 TiAl, Ni 2 TiAl, Cu 2 TiAl, Y 2 TiAl, Zr 2 TiAl, Nb 2 TiAl, Mo 2 TiAl, Hf 2 TiAl, Ta 2 TiAl, W 2 TiAl, Gd 2 TiAl, Dy 2 TiAl X 2 TiSi: Sc 2 TiSi, V 2 TiSi, Cr 2 TiSi, Mn 2 TiSi, Fe 2 TiSi, Co 2 TiSi, Ni 2 TiSi, Cu 2 TiSi, Y 2 TiSi, Zr 2 TiSi, Nb 2 TiSi, Mo 2 TiSi, Hf 2 TiSi, Ta 2 TiSi, W 2 TiSi, Gd 2 TiSi, Dy 2 TiSi X 2 TiGe: Sc 2 TiGe, V 2 TiGe, Cr 2 TiGe, Mn 2 TiGe, Fe 2 TiGe, Co 2 TiGe, Ni 2 TiGe, Cu 2 TiGe, Y 2 TiGe, Zr 2 TiGe, Nb 2 TiGe, Mo 2 TiGe, Hf 2 TiGe, Ta 2 TiGe, W 2 TiGe, Gd 2 TiGe, Dy 2 TiGe X 2 TiGa: V 2 TiGa, Cr 2 TiGa, Mn 2 TiGa, Fe 2 TiGa, Co 2 TiGa, Ni 2 TiGa, Cu 2 TiGa, Zr 2 TiGa, Ta 2 TiGa, W 2 TiGa, Gd 2 TiGa, Dy 2 TiGa X 2 TiSn: Co 2 TiSn X 2 YAl: Sc 2 YAl, Ti 2 YAl, V 2 YAl, Cr 2 YAl, Mn 2 YAl, Fe 2 YAl, Co 2 YAl, Ni 2 YAl, Cu 2 YAl, Zr 2 YAl, Nb 2 YAl, Hf 2 YAl, Ta 2 YAl, W 2 YAl, Gd 2 YAl, Dy 2 YAl X 2 YGe: Sc 2 YGe, Ti 2 YGe, V 2 YGe, Cr 2 YGe, Mn 2 YGe, Fe 2 YGe, Co 2 YGe, Ni 2 YGe, Cu 2 YGe, Zr 2 YGe, Nb 2 YGe, Hf 2 YGe, Ta 2 YGe, W 2 YGe, Gd 2 YGe, Dy 2 YGe X 2 YSi: Sc 2 YSi, Ti 2 YSi, V 2 YSi, Cr 2 YSi, Mn 2 YSi, Fe 2 YSi, Co 2 YSi, Ni 2 YSi, Cu 2 YSi Zr 2 YSi, Nb 2 YSi, Hf 2 YSi, Ta 2 YSi, W 2 YSi, Gd 2 YSi, Dy 2 YSi X 2 YGa: Sc 2 YGa, Ti 2 YGa, Cr 2 YGa, Mn 2 YGa, Fe 2 YGa, Co 2 YGa, Ni 2 YGa, Cu 2 YGa, Zr 2 YGa, Nb 2 YGa, Hf 2 YGa, Ta 2 YGa, W 2 YGa, Gd 2 YGa X 2 YBi: Sc 2 YBi, Ti 2 YBi, V 2 YBi, Cr 2 YBi, Mn 2 YBi, Fe 2 YBi, Co 2 YBi, Ni 2 YBi, Cu 2 YBi, Zr 2 YBi, Nb 2 YBi, Mo 2 YBi, Hf 2 YBi, Ta 2 YBi, W 2 YBi, Dy 2 YBi, Gd 2 YBi X 2 YSn: Sc 2 YSn, Ti 2 YSn, V 2 YSn, Cr 2 YSn, Mn 2 YSn, Fe 2 YSn, Co 2 YSn, Ni 2 YSn, Cu 2 YSn, Zr 2 YSn, Nb 2 YSn, Mo 2 YSn, Hf 2 YSn, W 2 YSn, Ta 2 YSn X 2 CuAl: Sc 2 CuAl, Cr 2 CuAl, Mn 2 CuAl, Fe 2 CuAl, Co 2 CuAl, Ni 2 CuAl, Nb 2 CuAl, Hf 2 CuAl X 2 CuSi: Cr 2 CuSi, Mn 2 CuSi, Fe 2 CuSi, Co 2 CuSi, Ni 2 CuSi X 2 CuGa: Sc 2 CuGa, Ti 2 CuGa, V 2 CuGa, Cr 2 CuGa, Mn 2 CuGa, Fe 2 CuGa, Co 2 CuGa, Ni 2 CuGa, Y 2 CuGa, Zr 2 CuGa, Nb 2 CuGa, Mo 2 CuGa, Hf 2 CuGa, Ta 2 CuGa, W 2 CuGa, Dy 2 CuGa, Gd 2 CuGa X 2 CuGe: Sc 2 CuGe, Ti 2 CuGe, V 2 CuGe, Cr 2 CuGe, Mn 2 CuGe, Fe 2 CuGe, Co2 2 CuGe, Ni 2 CuGe, Y 2 CuGe, Zr 2 CuGe, Nb 2 CuGe, Mo 2 CuGe, Hf 2 CuGe, Ta 2 CuGe, W 2 CuGe, Dy 2 CuGe, Gd 2 CuGe X 2 CuSn: Mn 2 CuSn X 2 CuSn: Mn 2 CuBi X 2 CoAl: Sc 2 CoAl, Ti 2 CoAl, V 2 CoAl, Cr 2 CoAl, Mn 2 CoAl, Fe 2 CoAl, Ni 2 CoAl, Cu 2 CoAl, Y 2 CoAl, Zr 2 CoAl, Nb 2 CoAl, Mo 2 CoAl, Hf 2 CoAl, Ta 2 CoAl, W 2 CoAl, Dy 2 CoAl, Gd 2 CoAl X 2 CoGa: Ti 2 CoGa, Cr 2 CoGa Fe 2 CoGa, Ni 2 CoGa, Gd 2 CoGa, Dy 2 CoGa X 2 CoGe: Cr 2 CoGe, Mn 2 CoGe, Fe 2 CoGe, Ni 2 CoGe, Gd 2 CoGe, Dy 2 CoGe X 2 CoSi: Cr 2 CoSi, Mn 2 CoSi, Fe 2 CoSi, Ni 2 CoSi, Cu 2 CoSi, Mo 2 CoSi, Gd 2 CoSi, Dy 2 CoSi X 2 CoBi: Cr 2 CoBi, V 2 CoBi X 2 CoSn: Cr 2 CoSn, V 2 CoSn X 2 MnAl: Sc 2 MnAl, Cr 2 MnAl, Fe 2 MnAl, Co 2 MnAl, Ni 2 MnAl, Cu 2 MnAl, Nb 2 MnAl, Hf 2 MnAl X 2 MnSi: Sc 2 MnSi, Ti 2 MnSi, V 2 MnSi, Cr 2 MnSi, Fe 2 MnSi, Co 2 MnSi, Ni 2 MnSi, Cu 2 MnSi X 2 MnGe: Sc 2 MnGe, Cr 2 MnGe, Co 2 MnGe, Ni 2 MnGe X 2 MnGa: Cr 2 MnGa, Co 2 MnGa, Ni 2 MnGa X 2 MnSn: Co 2 MnSn X 2 MnBi: Co 2 MnBi X 2 CrAl: Sc 2 CrAl, Fe 2 CrAl, Mn 2 CrAl, Ni 2 CrAl, Nb 2 CrAl, Hf 2 CrAl, Dy 2 CrAl, Gd 2 CrAl X 2 CrSi: Fe 2 CrSi, Ni 2 CrSi, Gd 2 CrSi, Dy 2 CrSi X 2 CrGa: Ni 2 CrGa, Gd 2 CrGa, Dy 2 CrGa X 2 CrGe: Ni 2 CrGe, Gd 2 CrGe, Dy 2 CrGe X 2 NiAl: Sc 2 NiAl, Cr 2 NiAl, Cu 2 NiAl, Nb 2 NiAl, Hf 2 NiAl, Ta 2 NiAl X 2 NiSi: Cr 2 NiSi X 2 NiGa: Cr 2 NiGa X 2 NiGe: Cr 2 NiGe X 2 NiBi: Cr 2 NiBi X 2 NiSn: Cr 2 NiSn X 2 ScAl: Cr 2 ScAl, Fe 2 ScAl, Ni 2 ScAl, Cu 2 ScAl, Nb 2 ScAl, Hf 2 ScAl X 2 ScSi: Cr 2 ScSi, Fe 2 ScSi, Ni 2 ScSi X 2 ScGa: Cr 2 ScGa, Fe 2 ScGa, Ni 2 ScGa X 2 ScGe: Cr 2 ScGe, Fe 2 ScGe, Ni 2 ScGe X 2 ScSn: Co 2 ScSn X 2 ZrAl: Cr 2 ZrAl, Fe 2 ZrAl, Ni 2 ZrAl, Nb 2 ZrAl X 2 ZrSi: Cr 2 ZrSi, Fe 2 ZrSi, Ni 2 ZrSi X 2 ZrGa: Cr 2 ZrGa, Fe 2 ZrGa, Ni 2 ZrGa X 2 ZrGe: Cr 2 ZrGe, Fe 2 ZrGe, Ni 2 ZrGe X 2 VAl: Sc 2 VAl, Ti 2 VAl, Cr 2 VAl, Mn 2 VAl, Fe 2 VAl, Co 2 VAl, Ni 2 VAl, Cu 2 VAl, Y 2 VAl, Zr 2 VAl, Nb 2 VAl, Mo 2 VAl, Hf 2 VAl, Ta 2 VAl, W 2 VAl X 2 VGa: Cr 2 VGa, Fe 2 VGa, Ni 2 VGa X 2 VGe: Cr 2 VGe, Fe 2 VGe, Ni 2 VGe X 2 VSi: Cr 2 VSi, Fe 2 VSi, Co 2 VSi, Ni 2 VSi X 2 NbAl: Fe 2 NbAl, Y 2 NbAl, Hf 2 NbAl, Ta 2 NbAl X 2 TaAl: Hf 2 TaAl X 2 HfGe: Co 2 HfGe X 2 HfAl: Co 2 HfAl X 2 MoAl: Fe 2 MoAl Limitations None. CRediT authorship contribution statement Ridwan Nahar: Conceptualization, Investigation, Data curation, Writing – original draft. Ka Ming Law: Investigation, Data curation. Thomas Roden: Investigation, Data curation. Michael Zengel: Investigation, Data curation. Justin Lewis: Investigation, Data curation. Sujan Budhathoki: Investigation, Data curation. Riley Nold: Investigation, Data curation. Harshil Avlani: Investigation, Data curation. Babajide Akintunde: Investigation, Data curation. Naomi Derksen: Investigation, Data curation. Adam J. Hauser: Conceptualization, Validation, Data curation, Writing – review & editing, Project administration, Funding acquisition.

Data Availability Dataset on Density Functional Theory investigation of Ternary Heusler Alloys (Original data) (Mendeley Data) Ethics Statement The proposed data does not involve any human subjects, animal experiments, or data collected from social media platforms. Acknowledgement The author gratefully acknowledges financial support from the 10.13039/100000001 National Science Foundation (NSF CAREER DMR-2047251) Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 18; 52:109971

oa_package/10/cd/PMC10788214.tar.gz

PMC10788215

38226035

In the pursuit of advancing research in continuous user authentication, we introduce COUNT-OS-I and COUNT-OS-II, two distinct performance counter datasets from Windows operating systems, crafted to bolster research in continuous user authentication. Encompassing data from 63 computers and users, the datasets offer rich, real-world insights for developing and evaluating authentication models. COUNT-OS-I spans 26 users in an IT department, capturing 159 attributes across diverse hardware and software environments over 26 h on average per user. COUNT-OS-II, on the other hand, encompasses 37 users with identical system configurations, recording 218 attributes per sample over a 48-hour period. Both datasets utilize pseudonymization to safeguard user identities while maintaining data integrity and statistical accuracy. The well-balanced nature of the data, confirmed by comprehensive statistical analysis, positions these datasets as reliable benchmarks for the continuous user authentication domain. Through their release, we aim to empower the development of robust, real-world applicable authentication models, contributing to enhanced system security and user trust. Keywords

Specifications Table Value of the Data • Real-World Applicability: The data collected from public organizations in Brazil offers a realistic environment for testing and validating biometric authentication models. • Comprehensive Data: These datasets encompass various performance counters obtained from the Windows OS, providing a broad perspective on system interactions. • Variety in System Configurations: COUNT-OS-I includes data from computers with diverse characteristics and configurations, ensuring the models are adaptable to various environments. Conversely, COUNT-OS-II comprises data from computers with similar characteristics and configurations. • Long-Term Behaviour Analysis: These datasets provide a significant amount of data, averaging 26 h per user for COUNT-OS-I and around 48 h for COUNT-OS-II, enabling the analysis of long-term user behaviour. • Pseudonymization: Through the application of pseudonymization, user privacy is maintained while preserving the integrity and statistical accuracy of the data. Data Description To advance the field of continuous user authentication, we have meticulously crafted two comprehensive datasets: COUNT-OS-I and COUNT-OS-II, each harboring unique characteristics while sharing common ground in their utility and design principles. These datasets encompass performance counters extracted from the Windows operating system, offering an intricate data set for evaluating and refining authentication models in real-world scenarios. Both datasets were derived from real-world settings within public organizations in Brazil, e ensuring their relevance and applicability to real-life situations. Volunteers from diverse professional backgrounds participated in the data collection, contributing to the richness and variability of the data. Furthermore, both datasets were collected at a sample rate of every 5 s, providing a dense and detailed view of user interactions and system performance. The commitment to preserving user confidentiality is unwavering across both datasets, with pseudonymization applied meticulously to safeguard individual identities while maintaining data integrity and statistical robustness. The statistical analysis of the number of instances per users in the COUNT-OS-I and COUNT-OS-II datasets can be found in Table 1 . COUNT-OS-I The COUNT-OS-I dataset was specifically generated in a real-world scenario to evaluate our work on continuous user authentication. This dataset consists of performance counters extracted from the Windows operating system of 26 computers, representing 26 individual users. The data were collected on the computers of the Information Technology Department of a public organization in Brazil. The participants in this study were volunteers, with aged between 20 and 45 years old, consisting of both males and females. Most of the participants were systems analysts and software developers who performed their routine work activities. There were no specific restrictions imposed on the tasks that the participants were required to perform during the data collection process. The participants used a variety of software applications as part of their regular work activities. This included web browsers such as Firefox, Chrome, and Edge, developer tools like Eclipse and SQL Developer, office programs such as Microsoft Office Word, Excel, and PowerPoint, as well as chat applications like WhatsApp. It's important to note that the list of applications mentioned is not exhaustive, and participants were not limited to using only these applications. For the COUNT-OS-I dataset, the data collected is based on computers with different characteristics and configurations in terms of hardware, operating system versions, and installed software. This diversity ensures a representative sample of real-world scenarios and allows for a comprehensive evaluation of the authentication model. During the data collection process, each sample was recorded at a frequency of every 5 s, capturing system data over a period of approximately 26 h, on average, for each user. This duration provides sufficient data to analyze user behaviour and system performance over an extended period. Each sample in the COUNT-OS-I dataset corresponds to a feature vector comprising 159 attributes. These attributes capture various aspects of system performance, including metrics related to CPU utilization, memory usage, disk activity, network traffic, software APIs and other relevant performance counters. Table 2 presents examples of performance counters that were collected from the different datasets. The complete list of performance counters collected from each dataset can be obtained from the dataset website [2] . We apply pseudonymization to hide users' sensitive information by replacing private identifiers with pseudonyms, ensuring the confidentiality of individuals' identities. This technique preserves statistical accuracy and data integrity. COUNT-OS-II This dataset comprises performance counters extracted from the Windows operating system installed on 37 computers. These computers possess identical hardware configurations (CPU, memory, network, disk), operating systems, and software installations. The data collection was conducted within various departments of a public organization in Brazil. The participants in this study (37 users) were voluntary administration assistants who performed various administrative tasks as part of their routine work activities. No restrictions were imposed on the specific tasks they were assigned. The participants commonly utilized programs such as the Chrome browser and office applications like Office Word, Excel, and PowerPoint, in addition to the WhatsApp chat application. The data were collected over six days (approximately 48 h), with sample collected at a 5-second interval. Each sample corresponds to a feature vector composed of 218 attributes. In this dataset, we also apply pseudonymization to hide users' sensitive information. Data Extraction To obtain the Performance Counters features presents in the COUNT-OS-I and COUNT-OS-II datasets, according to Fig. 1 , we follow the simple five-step process of turning basic computer performance information into a cleaned and valuable set of data: 1. Data Extraction: In the first step, we utilize Perfmon [1] , a native performance monitoring tool of the Windows operating system, to collect various system performance metrics. The data is gathered separately for each user, capturing intricate details about their system's performance. Each user's data is then saved into its own CSV file, creating a structured and accessible format for future analysis. 2. Anonymized Labeling: We added an anonymized label to each user within the dataset. This label serves as a unique identifier, ensuring user privacy while allowing us to track and analyze the data across different users. 3. Feature Standardization and Aggregation: In this phase, we remove any special characters that may cause discrepancies in automatic analysis of the data. Additionally, we aggregate performance values by device type. For instance, metrics like 'Network adapter 1 output rate' and 'Network adapter 2 output rate' are combined into a single feature named 'Network adapter output rate - Total,' representing the total network output rate across all network adapters. 4. Data Consolidation and Cleaning: we combine the individual CSV files of all users into a single comprehensive file. During the consolidation process, we began by removing uncommon features among users. This involved excluding attributes not shared by all users to maintain dataset uniformity and prevent user identification through unique attributes. Additionally, we removed features that exhibited no variance in values across users, as these attributes do not significantly contribute to the analysis and could impact the results. As a result of this process, we achieved a cleaned and consistent dataset, containing only the most relevant and diverse features. Limitations Limited Behavioral Context: The datasets focus on performance counters without detailed contextual information about the tasks being performed, which could provide additional insights into user behavior for authentication purposes. Ethics Statement The original authors of the source datasets extracted data samples from performance counters on different computers using the Windows operating system, which were generated by various human subjects. No personally identifiable information was collected as part of the data collection process. Participants provided informed consent for the publication of data and research, and the published data is anonymized. The data collection, conducted following the ethical guidelines established by the Amazonas State Department of Finance, was officially authorized with protocol number 0036–2019/GTEC. The authors declare that they have followed the general ethics rules of scientific research. CRediT authorship contribution statement Cesar Andrade: Conceptualization, Methodology, Software, Investigation, Data curation, Writing – review & editing. Hendrio Bragança: Conceptualization, Methodology, Investigation, Data curation, Formal analysis. Eduardo Feitosa: Conceptualization, Supervision, Writing – review & editing. Eduardo Souto: Conceptualization, Supervision, Writing – review & editing.

Data Availability Performance counter for biometrics authentication (Original data) (Figshare.com) Acknowledgements This research was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior—Brasil (CAPES-PROEX)—Finance Code 001. This work was partially supported by Amazonas State Research Support Foundation—FAPEAM—through the POSGRAD 2022–2023 project. This research, according to Article 48 of Decree no 6.008/2006, was partially funded by Samsung Electronics of Amazonia Ltda, under the terms of Federal Law no 8.387/1991, through agreement no 003/2019, signed with ICOMP/UFAM. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 21; 52:109999

oa_package/9b/c1/PMC10788215.tar.gz

PMC10788219

38226043

Experimental Design, Materials and Methods The population of the data consists of professionals within the construction industry namely Project Managers, Contractors, Engineers, Architects, and Consultants. A total of 150 questionnaires were distributed, 84 via the online SurveyMonkey tool. These included 15 email invitations with 5 responses (33.33%) and 79 web links. Additionally, 66 hard-copy questionnaires were administered and returned. Evidence from the literature provided 41 risk factors associated with cost overruns on public sector projects within the construction industry. The responses were rated on a 7-point Likert scale (1 = extremely low, 2 = very low, 3 = low, 4 = moderate, 5 = high, 6 = very high, 7 = extremely high), to determine the probability and severity of each risk factor. The data collected was analysed in the Statistical Package for Social Sciences (SPSS) IBM 25. Descriptive statistical tools such as frequency, percentage, and mean were used to present the data. Calculations of the risk impact (RI) values, normalisation values, and ranking were carried out in Microsoft Excel 2018. The respondents’ perceptions of the problematic issues related to cost overrun on public sector infrastructure development projects (PSIDPs) were ranked distinctly according to the sector of employment of the respondents ( Table 11 ). The 22 critical factors contributing to cost overrun within Trinidad and Tobago public sector projects were obtained through normalisation of the risk impact (RI) values of the 41 factors and ranked according to the normalised values obtained so that factors having values greater than 0.5 were deemed critical ( Table 13 ). Through the application of fuzzy logic, namely fuzzy synthetic evaluation, the 22 critical risk factors (CRFs) were classified under four critical risk groups (CRGs), namely, political, socio-economical, technical, and psychological, and ranked overall according to their category, based on the risk impact ( Table 14 ) [ 1 ]. The weighing function, of the CRFs, (second-level) and CRGs (first level) are calculated from the mean values, obtained through SPSS for both its probability and severity ( Table 15 ). Next, the membership functions of the CRFs & CRGs (level 1) along with the risk level of each CRG (MF level 2) were determined and presented in Table 16 . The obtained fuzzy evaluation matrixes, of the CRGs (level 2) were further normalized by considering their weighing functions to generate the final fuzzy evaluation matrix of overall risk level (ORL) of cost overrun of social housing development (i.e. level 1). The probability and severity matrixes of the PRFs are represented in column 3 of Table 17 . The overall risk level of cost overrun on public sector projects in developing countries is presented in Table 18 which illustrates that the political category has more risk compared to the others. The outcome of this study indicates that further studies could be conducted to evaluate the cost controlling and monitoring strategies for the identified risk factors of cost overrun on social housing projects and a study on cost planning and estimating mechanisms to mitigate the factors of cost overrun on social housing projects could also be carried out. Furthermore, similar types of studies can be conducted for the other types of building and infrastructure construction projects which will contribute greatly to the existing knowledge and the betterment of the industry.

This data article explores the factors that contribute to cost overrun on public sector projects within Trinidad and Tobago. The data was obtained through literature research, and structured questionnaires, designed using open-ended questions and the Likert scale. The responses were gathered from project actors and decision-makers within the public and private construction industry, mainly, project managers, contractors, engineers, architects, and consultants. The dataset was analysed using frequency, simple percentage, mean, risk impact, and fuzzy logic via the fuzzy synthetic evaluation method (FSE). The significance of the analysed data is to determine the critical root causes of cost overrun which affect public sector infrastructure development projects (PSIDPs), from being completed on time and within budget. The dataset is most useful to project and construction management professionals and academia, to provide additional insight into the understanding of the leading factors associated with cost overrun and the critical group in which they occur (political factors). Such understanding can encourage greater decisions under uncertainty and complexity, thus accounting for and reducing cost overrun on public sector projects. Keywords

Specifications Table Value of the Data • The data set is the first to provide a methodological classification of the leading root causes of cost overruns in public sector social development housing projects. This is useful in acquiring a deeper understanding of these leading root causes and validated against their theoretical ontologies. • The data can be used in decision making research to show the uncertainty, imprecision and complexity of perceptions and heuristics used in the construction industry and their major influences on the economic viability of social developmental projects. The data set shifts the current research agenda in cost overrun studies, exposing the lack of attention to the true leading root causes of cost overruns and adds to contemporary academic debate by encouraging project and construction practitioners to reflect, refocus, reframe, and reset the research agenda to uncover key tacit knowledge areas. • The data can be applied to develop forecasting models to demonstrate the misalignment in the construction housing industry and highlight the gaps in contemporary project practices leading to unsustainable delivery and practices of social housing. The data can be used as a basis of comparison with that of other Small Island Developing States and/or on a worldwide scale, in the field of construction project management. It further updates project management practices by uncovering and prioritising theoretical constructs critical to public sector projects. • The provided data can be utilized by academia and construction project practitioners to develop a multitude of risk assessment processes, models and pragmatic tools based on these critical risk factors for further testing to optimize cost performances and sustainability on this value driven socially dependent infrastructure projects. • The data can be used by policy makers and governmental bodies to analyse the latent effects of critical risk factors grouped under various root causes can have on overall developmental policies, and their emulation on the overall social housing value. These latent effects can be studied to develop strategies to mitigate wicked problems associated with social housing such as crime, unemployment, and income inequalities. Data Description The data was obtained through literature research, and structured questionnaires. A total of 150 questionnaires were distributed to Project Managers, Contractors, Engineers, Architects, and Consultants within the construction industry who have been involved in social housing projects [ 1 ]. The data received from the participants were presented as follows: The data on the highest level of education attained by the respondents is presented in Table 1 which illustrates that more than 70% of respondents have a minimum qualification of a Bachelor of Science degree, data on the professional role ( Table 2 ) which highlighted that respondents represent mainly five professional roles, sector of employment in which they are employed ( Table 3 ) either in the public sector or the private sector, types of projects mainly carried out by the organisations to which the participants belong ( Table 4 ) under main eight categories, the number of employees ( Table 5 ) where that most of the respondents are belonging to the organisations which are having more than 200 employees, number of projects participated in ( Table 6 ), annual estimated turn over ( Table 7 ), expected duration of projects ( Table 8 ), and the actual time spent ( Table 9 ). Table 10 presents data on the number of years of experience of each respondent in the field of project management, consultancy, contracting, engineering, and architecture. Data on the Risk Impact associated with cost overrun on construction projects compared between the private sector and public is presented in Table 11 . The data clearly show that the impact of the factors that contribute to the cost overruns is different between the public and private sectors. Furthermore, Table 12 presented factors contributing to cost overrun on public sector projects which were extracted through the existing literature. The analysis of the raw data (factors presented in Table 12 ), provides the 22 critical factors associated with a cost overrun on public sector projects ( Table 13 ) based on the severity and the probability of each risk whistle analysing the risk impact factor. The data in Table 14 (Data on the classification and ranking of critical risk factors), Table 15 (Data on the weightings for the 22 CRFs and 4 PRFs for Social Housing Program), Table 16 (Data on the membership function of all CRFs and PRFs for Risk Probability and Severity), Table 17 (Data on the membership function of the overall risk level), Table 18 (Data on the overall risk level) presents the levels to the fuzzy logic analysis approach implemented to rank the principal risk groups (Political, Socio-economical, technical and psychological) according to the risk index. Limitations None. Ethics Statement The proposed data does not involve any human subjects, animal experiments, or data collected from social media platforms. CRediT authorship contribution statement Aaron Anil Chadee: Conceptualization, Methodology, Software, Writing – original draft, Investigation. Chamari Allis: Validation, Writing – review & editing. Upaka Rathnayake: Validation, Writing – review & editing. Hector Martin: Validation, Writing – review & editing. Hazi M Azamathulla: Validation, Writing – review & editing.

Data Availability Data exploration on the factors associated with cost overrun on social housing projects in Trinidad and Tobago (Original data) (Mendeley Data) Acknowledgements This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 15; 52:109966

oa_package/11/76/PMC10788219.tar.gz

PMC10788220

38226168

Introduction Cancer is a major public health problem worldwide and has become one of the most common causes of death. 1 , 2 It can occur in various parts of the body and ultimately lead to serious consequences. For example, lung cancer, specifically non-small cell lung cancer (NSCLC), is the primary cause of respiratory system cancer-related deaths, accounting for 1.8 million deaths in 2020. 3 Digestive system tumors, such as esophageal cancer (EC), gastric cancer (GC), colorectal cancer (CRC), and hepatocellular carcinoma (HCC) have high morbidity and mortality rates, making them prominent among malignant tumors. 4 For reproductive system, prostate cancer and ovarian cancer are the most common cancer type in male and female, respectively. 5 , 6 , 7 Notably, ovarian cancer has the highest mortality rate among all gynecological malignancies. 6 Since these various cancers correspond to multiple systems, tumor biomarkers from multi-omics have also been examined. 8 , 9 , 10 , 11 , 12 Most biomarkers focus on the underlying mechanisms of cancer development, such as gene mutations or abnormal expression. However, their effectiveness is not always fairly acceptable due to the complex relationship between genes and phenotypes. Additionally, conventional diagnostics relying on proteomic/genomic biomarkers are limited considering throughput, analysis speed, and invasiveness of sampling. 13 It is urgently needed for novel approaches in cancer diagnosis, risk prediction, and treatment. Glycans, closely involved in the downstream cellular metabolism, can be a promising source for the discovery of new reliable tumor biomarkers, which are more distal over genomic and proteomic approaches for precision cancer diagnostics. Glycosylation is one of the most prevalent post-translational modifications and involves in many fundamental molecular and cell biology processes occurring in cancer, including tumor cell dissociation and invasion, tumor cell–matrix interactions, cell signaling transduction pathways, inflammation and immune response. 14 , 15 It was reported that protein glycosylation was associated with the pathogenesis of many cancers, such as gastric cancer, colorectal cancer, and hepatocellular carcinoma. 14 , 16 , 17 , 18 , 19 , 20 Particularly, glycans play important role in immunosurveillance, with most of the key factors involved in immune response are glycosylated. 15 , 21 , 22 , 23 Previous studies demonstrated that antitumor immunity was an important barrier to tumor formation and progression, while evading immune destruction have been listed as an emerging hallmark of cancer. 24 These findings linked the glycosylation with cancers, and the cancer states may be reflected in the appearance of abnormal glycans or altered quantitative proportions within the glycome. 7 , 14 , 25 The important role of glycans in the development and progression of cancer suggests that changes in serum glycosylation were promising cancer biomarkers, which can aid in the precision screening and diagnosis of the cancer disease, 15 , 21 as well as improve the survival rate and reduce the patient suffers or economic burden. Moreover, serum-glycan-based markers demonstrate superiority to traditional biopsy and computed tomography methods, because serum analysis is non-invasive and low-cost for point-of-care testing (POCT). 13 There are existing data analysis methods for glycome biomarkers identifying, such as differential analyses including T-test and Wilcoxon-Mann-Whitney U test for the diagnosis of prostate cancer, which have found that the N-glycopeptide IgG2-GP09 (EEQFNSTFR (H5N5S1)) was dramatically elevated in patients with prostate cancer. 26 , 27 In another study by Ruhaak at al. , partial least squares discriminant analysis (PLS-DA) was conducted to assess the potential of glycosylation profiles for the use of diagnosis of ovarian cancer. 28 However, the differential analyses were primarily directed toward the examination of individual glycans in isolation, whereas PLS-DA involved a straightforward linear amalgamation of diverse glycan characteristics for cancer diagnosis. Therefore, these traditional methods were often inadequate in characterizing the relationship between glycans and cancer, which hindered them from more accurate and pervasive cancer diagnosis, as well as the identification of promising biomarkers. The development of artificial intelligence provides a new and powerful approach for health care, including rapid and accurate image interpretation, potential cancer biomarkers identification, and effective early diagnosis. 29 , 30 , 31 , 32 Recent studies on the relationship between glycosylation and cancer diseases have utilized machine learning models to assist in the classification between patients with cancer and healthy individuals, as well as to screen for relevant glycan biomarkers. 33 Since the training data are the knowledge source of the machine learning models, it could be difficult to build models with decent performance for groups with small sample size. On the other hand, differences among datasets could potentially form challenges in maintaining comparable performance of a model trained on one cohort when applied to another cohort. These facts may have profound negative impacts on health care for the data-disadvantaged groups and generate new health care disparities. 34 , 35 , 36 In this study, we have investigated the utility of deep learning techniques in the discrimination power of glycome for multiple cancer diagnosis using the consecutive data from our previous studies (Unpublished). We have found that transfer learning scheme, in many cases, can enhance the performance of machine learning models for groups that have limited data. We focused on the ovarian cancer (OC) and investigated the role of transfer learning method in improving the performance of the serum N-glycome based model in data-disadvantaged cohorts. Our study also identified several Glycosylation biomarkers for ovarian cancer, some of which were specific to the data-disadvantaged cohorts. Finally, we conducted the experiments on other cancer groups including non-small cell lung cancer (NSCLC), gastric cancer (GC) and esophageal cancer (EC) to validate the universality of the transfer learning scheme. These results provided a perspective for an unbiased machine learning paradigm, which is essential for reducing health care disparities arising from the data inequality.

STAR★Methods Key resources table Resource availability Lead contact Further information and requests for resources should be directed to and will be fulfilled by the lead contact, Kang Ning (e-mail: [email protected] ). Materials availability This study did not generate new unique reagents. Data and code availability • All the raw data enrolled in this study have been deposited to the ProteomeXchange Consortium and are publicly available as of the date of publication. The accession number is listed in the key resources table . • This paper does not report original code. • Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request. Experimental model and study participant details Study population This study was performed in accordance with the principles of the Declaration of Helsinki criteria, and was approved by the Ethics Committee of Tongji Hospital affiliated to Tongji Medical College of Huazhong University of Science & Technology (study approval number: TJ-IRB20221109). All participants were recruited from the Tongji Hospital affiliated to Tongji Medical College of Huazhong University of Science & Technology. The study is exempt from informed consent, which has been approved by the Ethics Committee. There were four cancer groups, including ovarian cancer (OC) group, non-small cell lung cancer (NSCLC) group, gastric cancer (GC) group and esophageal cancer (EC) group. Cancer patients in OC group were enrolled with the following criteria: (i) age of ≥18 years, (ii) pathologically confirmed OC, and (iii) no history of other malignancies. Benign samples of OC group included patients with ovarian serous cystadenoma, mucinous cystadenoma and ovarian fibroma and followed the criteria: (i) age of ≥18 years, (ii) no history of other malignancies. Asymptomatic adults with non-OC were included as controls during the same recruitment period and the inclusion criteria were: (i) age of ≥18 years, and (ii) no history of malignancies. Subjects with NSCLC (or GC, EC) were enrolled with the following criteria: (i) age of ≥18 years, (ii) pathologically confirmed NSCLC (or GC, EC), and (iii) no history of other malignancies. The healthy controls of NSCLC (or GC, EC) group were recruited during the same recruitment period with the inclusion criteria: (i) age of ≥18 years, and (ii) no history of malignancies. Patients with severe cardiovascular diseases such as coronary heart disease and stroke were excluded. Each of the four cancer groups were divided into two cohorts according to the sample collection time: the discovery cohort and the validation cohort. The ovarian cancer group consisted of healthy controls (N discovery = 40, N validation = 20), benign neoplasm (N discovery = 50, N validation = 20), and cancer patients (N discovery = 52, N validation = 20). For non-small cell lung cancer (NSCLC), gastric cancer (GC) and esophageal cancer (EC) groups, there were only two classes: healthy controls and cancer patients. The NSCLC group consisted of 309 healthy controls (N discovery = 234, N validation = 75) and 275 cancer patients (N discovery = 200, N validation = 75). The GC group consisted of 63 healthy controls (N discovery = 34, N validation = 29) and 52 cancer patients (N discovery = 29, N validation = 23). And the EC group consisted of 63 healthy controls (N discovery = 34, N validation = 29) and 122 cancer patients (N discovery = 61, N validation = 61). The demographic and clinical characteristics of the participants were shown in Table 1 . Each group was divided into two cohorts according to the sample collection time: the discovery cohort and the validation cohort. There was a smaller number of samples in the validation cohort, which served as the data-disadvantaged cohort. Method details N-glycome data acquisition N-glycome profile was characterized by a MALDI-MS based high-throughput analytical assay. 55 Profiling of Serum N-glycans were performed with the same manner as reported in our recent study, 56 , 57 including N-glycan release, chemical derivatization, MALDI-MS analysis and MS data processing. In detail, human serum N-glycans were enzymatically released by PNGaseF digestion using Protein Deglycosylation Mix II (New England Biolabs, Ipswich, MA, USA), and subsequently separated and purified through solid-phase extraction (SPE). 20 First, serum samples (10 μL) were dissolved in ultra-purified water, followed by the addition of 3.6 μL of PNGaseF buffer and 2.4 μL of denaturing buffer to obtain a reaction mixture of 100 μL. The mixture was denatured in a Thermo-Shaker (Ningbo Hinotek Technology Co., Ltd) and then cooled to room temperature before 12 μL of NP-40 and 5 units of PNGaseF were sequentially introduced. The resulting mixture was incubated at 37°C overnight. The glycans were then captured by solid-phase extraction (SPE) with porous graphitized carbon (PGC, Sigma-Aldrich, St. Louis, MO, USA), and further concentrated using a vacuum concentrator (Eppendorf, Germany). To prevent the loss of sialic acid during MALDI-MS analysis, we performed methylamidation of the carboxyl group in the glycans according to our previous study. 58 In brief, we dissolved the sample in 25 μL of dimethyl sulfoxide (DMSO) containing 1 M methylamine hydrochloride and 0.5 M N-methylmorpholine. Next, we added 25 μL of DMSO containing 50 mM (7-azabenzotriazol-1-yloxy) trispyrrolidinophosphonium hexa-fluorophosphate (PyAOP). After incubating the mixture at room temperature for 30 min, the derivatized glycans were purified by solid-phase extraction (SPE) using microcrystalline cellulose (MCC, Sigma-Aldrich, St. Louis, MO, USA). 59 For MALDI-MS detection, the 5800 MALDI-TOF-MS (AB Sciex, Concord, Canada) was used to analyze the glycome spectrum. In brief, the purified sample was solubilized in 5 μL of 50% ACN aqueous solution, and then 0.5 μL MALDI sample was added to the MALDI plate. After air-drying, an equal volume of 10 mg/mL 2,5-Dihydroxybenzoic acid (DHB, Sigma-Aldrich, St. Louis, MO, USA) containing 50 mM sodium acetate (Sigma-Aldrich, St. Louis, MO, USA) was added. The range of m/z was monitored at 1000–4500, and a total of 1000 laser shots were applied to each sample spot. MS data were acquired in the positive ion reflector mode. The samples were randomly spotted onto the plate in triplicate to mitigate the batch effect. Glycan structures were assigned referring to GlycoMod and previous literatures. 60 , 61 , 62 Moreover, nanoLC-PGC-MS/MS analysis was employed to validate the N-glycans structures. 58 , 63 GlycoWorkBench 2.1 software was used for the visualization of glycan structures. All the raw data enrolled in this study have been deposited to the ProteomeXchange Consortium. Serum N-glycome profile and glycan derived traits have been illustrated in the supplemental information ( Tables S12 and S13 ). Machine learning for diseases diagnosis Machine learning models are implemented by Python 3.7.15 using standard libraries that are publicly available: pandas (1.2.4), PyTorch (1.11.0) and scikit-learn (1.0.2). For each cancer group, samples in discovery cohort were randomly divided into a training set (80% of samples) and a test set for independent evaluation (remaining 20%). Random forests (RF) and neural network (NN) were trained as classifier models for the diagnosis of cancer patients by using serum N-glycome profiles. We implemented the RF classifier with the following parameter: ntree = 500. For NN classifier, there were 4 layers: an input layer, a fully connected layer with 200 nodes, a fully connected layer with 100 nodes, and an output layer. A cross-validation procedure was applied to determine the within-training set performance by splitting data into training and test sets for 40-times repeated, fivefold-stratified cross-validation. The optimal models selected based on cross-validated results were then evaluated in the validation cohort dataset and AUROC value was calculated accordingly for the visualization of results. Transfer learning For transfer learning, we set the discovery cohort as the source context and the validation cohort as the target context. We randomly divided n% (n = 10, 20, 30, 40, 50, 60) of validation cohort samples as transfer set and the remaining samples as the test set. The base model was ab initio trained in the source context with the function RandomForestClassifier integrated in the scikit-learn package with default parameters, followed by applying transfer learning to the transfer set from the target context. We applied two random forest transfer algorithms to each decision tree in the base model: (1) The Structure Expansion Reduction (SER) algorithm searches greedily for locally optimal modifications of each tree structure by trying to locally expand or reduce the tree around individual nodes. 54 For each node v in the decision tree of base model, a set of all labeled points in the target data that reaches v was calculated. Then, the leaf v is expanded to a full tree with respect to the calculated sample set, and the internal node v will be decided whether to prune into a leaf node under the consideration of classification error. (2) The Structure Transfer (STRUT) algorithm does not modify structure, but only the parameter (thresholds) associated with decision nodes. 54 It adapts a decision tree trained on the source context to the target context by discarding all numeric threshold values in the tree and re-trained a new threshold for a node v using the subset of target samples that reach v . The two types of modified decision trees were then mixed together to obtain the final transfer random forest, which is defined as the transfer model and used for prediction. Cross-validation was repeated 40-times to obtain a distribution of transfer random forest prediction evaluations on the test set of validation cohort, representing the transfer models’ diagnosis performance. Potential biomarker identification To identify the N-glycans and glycan derived traits that contributed the most to diagnosis, two major aspects were considered for the model. Firstly, we ranked the glycans features according to the permutation importance, which is defined to be the decrease in a model score when a single feature value is randomly shuffled. 64 The calculation of permutation importance was repeated 40-times and the average value was determined as the final importance score. Next, we gradually selected the topN features according to the permutation importance to re-build the model and recorded the model performance for 40-times repeats. The glycans features used to achieve the best average performance were then identified to be the potential biomarkers. Quantification and statistical analysis Machine learning performance evaluation We included Macro AUROC and Micro AUROC to characterize the model performance in ovarian cancer group as our model provided outputs of probabilities for three disease phenotypes (control, benign, cancer), which formed a tri-classification task. Macro AUROC evaluated the average classification performance of three disease phenotypes, while Micro AUROC focused on whether each sample was classified accurately. In addition, we also considered AUROC for each of the three disease phenotypes (one versus all others) to understand the model’s ability in discriminating a particular class from others. For NSCLC, GC and EC groups, the model performance was evaluated using AUROC value, which considers the trade-offs between sensitivity and specificity at all possible thresholds. Another widely used machine learning performance metric is the AUPR value, which has been mathematically proven that the performance ranks of two models remain same in the ROC space and the PR space. However, linear interpolation in the precision–recall space is problematic, which may lead to inaccurate calculation of AUPR for datasets of small sample sizes. Thus, AUROC is a more robust metric for evaluating machine learning performance in our study. 65 Also, we have included other evaluation metrics including precision, recall, f1-score and accuracy as a reference for performance evaluation. PCS framework Throughout the statistical analysis, we adhered to the PCS (Predictability, Computability and Stability) framework for trustworthy data science, which has been used and proven valuable in many previous scientific discoveries including interpretable drug response prediction 66 and metabolomic signature of pancreatic cancer risk. 67 For the modeling stage as in this work, the PCS (Predictability, Computability and Stability) framework uses predictability as a reality check, and for reproducibility, it advocates for a stability analysis across different reasonable perturbations of the data and models that pass the prediction check. Under this framework, the entire discovery cohort was divided into training set and test set for base model training and selecting. The set-aside validation cohort was used for obtaining an unbiased evaluation of the three learning schemes. Moreover, a 40-times repeats procedures were applied for all model evaluation processes to guarantee the stability of the results.

Results Sample cohort and experimental design for ovarian cancer group Serum samples including N-glycans and glycan derived traits were taken from 142 participants, including 40 healthy controls, 50 patients with benign neoplasm and 52 patients with ovarian cancer ( Figure 1 A, discovery cohort, Table 1 ). Based on the subjects from discovery cohort, neural network (NN) and random forests (RF) models were trained and compared to obtain the optimal model defined as base model. A new transfer model was developed from base model using transfer learning algorithms consisted of SER and STRUT algorithms ( Figure 1 B). Finally, the transfer model was tested in the validation cohort including 60 participants and identified the potential serum glycans biomarkers ( Figure 1 C, validation cohort, Table 1 ). N-glycome abundance distribution in ovarian cancer group Principal component analysis (PCA) was used to visualize the serum N-glycome’ features distribution of healthy control, benign neoplasm, and cancer patient samples in discovery and validation cohorts of ovarian cancer group ( Figure 2 A). As shown in the figure, there was a clear distinction between the cancer group and other groups, while the control group and benign group were mixed together and difficult to distinguish, indicating that patients with ovarian cancer had obvious changes in N-glycome patterns compared with benign patients. In addition, the PCA results revealed some differences between the discovery cohort and the validation cohort, such as the separation of benign groups, which frequently renders the results obtained from the discovery cohort are not completely applicable to the validation cohorts. Development of N-glycome-based multi-class diagnosis model Based on the discovery cohort, we trained different machine learning multi-class classifiers (neural network (NN) and random forest (RF)) to classify the healthy control, benign neoplasm, and cancer patient samples using serum N-glycome’ abundance data from the training set (80% samples with the same class proportions as the cohort) and presented their final performance from the withheld test set (20% samples). 5-fold cross-validation method was used to evaluate their average classification performance. Compared to the NN model, whose AUROC scores were between 0.5 and 0.6, the RF multi-class model achieved a far better performance with AUROC scores around 0.8 ( Figure 2 B; Table S1 ). Therefore, the RF multi-class model was used as the base model for further analyses. Disparities in machine learning model performance We then applied the base model to the validation cohort and compared its performance with that in the discovery cohort. The RF multi-class model showed a decent classification performance in the discovery cohort, with a mean Macro AUROC score of 0.83, and mean Micro AUROC score of 0.83 ( Figure 2 C), suggesting that patients with ovarian cancer detection based on serum N-glycome is feasible. In the validation cohort, however, the model’s diagnostic performance deteriorated, as shown by Macro AUROC and Micro AUROC scores less than 0.82 ( Figure 2 C; Table S2 ). This decline was particularly evident in distinguishing the benign neoplasm patients (one versus the others), with an AUROC score of only 0.62 in the validation cohort compared to 0.79 in the discovery cohort, which corresponded to the obvious separation between the benign group of the two cohorts in the previous PCA results. In addition, we noticed that the RF multi-class model achieved an AUROC score of 0.96 for healthy control samples in the validation cohort. This is possibly due to the high similarity between the control groups in the discovery and validation cohorts, and the small sample size of the validation cohort may also contribute to it. Transfer learning for improving model performance To attain better diagnostic performance in the validation cohort, we tried the transfer learning scheme, which can provide improved performance by leveraging the knowledge learned from the discovery cohort. We randomly divided the validation cohort data of n% (n = 10, 20, 30, 40, 50, 60) as the transfer set and the remaining data as the test set, then performed assessments for three models: (1) Independent model: ab initio training and testing the RF multi-class model on the transfer set and the test set of the validation cohort, respectively. (2) Plus model: ab initio training the RF multi-class model using the Plus dataset including all the discovery cohort data and the validation cohort transfer set and testing it on the test set of the validation cohort. (3) Transfer model: ab initio training the RF multi-class model using discovery cohort data, followed by applying transfer learning to the transfer set from the validation cohort, and then testing it on the test set of the validation cohort. We found that the performance of the three models in the validation cohort was greatly improved over the base model even when only a small proportion of the validation cohort data was used as the transfer set, especially for the transfer models ( Figures 2 D and S1 ; Table S3 ). For example, with 10% of validation cohort data, transfer model could achieve a mean Macro AUROC of 0.84 and a mean Micro AUROC of 0.83 on the test set of validation cohort, which is comparable to the performance on the discovery cohort. This result suggests that the transfer model has successfully generalized its learned patterns to the validation cohort. It is an indication of transfer model’s ability to adapt to varying data distributions and implies its potential for practical applications in broader clinical settings. We then compared machine learning schemes on performance for the test set of validation cohort and found that transfer learning produced models with significantly better performance compared to the models from independent learning (e.g., P Macro AUROC = 3.6e-07, P Micro AUROC = 3.6e-08 for 30% partitions) and plus learning (e.g., P Macro AUROC = 2.0e-03, P Micro AUROC = 1.4e-02 for 30% partitions). Due to the absence of additional data from the discovery cohort, the performance of the independent learning is highly dependent on the sample size of the validation cohort. When trained with 10% of validation cohort data, the independent learning fails to produce reliable models, with a mean macro AUROC of only 0.75. As the dataset used for training increases, the independent model could achieve diagnostic performance similar to the plus model (e.g., P Macro AUROC = 0.51, P Micro AUROC = 0.18 for 50% partitions), but both are inferior to the transfer model. When using 60% of validation cohort data as the transfer set, the transfer model demonstrated excellent performance in the test set of the validation cohort, with Macro AUROC and Micro AUROC reaching approximately 0.90 and 0.89, respectively. Notably, transfer learning also significantly improved the classification performance of the benign group in the validation cohort compared to the base model (e.g., p = 8e-10 for 30% partitions, one versus the others), with the transfer model achieving a maximum mean AUC of 0.77 ( Figure S2 ). In summary, transfer learning significantly improved the diagnostic performance of the base model on the validation cohort and achieved decent performance compared to the plus learning and independent learning schemes, even only with a small amount of validation data. Identification of a serum N-glycans biomarker panel for ovarian cancer Next, we selected the serum N-glycans and glycan derived traits, which contributing for the transfer model, based on the feature permutation importance to identity clues to model interpretability. Taking the model’s Macro AUROC and Micro AUROC into account, we finally determined the top 22 features including 18 serum N-glycans and 4 glycan derived traits to be the potential biomarkers ( Figure 3 A). These serum N-glycans and glycan derived traits achieved a mean Macro AUROC of 0.95 and a mean Micro AUROC of 0.91 in the validation cohort. As expected, significant changes in the abundance of them were observed between controls and patients with ovarian disease in the validation cohort ( Figure S3 ). Interestingly, the 4 glycan derived traits were all sialylation, and two of them (tri-sialyation, tetra-sialylation) were significantly increased (p < 0.05) in patients with ovarian cancer, corresponding to previous research. 15 , 37 , 38 However, the total sialylation was substantially changed in patients with ovarian cancer in the opposite trend, possibly due to the decrease abundance of some low sialylated N-glycans. For the N-glycans, it was found that 10 N-glycans, including H6N5S2F1, H5N3F1, H3N4, H4N5S1F1, H6N5S1, H7N6S3, H6N3, H5N4, H3N5, and H6N5S3F1, were significantly increased (p < 0.05) in patients with ovarian cancer compared with healthy controls. Moreover, three of them (H6N5S2F1, H5N4, and H6N5S3F1) were still significantly increased from benign neoplasm to patients with cancer, implicating their contribution to OC metastasis and the potential to be the basis for distinguishing patients with ovarian cancer from healthy controls and benign neoplasm. Notably, two N-glycans (H5N3F1, H3N4) showed a discontinuous pattern of abundance changes, with a significantly increased abundance from healthy controls to benign neoplasm group, and then a significant decrease in patients with ovarian cancer. In addition, we also found three N-glycans (H5N4S2F1, H5N4S1F1, H5N4S2) were significantly down-regulated in patients with ovarian disease, implicating their inhibitory role in ovarian cancer progression. Specific N-glycans biomarker for the validation cohort Using the same criteria for selection, we obtained the top 16 features consist of 13 N-glycans and 3 glycan derived traits for the base model ( Figure S4 ). Comparing the contributing features between the base model and transfer model, it was found that there were 8 features that were unique to the transfer model ( Figure 3 B). We further compared the top 22 features from the base model with transfer model and the result showed that there were still 6 features that only contributed to the transfer model ( Figure 3 B). Specifically, the abundance of the 6 features, including S4, H5N3F1, H7N6S3, H5N4, H5N4S2, and H3N5, differed significantly between healthy controls and patients with ovarian cancer in the validation cohort ( Figure 3 C), while no such significant difference was observed in the discovery cohort except for H5N3F1 ( Figure S5 ). Overall, these results suggest that the transfer learning can capture N-glycans features that are specific to the validation cohort, which may explain the improved performance of the transfer model and is of great significance for the diagnosis of patients with ovarian cancer in the validation cohort. Validation of transfer scheme for other serum-glycan-based cancer prediction Furthermore, we applied the same strategies on the serum samples from non-small cell lung (NSCLC), gastric (GC) and esophageal cancers (EC), which only include patients with cancer and healthy controls ( Table 1 ). Principal component analysis results indicated obvious differences of validation cohorts from the discovery cohorts ( Figures 4 A, 4C, and 4E), corresponding to the decrease of performance of the base model on the validation cohort ( Figure S6 ; Table S2 ). Then, using transfer learning scheme, we successfully improved the classification performance for all three cancer groups with a higher mean AUC score ( Figures 4 B, 4D, and 4F; Tables S4 , S5 , and S6 ). Compared to independent model and plus model, transfer model exhibited the best performance in NSCLC groups, with the AUC score around 0.90 ( Figure 4 B). For GC and EC groups, transfer learning showed better performance than independent learning and plus learning when the transfer set was small, demonstrating the superiority of transfer learning in the data-disadvantaged situation. As the dataset size increased, the independent model got improved AUC score, approached or even exceeded that of the transfer model ( Figures 4 D and 4F). In the EC group, for example, with 30% or more of the validation cohort data for training, independent model achieved a mean AUC score of 0.9717, significantly higher than the transfer model’s AUC score (p = 3.3e-05). These results indicated that the ultimate solution to improve the model performance would be to increase the number of samples for model training. Notably, the control and cancer samples are not balanced in the validation cohort of EC group, which may influence the evaluation of the model performance. We resampled the cancer samples to form a data-balanced cohort to reevaluate the model performance and the result supported above conclusions ( Table S7 ). Moreover, we identified specific N-glycans and glycan derived traits that can classify different cancers based on the feature importance for transfer model ( Figure 4 G). The abundance of four N-glycans, H5N3F1, H7N6S2, H5N4S1, and H6N3, made great contributions to the classification of healthy controls and patients with NSCLC. For the subjects in the GC group, S3, B_neutral, and H6N5S3 showed higher importance, while H5N3F1 was found to significantly influence the model’s diagnostic performance for patients with EC. In summary, for the binary classification tasks of NSCLC, GC, and EC diseases, the transfer model can also achieve improved performance on the validation cohort and show superiority than independent and plus learning scheme in the data-disadvantaged situation. With transfer model, we can identify disease-specific N-glycans signatures, which are expected to become diagnostic biomarkers for the patients with cancer.

Discussion Previous studies have reported that the N-glycome-based classification models can accurately distinguish between healthy individuals and patients with cancer, 39 , 40 demonstrating reliable predictive performance. However, the heterogeneity of data between different cohorts can lead to a decrease in model performance when predicting across cohorts. We observed the AUC values of the base model trained on the discovery cohort decreased by 0.02–0.1 in the validation cohort ( Figures 2 C and S6 ), which affects the generalizability and applicability of the model. In this work, we showed the superiority of the transfer learning scheme in glycome based assisted cancer diagnosis. We found that transfer learning can provide a possible solution to enhance the performance of the model in new cohorts by transferring the shared knowledge learned from the training cohort. For example, it was found that the AUC score of the transfer model is around 0.05 to 0.1 higher than that of the base model in the ovarian cancer group ( Figure 2 D). Our results on various cancer groups had also supported that the transfer model achieved the highest AUC score in the validation cohort compared to the independent model and plus model. In addition, transfer learning demonstrated advantages in selecting glycan biomarkers associated with relevant cancers. We noticed that the transfer model was able to identify glycome biomarkers that are specific to the study cohort, which holds significant implications for advancing precision medicine. Through the transfer model, we have identified the glycome features that are closely associated with different cancers ( Figure 5 ). Sialoglycans, especially the tri-antennary and tetra-antennary sialylated N-glycans were significantly increased in the patients with OC. Several directly measured glycans containing sialic acid, such as H6N5S2F1, H7N6S3, and H6N5S3F1 also demonstrated an increased trend in patients with ovarian cancer compared with healthy controls. It has been reported that hypersialylation could enhance immune evasion and tumor cell survival, and stimulate tumor invasion and migration, 38 , 41 which may contribute to the development and progression of ovarian cancer. Additionally, we have also observed a significant increase in glycans H3N4 and H3N5, as well as the derived trait G0 (measuring all agalactosylated glycans), corresponding to previous research. 42 Malhotra et al. demonstrated that these particular agalactosyl G0 glycoform of IgG may elicit a pro-inflammatory response as a result of their ability to act as ligands for mannose binding lectin and subsequent complement activation. 43 , 44 , 45 In another study by Bones et al., it was described that increased the expression of IgG carrying core fucosylated G0 type glycans with specificity against an antigen on the tumor cell surface may result in subsequent complement activation with tumor cell lysis via the membrane attack complex. 46 It is noteworthy that, similar change in agalactosylated glycans was also reported implicated in other cancer progression, such as prostate 47 and stomach 46 cancer, indicating the great potential of the glycan derived trait G0 in cancer diagnosis and treatment. For NSCLC group, glycans including H5N3F1, H7N6S2, H5N4S1, and H6N3 demonstrated high feature importance in classifying the healthy controls and patients. Subjects with GC mainly have close connections with sialylated glycans and bisecting GlcNAcylated N-glycans, such as H6N5S3, H6N5S2F1, H6N5S3F1, H4N5, H3N5, and H4N5F1. A significant increase in bisecting GlcNAcylated N-glycans were found in the patients with gastric cancer (p = 8.2e-8, not shown). The bisecting GlcNAc plays an important role in retarding tumor progression through reducing galectin-lattice dependent growth factor signaling, 48 indicating that it may be a part of disease pathophysiology. In EC group, H5N3S1 glycan was the most potential biomarker with far greater importance than other glycan features. Using the top10 glycome features, transfer models demonstrated outstanding performance in the validation cohort of all four types of cancers. For all the four cancer groups, transfer models achieved an AUC score exceeding 0.90, which outperformed the traditional approach of PLS-DA ( Figures 5 and S7 ; Table S8 ). These results highlighted the immense potential of transfer learning method in the diagnosis of patients with cancer and identification of promising cancer biomarkers. Overall, our study indicated that the introduction of transfer learning contributes to enhancing the clinical applications of glycome-based cancer diagnostic model, which could serve as a powerful approach for cancer risk assessment and identification of potential biomarkers, providing new perspectives for the diagnosis and treatment of various cancers. Given the poor performance of other learning schemes (base model, independent model and plus model) in the validation cohort (serving as data-disadvantaged cohort), our efforts could largely diminish the negative impacts on health care for the data-disadvantaged groups, while bridge the gap of health care disparities. Previous studies have reported that glycosylation is strongly influenced by factors such as age and sex. 49 , 50 , 51 To determine their specific influence on our findings, we resampled the validation cohort of NSCLC group and formed the age (age between 55 and 60) and sex (male only) subsets to re-evaluate the model performance. As the results shown, we found that there is indeed a slight difference in the model performance, but the conclusion that the transfer learning is superior to other schemes is still upheld ( Tables S9 and S10 ). Another issue to be considered is the potential collinearity problems caused by the mixed features of single glycans and derived traits in the same model. We re-trained the independent, plus and transfer models using only the directly detected glycans and evaluated their performance in the validation cohort of the ovarian cancer group ( Table S11 ). The results demonstrated a similar model performance and transfer model still got the highest AUC score than the independent and plus model, supporting the conclusion that transfer learning could improve diagnosis performance. Despite of these limitations, this work represents the first of its kind to introduce transfer learning scheme in the building of glycome-based cancer diagnostic model, and have demonstrated its superiority in improving diagnosis performance and identifying cancer biomarkers. It has provided a novel reliable and universal perspective for future cancer diagnosis in clinics. Limitations of the study As a pilot study, our work has several limitations. Given the low incidence of ovarian cancer in the general population, 52 procurement of qualified specimens for modeling and biomarker discovery is challenging. While we acknowledge the limited sample size of subjects, we emphasize rigor in our statistical approach, adhering to the PCS framework 53 for modeling and evaluation of model stability and robustness, as well as the use of independent validation cohort and repeated procedures. In our future work, we might attempt to include more samples to complement our results. Secondly, the differences across cohorts in this study mainly stem from variations in sampling time, and inclusion of study cohorts from different sampling locations may further highlight the advantages of transfer learning. Though we could not validate it at this moment, this hypothesis should be tested in the future. Also, there is limited biological evidence available to substantiate the identified associations between glycome and cancer, and future work to uncover the mechanisms underlying these associations is required to enhance our comprehension of the precise role played by cancer-specific glycans in the pathogenesis of cancer.

These authors contributed equally Lead contact Summary Protein glycosylation is associated with the pathogenesis of various cancers. The utilization of certain glycans in cancer diagnosis models holds promise, yet their accuracy is not always guaranteed. Here, we investigated the utility of deep learning techniques, specifically random forests combined with transfer learning, in enhancing serum glycome’s discriminative power for cancer diagnosis (including ovarian cancer, non-small cell lung cancer, gastric cancer, and esophageal cancer). We started with ovarian cancer and demonstrated that transfer learning can achieve superior performance in data-disadvantaged cohorts (AUROC >0.9), outperforming the approach of PLS-DA. We identified a serum glycan-biomarker panel including 18 serum N-glycans and 4 glycan derived traits, most of which were featured with sialylation. Furthermore, we validated advantage of the transfer learning scheme across other cancer groups. These findings highlighted the superiority of transfer learning in improving the performance of glycans-based cancer diagnosis model and identifying cancer biomarkers, providing a new high-fidelity cancer diagnosis venue. Graphical abstract Highlights • Serum glycome is promising biomarkers for cancer diagnosis • Deep learning improved the performance of glycome-based cancer diagnosis model • Deep learning model identified potential serum glycan-biomarker panel for cancers • Glycan derived traits sialylation is an important biomarker for ovarian cancer Cancer; Glycomics; Machine learning Subject areas Published: December 13, 2023

Supplemental information Acknowledgments This work was partially supported by the 10.13039/501100012166 National Key R&D Program of China (Grant No. 2022YFC3400800, 2023YFA1800900 and 2018YFC0910502), the 10.13039/501100001809 National Natural Science Foundation of China (Grant Nos. 32071465, 31871334, 31671374, 81827901 and 82272418), and High-Level Talents Research Start-Up Project of 10.13039/501100013795 Fujian Medical University (Grant No. XRCZX2023015). Numerical computations were performed on the Hefei Advanced Computing Center. Author contributions H. -B. Z., study concept and design, analysis and interpretation of data, statistical analysis, drafting of the article,; S. L., research performance, acquisition and analysis of data, drafting of the article; Y. W., research performance, acquisition and analysis of data; H. -H. H., statistical analysis; L. -K. S., research performance; Y. -Y. Y., research performance; L.-M. C., provision of study material or patients; X. L., study concept and design, critical revision of the article for important intellectual content; K. N., study concept and design, administrative support, critical revision of the article for important intellectual content. All authors read and approved the final article. Declaration of interests The authors declare no conflict of interest. Inclusion and diversity We support inclusive, diverse, and equitable conduct of research.

CC BY

no

2024-01-16 23:40:18

iScience. 2023 Dec 13; 27(1):108715

oa_package/de/6a/PMC10788220.tar.gz

PMC10788223

38226324

Introduction Tea, as a beverage, is popular for its healthy benefits and pleasant flavor. Tea plant is a perennial economic crop which is susceptible to various kinds of destructive foliar diseases due to the warm and humid growth environment in tea plantations that provides a suitable microclimate for the breeding of pathogens ( Baby, 2002 ). Among these diseases, blister blight caused by Exobasidium vexans Massee is considered as the most serious disease in the world that mainly damages tender leaves, buds and young fruits, resulting in yield loss and quality decrease ( Punyasiri et al., 2005 , Sen et al., 2020 ). At the tea plant experimental plots of Sri Lanka, the disease first occurs through the formation of appressoria and penetration of the cuticle, after penetration visible translucent spots are formed, and with the development of spots, characteristic circular blisters appear on the surface of young tissues, finally resulting in the infected tea leaves distorted and infected stems broken off at the point of infection ( Punyasiri et al., 2005 ). Blister blight disease occurring in almost all the tea plantation regions of Asia, has become the major problem on tea plants that causes serious economic losses ( Sinniah et al., 2016 , Baby, 2002 , Guo et al., 2005 ). It is reported that blister blight disease is more serious in alpine tea gardens in southwest and south of China. In epidemic years, the incidence of blister blight in tea growing regions located in southwest of China can reach 40–50 %, and even reach 90 % in severe cases ( Chen and Sun, 2013 , Sen et al., 2020 ). It was estimated to cause about 33 % crop loss in Sri Lanka ( De Silva et al., 1992 ), 25 % in Indonesia and as high as 50 % in South India, respectively, in tea fields in the absence of control measures ( Radhakrishnan and Baby, 2004 ). Besides yield loss, blister blight disease negatively affects the growth of tea trees, resulting in the quality decrease of processed tea which is even exceeded the economic threshold level 35 % ( Ponmurugan et al., 2016 , Radhakrishnan and Baby, 2004 ). Tea has abundant non-volatile and volatile metabolites such as tea polyphenols, amino acids components, caffeine, soluble sugars, lipids, and aromatic compounds, which are proved to be closely associated with the flavor and quality of tea ( Cao et al., 2021 ). Some quality related metabolites such as caffeine, epicatechin (EC) and theanine (Thea) also have been reported to have potential effect against various kinds of viral or fungal diseases ( Sharma et al., 2021 , Punyasiri et al., 2005 ). The development of pathogen in various stages showed different effect on pathogenesis related protein, anti-oxidative enzymes and flavonoid pathway in tea, suggesting the possible role of some chemical compounds (i.e., reactive oxygen species, anthocyanins, lignins, catechins) and other synthesized compounds in acting as antifungal agents in different tea cultivars ( Nisha et al., 2018 ). A large amount of studies have validated that various kinds of biotic (ie, disease infection, insect attack) and abiotic (ie, mechanical damage, environment conditions) stresses can remarkably influence the quality and flavor of tea by significantly changing the composition and amount of metabolites, especially affect the aromatic compounds in tea plants ( Chen et al., 2017 , Cho et al., 2017 , Mei et al., 2017 , Zeng et al., 2019 ). Tea processed from blister blight infected fresh tea leaves is fragile and shows obvious bitter taste, and the level of chemical components related to tea quality has decreased significantly, especially tea polyphenols and catechins ( Guo et al., 2005 , Jayaswall et al., 2016 ). Premkumar et al. (2008) reported that after the infection of blister blight disease, there were remarkable reduction in the amount of tea polyphenols, catechins, sugar, nitrogen, proteins, amino acids and other substances in infected tea leaves. Mur et al. (2015) analyzed the chemical compounds including caffeine, flavan-3-ol, flavone and flavonol in blister blight infected tender tea leaves by using high performance liquid chromatography with online photodiode array detection and electrospray ionization-tandem mass spectrometry (HPLC-PDA-ESI/MS), and found that kaempferol and quercetin glucosides, kaempferol triglycosides and some catechin-class antioxidants were increased, while the level of caffeine and apigenin and myricetin glycosides were remarkably reduced as disease progressed. However, limited information are still available on the tea quality affected by blister blight disease. In this study, the development and spread of blister blight on tea leaves, and the composition and abundance of fungal community on leaf tissues were fully investigated. After the infection of the disease, major metabolite differences in healthy and diseased tea leaves were studied, centring on the metabolites that contributed to the tea taste, which is helpful to improve the understanding on the influence of blister blight on tea quality.

Materials and methods Materials During 2018 to 2020, tea leaves displaying blister blight symptoms were observed in high mountain tea plantations above 1000 m altitude located in Quxian county, Dazhou city, Sichuan province, China (30°85′N, 106°94′E). The typical blister blight symptoms were collected to show the symptom development. To analyze the characteristics of the blister blight disease, more than 100 fresh diseased tea shoots showing blister blight symptoms were harvested at April 2020, and used for microbial diversity analysis and microscopic analysis of pathogens including tissue section observation, transmission electron microscopy (TEM) and scanning electron microscopy (SEM) analysis. To investigate the effect of blister blight on tea quality and flavor, diseased tea shoots with one bud and two leaves (500 g) were sampled, while healthy tea leaves were used as a control. All the experiment had three replicates. Tea samples used for metabolites profile analysis were fixed by microwave (2–3 min), dried in a tea dryer machine (1 h, 80°C), finally milled and stored in a − 80°C freezer. Microscopic analysis Toluidine blue staining (TBS) Tea leaf tissues (1 × 1 cm) were firstly fixed with FAA (Formaldehyde-acetic acid–ethanol) (Solarbio, Beijing, China), then dehydrated with ethanol, embedded in paraffin and routinely sliced. The slices were placed in xylene I (Sinaopharm Group Chemical Reagent Co. LTD) for 20 min, xylene II (Sinaopharm Group Chemical Reagent Co. LTD) for 20 min, anhydrous ethanol I (Sinaopharm Group Chemical Reagent Co. LTD) for 5 min, anhydrous ethanol II (Sinaopharm Group Chemical Reagent Co. LTD) for 5 min, 75 % alcohol for 5 min, and washed with double distilled water (ddH 2 O). The dehydrated slices were stained in toluidine blue solution (Wuhan Google Biotechnology Co. LTD) for 6 min, then washed with ddH 2 O, finally the qualified slices were dried in an oven. Transparent sealing: the dried slices were treated in xylene (Sinaopharm Group Chemical Reagent Co. LTD) for 5 min, then taken out to dry, finally sealed with neutral gum for microscopic examination. Scanning electron microscopy (SEM) Tea leaf tissue blocks (3 mm 2 ) were harvested within 3 min and washed with phosphate buffer saline (PBS) (Servicebio, Wuhan, China) gently, then immediately immersed in electron microscopy fixative (Servicebio, Wuhan, China) at room temperature for 2 h, finally transferred into 4 °C for preservation. After fixation, tissue blocks were rinsed with 0.1 M phosphate buffer (PB, pH 7.4) for 3 times (15 min each), transferred into 1 % osmic acid (Ted Pella Inc.) for 1–2 h at room temperature, then washed in 0.1 M PB (pH 7.4) for 3 times (15 min each). Leaf tissue blocks were dehydrated with different ethanol concentrations (30 %, 50 %, 70 %, 80 %, 90 %, 95 %, 100 % and 100 %) in sequence for 15 min each, finally were placed in isoamyl acetate (Sinaopharm Group Chemical Reagent Co. LTD) for 15 min. After 0.1 M PB (pH 7.4) wash, the samples were dried with critical point dryer (Quorum, United Kingdom), then were attached to metallic stubs by using carbon stickers and sputter-coated with gold for 30 s, finally were observed with SEM (Hitachi, Japan). Transmission electron microscope (TEM) Sampled fresh tea leaves were cut into 1 mm 3 tissue blocks within 3 min, then quickly fixed in the electron microscopy fixative (Servicebio, Wuhan, China) at room temperature for 2 h, finally stored at 4 °C. After fixation, leaf tissue blocks were rinsed with 0.1 M PB (pH 7.4) for 3 times (15 min each). Post-fixation: tissue blocks were fixed in 1 % osmic acid (Ted Pella Inc.) that prepared in 0.1 M PB (pH 7.4) at room temperature for 7 h away from light, then washed three times with 0.1 M PB (pH 7.4), 15 min each time. Tissue dehydration: treated tissues were dehydrated in 30 %, 50 %, 70 %, 80 %, 95 %, 100 % alcohol in sequence, 1 h each time, and then in ethanol: acetone = 3: 1, ethanol: acetone = 1: 1 and ethanol: acetone = 1: 3, for 0.5 h each, respectively, finally in pure acetone for 1 h. EMBed 812 (SPI, USA) was used for osmotic embedding, then the embedded plates were placed in 65°C oven to polymerize for 48 h. Treated blocks were cut into thin slices (60–80 nm) with ultra-microtome (Leica, Germany). The copper mesh slices were dyed in the dark with uranium acetate saturated alcohol solution (2 %) for 8 min, rinsed in ethanol (70 %) for 3 times and in ddH 2 O for 3 times, dyed in lead citrate solution (2.6 %, without CO 2 ) for 8 min, washed with ddH 2 O for 3 times, finally put into a copper mesh box to dry overnight at room temperature. The prepared slices were observed with TEM (Hitachi, Japan). Microscopic examination Diseased tea leaves exhibiting blister blight with abundant basidiospore were observed with a U-TV0.5XC-3 microscope (Olympus, Japan). The shapes and sizes of basidiospore were recorded by measuring at least 30 randomly selected basidiospore. Microbial diversity analysis To analysis the composition of microbial communities in diseased lesion tissues showing blister blight in Quxian, Sichuan province, China, genomic DNA of lesion tissues was extracted by using HiPure Soil DNA Kits (Magen, Guangzhou, China) based on the manufacturer’s protocols. ITS gene region was used for the identification of fungal communities. Gene region of ITS2 was amplified with primer pair ITS3_KYO2 (5′-GATGAAGAACGYAGYRAA-3′) and ITS4 (5′-TCCTCCGCTTATTGATATGC-3′). The amplified products were purified and then quantified using Nanodrop procedures. Sequencing was conducted by Genedenovo Inc. (Guangzhou, China) by using Illumina Hiseq 2500 PE250 platform (Illumina, San Diego, CA, USA). The α-diversity indices such as Chao1, Simpson, and Shannon were quantified to show the fungal diversity in diseased tea leaves according to the Operational Taxonomic Units (OTUs) richness in QIIME (version 1.9.1) ( Caporaso et al., 2010 ). The species distribution river map of the fungal composition was plotted using R project ggplot2 package (version 2.2.1). The heatmap of species abundance was constructed based on pheatmap package (version 1.0.12) in R project. Functional group (guild) of the microbial communities were inferred by FUNGuild (version 1.0). The phylogenetic tree was constructed by Neighbor-Joining (NJ) method in MEGA6. The distances of evolutionary were computed by the method of Maximum Composite Likelihood. Sensory evaluation Quantitative descriptive analysis was conducted by a sensory panel with seven well-trained panelists (4 males and 3 females, 40–55 years old) from the Department of Tea Science, Southwest University, China. All the panelists had been informed of the sensory evaluation, and agreed to take part in this research and use their information. The rights and privacy of all the panelists were protected during the research, and ethical approval for the involvement of human subjects in this study was granted by Southwest university research ethics committee. Total 3.0 g tea sample was weighed, and brewed with 150 mL boiling water for 5 min, then tea infusion was used for taste evaluation. Taste evaluation included the common characteristics such as bitterness, astringency, umami and sweetness ( Cao et al., 2021 ). Sensory evaluation was conducted by the national sensory evaluation method for tea GB/T 23776–2018 ( Gong et al., 2018 ). Values of taste attributes were scored by a 10-point scale and 5 intensity were evaluated in given tea samples: 0–2 represents extremely weak, 2–4 represents weak, 5–6 represents moderate, 7–8 represents strong, and 9–10 represents extremely strong. The highest and lowest scores for each attribute were removed and the mean value was used for the evaluation. Sugars analysis by GC–MS The content and composition of sugars were analyzed using gas chromatography-mass spectrometry (GC/MS) ( Chen et al., 2023 ). Tea sample (20 mg) was mixed with 500 μL methanol: isopropanol: water (3: 3: 2, V/V/V) solution, vortexed for 3 min, ultrasound for 30 min, and then centrifuged at 12,210 g (3 min, 4°C). Agilent 8890 gas chromatograph coupled to a 5977B mass spectrometer with a DB-5MS column (J&W Scientific, USA) was used for GC/MS analysis. Helium was the carrier gas with a flow rate of 1 mL/min. Injections were made with a split ratio 5:1 and the total injection volume for each sample was 1 μL. The oven temperature was set at 170°C for 2 min, then increased to 240°C at 10 °C/min, finally raised to 280 °C at 5 °C/min, raised to 310 °C at 25 °C/min and kept for 4 min. Selective ion monitoring mode was used for sample analysis. The orthogonal projections to latent structure-discriminant analysis (OPLS-DA) were applied to distinguish the significantly differential metabolites between different tea samples. Significantly changed metabolites between healthy and diseased tea leaves were determined by absolute Log2FC (fold change), variable importance in project (VIP) and P value (Student’s t test). When the fold change ≥ 2 or ≤ 0.5, P value ≤ 0.5 and VIP ≥ 1 between two groups, the metabolites difference was considered significant. The metabolites annotated in Kyoto Encyclopedia of Genes and Genomes (KEGG) compound database ( https://www.kegg.jp/kegg/compound/ ), were then mapped into metabolic pathway based on database search ( https://www.kegg.jp/kegg/pathway.html ). The P value represented the metabolites enrichment in a pathway, and P ≤ 0.05 was indicative of significant enrichment. Catechins, caffeine and amino acids analysis by HPLC Catechins and caffeine contents were analyzed by using high performance liquid chromatography (HPLC) (Agilent 1200, Agilent Technologies, Santa Clara, CA, USA) ( Chen et al., 2023 ). Prepared tea infusions were analyzed by an Agilent Eclipse XD8 C18 column (250 mm × 4.6 mm i.d., 5 μm; Thermo Electron Corporation, Waltham, MA, USA). The volume of injected sample was 5 μL with 0.9 mL/min flow rate. The wavelength of UV detection was 278 nm. The composition and content of amino acids were analyzed with HPLC system ( Chen et al., 2023 ). The HPLC conditions were as follows: mobile phase A was 4 mM sodium acetate (pH = 5.5); mobile phase B was 80 % acetonitrile solution; column temperature was 35 °C; flow rate was set at 0.9 mL/min. The wavelength of UV detection was 360 nm. The results were expressed as mean value ± standard deviation (SD). The t test of independent sample was used to calculate the statistically significant difference between different groups (P ≤ 0.05) by IBM SPSS Statistics 20.0 (SPSS Inc., Chicago, USA).

Results and discussions Disease symptoms on tea caused by blister blight disease As an important tea foliar disease, blister blight mainly infected young organs and tissues including tender leaves, buds, petioles and stems ( Fig. 1 ), directly affecting tea industry qualitatively and quantitatively. The damage to tender shoots and leaves was the most serious, while mature leaves and stems were not susceptible to blister blight disease. The symptoms caused by blister blight were firstly described in details by Petch (1923) . Blister blight usually has a short life cycle of 11–28 days due to the different climatic conditions in different locations ( Sen et al., 2020 ). In this study, the initial symptoms of blister blight disease usually infected tender leaves that caused light green, yellow-green, light yellow or slightly red tiny water-soaked translucent spots after 5–15 days of infection, then the spots gradually developed into well-defined circular and larger lesions with the diameter of 1–13 mm ( Fig. 1 A). Simultaneously, the spots were sunken into slightly concaves, and on the downside, they bulged correspondingly, finally resulting in typical blister lesions, but there were also a few lesions that bulged towards the upper of the leaves ( Fig. 1 A). The concave upper surface of the blister was smooth and shiny, and the convex surface was generally thickened with powdery coating, then become pure white velvety due to the intensive growth of basidiospores ( Fig. 1 B). At the most serious stage, the adjacent blisters might fuse into large irregular lesions, and diseased tender shoots became distorted or irregularly rolled. At late stage, the mature basidiospores were released and spread by airflow, and the blisters turned dark brown or purple-red ulcer-like and dried up, or even form holes ( Fig. 1 A), and might be infected by other saprophytic fungi. When young stems and petioles were infected, the diseased tissues showed obvious swelling, which turned to dark brown withered spots in the later stage ( Fig. 1 C). The damage to stems was more serious, as the infected stems bent over and broke off at the affected spot, affecting the growth of tea plants. Blister blight is a low temperature and high humidity type disease, and the mycelium is latent in the living tissue of the diseased leaves for over-wintering and over-summering. When the climatic conditions were suitable, the basidiospores fell to the water droplets on the young leaves or new shoots of the tea trees with the wind, germinated and invaded the tea leaves in the water environment, and then the basidiospores reproduced in large numbers, so repeatedly infected, thus causing large yield and economic loss ( Sinniah et al., 2016 , Mur et al., 2015 , Sen et al., 2020 ). It has been reported that a white powder coating on blister can produce two million spores in 24 h which represents an enormous consumption of metabolites and energy of tea ( Baby, 2002 ). In Quxian county, Sichuan province of China, blister blight disease mainly occurred in high mountain tea plantations above 1000 m altitude, infection began to appear when the temperature was above 13°C, and with the increase of temperature and humidity, the disease aggravated. Higher temperature coupled with less rainfall and low humidity during August in summer resulted in the disappearance of blister blight disease, and during December to January in winter, blister blight disease also disappeared when the temperature was lower than 12°C ( Table S1 ). As shown in Table S1 , In the main tea growing areas worldwide, the occurrence period and damage degree of blister blight was closely associated with the different climatic conditions, therefore, the disease incidence were remarkably different in different tea plantations such as Sichuan, Guizhou, Anhui, Yunan, Zhejiang and Hunan province of China, and Sri Lanka, India, etc ( Wang et al., 2013 , Liu et al., 2021 , Liu, 2017 , Shi et al., 2016 , Sinniah et al., 2016 , Mur et al., 2015 , Ran et al., 2021 ). The relationships between rainfall, temperature and humidity to the intensity of blister blight show a strong linear regression pattern, which strongly supports that blister blight intensity decreases with reduced intensity of rainfall, rising temperatures and low humidity ( Mur et al., 2015 ). Although the crop loss changes with the nature of tea varieties and geographical locations, there are no cultivars that are completely resistant to blister blight disease in China, Sri Lanka, India or elsewhere. However, tolerant and susceptible tea varieties showed various degrees of physiological and biochemical changes during the infection of blister blight. Prevention and control of blister blight disease in early stage has not been highly effective due to the lack of suitable biological, chemical and cultural methods. So far, a few fungicides (ie. copper oxychloride, nickel chloride hexahydrate, ergosterol biosynthesis inhibitors) were found to be effective in blister blight disease control ( Baby, 2002 , Sen et al., 2020 ). However, for tea quality, the chemical elicitors may be an more eco-friendly approach for disease control, which can improve the innate immunity of tea plants by significantly increasing the level of defense molecule ( Sen et al., 2020 ). The development and spread of disease on tea leaf The pathogen Exobasidium usually infected young tissues and harvestable tea shoots resulting in serious crop loss, and the normal physiological metabolism and growth of tea was negatively affected by the infection of Exobasidium ( Fig. 1 ). To reveal the development and spread of the pathogen E. vexans on tea leaves, TBS, SEM and TEM analysis were applied. The structure of tea leaves from top to bottom was upper epidermis, palisade tissue, sponge tissue and lower epidermis. As shown in Fig. 2 A, in healthy tea leaves, the cells in the upper epidermis were closely arranged without gap between cells; palisade tissue was composed of closely arranged cylindrical cells, which was perpendicular to the upper epidermal cells; below the palisade tissue was sponge tissue that was consisted of loosely arranged parenchyma cells with irregular cell morphology and large cell gap; the lower epidermis was made up of a layer of flat cell, which was densely covered with villi and stomata. Generally, the pathogen of blister blight disease invaded through the stomata of epidermal cells and was stained dark blue ( Fig. 2 A). With the proliferation of pathogens, basidiospores grew in clusters by consuming nutrients in tea tissue cells ( Mur et al., 2015 ), broke through the epidermis, and a large number of hyphae expanded into the cells of sponge tissue, resulting in the necrosis of tea leaf cells and the destruction of leaf tissue structure ( Fig. 2 A). The results of SEM further confirmed that the pathogen causing blister blight disease mainly invaded from the stomata of the lower epidermis of tea leaves, and during infection, the basidiospores born on the top of the basidia propagated in large numbers and distributed throughout the stomata ( Fig. 2 B). The basidia were clavate, generally bearing two sterigmata, and basidiospores were oval or kidney shaped. In serious cases, the stomatal structure was destroyed, and the guard cells were damaged and deformed, affecting the tissue structure and the development of tea leaves, resulting in the destruction of leaf tissue structure and abnormal growth. In contrast, the lower surface of healthy tea leaves was covered with fine villi, and the stomata were distributed evenly. The cell ultrastructure of healthy tea leaves and infected tea leaves were observed by TEM. As shown in Fig. 2 C, the cell wall structure of healthy tea leaf was clear and complete, the organelles in the cell were intact, and the sponge tissue cells were loosely arranged with large intercellular space that was mainly occupied by vacuoles. Chloroplasts were spindle shaped, evenly distributed near the cell wall, with clear structure and membrane coating. The grana thylakoids were stacked neatly, and the stroma lamellae were arranged orderly ( Fig. 2 C). A small number of mitochondria were distributed around chloroplasts, and the vacuoles were in the center of cells with intact structure. However, the structure of diseased tea leaves was significantly affected by the infection of pathogens. After breaking through the epidermis of tea leaves, the pathogen mainly grew and proliferated in the intercellular space of sponge tissue, and then further infected sponge tissue cells, resulting in the extrusion and deformation of the internal structure of the cells ( Fig. 2 C). The membrane coating of chloroplasts and grana thylakoids disappeared, and the stroma lamellae were mixed with irregular structure ( Fig. 2 C). The whole cell was in a dissociated state, numerous of sediments were distributed in the cytoplasm and organelles were stacked ( Fig. 2 C). In general, after the infection of blister blight disease, the growth and development of tea leaves were negatively affected. Cool temperature (15–25 °C), higher relative humidity (>80 %), less sunshine (<4h) and long duration of surface wetness (>11 h) are conducive to the infection and spread of basidiospores ( Baby, 2002 , Mur et al., 2015 , Sen et al., 2020 ). Composition and abundance of fungal community on blister blight infected tea leaves Tea leaves infected with blister blight were generally thickened with powdery coating ( Fig. 1 ). To verify the pathogens on the lesions, the microbial diversity and abundance of tea leaves showing blister blight were detected. After quality filtering, a total of 124,120 sequences (raw tags) were obtained from three samples (ten lesion tissues for each sample, CB1, CB2 and CB3). The alpha diversity index including Shannon, Chao and Simpson index were calculated to determine the species richness and evenness, and there was no significant difference in each index among the three samples. The number of OTUs per sample ranged from 262 to 268 ( Fig. 3 A). On the base of SILVA taxonomic database, all the gene sequences were classified from phylum to species using analytical program QIIME. Although the distribution of each genus or species varied, the overall fungal composition were similar among samples. In all, the top ten species in mean abundance detected from the diseased lesions were shown in the species distribution river map, and the other known species were classified as others ( Fig. 3 B). The OTUs that cannot be annotated to defined species were defined as the Unclassified category, and approximately 9.29 % of reads have not been classified. Among the known species, Exobasidium was the most abundant fungus accounting for 85.84 % of the total as shown in Fig. 3 B. In addition, there were a small amount of Colletotrichum gloeosporioides (0.27 %), Bulleribasidium variabile (0.11 %) and Hannaella coprosmae (0.06 %). Most of the fungi were plant pathogens, followed by saprotrophs and endophytes ( Fig. 3 C). Our results confirmed the dominant microbial population and abundance of fungal community on tea leaves infected with blister blight disease. However, Barman et al. (2020) found that tea blister blight lesions were colonized by Nigrospora and Pestalotiopsis , and the two endophytes could also present as potent phytopathogen that could inflict serious damage to tea production, which was different with the results of this study. It may be that the fungal community is remarkably influenced by the surrounding environment and host factors. Among these fungi, the dominant Exobasidium is significantly influenced by the environmental factors. Previous studies indicated that a leaf wetness of 11 h was critical for tea blister blight incitation, while temperature more than 32°C was lethal for the basidiospores, sporulation was prevented at 35°C, and average 3.5 h of sunshine per day over 5 days could significantly reduce the incidence of disease ( Visser et al., 1961 , Baby, 2002 ). Species with abundance greater than 0.1 % in the samples were shown in the Fig. 3 D. Among them, OTU000001 was the most abundant fungus which was initially labeled as Exobasidium by microbial diversity analysis. For the phylogenetic analysis, the homologous multiple strains of the Exobasidium species were collected from GenBank database. The evolutionary tree was constructed by NJ method, and the optimal tree with the sum of branch length is 0.43469523. The phylogenetic analysis based on ITS rRNA region indicated that OTU000001 was grouped together with E. reticulatum and also clustered with E. vexans in a large clade ( Fig. 3 E). Blast searches showed that OTU000001 had 99.47–100.00 % identity with E. reticulatum strain SN1 (KY038487) and GP1 (KY038486), and 100 % identity with E. vexans strain MC32016 (MG827276). Both E. reticulatum strains SN1 and GP1 were isolated from blister blight infected tea leaves in India and the paper was unpublished until now. However, for a long time E. reticulatum was considered as the causal agent of tea net blister blight disease which had similar disease symptom and morphological characteristics with blister blight disease caused by E. vexans . E. vexans was the most common pathogen causing tea blister blight disease that had been frequently reported in the tea growing countries worldwide ( Sen et al., 2020 , Chen and Sun, 2013 ). Basidiospores were ellipsoid curved, hyaline and initially unicellular, but a distinct septation developed when the basidiospores mature (the black arrow). The size of the basidiospores was 12.7–25.8 × 4.0–6.9 um ( Fig. 3 E). Although the phylogenetic relationship of E. vexans and E. reticulatum were closely related, based on the disease symptoms ( Fig. 1 ) and morphological characteristics of Exobasidium ( Fig. 3 E), the dominant pathogen on blister blight lesions was identified as E. vexans that systematically placed under Exobasidium (Genus), Exobasidiaceae (Family), Exobasidiales (Order), Exobasidiomycetes (Class), Basidiomycota (Phylum). Sensory evaluation and changes of main non-volatile metabolites in diseased tea Blister blight disease plays an important role in the growth and development of tea trees, which directly influences the tea quality and flavor. So far, the molecular and biochemistry basis of blister blight disease resistance in tea plants have been well studied ( Nisha et al., 2018 , Premkumar et al., 2008 , Mur et al., 2015 ; Jayaswall et al., 2015), however, there are limited information on the tea quality and flavor influenced by this disease. Taste is the most important factor in determining the quality and flavor of tea. Sensory evaluation results showed that diseased tea leaves had strong sweetness with weak bitter, astringent and umami taste compared with healthy tea leaves ( Fig. 4 A), which is not in accordance with the previous results that diseased tea leaves showed obvious bitter taste ( Guo et al., 2005 , Jayaswall et al., 2016 ). To reveal the sensory difference, the taste related metabolites between healthy and diseased tea leaves were investigated. OPLS-DA was performed to distinguish the significantly differential metabolites between CK and CB based on the principle of VIP ≥ 1, P value < 0.05 and fold change ≥ 2 or ≤ 0.5 ( Fig. 4 B). As the main water-soluble carbohydrate in tea, soluble sugar significantly contributes to sweetness of tea ( Rolland et al., 2006 ). The sugars in tea leaves mainly include water-soluble monosaccharides and disaccharides, water-insoluble polysaccharides and a small amount of other saccharides. In this study, total 28 sugars were detected in CK and CB, both including 21 monosaccharides, 6 disaccharides, and 1 trisaccharide ( Fig. 4 C). The content of total sugars in infected tea leaves was remarkably increased by 131.96 % (increased from 19.26 to 44.66 mg/g) ( Fig. 4 C), indicating that the accumulation of total sugars is significantly induced by the infection of blister blight disease and the change is positively associated with the enhanced sweet taste of tea. Sugars not only play important roles in the development of tea plant, yield formation and stress response mainly by producing multiple sugars to fuel growth and synthesize essential compounds, but also can be used as carbon sources to provide material basis for the growth of heterotrophic microorganisms ( Lastdrager et al., 2014 , Rolland et al., 2006 ). 21 monosaccharides, 6 disaccharides and 1 trisaccharide accounted for 61.26 %, 36.22 % and 2.52 % of the total sugars in CK, respectively, while in CB, the three groups accounted for 43.93 %, 61.43 % and 4.64 %, respectively ( Fig. 4 D), showing a significant change by the infection of blister blight disease. Significantly changed metabolites between healthy and diseased tea were determined by VIP ≥ 1, P value < 0.05 and fold change ≥ 2 or ≤ 0.5. Among the 28 sugars, 10 up-regulated sugars (≥2 fold change) were detected in CB compared with healthy tea leaves, while the other 18 sugars showed no significant changes (0.5 < fold change < 2) ( Fig. 4 E). Among the 10 up-regulated sugars, sucrose and maltose belonging to disaccharide accounted for 50.40 % of the total sugars, 7 kinds of monosaccharides including d -xylose, d -arabinose, d -mannose, d -glucuronic acid, glucose, d -galactose and d -fructose only accounted for 15.49 %, and the other 1 trisaccharide (raffinose) accounted for 4.64 % in CB, indicating that monosaccharide and disaccharide were the most abundant differential soluble sugars in diseased tea, which may contribute to the enhanced sweet taste of diseased tea. However, Pius et al. (1998) found that sucrose and glucose contents were significantly decreased in the blistered lesions, and the content of fructose was remarkably increased during the initiation of sporulation and remained constant up to the end of sporulation in non-blistered and blistered regions. The decrease of the sugar accumulation in diseased tea leaves may be due to the increase in the utilization or consumption rate by the pathogen as material basis during the infection process. However, in this study, most of the soluble sugars including sucrose, glucose and fructose detected in raw materials (one bud and two leaves) were all significantly increased ( Fig. 4 C), which may be due to the different infection stages, tea varieties, etc. The metabolic pathway enrichment analysis Based on the KEGG functional annotations and pathway enrichment analysis, significantly differential soluble sugars including monosaccharide, disaccharide and trisaccharide were involved in 14 metabolic pathways ( Fig. 5 A). Among the metabolisms, galactose metabolism, amino sugar and nucleotide sugar metabolism, and starch and sucrose metabolism were the significant enrichment metabolisms involving in the response to blister blight disease ( Fig. 5 A). The greater the rich factor, the more significant the enrichment. The larger size of the dot represents the more differential metabolites enriched in the pathway. Moreover, five sugars (sucrose, d -glucose, d -galactose, d -fructose and raffinose) that significantly increased in diseased tea leaves were mainly involved in galactose metabolism ( Fig. 5 B). Sucrose, d -glucose, d -galactose, d -fructose and raffinose accounting for 69.63 % of the total sugars were the abundant differential soluble sugars showing 3.55, 11.25, 6.13, 5.34 and 4.27 fold higher in CB than those in CK, respectively ( Fig. 5 B). Although the fold change of sucrose is much lower than the other four sugars, sucrose was the most abundant soluble sugar in tea with the content increased by 255.09 %, accounting for 49.93 % of the total sugars in CB ( Fig. 5 B). Sucrose was converted from the raffinose family of oligosaccharides by α-galactosidase, and could be used as substrates for the synthesis of d -glucose and d -fructose. Glucose was the second abundant differential sugar which increased from 0.44 to 4.99 mg/g with the highest fold change in CB, followed by raffinose, d -fructose and d -galactose with the content of 2.07, 1.56 and 0.17 mg/g, respectively ( Fig. 6 B). Raffinose was converted into d -glucose and d -galactose through several intermediate products. The result indicated that the five significantly up-regulated metabolites in sugar metabolism are significantly induced by the infection of blister blight disease which plays a vital role in the tea growth. At present, there were no reports on related sugar metabolism pathway analysis involving in disease infection on tea, however, previous studies found that carbohydrate metabolism (starch and sucrose, fructose and mannose) can be activated by the damage of tea geometrid ( Wang, 2018 ), and sugar metabolism is closely related to the response of tea plants to low temperature stress ( Yue, 2015 ). The influence of blister blight disease on tea quality Generally, besides soluble sugars, the sweetness characteristic of tea is affected by the comprehensive effect of various kinds of metabolites, especially caffeine (bitterness), catechins (bitterness and astringency) and Thea (umami taste) ( Cao et al., 2021 , Cho et al., 2017 , Wang et al., 2010 ). The contents of catechins ( Fig. 6 A) and caffeine ( Fig. 6 D) in blister blight infected tea leaves were all decreased by 23.9 % and 19.22 %, respectively. Gulati et al. (1999) also showed that the significant decrease in catechins was highly related to high disease severity. During the infection of the pathogen E. vexans in tea, catechins can be polymerisated to oligomeric proanthocyanidins and 2,3-cis isomerisation ( Punyasiri et al., 2004 ). As important components of catechins, the contents of epigallocatechin gallate (EGCG) ( Fig. 6 B), epicatechin gallate (ECG) ( Fig. 6 E) and EC ( Fig. 6 F) were all remarkably decreased by 35.36 %, 57.91 % and 22.68 % respectively, while the content of EGC ( Fig. 6 C) was significantly increased by 39.15 %, indicating that EGC was induced by the infection of blister blight disease. Total free amino acids content in CB was significantly decreased by 24.67 % compared with that in CK ( Fig. 6 G), which was consistent with the results of Premkumar et al. (2008) . The decrease of chemical constituents may be due to the degradation by certain secreted metabolites or utilization by the pathogen. Amino acids including the most abundant Thea, glutamic acid (Glu) and aspartic acid (Asp) were thought to be closely associated with the umami taste of tea ( Cheng et al., 2017 ), and the contents of Thea ( Fig. 6 H), Asp ( Fig. 6 I) and Glu ( Fig. 6 J) were all significantly decreased in CB compared with that in CK. However, some amino acids such as arginine (Arg) ( Fig. 6 K), tyrosine (Tyr) ( Fig. 6 L) methionine (Met) ( Fig. 6 M), histidine (His) and phenylalanine (Phe) ( Fig. 6 N) were proved to have bitter taste ( Kirimura et al., 1969 ), and the five amino acids in this study were detected in low contents (<1.0 mg/g). The contents of Tyr, Met and His + Phe were all significantly decreased in CB compared with that in CK, while Arg was significantly enhanced in infected tea leaves ( Fig. 6 ). The increase in the content of Arg may be attributed to the degradation of protein. In summary, the development and spread of blister blight on tea are closely associated with the climatic conditions such as temperature, humidity and sunshine, which significantly influence the metabolites of tea, resulting the quality and flavor change ( Fig. 6 ). The accumulation of caffeine, catechins and its main derivatives (EGCG, ECG and EC), and amino acids and its main components (Thea, Glu and Asp) were all significantly decreased in blister blight infected tea leaves, while EGC, Arg and soluble sugars content were significantly increased ( Fig. 6 ). Therefore, under the suitable environmental conditions (cool temperature, higher relative humidity, etc.), after the infection of E. vexans , the metabolites showing sweet taste in the diseased leaves were significantly increased, while the most of the metabolites showing bitter and astringent taste were obviously reduced, indicating that the changes of these metabolites may be of great significance to enhance the sweet taste of tea.

Results and discussions Disease symptoms on tea caused by blister blight disease As an important tea foliar disease, blister blight mainly infected young organs and tissues including tender leaves, buds, petioles and stems ( Fig. 1 ), directly affecting tea industry qualitatively and quantitatively. The damage to tender shoots and leaves was the most serious, while mature leaves and stems were not susceptible to blister blight disease. The symptoms caused by blister blight were firstly described in details by Petch (1923) . Blister blight usually has a short life cycle of 11–28 days due to the different climatic conditions in different locations ( Sen et al., 2020 ). In this study, the initial symptoms of blister blight disease usually infected tender leaves that caused light green, yellow-green, light yellow or slightly red tiny water-soaked translucent spots after 5–15 days of infection, then the spots gradually developed into well-defined circular and larger lesions with the diameter of 1–13 mm ( Fig. 1 A). Simultaneously, the spots were sunken into slightly concaves, and on the downside, they bulged correspondingly, finally resulting in typical blister lesions, but there were also a few lesions that bulged towards the upper of the leaves ( Fig. 1 A). The concave upper surface of the blister was smooth and shiny, and the convex surface was generally thickened with powdery coating, then become pure white velvety due to the intensive growth of basidiospores ( Fig. 1 B). At the most serious stage, the adjacent blisters might fuse into large irregular lesions, and diseased tender shoots became distorted or irregularly rolled. At late stage, the mature basidiospores were released and spread by airflow, and the blisters turned dark brown or purple-red ulcer-like and dried up, or even form holes ( Fig. 1 A), and might be infected by other saprophytic fungi. When young stems and petioles were infected, the diseased tissues showed obvious swelling, which turned to dark brown withered spots in the later stage ( Fig. 1 C). The damage to stems was more serious, as the infected stems bent over and broke off at the affected spot, affecting the growth of tea plants. Blister blight is a low temperature and high humidity type disease, and the mycelium is latent in the living tissue of the diseased leaves for over-wintering and over-summering. When the climatic conditions were suitable, the basidiospores fell to the water droplets on the young leaves or new shoots of the tea trees with the wind, germinated and invaded the tea leaves in the water environment, and then the basidiospores reproduced in large numbers, so repeatedly infected, thus causing large yield and economic loss ( Sinniah et al., 2016 , Mur et al., 2015 , Sen et al., 2020 ). It has been reported that a white powder coating on blister can produce two million spores in 24 h which represents an enormous consumption of metabolites and energy of tea ( Baby, 2002 ). In Quxian county, Sichuan province of China, blister blight disease mainly occurred in high mountain tea plantations above 1000 m altitude, infection began to appear when the temperature was above 13°C, and with the increase of temperature and humidity, the disease aggravated. Higher temperature coupled with less rainfall and low humidity during August in summer resulted in the disappearance of blister blight disease, and during December to January in winter, blister blight disease also disappeared when the temperature was lower than 12°C ( Table S1 ). As shown in Table S1 , In the main tea growing areas worldwide, the occurrence period and damage degree of blister blight was closely associated with the different climatic conditions, therefore, the disease incidence were remarkably different in different tea plantations such as Sichuan, Guizhou, Anhui, Yunan, Zhejiang and Hunan province of China, and Sri Lanka, India, etc ( Wang et al., 2013 , Liu et al., 2021 , Liu, 2017 , Shi et al., 2016 , Sinniah et al., 2016 , Mur et al., 2015 , Ran et al., 2021 ). The relationships between rainfall, temperature and humidity to the intensity of blister blight show a strong linear regression pattern, which strongly supports that blister blight intensity decreases with reduced intensity of rainfall, rising temperatures and low humidity ( Mur et al., 2015 ). Although the crop loss changes with the nature of tea varieties and geographical locations, there are no cultivars that are completely resistant to blister blight disease in China, Sri Lanka, India or elsewhere. However, tolerant and susceptible tea varieties showed various degrees of physiological and biochemical changes during the infection of blister blight. Prevention and control of blister blight disease in early stage has not been highly effective due to the lack of suitable biological, chemical and cultural methods. So far, a few fungicides (ie. copper oxychloride, nickel chloride hexahydrate, ergosterol biosynthesis inhibitors) were found to be effective in blister blight disease control ( Baby, 2002 , Sen et al., 2020 ). However, for tea quality, the chemical elicitors may be an more eco-friendly approach for disease control, which can improve the innate immunity of tea plants by significantly increasing the level of defense molecule ( Sen et al., 2020 ). The development and spread of disease on tea leaf The pathogen Exobasidium usually infected young tissues and harvestable tea shoots resulting in serious crop loss, and the normal physiological metabolism and growth of tea was negatively affected by the infection of Exobasidium ( Fig. 1 ). To reveal the development and spread of the pathogen E. vexans on tea leaves, TBS, SEM and TEM analysis were applied. The structure of tea leaves from top to bottom was upper epidermis, palisade tissue, sponge tissue and lower epidermis. As shown in Fig. 2 A, in healthy tea leaves, the cells in the upper epidermis were closely arranged without gap between cells; palisade tissue was composed of closely arranged cylindrical cells, which was perpendicular to the upper epidermal cells; below the palisade tissue was sponge tissue that was consisted of loosely arranged parenchyma cells with irregular cell morphology and large cell gap; the lower epidermis was made up of a layer of flat cell, which was densely covered with villi and stomata. Generally, the pathogen of blister blight disease invaded through the stomata of epidermal cells and was stained dark blue ( Fig. 2 A). With the proliferation of pathogens, basidiospores grew in clusters by consuming nutrients in tea tissue cells ( Mur et al., 2015 ), broke through the epidermis, and a large number of hyphae expanded into the cells of sponge tissue, resulting in the necrosis of tea leaf cells and the destruction of leaf tissue structure ( Fig. 2 A). The results of SEM further confirmed that the pathogen causing blister blight disease mainly invaded from the stomata of the lower epidermis of tea leaves, and during infection, the basidiospores born on the top of the basidia propagated in large numbers and distributed throughout the stomata ( Fig. 2 B). The basidia were clavate, generally bearing two sterigmata, and basidiospores were oval or kidney shaped. In serious cases, the stomatal structure was destroyed, and the guard cells were damaged and deformed, affecting the tissue structure and the development of tea leaves, resulting in the destruction of leaf tissue structure and abnormal growth. In contrast, the lower surface of healthy tea leaves was covered with fine villi, and the stomata were distributed evenly. The cell ultrastructure of healthy tea leaves and infected tea leaves were observed by TEM. As shown in Fig. 2 C, the cell wall structure of healthy tea leaf was clear and complete, the organelles in the cell were intact, and the sponge tissue cells were loosely arranged with large intercellular space that was mainly occupied by vacuoles. Chloroplasts were spindle shaped, evenly distributed near the cell wall, with clear structure and membrane coating. The grana thylakoids were stacked neatly, and the stroma lamellae were arranged orderly ( Fig. 2 C). A small number of mitochondria were distributed around chloroplasts, and the vacuoles were in the center of cells with intact structure. However, the structure of diseased tea leaves was significantly affected by the infection of pathogens. After breaking through the epidermis of tea leaves, the pathogen mainly grew and proliferated in the intercellular space of sponge tissue, and then further infected sponge tissue cells, resulting in the extrusion and deformation of the internal structure of the cells ( Fig. 2 C). The membrane coating of chloroplasts and grana thylakoids disappeared, and the stroma lamellae were mixed with irregular structure ( Fig. 2 C). The whole cell was in a dissociated state, numerous of sediments were distributed in the cytoplasm and organelles were stacked ( Fig. 2 C). In general, after the infection of blister blight disease, the growth and development of tea leaves were negatively affected. Cool temperature (15–25 °C), higher relative humidity (>80 %), less sunshine (<4h) and long duration of surface wetness (>11 h) are conducive to the infection and spread of basidiospores ( Baby, 2002 , Mur et al., 2015 , Sen et al., 2020 ). Composition and abundance of fungal community on blister blight infected tea leaves Tea leaves infected with blister blight were generally thickened with powdery coating ( Fig. 1 ). To verify the pathogens on the lesions, the microbial diversity and abundance of tea leaves showing blister blight were detected. After quality filtering, a total of 124,120 sequences (raw tags) were obtained from three samples (ten lesion tissues for each sample, CB1, CB2 and CB3). The alpha diversity index including Shannon, Chao and Simpson index were calculated to determine the species richness and evenness, and there was no significant difference in each index among the three samples. The number of OTUs per sample ranged from 262 to 268 ( Fig. 3 A). On the base of SILVA taxonomic database, all the gene sequences were classified from phylum to species using analytical program QIIME. Although the distribution of each genus or species varied, the overall fungal composition were similar among samples. In all, the top ten species in mean abundance detected from the diseased lesions were shown in the species distribution river map, and the other known species were classified as others ( Fig. 3 B). The OTUs that cannot be annotated to defined species were defined as the Unclassified category, and approximately 9.29 % of reads have not been classified. Among the known species, Exobasidium was the most abundant fungus accounting for 85.84 % of the total as shown in Fig. 3 B. In addition, there were a small amount of Colletotrichum gloeosporioides (0.27 %), Bulleribasidium variabile (0.11 %) and Hannaella coprosmae (0.06 %). Most of the fungi were plant pathogens, followed by saprotrophs and endophytes ( Fig. 3 C). Our results confirmed the dominant microbial population and abundance of fungal community on tea leaves infected with blister blight disease. However, Barman et al. (2020) found that tea blister blight lesions were colonized by Nigrospora and Pestalotiopsis , and the two endophytes could also present as potent phytopathogen that could inflict serious damage to tea production, which was different with the results of this study. It may be that the fungal community is remarkably influenced by the surrounding environment and host factors. Among these fungi, the dominant Exobasidium is significantly influenced by the environmental factors. Previous studies indicated that a leaf wetness of 11 h was critical for tea blister blight incitation, while temperature more than 32°C was lethal for the basidiospores, sporulation was prevented at 35°C, and average 3.5 h of sunshine per day over 5 days could significantly reduce the incidence of disease ( Visser et al., 1961 , Baby, 2002 ). Species with abundance greater than 0.1 % in the samples were shown in the Fig. 3 D. Among them, OTU000001 was the most abundant fungus which was initially labeled as Exobasidium by microbial diversity analysis. For the phylogenetic analysis, the homologous multiple strains of the Exobasidium species were collected from GenBank database. The evolutionary tree was constructed by NJ method, and the optimal tree with the sum of branch length is 0.43469523. The phylogenetic analysis based on ITS rRNA region indicated that OTU000001 was grouped together with E. reticulatum and also clustered with E. vexans in a large clade ( Fig. 3 E). Blast searches showed that OTU000001 had 99.47–100.00 % identity with E. reticulatum strain SN1 (KY038487) and GP1 (KY038486), and 100 % identity with E. vexans strain MC32016 (MG827276). Both E. reticulatum strains SN1 and GP1 were isolated from blister blight infected tea leaves in India and the paper was unpublished until now. However, for a long time E. reticulatum was considered as the causal agent of tea net blister blight disease which had similar disease symptom and morphological characteristics with blister blight disease caused by E. vexans . E. vexans was the most common pathogen causing tea blister blight disease that had been frequently reported in the tea growing countries worldwide ( Sen et al., 2020 , Chen and Sun, 2013 ). Basidiospores were ellipsoid curved, hyaline and initially unicellular, but a distinct septation developed when the basidiospores mature (the black arrow). The size of the basidiospores was 12.7–25.8 × 4.0–6.9 um ( Fig. 3 E). Although the phylogenetic relationship of E. vexans and E. reticulatum were closely related, based on the disease symptoms ( Fig. 1 ) and morphological characteristics of Exobasidium ( Fig. 3 E), the dominant pathogen on blister blight lesions was identified as E. vexans that systematically placed under Exobasidium (Genus), Exobasidiaceae (Family), Exobasidiales (Order), Exobasidiomycetes (Class), Basidiomycota (Phylum). Sensory evaluation and changes of main non-volatile metabolites in diseased tea Blister blight disease plays an important role in the growth and development of tea trees, which directly influences the tea quality and flavor. So far, the molecular and biochemistry basis of blister blight disease resistance in tea plants have been well studied ( Nisha et al., 2018 , Premkumar et al., 2008 , Mur et al., 2015 ; Jayaswall et al., 2015), however, there are limited information on the tea quality and flavor influenced by this disease. Taste is the most important factor in determining the quality and flavor of tea. Sensory evaluation results showed that diseased tea leaves had strong sweetness with weak bitter, astringent and umami taste compared with healthy tea leaves ( Fig. 4 A), which is not in accordance with the previous results that diseased tea leaves showed obvious bitter taste ( Guo et al., 2005 , Jayaswall et al., 2016 ). To reveal the sensory difference, the taste related metabolites between healthy and diseased tea leaves were investigated. OPLS-DA was performed to distinguish the significantly differential metabolites between CK and CB based on the principle of VIP ≥ 1, P value < 0.05 and fold change ≥ 2 or ≤ 0.5 ( Fig. 4 B). As the main water-soluble carbohydrate in tea, soluble sugar significantly contributes to sweetness of tea ( Rolland et al., 2006 ). The sugars in tea leaves mainly include water-soluble monosaccharides and disaccharides, water-insoluble polysaccharides and a small amount of other saccharides. In this study, total 28 sugars were detected in CK and CB, both including 21 monosaccharides, 6 disaccharides, and 1 trisaccharide ( Fig. 4 C). The content of total sugars in infected tea leaves was remarkably increased by 131.96 % (increased from 19.26 to 44.66 mg/g) ( Fig. 4 C), indicating that the accumulation of total sugars is significantly induced by the infection of blister blight disease and the change is positively associated with the enhanced sweet taste of tea. Sugars not only play important roles in the development of tea plant, yield formation and stress response mainly by producing multiple sugars to fuel growth and synthesize essential compounds, but also can be used as carbon sources to provide material basis for the growth of heterotrophic microorganisms ( Lastdrager et al., 2014 , Rolland et al., 2006 ). 21 monosaccharides, 6 disaccharides and 1 trisaccharide accounted for 61.26 %, 36.22 % and 2.52 % of the total sugars in CK, respectively, while in CB, the three groups accounted for 43.93 %, 61.43 % and 4.64 %, respectively ( Fig. 4 D), showing a significant change by the infection of blister blight disease. Significantly changed metabolites between healthy and diseased tea were determined by VIP ≥ 1, P value < 0.05 and fold change ≥ 2 or ≤ 0.5. Among the 28 sugars, 10 up-regulated sugars (≥2 fold change) were detected in CB compared with healthy tea leaves, while the other 18 sugars showed no significant changes (0.5 < fold change < 2) ( Fig. 4 E). Among the 10 up-regulated sugars, sucrose and maltose belonging to disaccharide accounted for 50.40 % of the total sugars, 7 kinds of monosaccharides including d -xylose, d -arabinose, d -mannose, d -glucuronic acid, glucose, d -galactose and d -fructose only accounted for 15.49 %, and the other 1 trisaccharide (raffinose) accounted for 4.64 % in CB, indicating that monosaccharide and disaccharide were the most abundant differential soluble sugars in diseased tea, which may contribute to the enhanced sweet taste of diseased tea. However, Pius et al. (1998) found that sucrose and glucose contents were significantly decreased in the blistered lesions, and the content of fructose was remarkably increased during the initiation of sporulation and remained constant up to the end of sporulation in non-blistered and blistered regions. The decrease of the sugar accumulation in diseased tea leaves may be due to the increase in the utilization or consumption rate by the pathogen as material basis during the infection process. However, in this study, most of the soluble sugars including sucrose, glucose and fructose detected in raw materials (one bud and two leaves) were all significantly increased ( Fig. 4 C), which may be due to the different infection stages, tea varieties, etc. The metabolic pathway enrichment analysis Based on the KEGG functional annotations and pathway enrichment analysis, significantly differential soluble sugars including monosaccharide, disaccharide and trisaccharide were involved in 14 metabolic pathways ( Fig. 5 A). Among the metabolisms, galactose metabolism, amino sugar and nucleotide sugar metabolism, and starch and sucrose metabolism were the significant enrichment metabolisms involving in the response to blister blight disease ( Fig. 5 A). The greater the rich factor, the more significant the enrichment. The larger size of the dot represents the more differential metabolites enriched in the pathway. Moreover, five sugars (sucrose, d -glucose, d -galactose, d -fructose and raffinose) that significantly increased in diseased tea leaves were mainly involved in galactose metabolism ( Fig. 5 B). Sucrose, d -glucose, d -galactose, d -fructose and raffinose accounting for 69.63 % of the total sugars were the abundant differential soluble sugars showing 3.55, 11.25, 6.13, 5.34 and 4.27 fold higher in CB than those in CK, respectively ( Fig. 5 B). Although the fold change of sucrose is much lower than the other four sugars, sucrose was the most abundant soluble sugar in tea with the content increased by 255.09 %, accounting for 49.93 % of the total sugars in CB ( Fig. 5 B). Sucrose was converted from the raffinose family of oligosaccharides by α-galactosidase, and could be used as substrates for the synthesis of d -glucose and d -fructose. Glucose was the second abundant differential sugar which increased from 0.44 to 4.99 mg/g with the highest fold change in CB, followed by raffinose, d -fructose and d -galactose with the content of 2.07, 1.56 and 0.17 mg/g, respectively ( Fig. 6 B). Raffinose was converted into d -glucose and d -galactose through several intermediate products. The result indicated that the five significantly up-regulated metabolites in sugar metabolism are significantly induced by the infection of blister blight disease which plays a vital role in the tea growth. At present, there were no reports on related sugar metabolism pathway analysis involving in disease infection on tea, however, previous studies found that carbohydrate metabolism (starch and sucrose, fructose and mannose) can be activated by the damage of tea geometrid ( Wang, 2018 ), and sugar metabolism is closely related to the response of tea plants to low temperature stress ( Yue, 2015 ). The influence of blister blight disease on tea quality Generally, besides soluble sugars, the sweetness characteristic of tea is affected by the comprehensive effect of various kinds of metabolites, especially caffeine (bitterness), catechins (bitterness and astringency) and Thea (umami taste) ( Cao et al., 2021 , Cho et al., 2017 , Wang et al., 2010 ). The contents of catechins ( Fig. 6 A) and caffeine ( Fig. 6 D) in blister blight infected tea leaves were all decreased by 23.9 % and 19.22 %, respectively. Gulati et al. (1999) also showed that the significant decrease in catechins was highly related to high disease severity. During the infection of the pathogen E. vexans in tea, catechins can be polymerisated to oligomeric proanthocyanidins and 2,3-cis isomerisation ( Punyasiri et al., 2004 ). As important components of catechins, the contents of epigallocatechin gallate (EGCG) ( Fig. 6 B), epicatechin gallate (ECG) ( Fig. 6 E) and EC ( Fig. 6 F) were all remarkably decreased by 35.36 %, 57.91 % and 22.68 % respectively, while the content of EGC ( Fig. 6 C) was significantly increased by 39.15 %, indicating that EGC was induced by the infection of blister blight disease. Total free amino acids content in CB was significantly decreased by 24.67 % compared with that in CK ( Fig. 6 G), which was consistent with the results of Premkumar et al. (2008) . The decrease of chemical constituents may be due to the degradation by certain secreted metabolites or utilization by the pathogen. Amino acids including the most abundant Thea, glutamic acid (Glu) and aspartic acid (Asp) were thought to be closely associated with the umami taste of tea ( Cheng et al., 2017 ), and the contents of Thea ( Fig. 6 H), Asp ( Fig. 6 I) and Glu ( Fig. 6 J) were all significantly decreased in CB compared with that in CK. However, some amino acids such as arginine (Arg) ( Fig. 6 K), tyrosine (Tyr) ( Fig. 6 L) methionine (Met) ( Fig. 6 M), histidine (His) and phenylalanine (Phe) ( Fig. 6 N) were proved to have bitter taste ( Kirimura et al., 1969 ), and the five amino acids in this study were detected in low contents (<1.0 mg/g). The contents of Tyr, Met and His + Phe were all significantly decreased in CB compared with that in CK, while Arg was significantly enhanced in infected tea leaves ( Fig. 6 ). The increase in the content of Arg may be attributed to the degradation of protein. In summary, the development and spread of blister blight on tea are closely associated with the climatic conditions such as temperature, humidity and sunshine, which significantly influence the metabolites of tea, resulting the quality and flavor change ( Fig. 6 ). The accumulation of caffeine, catechins and its main derivatives (EGCG, ECG and EC), and amino acids and its main components (Thea, Glu and Asp) were all significantly decreased in blister blight infected tea leaves, while EGC, Arg and soluble sugars content were significantly increased ( Fig. 6 ). Therefore, under the suitable environmental conditions (cool temperature, higher relative humidity, etc.), after the infection of E. vexans , the metabolites showing sweet taste in the diseased leaves were significantly increased, while the most of the metabolites showing bitter and astringent taste were obviously reduced, indicating that the changes of these metabolites may be of great significance to enhance the sweet taste of tea.

Conclusions Blister blight is a low temperature and high humidity type disease, which not only endangers the growth and development of tea trees, but also affects tea quality and flavor. The disease symptoms, the spread of disease on tea leaves, and the composition and abundance of fungal community on blister tissues were fully described, which greatly advances the understanding of the influence of blister blight disease on tea plants and can provide valuable information to develop an integrated management strategy for the control of blister blight disease. Microbial diversity analysis showed Exobasidium was the most abundant fungus (85.84 %) in blister tissues. Tea quality is affected by the comprehensive effect of various non-volatile substances which is significantly changed by disease infection. The main monosaccharides, disaccharide and trisaccharide in sugar metabolism were significantly induced by the infection of blister blight, and sucrose, d -glucose, d -galactose, d -fructose and raffinose (accounting for 69.63 % of the total sugars) were the abundant differential sugars showing 3.55, 11.25, 6.13, 5.34 and 4.27 fold higher in infected tea leaves, respectively. Furthermore, the bitter taste related metabolites including caffeine, catechins (EGCG, ECG and EC) and some amino acids (Thea, Asp and Glu) were significantly decreased, which also greatly contributes to the enhanced sweet taste of diseased tea.

Highlights • Spread and development of blister blight disease on tea leaves were fully described. • Composition and abundance of fungal community on infected tea leaves were revealed. • Soluble sugars significantly contributed to the enhanced sweetness of diseased tea leaves. • Monosaccharides were the main differential sugars significantly induced in infected tea leaves. Blister blight, as one of the most threatening and damaging disease worldwide, mainly infects young organs and tissues seriously affecting tea growth and quality. In this study, the spread of pathogen on tea leaves were examined by toluidine blue staining, scanning electron microscope and transmission electron microscope analysis. The composition and abundance of fungal community on leaf tissues were firstly analyzed. Sensory evaluation and metabolites analysis indicated that diseased tea leaves had strong sweet taste and soluble sugars contributed significantly to the taste, while metabolites showing bitter and astringent taste (caffeine, catechins) were significantly decreased. According to the biological functions of differential metabolites, sugars including 7 monosaccharides ( d -xylose, d -arabinose, d -mannose, d -glucuronic acid, glucose, d -galactose and d -fructose), 2 disaccharide (sucrose and maltose) and 1 trisaccharide (raffinose) were the main differential sugars increased in content (>2 fold change), which was of great significance to sweet taste of diseased tea. Keywords

Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Supplementary data The following are the Supplementary data to this article: Data availability Data will be made available on request. Acknowledgements This work was supported by the 10.13039/501100001809 National Natural Science Foundation of China (NSFC) (32272764) and Chongqing Technology Innovation and Application Development Project (cstc2021jscx-tpyzxX0010).

CC BY

no

2024-01-16 23:40:18

Food Chem X. 2023 Dec 18; 21:101077

oa_package/e2/a7/PMC10788223.tar.gz

PMC10788226

38226032

Experimental Design, Materials and Methods Rearing of S. frugiperda In this study, S. frugiperda (instar L4-L5) were freshly collected from maize field crops during 2022, transported in containers, and reared using artificial diet under laboratory conditions [1 , 2] . DNA Extraction For DNA extraction and metagenomics analysis, two complete F3 larvae (instar L3) were selected to obtain only one DNA sample. DNA was isolated using the ZymoBIOMICS DNA Miniprep Kit (Zymo Research, Irvine, CA) following the manufacturer's instructions. The genomic DNA was processed and analyzed with the Shotgun Metagenomic Sequencing Service (Zymo Research, Irvine, CA). Sequencing and Assembly Sequencing libraries were prepared with Illumina® DNA Library Prep Kit (Illumina, San Diego, CA) and the final library was sequenced on the platform NovaSeq® (Illumina, San Diego, CA). Generating 1.1 Gb of paired-end reads of 150 bp in length. Bioinformatics analyses were made using the pipelines at Bacterial and Viral Bioinformatics Resource Center (BV-BRC) and was submitted to the Metagenomic Binning Service [3 , [7] , [8] , [9] , [10] . Each set of binned contigs was annotated using RAST tool kit (RASTtk) [10] . All software were run with default parameters.

Spodoptera frugiperda (Smith) (Lepidoptera: Noctuidae), also known as the fall armyworm, is an economically important and widespread polyphagous pest. Microorganisms associated to this insect during life cycle play important ecological roles. We report 3 metagenome-assembled bacterial genomes reconstructed from a metagenome dataset obtained from S. frugiperda larvae F3 3rd-instar reared using artificial diet under laboratory conditions. Genome data for Enterococcus casseliflavus indicated a genome length of 3,659,8333 bp and GC content of 42.54%. Genome data for E. mundtii indicated a genome length of 2,921,701 bp and GC content of 38.37%. Finally, genome data for Lactiplantibacillus plantarum indicated a genome length of 3,298,601 bp, GC content of 44.31%. Genome analysis allowed us to identify genus-specific protein families (PLFams), transporters and antibiotic resistance-related genes among others. DNA sequences were deposited in National Center for Biotechnology Information ( https://www.ncbi.nlm.nih.gov/ ) as Bioproject accession PRJNA899064. Keywords

Specifications Table Value of the Data • The Genomes of E. casseliflavus, E. mundtii and L. plantarum can provide insights for the understanding of bacterial interaction with S. frugiperda. • These bacterial genomes data are applicable for comparative genomic and taxonomic purposes. • These data are valuable resources for researchers working in the field of S. frugiperda microbiome to understand ecological interactions and use of biological control agents. • Data will help to expand the knowledge of bacteria associated to healthy larvae under laboratory-rearing conditions or their interactions with the artificial diet. Objective Healthy colonies of insects are a mandatory requirement for biocontrol experiments. In this regard, endogenous microbiota of S. frugiperda might influence growth development and overall state of the insect. However, little is known about S. frugiperda microbiota during rearing using artificial diet under laboratory conditions. Therefore, the aim of the present work was to identify relevant genomic features and functional genes from 3rd-instar larvae of S. frugiperda -related bacteria with a potential ecological role, through a metagenome-assembled bacterial genome approach. Data Description This data contains metagenome-assembled bacterial genome using shotgun metagenomic sequencing of two 3rd-instar larvae of S. frugiperda reared using artificial diet under laboratory conditions [1 , 2] . The sequencing result was of 1.1 Gb paired-end reads of 150 bp in length. Table 1 provides the MAGs available in the dataset. Bacterial binning analyzed in CheckM [3] with high-quality produced with > 99.8 % completeness and < 2.1% contamination ( Table 1 ). The BV-BRC metagenomic binning service [3] show that the genome ( Fig. 1 ) for E. casseliflavus [4] contains 41 contigs with genome length of 3,659,833 bp, a mean coverage of 272.59 and GC content of 42.54%. The annotated genome identifies 3338 proteins belong to genus-specific protein families (PLFams) and 3664 protein coding sequence (CDS), 1 virulence factor according to VFDB source, 33 transporters and 41 antibiotic resistance-related genes. The genome for E. mundtii [5] contains 47 contigs with genome length of 2,921,701 bp, a mean coverage of 378.03 and GC content of 38.37%. The annotated genome identifies 2739 proteins belong to genus-specific protein families (PLFams) and 2923 protein coding sequence (CDS), 2 virulence factor according to VFDB source, 15 transporters and 38 antibiotic resistance-related genes. The genome for L. plantarum [6] contains 122 contigs with genome length of 3,298,601 bp, a mean coverage of 14.81 and GC content of 44.31%. The annotated genome identifies 2893 proteins belong to genus-specific protein families (PLFams) and 3258 protein coding sequence (CDS), not detected virulence factor, 16 transporters and 28 antibiotic resistance-related genes. Table 2 lists the antibiotic resistance genes present in each bacteria specie. Data Accessibility The raw sequence data were deposited at the National Centre for Biotechnology Information (NCBI) database under the project number PRJNA899064. The sequences of MAGs are available at GenBank under the genome accessions summarized in Table 1 . Ethics Statements This work did not involve any human subjects, animals or species that require ethical approval. CRediT authorship contribution statement Francisco Javier Flores Gallardo: Methodology, Software. José Luis Hernández Flores: Conceptualization, Methodology, Software, Writing – original draft. Selene Aguilera Aguirre: Methodology, Software, Formal analysis. Miguel Ángel Ramos López: Writing – original draft. Jackeline Lizzeta Arvizu Gómez: Investigation, Formal analysis. Carlos Saldaña Gutierrez: Investigation. María Carlota García Gutiérrez: Methodology. José Alberto Rodríguez Morales: Methodology. Juan Campos Guillén: Conceptualization, Methodology, Software, Writing – original draft.

Data Availability PRJNA899064 (Original data) (NCBI GenBank). Acknowledgments Funding: This study was partially financed by the 10.13039/100008989 Universidad Autónoma de Querétaro (FONDEC-UAQ-2022; FOPER-2022). Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:40:18

Data Brief. 2023 Dec 18; 52:109989

oa_package/4b/e7/PMC10788226.tar.gz

PMC10788227

38226171

Introduction Autoimmune diseases affect approximately 5% of the global population. 1 Biological drugs targeting proinflammatory cytokines are effective; however, disease recurrence during or after treatment is often problematic. 2 One of the cellular targets of such drugs contain interleukin-17 (IL-17)-producing helper T (Th17) cells. 3 , 4 , 5 Th17 cells regulate inflammation by producing proinflammatory effector cytokines, such as IL-17, tumor necrosis factor alpha (TNF-α), IL-22, interferon gamma (IFNγ), and granulocyte-macrophage colony-stimulating factor (GM-CSF). Furthermore, Th17 cells survive longer than other T cells. Memory-like Th17s can replicate and produce a new effector cell population, which resembles stem cells, and contributes to their persistence. 6 , 7 , 8 Understanding the mechanisms underlying long-term survival of autoreactive Th17 will be useful for deciding the optimal treatment of chronic autoimmune diseases. CD28, the best-known costimulatory receptor, 9 mediates T cell activation through IL-2 production 10 and resistance to apoptosis. 11 At the molecular level, CD28 ligation activates serine/threonine kinase Akt, nuclear factor kB (NF-kB), and the mechanistic Target of Rapamycin (mTOR) that stimulate uptake and metabolism of glucose for full activation and differentiation into effector cells. 12 , 13 Abatacept, a fusion protein of CTLA-4 and immunoglobulin (Ig) antagonizes CD28 by competing with costimulatory ligands (CD80 and CD86), acts as a strong inhibitor of T cells, and has been used for the treatment of autoimmune diseases. CD28 serves as a primary checkpoint in T cell activation but also may be involved in the peripheral maintenance of T cell homeostasis such as Th17 cells. Desmoglein 3 (DSG3) is an adhesion molecule that is primarily expressed on keratinocytes. It is also the target autoantigen in pemphigus vulgaris, an autoimmune blistering disease. 14 T cells extracted from mice carrying a DSG3-specific T cell receptor (Dsg3H1 TCR Tg mouse; hereafter simply designated as Dsg3H1) are known to directly infiltrate the epidermis and induce cellular immunity in DSG3-bearing keratinocytes and cause interface dermatitis after adoptive transfer into Rag2−/− mice. 15 , 16 Using the modified chronic experimental autoimmune dermatitis (EAD) model, we demonstrated that CD28 signal plays a key role in activation and effector function of Th17. Abatacept treatment completely blocked the development of skin inflammation by inhibiting activation and proliferation of effector T cells. In contrast, IL-7 receptor (IL-7R)-positive Th17 cells with memory-like phenotype were resistant to abatacept and remained in the body. To inhibit abatacept-resistant remaining Th17 cells in vivo , we extensively characterized this population and discovered that ALDH inhibitors can prevent the formation of this memory population.

STAR★Methods Key resources table Resource availability Lead contact Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Shunsuke Chikuma ( [email protected] ). Materials availability All unique reagents used in this study are available from the lead contact upon reasonable request. Data and code availability (1) RNA-seq data have been deposited at DDBJ and are publicly available as of the date of publication. (2) Accession numbers are listed in the key resources table . This paper does not report original code. (3) Any other information required to reanalyze the data reported in this paper is available from the lead contact upon request. Experimental model and study participant details Animals Dsg3H1 TCR Tg, 15 B6 CD45.1 congenic mice (Jackson Laboratory), and CD3 epsilon knockout mice were used in this study. Female mice, 7 weeks old, of the wild-type C57BL/6NCrSlc strain were obtained from Sankyo Laboratory (Tokyo, Japan). All mice were housed in SPF facilities at Keio University under standard conditions, which included a12-h light–dark cycle (lights-on at 7:30 a.m.) at a temperature of 24°C ± 1°C, with food and water provided ad libitum . All animal experiments were conducted according to the approved protocol (#80006) of the Animal Ethics Committee of Keio University Medical School. Methods details Th17-mediated EAD model The in vitro activation procedure of Dsg3H1 T cells was modified from Nishimoto et al. 16 Lymph node cells from 4–6-week-old Dsg3H1 mice were subjected to magnetic sorting using naive CD4 microbeads (Miltenyi) and LS columns (Miltenyi), following the manufacturer’s protocol. The sorted T cells were cultured in a 24-well tissue culture plate (Corning) coated with anti-CD3 and anti-CD28 mAbs (2 μg/mL each; Biolegend) at a density of 2 × 10 5 /mL in 1 mL of T cell culture media (RPMI medium supplemented with 10% fetal calf serum (FCS), penicillin/streptomycin, non-essential amino acid solution, HEPES solution, sodium pyruvate solution, and 55 μM 2-mercaptoethanol [RPMI and supplements were all procured from Nacalai Tesque, Kyoto, Japan, except 2-ME: Gibco]). To induce pathogenic Th17, mouse IL-6 (20 ng/mL), mouse IL-23 (20 ng/mL), human TGF-β (2 ng), mouse IL-1β (10 ng/mL), anti-mouse IFN-γ (5 μg/mL), and anti-mouse IL-4 (5 μg/mL) were added to the culture. Three days later, cells were harvested from the plate and further expanded in the presence of mouse IL-23 (20 ng/mL) and mouse IL-2 (20 ng/mL) for another 3 days. To induce control Th1, IL-12 (20 ng/mL) and anti-mouse IL-4 (5 μg/mL) were added to the culture. Three days later, cells were harvested from the plate and further expanded in the presence of mouse IL-2 (20 ng/mL) for another 3 days. On the day of transfer, 4–5 million expanded T cells were resuspended in phosphate-buffered saline (PBS) and intravenously injected into 5Gy-irradiated wild-type C57BL/6NCrSlc or non-irradiated B6 CD3e KO mice. In some experiments, donor Dsg3H1 TCR Tg mice were further bred to B6 CD45.1 congenic mice to allow the identification of donor T cells (CD45.1 + ) from recipient cells (CD45.2) ex vivo . To evaluate T cell expansion in vivo , T cells were labeled with CellTrace Violet dye right before the transfer according to the manufacturer’s instructions. As a humane endpoint, mice with dermatitis were observed for up to 1.5 months and then euthanized. Abatacept and cyanamide treatment Mice received intraperitoneal injections of abatacept (Bristol-Myers Squibb; 100–200 μg/body) or an equivalent amount of human immunoglobulin (Jackson ImmunoResearch Lab) at the time of transfer and then again on days 2 and 4. In some experiments, mice were administered daily cyanamide (80 mg/kg body weight) dissolved in water via oral gavage. Control mice received water. Preparation of skin infiltrating T cells For the isolation of skin cells, the pinna of sacrificed mice was mechanically separated into skin and cartilage. The skin was incubated in a 2.5 mg/mL Trypsin/1 mM EDTA solution (Nacalai Tesque) with the epidermal side facing up at 37°C for 1 h and then separated into epidermis and dermis. The epidermal sheet was gently rubbed with the plunger end of a disposable plastic syringe against a 100 mM cell strainer to obtain single cells. The dermis was further digested in RPMI 10% containing FCS, 2 mg/mL collagenase D (Roche), 1.2 mg/mL hyaluronidase (Fujifilm-Wako Pure Chemicals), and 100 μg/mL DNase-I (Roche) at 37°C for 1 h, and single cells were prepared. FACS Cells from the skin or lymph nodes were stained in FACS buffer (PBS containing 1% BSA and 0.05% sodium azide) with fluorochrome-conjugated antibodies against mouse T cells and a flexible viability dye (Fixable Viability Dye; FVD, Thermo Fisher Scientific). All the antibodies used in this study were obtained from Thermo Fisher Scientific (Tokyo, Japan) or BioLegend (Tokyo, Japan). The stained cells were analyzed using a FACSCanto II analyzer (BD Bioscience) or a CytoFlex S (Beckman), and the data were analyzed using Flowjo software (BD Bioscience). All gating strategy in this study for FACS was presented in Figures S3–S11 . For the detection of intracellular cytokines, cells were cultured at a density of 4 × 10 6 /mL in 1 mL of T cell culture media and stimulated with PMA (50 ng/mL) and ionomycin (1 μg/mL) for 5 h. During the last 2 h of stimulation, brefeldin A (3 μg/mL) and monensin (2 μM) were added to the culture. After the culture, cells were first stained with cell surface antigens, fixed, and permeabilized using a fixation/permeabilization buffer (BD), and then stained with anticytokine antibodies. For the detection of GLUT1, cells were fixed without stimulation, permeabilized in the same way as for cytokines, and then stained with Alexa Fluor 647 anti-GLUT1 antibody. Recovery of donor cells from recipients For RNA sequencing ( Figures 4 C and 5 A), the restimulation ( Figures 4 E and 4F) and secondary transfer assay of donor cells ( Figures 4 G and 4H), spleen and lymph node cells from recipient mice were sorted using CD4 microbeads (Miltenyi) and LS columns (Miltenyi). The sorted CD4 + T cells were labeled with a donor T cell marker (CD45.1 + ) and other markers, and then sorted using an FACS ARIA III. T cell restimulation assays For restimulation by plate-bound antibodies ( Figure 3 A), Dsg3H1-pTh17 T cells were plated in a 24-well tissue culture plate coated with anti-CD3 and anti-CD28 mAbs (2 μg/mL each) at a density of 1 × 10 6 /mL in 1 mL of T cell culture media. After 2 h, the cells were recovered for RNA extraction. For restimulation by Dsg3H1 peptide ( Figures 4 and S2 ), 2 × 10 4 donor T cells were cocultured with 2 × 10 5 splenocytes from wild-type mice irradiated with 20 Gy in a 96-well flat plate. The co-culture was performed in the presence of 2 μg/mL DsgH1 mimotope peptide (RNKAEFHQSVISQYR) in 0.2 mL of T cell culture media. In Figures 2 and 2 μg/ml anti-CD28 was additionally added to the indicated wells. After three days, cytokine production and cell proliferation were measured using Cytometric Bead Array (BD) and Cell Count Reagent SF, respectively, following the manufacturers’ instructions. Serial transfer model For the experiment presented in Figures 4 G and 4H, spleen and lymph node cells were pooled from 3 to 5 recipients 1 week after transfer. Donor cells were sorted as described above, and then 3 × 10 5 cells were intravenously transferred to 5Gy-irradiated C57BL/6N mice. The mice were analyzed 1 week later. RNA sequencing Total RNA was isolated using the RNeasy Plus Micro Kit (Qiagen). Libraries were prepared using the TruSeq stranded mRNA Library kit and sequenced on a NovaSeq 6000 (Illumina) to obtain 150-bp paired-end reads. HISAT2 version 2.1.0 was used to map the RNA-seq data to the mouse genomic DNA sequences (mm10). Read counts, fragments per kilobase of exon per million mapped fragments, and transcripts per million were calculated using featureCounts version 1.6.3. The samples were clustered using the Wald method based on Euclidean distances of the normalized counts, utilizing the stats (Version 3.6.1) and ggplots (Version 3.0.1.1) R packages. Then, DEGs were identified using DESeq2 version 1.30.1. Quantitative reverse transcription-PCR Total RNA was reverse-transcribed by High-Capacity cDNA Reverse Transcription Kit (Thermo Fisher.) Resulting cDNA was amplified with SsoFast EvaGreen Supermix (Bio-Rad) and CFX Connect Real-time PCR system by according to manufacturers’ protocols. Primers used for PCR are listed in the key resources table . Bioinformatics Enrichment of GO biological processes was performed using Metascape. 42 GSEA 43 was conducted using the GSEA desktop application (ver. 4.2.3). Pathway analysis and molecular characterization information were obtained using Ingenuity Pathway Analysis (IPA; Qiagen). Analysis of human cancer database Analysis of The Cancer Genome Atlas (TCGA) database for gene correlation was conducted with the assistance of the Timer 2.0 resource. 44 , 45 Data deposition All RNA sequence data have been publicly deposited on NCBI under accession #DRA016062 (run numbers DRR457517-DRR457541). Quantification and statistical analysis All statistical analyses, except for RNA-seq data, were performed using GraphPad PRISM software (ver. 8.4.3). Student’s T -test was used for comparing two groups, while multiple comparisons of one-way ANOVA were used for databases involving more than three groups. Cumulative incidence of dermatitis in experiments was analyzed using the Kaplan–Meier method with log rank tests. For gene correlation analysis, partial Spearman’s correlation was determined through TIMER 2.0 analysis. Detailed statistical information for each experiment and the number of replicates can be found in the corresponding figure or figure legends.

Results Pathogenic Th17-dependent mouse model of chronic skin inflammation We previously reported a Th17-dependent EAD model in mice. 16 Briefly, naive CD4 + T cells extracted from Dsg3H1 mice 15 were purified and activated in vitro under Th17 polarizing condition. Subsequently, when these Th17 cells were transferred into lymphocyte-deficient Rag2 knockout mice, they induced IL-17-dependent subacute skin inflammation, both histologically and immunologically resembling psoriasis. 16 Unfortunately, severe weight loss and rapid death of recipient mice following cell transfer prevented us from analyzing persistent autoimmune disease in this model ( 16 and unpublished). To investigate long-term Th17 cell survival in the persistent EAD model, we used sublethally irradiated wild-type mice as recipients. Additionally, during T cell differentiation culture, we added IL-1β, known to promote the development of long-lived pathogenic Th17 cells (pTh17; Figure 1 A and STAR methods section). The resulting pTh17 cells exhibited stronger production of IL-17A compared with normally skewed Th17 cells (nTh17). Furthermore, pTh17 cells produced IL-17F ( Figure S1 A), which is suggested to be produced by more epigenetically committed Th17 cells. 17 Genes encoding IL-23 receptor and GM-CSF ( Il23r and Csf2 , respectively) are higher in pTh17 than nTh17 ( Figure S1 B). Consequently, we termed the induced cells “Dsg3H1-pTh17,” representing pathogenic Th17-skewed Dsg3H1 cells. When Dsg3H1-pTh17 cells were transferred into irradiated syngeneic wild-type C57BL/6 mice, they induced dermatitis at a slower rate than in Cd3e KO recipients (which lack endogenous T cells and showed EAD with kinetics similar to Rag2 KO recipients) ( Figure 1 B). The skin inflammation, typically affecting the ears, back, neck, and/or tail, persisted for at least a month without the death of recipient mice ( Figure 1 C and data not shown). Thickened skin with a massive infiltration of mononuclear cells in epidermal and dermal tissues was evident ( Figure 1 C). In the affected skin, strong expression of cytokines, such as IL-17A, IFNγ, IL-6, and TNF-α, were detected, indicating severe inflammation caused by Dsg3H1-pTh17 ( Figure 1 D). Using congenically labeled donor T cells (pTh17 prepared from Dsg3H1 transgenic, CD45.1 congenic mice) allowed us to discriminate transferred cells via fluorescence-activated cell sorting (FACS) analyses in the recipients (CD45.2). In support of Th17-dependent inflammation, the transferred donor Dsg3H1-pTh17 cells (CD45.1 + ) were detected in epidermal and dermal tissues after 2 weeks, as well as in skin-draining lymph nodes and spleen ( Figure 1 E, upper panels). In contrast, Dsg3H1 T cells skewed into Th1 (Dsg3H1 Th1; Figure 1 E lower panels) did not persist in the skin nor induce skin inflammation in vivo . Moreover, Dsg3H1-pTh17 cells neither caused any inflammation nor exhibited survival beyond 2 weeks in vivo (data not shown) in non-irradiated wild-type mice. Taken together, we have successfully developed a chronic EAD model induced by the injection of pathogenic Th17 cells reactive to a defined autoantigen in the skin. CD28 blockade by abatacept prevents Th17-mediated skin inflammation Abatacept, a human CTLA-4 Ig that inhibits CD28 signaling, was previously shown to ameliorate human psoriasis. 18 , 19 However, its impact on skin-reactive helper T cells remains unknown. Therefore, we conducted tests using abatacept in our model. Remarkably, mice that received Dsg3H1-pTh17 cells and were treated with abatacept showed nearly complete prevention of skin lesions, indicating the critical role of CD28 signaling in skin inflammation ( Figures 2 A and 2B). To facilitate tracking of transferred cells and assessment of proliferative responses, we utilized congenic marker (CD45.1) and a proliferation reporter dye (CTV) ( Figures 2 C and 2D). Abatacept-treated recipients showed fewer donor cells compared with control mice ( Figure 2 C). While a significant increase in the proliferation of donor CD4 + cells was observed in mice treated with control Ig, abatacept-treated mice exhibited inhibition of proliferative responses in transferred pTh17 cells ( Figure 2 D). Glucose transporter 1 (GLUT1), a target gene of CD28 and a hallmark indicator of glucose metabolism and extensive T cell proliferation, 12 , 13 showed significantly lower expression in donor cells derived from abatacept-treated recipients ( Figure 2 E). These findings suggest that abatacept exerts an inhibitory effect on pTh17 cells. CD28 drives expression of proinflammatory genes in pathogenic Th17 The robust effects of abatacept prompted further investigation into the fundamental roles of CD28 signaling in pTh17 cells. Therefore, we performed an RNA-sequencing analysis on fully differentiated Dsg3H1-pTh17 cells that were restimulated with plate-immobilized antibodies ( Figure 3 A). Importantly, the “CD3-stimulated” and “CD3+CD28-stimulated” samples showed distinct clustering in principal-component analysis of RNA-sequencing data, as early as 2 h after restimulation ( Figure 3 B). Analysis of differentially expressed genes (DEGs) revealed that the rapid induction (at 2 h) of effector cytokines and transcriptional factors critically depend on CD28 signaling ( Figure 3 C). Specifically, we found that most CD28-dependent genes were cytokines (including those encoding IL-21, IL-2, TNF-α, GM-CSF, CCL4, XCL1, CCL20, IL-2, IL-31, and Tnfrsf4) and transcriptional regulators ( Nfkbid , Fos , Nfkbia , Maff , and Atf3 ) that play pivotal roles in inflammation. These genes coexisted with an antiapoptotic protein, Bcl-XL (encoded by Bcl2l1 ) ( Figure 3 C). A gene set enrichment analysis (GSEA) demonstrated that “CD3+CD28-stimulated cells” exhibited a strong bias toward the “INFLAMMATORY RESPONSE” signature compared with cells stimulated by CD3 alone ( Figure 3 D). We confirmed these findings by reactivating Dsg3H1-pTh17 cells using the physiologic cognate peptide recognized by Dsg3H1 T cells 15 ( Figure S2 A). We observed that secondary proliferation ( Figure S2 B) and the production of effector cytokines ( Figure S2 C) were significantly augmented by CD28 signaling. These findings indicate that signals mediated by CD28 are critically involved in the secondary response of pTh17 cells. Abatacept inhibits IL-7R neg inflammatory T cells but not IL-7R pos memory T cells Our model allowed us to examine the phenotypes of transferred pTh17 cells following CD28 blockade by abatacept. One week after transfer, we conducted an extensive analysis of donor Dsg3H1-pTh17 cells isolated from lymph nodes using a combination of congenic markers (CD45.1), a proliferation reporter dye (CTV), and other markers associated with effector/memory responses. We observed that the majority of proliferated CTV-diluted (CTV dil ) cells exhibited a low/negative ( neg ) phenotype for the IL-7 receptor α (hereafter referred to as IL-7R), and those from abatacept-treated mice were much fewer in number compared with control mice ( Figures 4 A and 44B left panels). In contrast, donor-derived cells that survived in abatacept-treated mice were mostly IL-7R positive ( pos ). Moreover, the number of CTV dil IL-7R pos cells was not affected by abatacept, indicating that these cells proliferated independently of CD28 signaling ( Figure 4 B right). We compared transcriptome data between the CTV dil IL-7R pos and CTV dil IL-7R neg populations in donor cells isolated from control mice ( Figure 4 C). Genes highly expressed by CTV dil IL-7R neg cells included effector molecules (such as Gzmb and Fasl ), cytokines ( Il21 , Ifng , Spp1 ; osteopontin), chemokine receptors ( Cxcr5 and Cx3cr1 ), transcription factors (TF) ( Nr4a1 and Nr4a2 ), and cell surface molecules ( Havcr2 and Tigit ), all of which suggested strong T cell activation ( Figure 4 D). In contrast, IL-7R pos cells expressed genes related to Th17 cells ( Satb1 , Mafb , Ccr4 , Ccr6 , and Rara ; Figure 4 D). These findings suggested that inflammatory cells in the IL-7R neg population were inhibited by abatacept. To directly examine cytokine expression, we sorted CTV dil IL-7R pos and CTV dil IL-7R neg cells from recipients treated with control Ig or abatacept. An equal number of sorted cells were then restimulated with the cognate antigenic Dsg3H1 peptide. Regardless of the treatment, CTV dil IL-7R neg cells produced higher levels of IFNγ, whereas CTV dil IL-7R pos cells expressed IL-2 ( Figures 4 E and 4F). Both populations produced a comparable amount of IL-17A, indicating that they are subpopulations of Th17 cells ( Figure 4 E). Importantly, the sorted cells proliferated similarly ( Figure 4 F), indicating that unlike naive T cells, Th17 cells did not become anergic following CD28 blockade. We conducted experiments to investigate whether IL-7R-positive and -negative pTh17 cells can transdifferentiate into each other and set up a transfer experiment into secondary recipients ( Figure 4 G). We observed that most CTV dil IL-7R pos cells, when transferred into secondary recipients, became IL-7R neg , whereas most CTV dil IL-7R neg cells remained IL-7R neg ( Figure 4 H). Taken together, these results suggested that pTh17 cells comprise two distinct populations. Abatacept appeared to block the IL-7R neg effector-like Th17 cells but allowed the survival of IL-7R pos memory-like cells that have the potential to transdifferentiate into IL-7R neg effector cells. Abatacept inhibits effector signature but allows survival of persistent memory cells We conducted RNA sequencing, to directly compare the entire donor T cell population isolated from recipients treated with either control Ig or abatacept ( Figures 5 A and 5B). As anticipated from the in vitro data, cells from abatacept-treated mice showed decreased expression of cytokines and transcription factors (TFs) associated with effector function ( Figures 5 C and 5D). The cytokines downregulated by abatacept treatment included those typically expressed by activated helper T cells, such as Ifng (Th1), Il4 (Th2), Il21 (T follicular helper; Tfh), and Spp1 . The TFs downregulated by abatacept also encompassed lineage-specific TFs such as Tbx21 (T-bet; Th1), Foxp3 (Treg), Bcl6 (Tfh), and those involved in activation and function ( Eomes , Tox , Tox2 , Nfatc1 , Nr4a2 , Ezh2 , and Batf ), indicating a clear inhibition of effector function. Conversely, T cells recovered from abatacept-treated recipients did not show significant upregulation of cytokines or chemokines, except for Ccl1 . However, they exhibited upregulation of unique transcriptional factors ( Nr1d1 , Satb1 , Bhlhe41 , Myb , and Foxq1 ). SATB1 20 and NR1D1 (REV-ERBα) 21 , 22 , 23 have been reported to be involved in the development and function of Th17 cells. Myb is known to be essential in CD62L pos stem cell memory development. 24 , 25 These findings collectively support the notion that CD28 blockade by abatacept inhibited the effector function of Th17 cells while preserving a unique IL-7R pos memory population. Abatacept-resistant memory Th17 cells exhibit genes for aldehyde dehydrogenases We further investigated the genes that may function on T cells extracted from abatacept-treated mice. Cells from control mice showed genes associated with “glycoprotein metabolic process,” “response to virus,” and “carbohydrate derivative catabolic process” signatures, suggesting the reliance of the cells depends on glycolysis for proliferation and effector function ( Figure 6 A). In contrast, cells from abatacept-treated mice showed genes linked to “cholesterol metabolism pathways,” “carbohydrate biosynthesis process,” and “amino acid metabolism process,” including Dhcr24 , Acsl3 , Them4 , and Acss2 ( Figures 6 A and 6B). These findings align with previous reports that highlight the importance of cholesterol and lipid metabolism in the survival and pathogenicity of Th17 cells. 26 , 27 In addition to genes involved in energy acquisition, we observed a significant upregulation of genes related to alcohol metabolism in donor cells treated with abatacept (categorized in “ethanol oxidation”; Aldh2 , Acss2 , Aldh1b1 , Acat1 , Fpgs , Uros , Mthfd1 , Acsl3 , Aldoc , Aldh6a1 , etc.; Figure 6 C). We were particularly intrigued by the upregulation of aldehyde dehydrogenase (ALDH) genes ( Aldh2 , Aldh1b1 , and Aldh6a1 ) for several reasons. First, ALDH plays a role in regulating stemness in hematopoiesis 28 and malignancy. 29 Second, ALDH is involved in mitochondrial function. 30 Third, ALDH may contribute to T cell survival. 31 Fourth, ALDH expression has been reported in supporting Treg survival in humans. 32 Importantly, we also found that Aldh2 was upregulated in IL-7R pos donor T cells extracted from both control and abatacept-treated mice (compared with IL-7R neg donor T cells.) ( Figure 6 D) Consequently, Aldh2 was upregulated in both IL-7R pos cells and T cells isolated from abatacept-treated mice (that are enriched in IL-7R pos memory cells in independent cohorts ( Figure 6 E). Therefore, our model reveals that memory-phenotype pTh17 cells exhibit a unique metabolic pathway that may involve ALDH for both survival and function. Abatacept together with ALDH inhibitor targets memory Th17 cells We aimed to explore the role of ALDH in Th17 activity both in vitro and in vivo . We used cyanamide and disulfiram, which are traditionally used as anti-alcoholic drugs. Treating DsgH1-Th17 cells with cyanamide or disulfiram inhibited cytokine production at lower doses and induced cell death at higher doses ( Figure 7 A). Subsequently, we treated EAD mice with cyanamide, either alone or in combination with abatacept ( Figure 7 A). We observed that mice treated with cyanamide showed a marked reduction in both IL-7R pos CTV dil and IL-7R neg CTV dil donor cells ( Figure 7 B). In contrast, abatacept alone selectively reduced the IL-7R neg CTV dil population (but not IL-7R pos CTV dil ) ( Figure 7 C). Notably, recipient mice treated with a combination of abatacept and cyanamide showed fewer CTV dil cells than those receiving single treatments, suggesting an additive effect. Finally, we confirmed the effects of the two drugs by examining actual cytokine expression in CTV dil cells. As presented in Figure 7 D, mice treated with abatacept or cyanamide alone showed a significant reduction in IL-17-producing cells, and those treated with the combination showed an additive effect. In contrast to IL-17, IFNγ production was almost completely inhibited by abatacept alone, whereas cyanamide alone or in combination with abatacept had minimal effects on IFNγ ( Figure 7 E). These findings suggested that the inhibition of ALDH and CD28 affects self-reactive pathogenic Th17 cells through distinct mechanisms. ALDH inhibition had inhibitory effects on the survival and IL-17 expression of IL-7R pos memory Th17 cells, whereas CD28 inhibition primarily affected the differentiation into effector Th17 cells. Importantly, the combination of both treatments had the most pronounced effect in reducing both memory and effector Th17 cell populations. ALDH2 correlates with IL-17 production in human cancer Lastly, we investigated whether ALDH2 expression is functionally correlated with Th17 activity in humans, particularly in a cancer context. Upon re-examining of The Cancer Genome Atlas data, we observed a weak but significant correlation between ALDH2 and the IL-17A gene in certain types of cancers ( Figure 8 ). Notably, IL-17A expression was evident in a limited fraction of patients, and it showed a significant correlation with high ALDH2 expression. Although these findings are preliminary, they suggest that ALDH2 may play a role in Th17 activity in humans.

Discussion We demonstrated that CD28 blockade selectively inhibits effector Th17 cells that are highly differentiated, leading to the complete inhibition of dermatitis. Traditionally, it has been thought that naive T cells that receive TCR signals without CD28 activation become anergic or unresponsive, thereby contributing to tolerance. 9 Given that the CD28/PI-3K/AKT axis is a hallmark of glycolysis, 12 it is plausible that abatacept inhibits aerobic glycolysis, which is essential for extensive proliferation and the expression of effector cytokines. However, we observed that abatacept did not inhibit the proliferation of memory-like T cells. This abatacept-resistant memory-phenotype Th17 population may explain the persistence of the disease, leading to recurrence during or after treatment. Recently, certain immunosuppressants, such as rapamycin, 33 MEK inhibitors, 34 and tyrosine kinase inhibitors, 35 have been found to induce long-term memory populations and sustain chronic immune responses. Therefore, in the context of autoimmunity, a potential drawback of CTLA-4 Ig is that although it may ameliorate inflammation by blocking CD28, it could also generate persistent, long-term memory Th17 cells by preventing exhaustion. Indeed, our data, along with previous research, indicate that memory-like Th17 cells can give rise to pathogenic effector cells (as shown in our data and by others 6 ). Regarding the mechanisms underlying the persistence of Th17 cells, Muranski et al. 8 have suggested the possibility of stemness, whereas Karmaus et al. 7 have proposed the existence of two metabolically distinct populations. Our data support the existence of two distinct Th17 populations that show different responses to CD28 blockade. Interestingly, in our study, the remaining memory-like Th17 cells expressed ALDH genes and can be targeted through systemic inhibition of ALDH. Therefore, ALDH not only controls stem cells but is also involved in the detoxification of endogenously produced aldehydes. Notably, the failure to detoxify endogenously produced aldehydes in patients with a combination of the ADH5 allele and ALDH2 causes Fanconi anemia, underscoring the critical role of ALDH in the hematopoietic system. 36 , 37 , 38 Aldehydes are known to inhibit T cells, as exemplified by excessive alcohol consumption negatively impacting follicular helper T cells and attenuating immune responses. 39 , 40 Aldh2 -deficient mice show impaired T cell responses, which are associated with altered metabolism. 31 Our data suggested that the inhibition of ALDH may lead to increased intracellular aldehyde concentrations in memory T cells, potentially affecting immune function. An important question arises: does the genetic diversity of ALDH genes influence T cell immune responses in humans? The most well-known single-nucleotide polymorphism (SNP) in the ALDH2 gene causes loss of function and is predominantly found in the East Asian population. 41 Despite the population with this SNP having low ethanol consumption, it is associated with cancer susceptibility and progression. This suggests that the detoxification of endogenous aldehydes by ALDH is not negligible in tumorigenesis and progression. Furthermore, our preliminary data demonstrated a positive correlation between ALDH2 and IL-17A expression in certain cancers, such as head and neck carcinoma, colon adenocarcinoma, and esophageal cancer. This finding suggested a potential contribution of ALDH activity to memory or effector Th17 responses, which may be beneficial in the context of cancer. In conclusion, we demonstrated a unique role of ALDH regulating Th17 cell responses. This systemic control of ALDH may hold promise for designing future treatment strategies for diseases involving T cell responses. Limitations of the study Firstly, we used irradiated wild-type mice as recipients rather than Rag2 knockout mice. Consequently, we did not address the potential contributions of recipient-derived lymphocytes in the establishment of EAD, including epitope spreading, autoantibody production, and T regulatory cell activity. Secondly, although we demonstrated the persistence of CTV dil IL-7R pos cells after abatacept treatment and their ability to produce IL-7R neg effector cells, the limited cell numbers prevented us from directly confirming whether this phenomenon contributed to disease recurrence after discontinuing treatment. Thirdly, the effects of abatacept and ALDH inhibitors on tissue resident memory T cells remain unclear and warrants further investigation. Fourthly, in terms of clinical relevance, comprehensive analyses examining whether ALDH expression is functionally correlated with Th17 cells in human autoimmune dermatitis or other autoimmune diseases are currently lacking. Future studies should address these points to provide a more comprehensive understanding of the topic.

Lead contact Summary IL-17-producing helper T (Th17) cells are long-lived and serve as central effector cells in chronic autoimmune diseases. The underlying mechanisms of Th17 persistence remain unclear. We demonstrated that abatacept, a CD28 antagonist, effectively prevented the development of skin disease in a Th17-dependent experimental autoimmune dermatitis model. Abatacept selectively inhibited the emergence of IL-7R-negative effector-phenotype T cells while allowing the survival and proliferation of IL-7R + memory-phenotype cells. The surviving IL-7R + Th17 cells expressed genes associated with alcohol/aldehyde detoxification and showed potential to transdifferentiate into IL-7R-negative effector cells. Inhibiting aldehyde dehydrogenase reduced IL-7R + Th17 cells in vivo , independently of CD28, and exhibited additive effects when combined with abatacept. Our findings suggest that CD28 blockade prevents inflammation without eliminating persistent memory cells. These remaining memory cells can be targeted by other drugs, such as aldehyde dehydrogenase inhibitors, to limit their survival, thereby facilitating the treatment of chronic autoimmune diseases. Graphical abstract Highlights • CD28 blockage by abatacept prevents dermatitis but does not eliminate memory Th17 cells • Pathogenic memory Th17 cells utilize aldehyde dehydrogenases for survival • Abatacept together with ALDH inhibitor reduce pathogenic memory Th17 cells in vivo Molecular biology; Immunity; Cell Subject areas Published: December 9, 2023

Supplemental information Acknowledgments The authors thank Akihiro Nohmi and Kazumi Yoshinaga for animal care and Yuko Tanishita and Kaori Yanai for secretarial assistance. We thank Dr. Jeffrey Bluestone (Sonoma Biotherapeutics, USA) for reagents and Dr. Masahiro Okada (RIKEN, Japan), Dr. Kenji Kabashima (Kyoto University, Japan), and Dr. Hiroyuki Yoshitomi (Kyoto University, Japan) for valuable discussion. JSPS KAKENHI (#16H06276: AdAMS) and (#JP20am0101001: BINDS) provided chemicals and mice. This work was supported by KAKENHI (#22K19446, #23H04789, 22H02852, #21H00439, #19K07488, and 19H05431 [to S.C.] and #21H05044 and #22K1944 [to A.Y.]), 10.13039/100009619 AMED ( 10.13039/501100003382 CREST JP22gm1110009, Moonshot JP22zf0127003 [to A.Y.]), 10.13039/100007449 Takeda Science Foundation , 10.13039/501100005926 KOSE Cosmetology Research Foundation (to S.C.), 10.13039/501100001697 Keio University Fukuzawa Fund (to S.C.), and 10.13039/501100001697 Keio University Academic Development Fund (to S.C.). Author contributions S.C. designed research; S.C., Y.T., and H.H. performed experiments; S.C., Y.T., T.S., and K.H. analyzed data; H.T., M.H-C., and M.A. provided necessary materials; S.C., K.H., H.T., and A.Y. wrote the paper. Declaration of interests The authors declare no competing interests. Inclusion and diversity We support inclusive, diverse, and equitable conduct of research.

CC BY

no

2024-01-16 23:40:19

iScience. 2023 Dec 9; 27(1):108646

oa_package/37/d3/PMC10788227.tar.gz

PMC10788234

0

Abstract Introduction Several studies mentioned parenchymal findings after SARS‐CoV‐2 pneumonia, but few studies have mentioned alterations in the airways. The aim of this study was to estimate the prevalence of tracheomalacia and to analyse the clinical characteristics in a cohort of patients with SARS‐CoV‐2. Methods The study population consisted of all patients with SARS‐CoV‐2 admitted a hospital serving a population of 500 000 inhabitants. Patients were visited between 2 and 6 months after hospital discharge. In this visit, all patients were subjected to an exhaustive clinical questionnaire and underwent clinical examination, pulmonary function tests and chest CT. Results From February 2020 to August 2021, 1920 patients were included in the cohort and tracheomalacia was observed in 15 (0.8%) on expiratory HRCT imaging. All patients with tracheomalacia also presented ground glass opacities in the CT scan and 12 patients had airway sequelae. Conclusions Tracheomalacia is an exceptional sequela of SARS‐CoV‐2 survivors. Espejo D , Zapata M , Omari S , Muñoz X , Cruz M‐J , Se‐COVID‐19 team . Acquired tracheomalacia due to SARS‐CoV‐2 pneumonia . Clin Respir J . 2024 ; 18 ( 1 ): e13719 . doi: 10.1111/crj.13719 .

To the Editor: Several studies have found that SARS‐CoV‐2 pneumonia has many respiratory consequences, including clinical, functional and radiological sequelae. 1 Radiologically, numerous parenchymal findings such as ground‐glass opacities, consolidations, reticulation patterns or pulmonary fibrosis have been recorded. 2 Nevertheless, few studies have mentioned alterations in the airways: among the alterations reported, bronchiectasis is the most frequent. 3 , 4 Tracheomalacia is defined as a diffuse or segmental weakness of the trachea due to the loss of structural integrity of the cartilage. Clinical presentation of tracheomalacia is nonspecific and its most frequent symptoms are dyspnoea and cough. Classically, it has been defined as primary (congenital) or secondary (acquired). Airway inflammation is a secondary cause of acquired forms of tracheomalacia including airway infections, airway or lung malignancy, inhalation of chemical irritants and prolonged intubation or tracheostomy with mechanical ventilation. 5 , 6 , 7 The aim of this study was to estimate the prevalence of tracheomalacia and to analyse the clinical characteristics in a cohort including all patients with SARS‐CoV‐2 pneumonia confirmed by PCR admitted to a reference centre in a hospital serving a population of 500 000 inhabitants at the beginning of the pandemic. To identify the respiratory sequelae, all patients were administered an exhaustive clinical questionnaire and underwent clinical examination, pulmonary function tests, and chest CT between 2 and 6 months after hospital discharge. From February 2020 to August 2021, 1920 patients were included in the cohort. Tracheomalacia was observed in 15 (0.8%) on expiratory HRCT imaging. The median age of these patients was 60 years, and their baseline characteristics are summarized in Table 1 . The treatment for SARS‐CoV‐2 infection in these patients was modified in the light of the new information obtained during the different waves of the pandemic. Several treatment protocols were applied including antiviral therapies as well as corticoids and empirical antibiotherapy with cephalosporins and quinolones. In this group, 10 patients received hydroxychloroquine, nine lopinavir/ritonavir, two tocilizumab, five dexamethasone and 13 azithromycin (some patients received several treatments at the same time). Regarding the symptoms identified on clinical examination, only three (20%) patients were asymptomatic; 10 (67%) had dyspnoea and two (13%) had cough. Table 1 also shows the findings reported at the follow‐up visit for pulmonary respiratory function and chest CT. In 14 patients, the tracheomalacia was diffuse, affecting the entire trachea. Of these, in six cases, there was clear associated bronchomalacia. In one patient, the malacia was segmental at the level of the upper third of the trachea. No tracheal stenosis was observed. In nine patients, tracheomalacia affected both at the membranous and cartilaginous level, and in six cases, the membranous part was only affected. 8 Ground glass opacities in the CT scan were found in all patients with tracheomalacia. Six patients (40%) also presented other parenchymatous alterations such as septal thickening and reticulation. Twelve patients also had airway sequelae such as bronchiectasis, bronchiolitis or airway bronchial thickening. These results suggest that the probability of presenting tracheomalacia due to SARS CoV2 pneumonia is low. In our opinion, there are three possible explanations for the appearance of tracheomalacia in the SARS CoV‐2 survivor population. The first one is orotracheal intubation. Although the prevalence of primary or secondary tracheomalacia is unknown in the general population, its most common acquired cause is orotracheal intubation or tracheostomy. In these cases, tracheomalacia is related to increases in the respiratory airway pressure, oxygen toxicity and recurrent infections. 6 Recently, Guarnieri et al. 9 reported tracheomalacia in 8 of 151 patients with SARS‐CoV‐2 acute respiratory distress syndrome who required intubation or tracheostomy and mechanical ventilation. In the present series, however, intubation was only required in five patients. Secondly, respiratory infections are a well‐known cause of tracheomalacia and were present in 67% of our patients. It should be highlighted that chronic infections are more strongly associated with tracheomalacia than acute presentations. However, in our experience, tracheomalacia is not an isolated finding, because all patients had parenchymal alterations and 80% had other airway sequelae, as shown in Table 1 . Therefore, high levels of inflammation, such as those related with SARS CoV‐2 infection, may be responsible for these alterations. In fact, Borczuk et al. found airway inflammation in the form of chronic diffuse inflammation in 41% of cases in a series of 68 necropsies in the initial stages of the pandemic. 10 The third possible explanation for the tracheomalacia in these patients is the presence of previous respiratory diseases. Five patients were affected by asthma, COPD or OSA. Although seldom described, some authors estimate the incidence of tracheomalacia in adult patients with respiratory airway disease to be approximately 12.6%. 6 The present study has some limitations. Probably, bronchoscopy should be used as a gold‐standard diagnostic tool for this disease, 5 but CT scan was used in the context of the pandemic, and also in severe stages of the disease in which the use of more invasive techniques was not justified. In fact, emerging evidence in paediatric populations suggests that CT scan can effectively diagnose tracheomalacia and should be considered as a less invasive alternative to bronchoscopy. 11 Another limitation is that tracheomalacia may also have existed before the SARS‐CoV‐2 infection, especially since most series established the mean age of presentation at 40 years old. 6 However, in the present series of patients, a guided clinical interview did not reveal any respiratory symptoms previous to the SARS CoV‐2 infection that might have been related to a possible tracheomalacia or any predisposing factor that might explain its presence. Finally, we cannot rule out the development of cicatricial stenosis in the long term. In conclusion, our results indicate that tracheomalacia is an exceptional sequela of SARS‐CoV‐2 survivors, and is always associated with parenchymal and other airway findings. Nevertheless, the early detection of this condition is crucial in clinical practice. Tracheomalacia causes respiratory symptoms that may have a major impact on patients' lives. It is also a well‐known predisposing factor for repetitive respiratory infections and may entail severe future comorbidity. AUTHOR CONTRIBUTIONS The authors of the manuscript have all contributed significantly to the research, preparation, revision and final production of the manuscript. CONFLICT OF INTEREST STATEMENT The authors have no competing interests to declare in relation to this study. ETHICS STATEMENT The study was approved by the local Ethics Committee, and all the subjects included gave informed consent prior to participation (PR [AG]222/2020).

ACKNOWLEDGEMENTS This project was supported by the Fundació Catalana de Pneumologia (FUCAP), Instituto de Salud Carlos III (PI21/01046) and European Regional Development Fund (FEDER). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. DATA AVAILABILITY STATEMENT The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.

CC BY

no

2024-01-16 23:40:19

Clin Respir J. 2023 Dec 20; 18(1):e13719

oa_package/93/b2/PMC10788234.tar.gz

PMC10788235

38226094

Introduction Stercoral colitis is characterized by inflammation and ulceration of the colonic wall due to fecal impaction [ 1 ]. Patients with stercoral colitis often present with abdominal pain, distension, constipation, and sometimes rectal bleeding. These symptoms are nonspecific and can mimic other colonic diseases, necessitating a high index of suspicion for diagnosis. The pathogenesis involves increased intraluminal pressure due to fecal impaction, leading to ischemia and ulceration of the colonic wall. Stercoral perforation occurs when pressure necrosis results from the fecaloma [ 2 ]. The sigmoid colon is most commonly affected due to its narrow lumen and vulnerability to pressure-induced ischemia. Microscopic changes are observed in stercoral colitis, including transmural inflammation and necrosis. Risk factors for stercoral colitis include any condition that cause fecal impaction, including chronic constipation, neurogenic bowel, psychiatric illnesses, and opiate use. Additionally, advanced age, immobility, and dehydration are also risk factors [ 3 ]. Treatment focuses on relieving fecal impaction and managing colonic inflammation. A combination of disimpaction, laxatives, enemas, bowel rest, intravenous hydration, and electrolyte correction is typical for non-perforated cases. Surgical intervention is considered in cases of perforation or failure of conservative management. Complications include colonic perforation, sepsis, and shock. Delayed treatment significantly increases the risk of complications, including fecal peritonitis and abscess formation. The prognosis depends on early recognition and treatment. Timely management generally results in favorable outcomes, but delayed treatment can lead to high mortality, especially in cases with complications.

Discussion A retrospective review of 49 patients noted that the rectosigmoid colon was the most frequently involved segment in stercoral colitis. CT findings typical of stercoral colitis include dilatation >6 cm, wall thickening >3 mm of the affected colon segment, pericolonic fat stranding, mucosal discontinuity, and presence of free air, free fluid, and pericolonic abscess. The sign most related with mortality was the length of the affected colon segment >40 cm [ 4 , 5 ]. A cohort study of 452 mild and 93 moderate-severe cases confirms that the sigmoid is the most frequently involved segment. Factors associated with perforation included slightly increased wall thickness (6.4 vs. 5.7 mm, p=0.03), opiate use (50 vs. 23%, p=0.04), and disease-specific mortality (11 vs. 0%, p=0.04) [ 6 ]. Mortality in this cohort was 11%. Management of stercoral colitis includes disimpaction (either manual or endoscopic), laxatives, and enemas. Mortality from stercoral colitis has been reported to be between 17% and 30% [ 6 - 10 ], and this is driven by cases with perforation, which is the single biggest risk factor for mortality. Once perforation occurs, surgical intervention is necessary to resect the segment of the colon involved. Most often, this involves resection of the sigmoid colon followed by an ileostomy. Stercoral colitis is actually not an uncommon ED presentation. A recent retrospective cohort of 269 patients spanning three years from a regional health system found that while the most common chief complaint was abdominal pain (present in more than one third of the cohort), abdominal pain was actually documented as absent in two thirds of cases [ 10 ]. In these cases, if a CT were not done (due to lack of abdominal pain), the diagnosis of stercoral colitis would have been missed. Interestingly, 10% of patients discharged home returned to the ED within 72 hours. The authors note that in general, patients were not given specific instructions on how to manage the root cause, their constipation. This is an important point, as we will likely see more stercoral colitis given the increasing numbers of at-risk populations including the elderly, the infirm, and those on opiates.

Conclusions Stercoral colitis is a potentially fatal condition that requires prompt recognition and management. Opiate use is a significant risk factor that has become more prevalent. Early intervention is crucial to improve patient outcomes and reduce the risk of life-threatening complications.

The authors present the case of a 62-year-old woman who had stercoral colitis secondary to opiate use for rheumatoid arthritis leading to chronic constipation. Computed tomography imaging demonstrated stool along a significant length of the colon. Stercoral colitis is a seldom suspected cause of severe abdominal pain. Although constipation may seem benign, when it gets to the level of a stercoral colitis, mortality due to colonic perforation is a very real concern. The authors review the presentation, risk factors, and management of stercoral colitis.

Case presentation A 62-year-old female with a past medical history of rheumatoid arthritis, hemorrhoids, and chronic constipation secondary to opiate pain medication use presented to the emergency department (ED) for constipation. She stated her last bowel movement was three days prior; however, her stool was hard, and she felt like she was incompletely evacuating. She explained while she had the urge to go, she was unable to have a bowel movement. She reported taking oxycodone daily for pain secondary to her rheumatoid arthritis. Her other medications were methotrexate and loratadine. She also reported taking bisacodyl laxative as needed, and her last dose was three days ago. The patient endorsed associated lumbar back pain and nausea; however, she denied fevers, chills, chest pain, shortness of breath, vomiting, abdominal pain, rectal pain, blood in stool, or dysuria. Her vital signs were as follows: blood pressure 171/94 mmHg, pulse 99 beats per minute, pulse oximetry 99% on room air, temperature 97.9°F, and respirations 18 breaths per minute. Her laboratory tests are summarized in Table 1 and were essentially unremarkable. Urinalysis revealed normal result. Computed tomography (CT) scan with contrast of the abdomen and pelvis revealed large rectal stool burden with mural thickening and perirectal inflammatory changes, consistent with stercoral colitis (Figure 1 ). Also noted is air in the rectal wall (pneumatosis). Due to the extent of the obstruction, the patient was admitted to the hospital. She was initially made nil per os (NPO), hydrated with intravenous saline. The patient also received an enema and underwent manual disimpaction, followed by polyethylene glycol administration to assist with fecal evacuation. The patient did have bowel movements after this regimen. The patient was monitored for signs of sepsis, which she did not develop. A pain management consult was obtained to wean the patient off of daily opiates. The patient was discharged on hospital day 4, without any complications.

CC BY

no

2024-01-16 23:40:19

Cureus.; 15(12):e50511

oa_package/7c/5e/PMC10788235.tar.gz

PMC10788236

38226084

Introduction Rheumatoid arthritis (RA) is a systemic autoimmune disease that mainly affects the joints, which can lead to joint deformity. Since the disease is systemic, it affects many organs, including the heart, which can lead to coronary artery disease, pericarditis, and heart failure [ 1 ]. The most common cardiac complication from RA is pericarditis, which accounts for almost 40% of cases and leads to fluid collection in the pericardium [ 1 , 2 ]. Most patients are asymptomatic; at most, 10% have the symptoms of pericarditis [ 3 ]. Additionally, if the anti-cyclic citrullinated peptide titer strands are elevated, the patient is at risk of developing pericardial effusion [ 4 ]. Sometimes, a patient can mask pericardial effusion symptoms when they are physically fit [ 4 ]. Pericarditis was associated with disease activity in RA [ 5 ]. In the United States, pericardial effusion affects nearly 6.5% of the general population. It can be present secondary to other diseases, such as infection, trauma, RA, systemic lupus erythematosus, and malignancy [ 6 ]. In this case, the patient had pericarditis and pericardial effusion as the first presentation of RA, which is a rare condition.

Discussion Pericarditis is a well-known extra-articular manifestation of RA. However, it is unusual for pericarditis to present in undiagnosed RA and as the first presentation of the disease. Few reported cases are similar to this patient’s. RA can cause a variety of cardiac manifestations, although pericarditis is the most common [ 1 ]. Even if the patient does not complain of typical joint pain, RA cannot be ruled out [ 6 ]. Moreover, other causes, including malignancy, infectious causes, and hypothyroidism, must be ruled out [ 3 ]. Aspirin and NSAIDs are the cornerstones of treating pericarditis [ 2 ]. After this patient began NSAIDs and colchicine, he showed significant improvement. Conversely, another study has shown that, after starting the patient on diclofenac (75 mg/day) and colchicine (1.2 mg/day), the patient did not improve and developed cardiac tamponade [ 5 ]. It is essential to start the patient with the appropriate dose of NSAIDs and colchicine [ 2 ]. Pericardiocentesis is part of the diagnostic workup [ 6 ], which is unnecessary if we have an apparent reason for the pericardial effusion. It is essential to do this when the patient has developed cardiac compression symptoms, such as lower limb pitting edema, elevated jugular vein pressure [ 7 ], or developed cardiac tamponade [ 5 ].

Conclusions We reported an unusual presentation of RA in which the patient presented with shortness of breath and chest pain. Extensive workup was done for the patient, and he was diagnosed with RA complicated with pericardial effusion. The patient did not have typical joint pain from RA. The patient underwent pericardiocentesis in another hospital, which failed. Then, the patient was treated medically with NSAIDs and colchicine and showed significant improvement. This case highlights the importance of recognizing pericardial effusion as a potential sign of undiagnosed RA and the need for further investigation.

Rheumatoid arthritis (RA) is a systemic autoimmune disease that mainly affects the joints, which can lead to joint deformity. Since the disease is systemic, it affects many organs, including the heart, which can lead to pericarditis, coronary artery disease, and heart failure. We are reporting on a male patient, 34 years of age and Sudanese, who complained of shortness of breath and chest pain that started weeks before he came to the hospital, with no other associated symptoms. The patient was admitted to the hospital, and extensive work was done for the patient, which revealed that he had pericardial effusion secondary to RA, which is the first presentation of the disease. RA rarely presents as a first presentation with pericarditis and pericardial effusion. The patient was managed medically, and he showed significant improvement.

Case presentation The 34-year-old Sudanese male, who was not known to have a chronic medical history, was referred from another hospital as a case of pericardial effusion and failed pericardiocentesis without reaching any diagnosis. The patient complained of shortness of breath and chest pain that started three weeks before his first admission, with no aggravating or relieving factors. In addition, the patient reported new mild left knee pain with no morning stiffness while admitted to the hospital. The patient denied any history of fever, cough, syncope, nausea, vomiting, diarrhea, skin rash, recent surgery, and recent travel. Upon presentation, the patient had a temperature of 36.8 °C, a heart rate of 107 beats per minute, a respiratory rate of 20 breaths per minute, a blood pressure of 132/88 mmHg, and an oxygen saturation of 98% in room air. He was conscious, oriented, and alert with a Glasgow coma scale of 15/15; he was not in distress, and there was no cyanosis or clubbing. His chest examination was clear to auscultation and percussion, and a CVS examination revealed normal heart sounds, presenting S1 and S2 with a normal rate and regular rhythm. An abdominal examination revealed a soft and lax abdomen with no tenderness or organomegaly. He had mild left knee swelling with no lower limb edema, redness, or hotness. The laboratory work (Table 1 ) also revealed that his kidney function, liver function, and enzymes were normal. Electrocardiography showed sinus tachycardia, and the chest X-ray was unremarkable. A computed tomography pulmonary angiogram was done, and pulmonary embolism was ruled out; an echocardiogram showed a large, highly trabeculated pericardial effusion (Figures 1 - 2 ). The patient was diagnosed with RA complicated with pericardial effusion. The cardiology and rheumatology teams met and decided to treat the pericardial effusion medically without the need for an intervention. The patient began taking colchicine 0.5 mg twice a day and ibuprofen 400 mg three times a day; the patient showed improved symptoms and was discharged after four days of admission. In the outpatient department, he followed up with the Cardiology Department, and another echocardiography was done after 10 days, which showed a significant regression of the effusion. The patient also followed up with rheumatology and began taking hydroxychloroquine 400 mg daily.

CC BY

no

2024-01-16 23:40:19

Cureus.; 15(12):e50489

oa_package/dc/f4/PMC10788236.tar.gz

PMC10788237

38226130

Introduction and background Avascular necrosis (AVN), also known as osteonecrosis, is a debilitating condition characterized by the death of bone tissue due to a lack of blood supply. This phenomenon predominantly affects weight-bearing joints, such as the hip and knee, and can lead to significant pain, joint dysfunction, and, if left untreated, irreversible damage. This review aims to comprehensively explore treatment strategies specifically focused on the early stages of AVN [ 1 ]. AVN involves the compromised blood supply to bone tissue, resulting in ischemia and subsequent cellular death. Typically occurring in joints, this condition manifests as the gradual deterioration of bone structure, ultimately affecting joint function. Understanding the underlying pathophysiology is crucial for developing effective treatment strategies and interventions that can halt or slow the progression of AVN [ 2 ]. Early detection of avascular necrosis is paramount for achieving favourable treatment outcomes. The insidious nature of AVN often means that symptoms may not become apparent until the disease has advanced, emphasizing the need for vigilant monitoring and diagnostic measures. Timely intervention during the initial stages offers the best chance to preserve joint integrity, minimize pain, and prevent the need for more invasive procedures like joint replacement [ 3 ]. This review will delve into various aspects of early avascular necrosis, including its pathophysiology, aetiology, and risk factors. The clinical presentation of AVN and diagnostic imaging techniques will be explored to provide a comprehensive understanding of how healthcare professionals can identify the condition in its nascent stages. Moreover, the review will thoroughly examine conservative and surgical treatment approaches, encompassing pharmacological interventions, physical therapy, and advanced surgical techniques. Emerging therapies, rehabilitation protocols, and long-term management strategies will also be discussed.

Conclusions In conclusion, AVN presents a multifaceted challenge in orthopaedic and rheumatologic practice. This comprehensive review has highlighted vital findings, emphasizing the importance of early detection and a multidisciplinary approach to care. From recognizing aetiological factors and clinical presentations to exploring conservative and surgical interventions, the varied nature of AVN requires individualized treatment plans. Emerging therapies, including regenerative medicine and stem cell therapy, offer promising avenues for future exploration alongside ongoing research into genetic and molecular pathways. The implications for clinical practice stress the need for personalized care, staying informed about advancements, and incorporating patient education. Recommendations for future management strategies underscore the importance of advancing imaging technology, integrating regenerative therapies, and empowering patients in their healthcare journey. As we navigate the complexities of AVN, translating research findings into clinical practice holds the key to improving outcomes and enhancing the overall quality of life for those affected by this challenging condition.

Avascular necrosis (AVN), characterised by compromised blood supply leading to bone necrosis, poses a significant challenge in orthopaedic and rheumatologic practice. This review comprehensively examines early AVN treatment strategies, including aetiology and risk factors, clinical presentation, conservative and surgical approaches, emerging therapies, and rehabilitation. Key findings underscore the importance of early detection, personalised treatment plans, and a multidisciplinary approach involving orthopaedic specialists, rheumatologists, and physical therapists. The implications for clinical practice emphasise individualised care, staying abreast of emerging therapies, and patient education. Recommendations for future management strategies highlight the need for imaging technology advancements, regenerative therapies integration, and ongoing research into genetic and molecular pathways. As the field continues to evolve, translating research findings into clinical practice holds promise for improving outcomes and enhancing the overall quality of life for individuals affected by AVN.

Review Aetiology and risk factors Traumatic Causes Fractures and dislocations: Traumatic fractures and dislocations pose substantial risks for the development of AVN. These events have the potential to disrupt the blood vessels supplying the affected bone, resulting in ischemia, a condition where blood flow to the bone is compromised. Notably, specific instances such as fractures of the femoral neck or traumatic hip dislocations are well-documented as high-risk scenarios for subsequent AVN. The severity of the injury and the extent of vascular damage play pivotal roles in influencing the likelihood of AVN. For instance, fractures that involve significant displacement or dislocations causing vascular compromise increase the risk substantially. Timely and appropriate management of these fractures is crucial for minimizing the risk of avascular events. Swift interventions, including surgical stabilization or reduction, are paramount in restoring proper alignment and blood flow to the injured area, thereby mitigating the potential for AVN development. To illustrate, studies have shown that prompt reduction and fixation of a displaced femoral neck fracture significantly reduce the incidence of AVN. Additionally, rehabilitation strategies emphasizing early mobilization and joint protection contribute significantly to a comprehensive approach aimed at preventing the progression of AVN [ 3 ]. Joint trauma: Injuries directly affecting the joints, such as severe contusions or crush injuries, can have profound implications for the circulation within the joint structures. The vulnerability of joints to avascular changes underscores the critical importance of vigilant monitoring and early intervention following joint trauma. Joint trauma can lead to microvascular damage, compromising blood supply and initiating the cascade of events that may culminate in AVN. Swift recognition of joint trauma, often facilitated by advanced imaging techniques, enables healthcare professionals to assess the extent of injury and determine appropriate interventions. Early measures may include joint aspiration to relieve intra-articular pressure, immobilization to minimize stress on the injured joint, and pharmacological interventions to manage inflammation and pain. The collaborative efforts of orthopaedic specialists and rehabilitation professionals are essential in orchestrating a timely and tailored approach to mitigate the risk of AVN development in the aftermath of joint trauma [ 4 ]. Non-Traumatic Causes Steroid use: Prolonged or high-dose steroid therapy stands as a well-established non-traumatic cause of AVN. Steroids exert their influence by inducing lipid microembolism and intravascular coagulation, which collectively contribute to compromised blood flow to the bone. The risk of AVN is particularly heightened in individuals undergoing extended courses of steroid treatment, such as those with autoimmune disorders or organ transplants. Clinicians must be vigilant in understanding the risks associated with steroid use, recognizing that specific patient populations may be predisposed to AVN. Monitoring patients on long-term steroid therapy, implementing dose reduction strategies when feasible, and considering alternative treatments are essential aspects of mitigating the risk of AVN in this context [ 5 ]. Alcohol abuse: Chronic alcohol consumption has been identified as a significant risk factor for AVN, mainly affecting weight-bearing joints such as the hip. The mechanisms through which alcohol contributes to AVN are multifaceted, involving vascular changes, alterations in lipid metabolism, and impaired bone remodelling. The impact of alcohol on blood vessels can lead to decreased blood supply to the bones, initiating a cascade of events culminating in AVN. Clinicians should actively inquire about and consider the history of alcohol abuse in patients presenting with AVN, particularly in cases involving the hip joint. This awareness allows for tailored interventions, including lifestyle modifications and substance abuse counselling, to address the underlying risk factor and potentially halt the progression of AVN [ 6 ]. Coagulation disorders: Conditions affecting blood coagulation, such as thrombophilia and hypercoagulable states, significantly elevate the risk of AVN. In individuals with coagulation disorders, an imbalance in blood coagulation factors can lead to microvascular thrombosis, compromising the blood supply to the bone. Thrombotic events within the small vessels supplying bone tissue contribute to the pathogenesis of AVN. Clinicians should maintain a heightened awareness of coagulation disorders when assessing patients at risk for or presenting with AVN. This involves thoroughly investigating the patient's medical history, family history, and appropriate laboratory testing to identify coagulation abnormalities. Timely recognition and management of coagulation disorders are crucial in reducing the risk and severity of AVN in affected individuals [ 7 ]. Common Risk Factors Age and gender: A prominent demographic factor influencing the manifestation of AVN is age, with a typical onset occurring in individuals aged 30-50. Additionally, there is a noteworthy preference for males in the incidence of AVN. Hormonal fluctuations and age-related changes in bone metabolism during this specific life stage may contribute to the heightened susceptibility observed in males and individuals within this age range. Understanding the intersection of hormonal influences and age-related factors is crucial in elucidating the mechanisms that render this demographic group more prone to AVN. Clinicians should consider these demographic characteristics when assessing individuals for potential risk factors and early signs of AVN, enabling proactive interventions and personalized management strategies [ 8 ]. Joint loading and weight-bearing: The mechanical stresses associated with weight-bearing joints, notably the hip and knee, play a pivotal role in developing AVN. Daily activities and excessive joint loading increase the risk of compromised blood supply to these weight-bearing joints. Obesity further exacerbates this risk, emphasizing the critical need for weight management in individuals susceptible to AVN. Excessive body weight places additional strain on weight-bearing joints, creating an environment conducive to the development or progression of AVN. Clinicians should prioritize weight management strategies in susceptible individuals, incorporating lifestyle modifications and personalized interventions to alleviate mechanical joint stress. This multifaceted approach addresses both the mechanical and metabolic factors contributing to AVN, promoting joint health and mitigating the risk of its occurrence [ 9 ]. Relationship with Systemic Conditions Autoimmune disorders: Individuals with autoimmune disorders, notably systemic lupus erythematosus (SLE) and rheumatoid arthritis (RA), face an elevated risk of developing AVN. The complex interplay of inflammatory processes underscores the intricate relationship between autoimmune-mediated vascular changes and AVN. In autoimmune conditions, the immune system's aberrant response can lead to vascular compromise, diminishing blood supply to the affected bone. Furthermore, the use of immunosuppressive medications, often employed in the management of autoimmune disorders, contributes to the multifactorial landscape of AVN development. Clinicians should remain vigilant in recognizing and managing AVN in individuals with autoimmune disorders, emphasizing regular monitoring, early detection of symptoms, and a collaborative approach between rheumatologists and orthopaedic specialists to optimize patient outcomes [ 10 ]. Hematologic disorders: Hematologic disorders, including sickle cell disease and other conditions affecting blood composition, pose a significant risk for developing AVN. These disorders can lead to vascular occlusion and ischemia, creating a milieu conducive to AVN. Individuals with sickle cell disease, in particular, experience a heightened susceptibility due to the characteristic sickling of RBCs, leading to impaired blood flow. Regular monitoring and early intervention are paramount in managing AVN in patients with underlying hematologic disorders. Clinicians must adopt a proactive approach, conducting thorough assessments and implementing preventative measures to minimize the risk of AVN in individuals with hematologic conditions. This includes collaborative care involving haematologists, orthopaedic specialists, and other relevant healthcare professionals to provide comprehensive and tailored management strategies [ 11 ]. Clinical presentation Signs and Symptoms Joint pain: The primary and hallmark symptom of AVN is progressive joint pain, typically localized to the affected area. Initially, the pain may manifest as mild and intermittent, often exacerbated by weight-bearing activities. However, as AVN advances, the pain tends to worsen over time, becoming more constant and severe. This pain is indicative of the compromised blood supply and subsequent necrosis of the bone, triggering inflammatory responses and nerve stimulation. Understanding the evolving nature of joint pain in AVN is essential for early detection and timely intervention to alleviate symptoms and prevent further joint damage [ 12 ]. Limited range of motion: As AVN progresses, individuals commonly experience a decreased range of motion in the affected joint. This limitation arises from the compromised structural integrity of the bone and surrounding tissues, leading to stiffness and reduced flexibility. The restriction in joint motion can significantly impact daily activities, affecting mobility and overall quality of life. Monitoring changes in the range of motion is a critical clinical indicator of disease progression and guides therapeutic interventions to preserve joint function [ 13 ]. Joint stiffness: Stiffness, particularly noticeable after periods of inactivity, is a prevalent complaint in individuals with AVN. Patients often report increased difficulty initiating movement, and morning stiffness is characteristic. The stiffness results from the inflammatory processes associated with AVN and the compromised lubrication within the joint. Recognizing and addressing joint stiffness early during AVN is crucial for implementing targeted interventions, including physical therapy and joint-preserving strategies [ 14 ]. Muscle atrophy: Progressive AVN can lead to muscle atrophy around the affected joint. The combination of pain and reduced use of the compromised joint contributes to muscle wasting. Muscle atrophy not only exacerbates weakness but also further compromises joint function. Rehabilitation strategies focusing on muscle strengthening and targeted exercises become integral components of AVN management to counteract atrophy and enhance overall joint stability [ 15 ]. Joint instability: In advanced cases, AVN may progress to joint instability, characterized by a sense of the joint giving way or feeling unstable, particularly during weight-bearing activities. Joint instability poses a significant challenge, affecting balance and increasing the risk of falls and injury. Early recognition of signs of joint instability prompts appropriate interventions, including orthopaedic evaluation and consideration of surgical options, to address the underlying structural issues and restore joint stability. This aspect highlights AVN's evolving and dynamic nature, emphasizing the importance of a comprehensive symptom assessment and management approach [ 16 ]. Diagnostic Imaging Techniques X-rays: X-rays serve as the initial and fundamental imaging modality in diagnosing suspected cases of AVN. This non-invasive technique is pivotal for assessing bone structure and detecting advanced stages of AVN. Characteristic findings on X-rays include joint space narrowing, indicative of degenerative changes, subchondral sclerosis marked by increased bone density, and the formation of crescent-shaped areas of necrotic bone known as crescent signs. X-rays play a crucial role in providing an overview of the skeletal anatomy, aiding in the identification and staging of AVN. While particularly valuable in advanced cases, X-rays may not capture early-stage changes, prompting the use of more sensitive imaging modalities for timely diagnosis [ 17 ]. MRI: MRI stands out as a susceptible and specific imaging technique, particularly adept at detecting the early stages of AVN. MRI provides detailed images of soft tissues, offering a comprehensive view of changes in bone marrow and the early signs of necrosis. This modality is particularly valuable in cases where early intervention is crucial, allowing clinicians to visualize the extent of involvement and make informed treatment decisions. MRI's superior soft tissue contrast enhances its ability to identify subtle changes in bone structure, making it an essential tool for the early diagnosis and ongoing monitoring of AVN [ 18 ]. CT scans: CT scans complement the diagnostic armamentarium for AVN by offering a three-dimensional visualization of the affected joint's structure. While not as sensitive as MRI in detecting early-stage changes, CT scans provide valuable information, especially when precise anatomical details are essential. CT scans benefit surgical planning for joint interventions, offering detailed insights into the spatial relationships between bones and facilitating targeted procedures. The ability of CT scans to capture fine anatomical details makes them a valuable adjunct to X-rays and MRI in the comprehensive assessment and staging of AVN, ensuring a holistic understanding of the condition for optimal clinical management [ 19 ]. Staging of AVN Importance of Staging Treatment planning: Staging is pivotal in formulating effective treatment plans for individuals with AVN. By categorizing the condition into distinct stages, clinicians can tailor interventions to the specific needs and severity of the disease. In the early stages, conservative measures such as activity modification and pharmacological treatments may be sufficient to alleviate symptoms and slow disease progression. As AVN advances, staging guides clinicians in considering more aggressive interventions, such as core decompression or joint replacement surgery. This staged approach ensures that treatments are appropriately matched to the evolving nature of the condition, optimizing outcomes and preserving joint function [ 19 ]. Prognostication: The stage of AVN serves as a crucial factor in prognostication, offering valuable insights into the likely course and outcomes of the condition. Early-stage AVN may have a more favourable prognosis when conservative measures are implemented promptly. In contrast, advanced stages may necessitate more intensive interventions and carry different prognostic considerations. Prognostic information derived from staging enables clinicians to communicate effectively with patients and caregivers, setting realistic expectations regarding the potential outcomes of treatment. This informed approach enhances patient understanding and engagement in the decision-making process [ 20 ]. Monitoring disease progression: Staging facilitates systematic and longitudinal monitoring of disease progression in individuals with AVN. Periodic imaging assessments, guided by the staged classification, allow clinicians to track changes in the affected joint over time. This monitoring is particularly critical in managing chronic conditions like AVN, where disease progression may be gradual. Adjustments to treatment plans can be made based on observed changes, ensuring that interventions remain aligned with the evolving nature of the disease. Regular monitoring enhances the ability to detect subtle alterations in joint structure and function, enabling timely interventions to mitigate disease progression [ 21 ]. Research and clinical trials: Staging provides a standardized framework invaluable in research and clinical trials focused on AVN. Consistent staging criteria allow for accurately categorizing patients into specific disease stages, facilitating comparisons of treatment outcomes across diverse patient populations. The reliability and generalizability of research findings are enhanced through standardized staging, contributing to the development of evidence-based guidelines for AVN management. The systematic approach provided by staging criteria ensures that research outcomes apply to a broader patient demographic, fostering advancements in treatment strategies and contributing to the collective knowledge in the field [ 22 ]. Commonly Used Staging Systems Several staging systems have been developed to classify AVN based on clinical and radiographic criteria. The choice of staging system may depend on the affected joint and the preferences of the treating physician. Some commonly used staging systems are described in Table 1 . Conservative treatment approaches Rest and Activity Modification Protected weight-bearing: Implementing a strategy of protected weight-bearing is a fundamental component of the conservative treatment approach for individuals with AVN. This intervention aims to reduce mechanical stress on the affected joint, providing an environment conducive to healing compromised bone tissue. Restricting weight-bearing activities is achieved by advising individuals to minimize the use of the affected joint for bearing their body weight. Assistive devices such as crutches or walkers may be recommended to further alleviate the load on the compromised joint during activities like walking. This protective measure not only aids in pain management but also plays a crucial role in preventing further damage to the bone and promoting the potential for natural healing [ 26 ]. Activity modification: Activity modification is a crucial aspect of managing AVN, involving adjustments to daily activities to minimize strain on the affected joint. Individuals are guided to avoid activities that impose excessive stress on the compromised joint, such as prolonged standing or sitting, which could exacerbate symptoms and hinder healing. Moreover, limiting activities that involve repetitive joint movements is emphasized to prevent additional wear and tear on the affected area. Occupational and lifestyle adjustments are often tailored to the individual's specific condition, considering the location and severity of AVN. This personalized approach ensures that the activity modification plan aligns with the individual's daily routines and contributes to mitigating joint stress, managing symptoms, and supporting the healing process. The collaborative efforts of healthcare professionals, including orthopaedic specialists and physical therapists, are integral in guiding individuals through these modifications and optimizing their effectiveness for long-term joint health [ 9 ]. Pharmacological Interventions Nonsteroidal anti-inflammatory drugs (NSAIDs): NSAIDs, including commonly used medications such as ibuprofen and naproxen, play a crucial role in the pharmacological management of AVN. These medications are employed to alleviate pain and reduce inflammation associated with the condition, providing symptomatic relief to individuals affected by AVN. By inhibiting the activity of enzymes responsible for producing inflammatory compounds, NSAIDs help mitigate pain and swelling in the affected joint. However, it is essential to note that the long-term use of NSAIDs may be associated with potential risks, particularly concerning gastrointestinal and cardiovascular health. Clinicians must carefully monitor patients using NSAIDs, considering individual health factors, pre-existing conditions, and the duration of treatment to minimize potential side effects and ensure the overall well-being of the patient [ 27 ]. Bisphosphonates: Bisphosphonates, which include medications such as alendronate and pamidronate, represent another class of drugs with potential implications for the management of AVN. These drugs are primarily known for preventing bone loss and maintaining bone density. In the context of AVN, bisphosphonates may slow the condition's progression by inhibiting bone resorption. By preserving bone structure, bisphosphonates aim to contribute to the overall management of AVN and potentially delay the need for surgical interventions. However, it is crucial to recognize that using bisphosphonates in AVN management is an area of ongoing research, and their effectiveness remains a subject of investigation. Furthermore, like any medication, bisphosphonates are associated with potential side effects, and their use should be carefully considered based on an individual's specific circumstances, including overall health, risk factors, and the stage of AVN. Regular monitoring and collaboration between healthcare professionals are essential to assess the benefits and risks of bisphosphonate therapy in AVN [ 28 ]. Physical Therapy Range of motion exercises: Range of motion exercises are a cornerstone of the rehabilitation process for individuals with AVN. These gentle exercises are designed to maintain joint flexibility, prevent stiffness, and optimize overall joint function. Physical therapists are pivotal in developing individualized exercise programs tailored to each patient's needs and limitations. These programs may involve controlled movements that guide the affected joint through its full range, promoting joint health and preventing the development of contractures. Regularly implementing range of motion exercises contributes to the preservation of joint mobility and assists in mitigating the impact of AVN on daily activities [ 29 ]. Strengthening exercises: Targeted strengthening exercises form an integral part of rehabilitation for AVN, aiming to enhance the strength of the muscles surrounding the affected joint. By bolstering muscle support, these exercises help compensate for joint instability and reduce the overall load on the compromised bone. Strengthening programs address specific muscle groups related to the affected joint, promoting joint stability and optimizing functional capacity. Physical therapists work closely with patients to ensure that strengthening exercises are appropriately tailored, gradually progressing in intensity to avoid overexertion and accommodate the patient's capabilities [ 30 ]. Modalities for pain management: Pain management modalities are incorporated by physical therapists to alleviate discomfort associated with AVN and facilitate tissue healing. These may include heat or cold therapy, ultrasound, and electrical stimulation. Heat therapy can enhance blood flow, reduce muscle tension, and promote relaxation, while cold therapy can help alleviate inflammation and numb pain. Ultrasound and electrical stimulation may contribute to pain relief and tissue healing by affecting cellular function. The selection of specific modalities is based on the individual's needs, preferences, and stage of AVN, providing a comprehensive and personalized approach to pain management [ 31 ]. Functional training: Functional training is a critical component of rehabilitation for individuals with AVN, focusing on improving the patient's ability to perform daily activities with greater ease and efficiency. This form of training goes beyond isolated exercises and incorporates movements that mimic real-life activities. Physical therapists guide patients in practising proper body mechanics and joint protection techniques during functional tasks. By enhancing functional capacity, individuals can regain independence in their daily lives and minimize the impact of AVN on activities such as walking, reaching, and lifting. Functional training aims to optimize overall function and quality of life, promoting long-term joint health and mitigating the challenges posed by AVN [ 32 ]. Surgical treatment options Surgical interventions play a crucial role in the management of AVN, especially when conservative measures fail to halt disease progression. The choice of surgical approach depends on factors such as the stage of AVN, the affected joint, and the patient's overall health. Several surgical treatment options are available, each addressing specific aspects of AVN. Core Decompression Technique: Core decompression is a surgical procedure for managing AVN. The technique involves making a small incision at the site of the affected joint, typically the femoral head, in the case of hip AVN. During the procedure, a core or plug of bone is carefully removed from the necrotic region. This deliberate removal of bone serves a dual purpose: first, it creates a void or channel within the affected area, and second, it facilitates the removal of the necrotic tissue. The goal is to create a pathway for new blood vessels to invade the necrotic region, potentially enhancing the blood supply to the compromised bone. Introducing fresh blood flow to the area is vital for promoting bone healing and regeneration. Core decompression is often considered in the early stages of AVN to intervene before extensive bone damage occurs. The success of this technique relies on the restoration of adequate blood circulation, contributing to the overall preservation of joint integrity [ 33 ]. Outcome: Core decompression has demonstrated success in clinical practice, offering relief from symptoms associated with AVN and, in some cases, slowing down the condition's progression. The procedure aims to provide a conducive environment for bone healing by creating a channel for improved blood supply. However, the effectiveness of core decompression can vary depending on factors such as the AVN stage, the affected joint's location, and the patient's overall health. In some instances, core decompression may be combined with additional therapeutic approaches to enhance its outcomes. This can include bone grafting, where healthy bone tissue is transplanted to the affected area to support regeneration further. The success of core decompression underscores the importance of early intervention and a multidisciplinary approach in the comprehensive management of AVN, with the ultimate goal of preserving joint function and improving the quality of life for affected individuals [ 34 ]. Vascularized Bone Grafting Technique: Vascularized bone grafting is a surgical technique to treat AVN, particularly in advanced stages with extensive bone damage. The procedure involves harvesting a piece of bone, along with its accompanying blood vessels, from one area of the body and transplanting it to the site of the necrotic bone. The key distinction of vascularised bone grafting lies in preserving the blood supply to the transplanted bone. By ensuring the inclusion of blood vessels, the graft brings along a direct source of nourishment, facilitating the integration of the transplanted bone with the recipient site. This vascularised approach is particularly beneficial in cases where AVN has progressed, and there is a need for robust blood flow to support the healing and regeneration of the compromised bone. The technique is intricate and requires surgical expertise to carefully connect the blood vessels of the graft to those of the recipient site, ensuring optimal vascularisation and promoting successful graft integration [ 35 ]. Outcome: Vascularized bone grafting has demonstrated success in clinical practice, especially in cases with advanced stages of AVN characterized by extensive bone damage. Restoring a direct and immediate blood supply to the transplanted bone enhances the potential for successful healing and integration. This approach aims to preserve joint function by providing structural support to the compromised bone. While vascularised bone grafting is considered an effective intervention, it is more complex and involved than core decompression. As such, it is typically reserved for cases with significant bone involvement and when less invasive treatments may not be sufficient. The success of vascularised bone grafting underscores its role as a valuable option in the comprehensive management of AVN, particularly in situations where preserving joint integrity is critical for the patient's mobility and quality of life [ 36 ]. Total Joint Replacement Indications: Total joint replacement, also known as arthroplasty, is a surgical intervention commonly indicated in cases of AVN with severe joint degeneration and functional impairment, particularly in weight-bearing joints such as the hip and knee. The decision to perform total joint replacement is typically made when conservative measures have proven ineffective, and the progression of AVN has resulted in significant damage to the joint structure. Indications for total joint replacement include persistent pain, loss of joint function, and a diminished quality of life due to the impact of AVN on daily activities. The procedure aims to alleviate pain, restore joint function, and enhance the overall mobility and well-being of the affected individual. Total joint replacement becomes a viable option when the joint damage is beyond the scope of other interventions, and the goal is to provide a durable and functional artificial joint that can mimic the natural joint's movement and stability [ 37 ]. Outcome: Total joint replacement has proven effective in providing long-term relief for individuals with end-stage AVN. Advances in prosthetic technology, surgical techniques, and postoperative care have contributed to improved outcomes, with many patients experiencing significant reductions in pain and restoration of functional capacity. The artificial joint, composed of metal, plastic, or ceramic materials, is designed to replicate the natural joint's movement and weight-bearing functions. Patients typically undergo extensive rehabilitation following the surgery to optimize recovery and regain joint function. The success of total joint replacement in AVN cases is reflected in improved quality of life, enhanced mobility, and a return to daily activities with reduced pain and increased joint stability. While the procedure is associated with a notable success rate, individual outcomes may vary, and careful consideration of factors such as patient health, age, and lifestyle is essential in the decision-making process. Overall, total joint replacement remains a critical and transformative intervention for individuals suffering from the debilitating effects of AVN in weight-bearing joints [ 38 ]. Arthroplasty Techniques Hemiarthroplasty: Hemiarthroplasty is a surgical procedure in which only one part of the joint is replaced with a prosthetic component. This approach is commonly employed when only a specific portion of the joint is affected by AVN, and the remaining joint structures are relatively healthy. For example, in hip hemiarthroplasty, the femoral head, which may be the region impacted by AVN, is replaced with a prosthetic component. In contrast, the acetabulum (socket) of the hip joint remains intact. This technique addresses the localized damage within the joint, providing pain relief and restoring function without replacing the entire joint. Hemiarthroplasty is often considered in cases where the disease is asymmetrically distributed within the joint, and the preservation of native joint structures is deemed beneficial [ 39 ]. Resurfacing arthroplasty: Resurfacing arthroplasty involves removing and replacing the damaged surfaces of the joint while preserving more of the native bone compared to traditional joint replacement procedures. This technique is commonly employed in hip arthroplasty, specifically in cases where AVN affects the femoral head. In resurfacing, rather than entirely replacing the femoral head, the damaged surface is reshaped and capped with a metal prosthesis, preserving more of the patient's natural bone. This approach is designed to mimic the joint's natural anatomy more closely, potentially allowing for a more excellent range of motion and stability. Resurfacing arthroplasty is often considered in younger, more active patients, where preserving native bone may be advantageous for future revision surgeries [ 25 ]. Revision arthroplasty: Revision arthroplasty is a surgical procedure performed when previous joint replacement surgeries require correction, modification, or updating. This procedure involves removing and replacing existing prosthetic components to address complications, such as implant wear, loosening, or instability, and improve overall joint function. In the context of AVN, revision arthroplasty may be indicated if there are issues with the initially implanted joint replacement components or if the disease progresses following an earlier intervention. This complex procedure aims to rectify problems associated with the initial joint replacement, often involving new or revised prosthetic components to optimize joint stability and function. Revision arthroplasty requires careful planning and expertise to achieve successful outcomes and may involve more extensive surgical manoeuvres than primary joint replacement procedures [ 40 ]. Emerging therapies and research Regenerative Medicine Platelet-rich plasma (PRP): PRP is a regenerative therapy that involves the extraction and concentration of platelets from the patient's blood. The process begins with collecting a blood sample, which is then processed to separate the platelets from other blood components. The resulting platelet-rich plasma, abundant in growth factors, is injected into the affected joint. The high concentration of growth factors in PRP is believed to stimulate tissue repair and regeneration. In the context of AVN, PRP holds promise as a potential treatment option. Research is ongoing to explore its efficacy in promoting angiogenesis (forming new blood vessels) and reducing inflammation within the affected joint. By harnessing the body's natural healing mechanisms, PRP aims to improve blood flow to the necrotic area and create an environment conducive to tissue healing [ 41 ]. Bone marrow aspirate concentrate (BMAC): BMAC is a regenerative therapy that involves the extraction of bone marrow from the patient and the subsequent concentration of mesenchymal stem cells and growth factors. Like PRP, the process begins with collecting a bone marrow sample, typically from the patient's hip or pelvis. The harvested material is then processed to concentrate the mesenchymal stem cells and growth factors before being injected into the affected joint. BMAC aims to enhance tissue repair and promote bone regeneration in conditions such as AVN. Early studies suggest potential benefits, particularly in the early stages of AVN, where the goal is to intervene before extensive bone damage occurs. The concentrated solution, rich in regenerative components, is believed to contribute to the repair of the necrotic bone and the restoration of joint function. While research on the efficacy of BMAC in AVN is ongoing, the therapy represents a promising avenue in regenerative medicine for addressing the challenges posed by this condition [ 42 ]. Stem Cell Therapy Mesenchymal stem cells (MSCs): MSCs, derived from sources such as bone marrow or adipose tissue, have emerged as a promising avenue in regenerative medicine for the treatment of AVN. MSCs possess the remarkable ability to differentiate into various cell types, including bone-forming cells (osteoblasts). In the context of AVN, where compromised blood supply leads to bone tissue death, MSCs offer a potential solution. Preclinical studies and early clinical trials have explored using MSCs to promote bone regeneration and repair necrotic tissue within the affected joint. The idea is that introducing MSCs to the necrotic area may enhance the body's natural healing processes, contributing to the formation of new, healthy bones. While research is ongoing, the regenerative potential of MSCs makes them a compelling candidate for future therapeutic applications in AVN [ 43 ]. Embryonic stem cells (ESCs) and induced pluripotent stem cells (iPSCs): Research on the use of pluripotent stem cells, including ESCs and iPSCs, for the treatment of AVN, is still in the experimental stages. ESCs are derived from embryos, and iPSCs are generated by reprogramming adult cells to a pluripotent state. Both pluripotent stem cells can differentiate into various cell types, including bone-forming cells (osteoblasts). This unique ability to become various cell types positions pluripotent stem cells as potentially transformative in regenerative therapies for conditions like AVN. While the research is in its early phases, the concept involves directing these cells to differentiate into bone-forming cells, providing a targeted approach to regenerate the necrotic bone tissue. Challenges related to safety, ethical considerations, and the optimization of differentiation protocols are essential aspects of ongoing research. The potential of pluripotent stem cells in AVN treatment opens avenues for innovative regenerative strategies, but their clinical application is likely to evolve as research progresses and addresses existing challenges [ 44 ]. Novel Pharmacological Approaches Angiogenesis modulators: Angiogenesis modulators are medications currently being investigated for their potential in treating AVN. These agents aim to promote angiogenesis, which is the formation of new blood vessels. In the context of AVN, where compromised blood supply leads to ischemia and necrosis of bone tissue, promoting the growth of new blood vessels holds promise for improving blood flow to the affected area. By enhancing angiogenesis, these medications seek to create a more favourable environment for tissue healing and regeneration. Research is ongoing to assess the effectiveness of angiogenesis modulators in mitigating the ischemic effects of AVN and potentially slowing disease progression [ 45 ]. Bone-targeted therapies: Novel pharmacological approaches in AVN treatment focus on bone-targeted therapies that specifically address bone remodelling and repair pathways. These drugs aim to modulate bone metabolism and enhance bone formation, offering a targeted approach to address the underlying pathology of AVN. The goal is to support the regeneration of necrotic bone tissue and potentially prevent further deterioration. As research progresses, the development of bone-targeted therapies may provide additional options for managing AVN, especially in cases where the disease has not progressed to advanced stages [ 46 ]. Anti-inflammatory agents: Recognizing the inflammatory component of AVN, new anti-inflammatory medications are being explored as potential interventions. These drugs aim to reduce inflammation within the affected joint, potentially alleviating symptoms and slowing disease progression. In AVN, inflammation plays a role in the pathogenesis and contributes to the destruction of bone tissue. Therefore, anti-inflammatory agents may offer a complementary approach to managing the condition, particularly with other treatment modalities. Ongoing research aims to evaluate the efficacy of these medications in addressing the inflammatory aspects of AVN and their potential role in a comprehensive treatment strategy [ 47 ]. Rehabilitation and post-treatment care Physical Therapy Protocols Range of motion exercises: Range of motion exercises play a pivotal role in the early stages of rehabilitation, aiming to improve joint flexibility and prevent stiffness. These exercises involve controlled movements that guide the affected joint through its full range of motion. By promoting flexibility, range of motion exercises restore normal joint movement, reduce the risk of contractures, and optimize overall joint function [ 48 ]. Strengthening exercises: Progressive strengthening exercises form a crucial component of rehabilitation, explicitly targeting the muscles around the affected joint. These exercises enhance muscle strength, providing stability and support to the treated joint. Strengthening the muscles becomes particularly important in cases where joint instability or muscle atrophy has occurred due to conditions like AVN. The gradual progression of strengthening exercises contributes to overall rehabilitation, promoting joint stability and preventing further functional decline [ 49 ]. Weight-bearing activities: The reintroduction of weight-bearing activities is a carefully monitored aspect of rehabilitation, aiming to avoid excessive stress on the treated joint. This phase involves a gradual progression of weight-bearing exercises, starting with low-impact activities and advancing to more demanding tasks as tolerated by the patient. The controlled reintroduction of weight-bearing helps assess the joint's response to increased load, ensuring a safe and effective rehabilitation process while minimizing the risk of complications [ 50 ]. Functional training: Functional training is a targeted approach in rehabilitation, focusing on enhancing the patient's ability to perform daily activities with improved efficiency and reduced strain on the treated joint. This form of training goes beyond isolated exercises and incorporates movements that mimic real-life activities. Occupational considerations and biomechanical principles are integrated to ensure the patient can safely and effectively engage in functional tasks, promoting independence and return to daily activities [ 51 ]. Pain management techniques: Physical therapists employ various pain management modalities as part of rehabilitation to alleviate discomfort and enhance the patient's ability to engage in exercises. Modalities such as heat or cold therapy, massage, and transcutaneous electrical nerve stimulation (TENS) address pain and discomfort associated with the treated joint. Integrating pain management techniques aims to create a more comfortable and conducive environment for the patient to participate actively in the rehabilitation process [ 52 ]. Follow-up Monitoring Imaging studies: Regular imaging studies, such as X-rays or MRI, are pivotal in post-treatment care to assess the healing progress and identify potential complications. These studies allow healthcare providers to monitor changes in bone structure and joint integrity, providing valuable insights into the effectiveness of the treatment. Periodic imaging helps detect any signs of relapse early, ensuring timely intervention and adjustment of the patient's care plan [ 53 ]. Clinical assessment: Ongoing clinical assessments are fundamental in post-treatment care, encompassing evaluations of joint function, range of motion, and strength. These assessments are essential for tracking the patient's rehabilitation progress and identifying areas that may require further attention or modification in the rehabilitation plan. Clinicians use hands-on evaluations and functional tests to gauge the individual's response to treatment, enabling them to tailor rehabilitation strategies to each patient's specific needs and challenges [ 54 ]. Pain assessment: Periodic pain levels and symptomatology assessment are critical in post-treatment care. Monitoring changes in pain patterns or intensity provides essential feedback on the patient's response to rehabilitation and the overall healing process. Pain assessment guides healthcare providers in making informed decisions regarding the continuation of specific exercises, the need for adjustments in the rehabilitation plan, or the consideration of additional medical interventions to manage pain effectively. A comprehensive understanding of the patient's pain experience is integral to optimizing post-treatment care and ensuring a successful recovery [ 55 ]. Long-term Management Strategies Lifestyle modifications: Lifestyle modifications play a crucial role in managing AVN in the long term and aim to reduce stress on the affected joint. Patients are often advised to use weight management strategies, as excess body weight can increase joint stress. Adopting joint-friendly exercise routines, such as low-impact activities, can help maintain joint health without exacerbating symptoms. Lifestyle modifications are tailored to the individual's needs and are designed to promote overall joint well-being while minimizing the risk of further damage [ 56 ]. Joint protection techniques: Educating individuals on joint protection techniques is integral to managing AVN. These techniques empower patients to optimize their daily activities without putting undue stress on the affected joint. Proper body mechanics, ergonomic adjustments in work and daily environments, and adaptive devices, when necessary, contribute to joint protection. By incorporating these techniques into daily routines, individuals with AVN can enhance their ability to perform activities while minimizing the risk of exacerbating joint symptoms [ 57 ]. Medication management: Long-term pharmacological management is often necessary to address pain and inflammation associated with AVN. Medications such as analgesics (pain relievers) or disease-modifying anti-rheumatic drugs (DMARDs) may be prescribed based on individual needs. Analgesics help manage pain symptoms, while DMARDs may be used to modify the disease process in cases where AVN is associated with autoimmune or inflammatory conditions. Medication management is tailored to the specific requirements of each patient and is an essential component of comprehensive AVN care [ 58 ]. Regular follow-up with healthcare providers: Regular follow-up appointments with orthopaedic specialists or rheumatologists are essential for ongoing joint health monitoring in individuals with AVN. These appointments allow for continuous assessment of the joint's condition and the early detection of any recurrent symptoms or signs of disease progression. This proactive approach enables healthcare providers to intervene promptly, adjusting treatment plans or recommending additional interventions to prevent further joint deterioration [ 59 ]. Patient education and empowerment: Patient education is a cornerstone of effective AVN management, promoting active participation in maintaining joint health. Empowering patients with knowledge about their condition, treatment options, and self-management strategies is vital. Education covers the importance of adherence to rehabilitation protocols, the implementation of lifestyle modifications, and the significance of regular follow-up care. By understanding their condition and actively participating in their care, patients can make informed decisions and take proactive steps to preserve joint function and overall well-being. Patient empowerment fosters a collaborative approach between healthcare providers and individuals, optimizing the long-term management of AVN [ 60 ]. Challenges and future directions Limitations of Current Treatment Strategies Stage-specific interventions: The current landscape of AVN treatment strategies recognizes the importance of stage-specific interventions, tailoring approaches based on the severity of the disease. While more established interventions exist for early-stage AVN, challenges persist in addressing advanced-stage cases, especially in weight-bearing joints. Research advances are crucial to developing effective interventions for late-stage disease, focusing on preserving joint function, alleviating symptoms, and preventing further degeneration. The refinement and expansion of treatment options for advanced-stage AVN remain areas of active investigation to enhance patient outcomes [ 9 ]. Risk of complications: Although effective in many instances, surgical interventions have inherent risks and potential complications. Complications such as infection, implant failure, or incomplete resolution of symptoms underscore the need for continuous improvement in surgical techniques and postoperative care. Research and innovation in the field aim to minimize these risks, enhance the safety profile of surgical interventions, and optimize the overall success of treatment. Ongoing efforts focus on refining surgical approaches and developing strategies to mitigate the potential complications associated with AVN interventions [ 61 ]. Limited pharmacological options: Current pharmacological interventions, including bisphosphonates and NSAIDs, primarily provide symptomatic relief and may not address the underlying causes of AVN. The quest for more targeted pharmacological agents is a crucial focus of ongoing research. Developing medications that alleviate symptoms and promote bone regeneration and vascular health is a critical area of exploration. Advancements in pharmacological options aim to offer more comprehensive and disease-modifying treatments, addressing the root causes of AVN and improving long-term outcomes for affected individuals [ 27 ]. Heterogeneity in patient response: The heterogeneity in patient response to conservative and surgical treatments poses a complex challenge in AVN management. Variability in treatment outcomes among individuals underscores the importance of identifying predictive factors influencing responses to different interventions. Personalized medicine approaches, guided by patient-specific characteristics, genetic factors, and disease mechanisms, are under investigation to tailor interventions to individual profiles. Understanding the factors contributing to diverse treatment responses is essential for refining treatment algorithms and optimizing the effectiveness of interventions for each patient with AVN. Ongoing research explores the complexities of individualized treatment responses in this multifaceted condition [ 62 ]. Areas for Further Research Biomarkers for early detection: Advancing research should prioritize the identification of reliable biomarkers for the early detection of AVN. Biomarkers that signal a predisposition to AVN or indicate early vascular changes could revolutionize clinical practice by enabling timely intervention and preventive strategies. Developing robust and sensitive biomarkers can transform AVN diagnosis and management, allowing healthcare providers to identify at-risk individuals and implement targeted interventions before irreversible damage occurs [ 63 ]. Genetic and molecular pathways: A deeper exploration of the genetic and molecular pathways involved in AVN development is crucial for gaining insights into targeted therapeutic approaches. Investigating the genetic factors contributing to AVN susceptibility provides an avenue for personalized treatment strategies. Understanding the intricate molecular mechanisms underlying AVN pathogenesis could lead to developing interventions that address the condition's root causes. Research in this area has the potential to unlock new therapeutic targets and guide the development of precision medicine approaches tailored to individual patients [ 64 ]. Regenerative therapies: Further research into regenerative therapies, including stem cell therapy and tissue engineering, is essential for advancing treatment options for AVN. Enhancing the regenerative potential of these therapies and optimizing their application for different joints and AVN stages is an ongoing exploration area. Investigating the mechanisms by which regenerative therapies promote tissue repair and exploring their long-term efficacy will contribute to the development of innovative and sustainable interventions for AVN [ 65 ]. Improved imaging techniques: Advancements in imaging techniques, particularly advanced MRI and molecular imaging, are crucial for enhancing our ability to visualize early changes in AVN bone vascularity and tissue health. Improved imaging modalities can provide more accurate and detailed information, enabling precise diagnosis and staging of AVN. Early vascular and structural changes detection through advanced imaging can facilitate timely intervention and guide treatment decisions. Research efforts focused on refining and developing imaging technologies will improve diagnostic capabilities and AVN management [ 66 ]. Integrating Multidisciplinary Approaches Collaboration among specialities: Effective AVN management necessitates seamless collaboration among various specialities, including orthopaedic surgeons, rheumatologists, radiologists, and physical therapists. Establishing multidisciplinary clinics where specialists work collaboratively ensures that individuals with AVN receive comprehensive and coordinated care. This collaborative approach enables the integration of diverse expertise, fostering a holistic understanding of the condition and optimizing treatment strategies tailored to each patient's needs [ 67 ]. Patient-centred care: Embracing a patient-centred approach is integral to AVN care, emphasizing the importance of considering individual preferences, lifestyles, and treatment goals. Incorporating patient perspectives into treatment decision-making and care planning enhances treatment adherence and satisfaction. By actively involving patients in their care, healthcare providers can build a partnership that empowers individuals to participate in their treatment journey, leading to more personalized and effective interventions [ 68 ]. Rehabilitation and lifestyle interventions: Integration of lifestyle interventions, including weight management, joint protection strategies, and patient education, is critical for achieving long-term success in AVN management. Multidisciplinary teams can collaborate to develop comprehensive rehabilitation and lifestyle programs tailored to individual needs. This collaborative effort addresses not only the immediate symptoms but also focuses on enhancing overall joint health and minimizing the risk of disease progression through proactive lifestyle modifications [ 69 ]. Telemedicine and remote monitoring: Exploring telemedicine and remote monitoring technologies represents a promising avenue to facilitate ongoing care and follow-up for individuals with AVN. Especially beneficial for patients in remote locations or those with mobility limitations, these technologies can enhance accessibility to multidisciplinary care. Telemedicine enables virtual consultations, allowing patients to connect with specialists and receive timely guidance. Remote monitoring tools can track key indicators of joint health, providing valuable data for healthcare providers to assess treatment effectiveness and make informed adjustments, thereby ensuring continuous and responsive care [ 70 ].

CC BY

no

2024-01-16 23:40:19

Cureus.; 15(12):e50510

oa_package/d9/ea/PMC10788237.tar.gz

PMC10788238

38226089

Introduction The female genital system is an intricate but intriguing structure that includes the external and internal genitalia, uterus, ovaries, fallopian tubes, and vagina. One of the most significant female reproductive organs is the uterus, which is also known as the womb and cervix. They are susceptible to both neoplastic and non-neoplastic diseases. The endometrium and myometrium, which make up the uterus, are constantly driven by hormones, inhabited by fetuses, and undergo a monthly loss of endometrial mucosa [ 1 ]. Along with cervix lesions, endometrial and corpus of the uterus are the most common causes of patient visits to gynecologists [ 2 ]. Despite a variety of treatment options, such as medication and conservative surgical techniques, hysterectomy remains the most common gynecological procedure performed globally [ 3 ]. The first partial hysterectomy was performed in Manchester, England, in 1843 by Charles Clay, and in 1929, the first total abdominal hysterectomy was performed [ 4 ]. Numerous conditions such as abnormal uterine bleeding, pelvic pain, pelvic inflammatory disease (PID), prolapse of the uterus, adenomyosis, endometriosis, fibroids, gynecological malignancies, and obstetric problems have been reported. Every hysterectomy sample must be examined histopathologically because histology is the only source of the final diagnosis [ 5 ]. Histopathological examination of the specimens obtained after hysterectomy is important for both diagnosis and treatment. The current work aimed to identify the various clinical indications, analyze the clinicopathological correlation in hysterectomy specimens, and analyze the patterns of lesions in hysterectomy specimens.

Materials and methods Study design This is a hospital-based retrospective observational study. Place and duration of study This study was carried out in the Pathology Department at Datta Meghe Medical College, Wanadongari, Nagpur, from February 2022 to January 2023. Sample size A total of 110 women who presented to the obstetrics and gynecology department of the Datta Meghe Medical College, Wanadongari, Nagpur, with a clinical diagnosis of female genital tract lesions, were enrolled in the study. Inclusion criteria The study included all forms of hysterectomy, including abdominal, vaginal, laparoscopic, and total abdominal hysterectomy, with or without unilateral or bilateral salpingectomy or salpingo-oophorectomy. Exclusion criteria Obstetric hysterectomy was the only exclusion criterion for this study. Data collection The documentation included the patient's name, age, sex, clinical appearance, and differential diagnosis, along with a detailed clinical history and information from the gynecological request form. Study procedure The samples were gathered and sent to the histopathology section of the pathology department. From the patient's case report, a succinct summary of the pertinent clinical history and results was taken. The samples received were fixed with 10% neutral buffered formalin. The samples were grossly examined and sectioned. In case if required, representative sections from anomalous regions were also taken. H&E staining was performed on the sections after they had been embedded and processed. The slides were all examined under a microscope. Statistics were conducted after taking note of the results. Ethical consideration As this study was retrospective and observational rather than interventional, there were no risk variables. Information was collected from the patient's request form and histopathological findings of the pathology department. The appropriate authorization was obtained from the Institutional Ethics Committee (SMHRC/IEC/2022/12-15). The identities of patients and doctors were not noted. Statistical analysis The tabulated results were subjected to statistical analyses. Statistical analysis was used to determine the type of lesion, the incidence rate, and the percentage in each age group. After entering the data into Microsoft Excel (Microsoft Corporation, Redmond, Washington, United States), statistical analysis was performed.

Results This study included 110 patients. The age distribution of the hysterectomy specimens is presented in Table 1 . Hysterectomies were performed in women aged between 25 and 75 years of age. The majority of cases, 47 (42.72%) of these 110 cases, occurred between the ages of 35 and 45 years, followed by 32 (29.09%) cases between the ages of 45 and 55 years. The least number of cases, four (3.64%), were between the ages of 65 and 75 years. According to Table 2 , vaginal hysterectomy was the second most prevalent type of hysterectomy, with 31 cases (28.19%), while the most frequent type of hysterectomy performed was total abdominal hysterectomy with unilateral or bilateral salpingo-oophorectomy, which represented 79 cases (71.82%). The indications for hysterectomies range from irregular menstruation to possible pelvic malignancies. Table 2 shows the number of hysterectomy indications. Most of the patients had a fibroid uterus, which accounted for 33 (30%) cases, followed by uterovaginal prolapse, comprising 31 (28.19%) cases, and dysfunctional uterine bleeding, comprising 25 (22.73%) cases. Table 3 shows the distribution of histological findings of the endometrium. The most common finding was the endometrium of the proliferative phase in 43 (39.09%) cases (Figure 1 ). The atrophic endometrium was then more frequently associated with uterovaginal prolapse, seen in 35 (31.82%) cases. Secretory phase endometrium was observed in 21 cases (19.10 %). In two (1.82%) cases, the endometrium showed a polyp. In seven (6.37%) cases, simple endometrial hyperplasia was observed, and in one (0.91%), atypia-related endometrial hyperplasia was noted. Out of 110 cases, a single case (0.91%) was diagnosed as endometrial endometrioid carcinoma, which on microscopic examination, showed a back-to-back arrangement of endometrial glands with cytological dysplasia along with stromal invasion. Table 4 illustrates the distribution of myometrial lesions, among which leiomyoma was the most common histopathological finding in 52 cases (47.28 %). Microscopy revealed well-defined tumors composed of oval to spindle-shaped cells with elongated blunt-ended nuclei and a modest amount of eosinophilic cytoplasm. The cells were arranged in the form of interlacing fascicles and bundles (Figure 2 ). Some leiomyomas exhibited secondary modifications such as hyaline degeneration and myxoid degeneration. The next most common finding was adenomyosis, which was observed in 23 (20.91%) cases. Few cases had both leiomyoma and adenomyosis, seen in 20 (18.19%). As shown in Table 5 , chronic cervicitis was the most common cervical lesion, comprising 85 (77.28%) cases (Figure 3 ). Chronic cervicitis with squamous metaplasia was observed in 13 patients (11.82 %). Six (5.46%) cases were papillary endocervicitis and three (2.73%) cases were cervical fibroids. Table 6 shows that in the present study, most of the cases (72 of 110) showed normal ovarian histology. There were 28 (25.46%) cases of non-neoplastic lesions and 10 (9.10%) cases of neoplastic lesions in the ovaries. Non-neoplastic lesions included follicular cysts, the most common finding observed in 22 (20%) cases (Figure 4 ), followed by luteal cysts in 6 (5.46%). Neoplastic lesions included serous cystadenoma as the most common finding in seven (6.37%) cases, followed by one (0.91%) case each of mature teratoma, mucinous cystadenoma, and mucinous cystadenocarcinoma. The correlation between the preoperative clinical diagnosis and the histopathological diagnosis is shown in Table 7 . In 110 patients, a preoperative clinical diagnosis was available. In most cases, ranging from 70% to 100%, the final histopathological diagnosis supports the preoperative clinical diagnosis. In the current study, a total of two cases of malignant tumors were observed, one case of endometrial carcinoma and one case of mucinous cystadenocarcinoma.

Discussion Hysterectomy is the most prevalent gynecological operation worldwide. It is an effective procedure to alleviate symptoms and provide patient contentment and offers a permanent solution for many disorders affecting the uterus and adnexa [ 6 ]. In this study, the age range of the patients was 25 to 75 years, with a mean age of 50.86 +/- 6.9 years. According to Verma et al. [ 7 ] the mean age was 50.1 years, while Adelusola et al. [ 8 ] study had a mean age of 49.1 years. In the current study, women between the ages of 35 and 45 years were the most frequently subjected to hysterectomies, which is similar to other studies [ 9 - 12 ]. In the present study, abdominal hysterectomy represented 79 (71.82%) cases and was the surgical procedure performed most frequently, while vaginal hysterectomy accounted for 31 cases (28.19%). In a study by MacKenzie et al. [ 13 ], abdominal hysterectomy was preferred in 79% of the cases and vaginal hysterectomy in 17% of the cases. Studies by Sachin et al. [ 14 ], Pandey et al. [ 15 ], Sujatha et al. [ 16 ], and Gupta et al. [ 17 ] revealed that the most common hysterectomy procedure was total abdominal hysterectomy. Data from the United Kingdom reveal that abdominal hysterectomy procedures are five to six times more common than vaginal hysterectomy procedures. In a study by Pandya et al. [ 18 ], vaginal hysterectomy was the surgical procedure most commonly used in comparison to abdominal hysterectomy. In our study, fibroids were the most frequent indication of hysterectomy, followed by uterovaginal prolapse, abnormal menstrual cycles, and abdominal masses. According to a study conducted in the United States by Broder et al. [ 19 ], fibroids (60%) and prolapse (11%) were the two most common indications. Similar results were found in studies conducted by Jandial [ 20 ] and Ullah et al. [ 21 ]. Even according to studies by Butt et al. [ 22 ], Tiwana et al. [ 23 ], Abe et al. [ 24 ], and Leung et al. [ 25 ], uterine fibroid was the most common indication of hysterectomy. In a study conducted by Verma et al. [ 7 ] in Uttar Pradesh, India, uterovaginal prolapse (37.5%) and fibroid uterus (25.6%) were shown to be the most common indications. However, a study by Canadian researchers Toma et al. [ 26 ] found that dysfunctional uterine bleeding was the most common indication, followed by uterine fibroid. The proliferative endometrium (39.7%), which is frequently associated with pathological lesions such as fibroids and adenomyosis, was the most frequent endometrial lesion identified in the present study, followed by the atrophic endometrium (32%), which was frequently observed in postmenopausal women with uterovaginal prolapse. This finding is similar to that of Patil et al. [ 27 ], in which the proliferative phase endometrium was the most common endometrial lesion, followed by the atrophic endometrium. Atrophic endometrium was the most prevalent endometrial pathology identified in the study by Kleebkaow et al. [ 28 ], who estimated its frequency to be 3.8%. However, Awale et al. [ 29 ] observed a greater frequency of atrophic endometrium in their investigation, which was 26.53%. In our study, leiomyoma was shown to be more frequent than adenomyosis, which has also been observed in studies by Neelgund et al. [ 30 ] and Khurshid et al. [ 31 ]. In our study, the most frequent incidental finding, chronic cervicitis, was observed in 77.28% of the patients. According to the studies done by Patil et al. [ 27 ], Talukder et al. [ 32 ], and Khunte et al. [ 33 ], the most frequent finding among cervical lesions is chronic cervicitis. The most frequent ovarian lesion observed was a simple follicular cyst, which is consistent with other previous studies by Nausheen et al. [ 3 ], Pandey et al. [ 15 ], and Perveen et al. [ 34 ]. The most frequent benign tumor was a simple serous cystadenoma. The mature cystic teratoma and the mucinous cystadenoma had one case each. There was a case of malignant mucinous cystadenocarcinoma. In the current study, a histopathological evaluation of the fallopian tubes revealed no abnormal lesions. Other studies revealed that the most frequent ovarian abnormalities in their studies were cysts with varied morphologies [ 30 , 34 , 35 ]. Most of the preoperative clinical diagnoses in our study were supported by histopathological reports, with a proportion ranging from 70% to 100%. Jaleel et al. [ 36 ] reported findings that are almost identical to ours. The preoperative clinical diagnosis of adenomyosis, dermoid cyst, uterovaginal prolapse, and cervical fibroid shows 100% correlation with histopathological reports. The only limitation of the current study was the lack of follow-up.

Conclusions The present study offers a good understanding of the histopathological patterns of lesions in hysterectomy specimens from our institution. The most prevalent uterine pathology is leiomyoma; the most prevalent ovarian lesion is a follicular cyst; and chronic cervicitis is the most frequently found incidental finding in the cervix in hysterectomy specimens. A total of two cases of malignant tumors are noted: one case of endometrial carcinoma and one case of mucinous cystadenocarcinoma of the ovary. Few lesions, including chronic cervicitis and adenomyosis, are discovered purely as incidental findings, although histopathological analysis and clinical diagnoses generally correlate well. To ensure better postoperative management, it is imperative that every hysterectomy specimen, even if it superficially appears to be normal, be subjected to a thorough histopathological examination.

Introduction The uterus is a crucial reproductive organ that is susceptible to the development of several non-neoplastic and neoplastic diseases in women, greatly increasing morbidity and mortality. Although there are various therapeutic options, hysterectomy is still a popular treatment option throughout the world. Abnormal uterine bleeding, pelvic pain, pelvic inflammatory disease (PID), prolapse of the uterus, adenomyosis, endometriosis, fibroids, gynecological malignancies, and obstetric problems that require hysterectomy, all samples must be examined histopathologically. Histopathological examination of the specimens obtained after hysterectomy is important for both diagnosis and treatment. The current work aimed to identify the various clinical indications, analyze the clinicopathological correlation in hysterectomy specimens, and analyze the patterns of lesions in hysterectomy specimens. Materials and methods This study was conducted in the Department of Pathology at the Datta Meghe Medical College, Wanadongari, Nagpur, from February 2022 to January 2023. All types of hysterectomy specimens received during this year were examined, and the tissues were processed and stained with H&E. Histopathological examination was performed, and various lesions in the hysterectomy specimens were examined. The study included all forms of hysterectomy, including abdominal, vaginal, laparoscopic, and total abdominal hysterectomy. Results An analysis of 110 cases of hysterectomy revealed that abdominal hysterectomy was the type of hysterectomy in 79 (71.82%) cases, with a maximum age range of 35 to 45 years (42.72%). The proliferative phase endometrium was the most common endometrial pathology, accounting for 43 (39.09%) cases, followed by the atrophic endometrium in 35 (31.82%) cases. Leiomyoma was the most prevalent myometrial lesion, accounting for 52 (47.28%) cases, followed by adenomyosis, accounting for 23 (20.91%) cases. Chronic cervicitis was the most common incidental finding in the hysterectomy samples, accounting for 85 (77.28%) cases. Follicular cysts, representing 22 (20%) cases, were the most common ovarian lesions, followed by serous cystadenoma in seven (6.37%) cases. Two cases of malignant tumors were noted: one case of endometrial carcinoma and one case of mucinous cystadenocarcinoma of the ovary. In most cases, ranging from 70% to 100%, the final histopathological diagnosis supports the preoperative clinical diagnosis. Conclusion Hysterectomy is the most common major gynecological surgery performed under elective conditions. Although histological studies and clinical diagnoses are closely correlated, several lesions, including chronic cervicitis and adenomyosis, were discovered incidentally. Therefore, every hysterectomy specimen must undergo a thorough histological investigation, even if it appears superficially normal, to confirm the diagnosis and improve postoperative care.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50497

oa_package/79/8e/PMC10788238.tar.gz

PMC10788239

38226120

Introduction Esophageal cancer is known to be a deadly malignancy and represents the fifth most common gastrointestinal cancer in the United States with approximately 16,940 new cases per year [ 1 ]. Histologically, esophageal cancers are mainly identified as squamous cell carcinoma (SCC) or adenocarcinoma (ACA). There are many risk factors for SCC including smoking, alcohol consumption, a diet that is low in fruits and vegetables, as well as drinking beverages at high temperatures. Notably, human papillomavirus (HPV) has also been correlated with an increased incidence of SCC in the upper esophagus. However, in general, the United States is considered a low-risk area for esophageal SCC and instead has been witnessing an increase in incidence of esophageal ACA. This may be due to the upsurge of obesity and gastroesophageal reflux disease (GERD). Conversely, SCC is mainly found in the "esophageal cancer belt" of northern Iran, southern Russia, central Asian countries, and northern China [ 1 ]. Initially, early lesions may be subtle, but more advanced lesions may be ulcerated and circumferential and infiltrate the submucosa. Spread may occur via the lymphatic system to regional lymph nodes. Distant metastases may subsequently involve the liver, lung, and bone marrow. Patients typically present with complaints of dysphagia, substantial weight loss, and other non-specific symptoms such as retrosternal discomfort or burning sensation. To evaluate, clinicians may start with a barium swallow, but an upper endoscopy with biopsy is performed to confirm the diagnosis. Computed tomography (CT) scans of the thorax and abdomen may be performed to determine the extent of the primary tumor as well as look for any potential liver metastasis and celiac lymphadenopathy. However, it should be noted that endoscopic ultrasound (EUS) has now become the standard for locoregional staging and for assessing tumor depth and mediastinal lymph node involvement. It also allows a fine-needle aspiration biopsy of lymph nodes. Management typically involves either endoscopic resection for superficial, limited mucosa disease or surgical resection for lesions penetrating the submucosa. Neoadjuvant chemoradiation of resectable lesions and palliative systemic therapy for unresectable or metastatic disease also play a role [ 1 ]. We present a case of newly diagnosed esophageal SCC requiring an esophageal stent to improve per os (PO) intake.

Discussion Dysphagia is typically the predominant symptom in patients with SCC of the esophagus and often leads to weight loss and malnutrition [ 2 ]. Because esophageal cancer is often first diagnosed at an advanced stage, a great number of patients require some form of palliative treatment to relieve the esophageal stricture, maintain caloric intake, and improve the quality of life. Among various treatment options for malignant dysphagia, endoscopically inserted stent is the most widely used method today. The first stents used were rigid plastic stents that were effective but accompanied complications such as perforation, migration, and food impaction. Self-expanding metal stents (SEMSs) were then developed in the 1990s and provided immediate dysphagia palliation in more than 85% of patients. They were also associated with significantly reduced stent-related mortality, perforation, and migration compared to plastic stents [ 2 ]. An array of self-expanding stents have since materialized to reduce tumor in-growth through the open mesh and minimize stent migration. Currently available options include uncovered SEMSs, partially covered SEMSs, fully covered SEMSs, and self-expandable plastic stents (SEPSs) [ 3 ]. Partially covered SEMSs have been the most frequently used stent clinically worldwide; however, stent selection should be tailored to the individual patient's characteristics [ 2 - 4 ]. In addition to patients with unresectable esophageal carcinoma, placement of SEMSs is also indicated for patients on neoadjuvant therapy awaiting definitive surgical treatments [ 2 ]. While SEMSs are widely used for palliation of dysphagia, there is no standardized optimal method, and other modalities exist [ 2 ]. A randomized trial comparing the efficacy of brachytherapy versus partially covered SEMS placement found brachytherapy to have a more latent onset of dysphagia improvement and a superior long-term dysphagia relief after three months of follow-up. Brachytherapy was also associated with fewer major complications compared to SEMSs. However, limited availability of brachytherapy and the uncertainty of optimal dose schedule hinder it from widespread use [ 2 - 3 ]. While endoscopic ablation methods such as photodynamic therapy (PDT) and laser therapy provided comparable palliation to SEMS, meta-analysis concluded ablative therapies required more re-interventions [ 2 - 3 ]. In our case, these factors were considered when placing a self-expanding covered metal esophageal stent for dysphagia palliation. The patient's improved PO intake after stent placement supports esophageal stent as a first-line modality for immediate malignant dysphagia relief. Treatment for patients with severe dysphagia in the setting of esophageal SCC should center around improving dysphagia symptoms and maintaining adequate caloric intake while definitive treatment options are determined.

Conclusions Esophageal SCC should be suspected in the setting of non-specific symptoms such as dysphagia, weight loss, and retrosternal discomfort. Typically, treatment options include esophageal or surgical resection along with neoadjuvant chemoradiation of resectable lesions. However, the placement of an esophageal stent may be considered to provide patients with palliative treatment. Stents may relieve the esophageal stricture, maintain caloric intake, and improve the quality of life.

Esophageal cancer is typically identified as squamous cell carcinoma or adenocarcinoma. There are multiple risk factors that may contribute to esophageal squamous cell carcinoma including smoking, alcohol consumption, and the human papillomavirus. Lesions may appear ulcerated, friable, and circumferential and may obstruct the esophagus. Therefore, patients may complain of non-specific symptoms including dysphagia, weight loss, and retrosternal discomfort. Clinicians often rely on an upper endoscopy with biopsy to confirm the diagnosis. Computed tomography scans and endoscopic ultrasounds are also employed to assess the extent of malignant spread. Management may involve endoscopic resection for superficial lesions or surgical resection for lesions penetrating the submucosa. Esophageal stents may play a role, specifically as a palliative measure for enhancing oral intake. We present an instance of utilizing a self-expandable, metal-covered esophageal stent with balloon dilation in the setting of a newly diagnosed esophageal squamous cell carcinoma lesion in a 73-year-old female. Ultimately, the use of an esophageal stent in this patient helped improve the patient's oral intake during her course of hospitalization. Her diet was slowly advanced to clear liquids and progressively to a low-residue diet before being discharged to follow-up with her diagnosis as outpatient with gastroenterology.

Case presentation A 73-year-old female with a past medical history of chronic obstructive pulmonary disease (COPD) and hypertension presented to the emergency department due to symptoms of exacerbated dysphagia for the past three months. She endorsed dysphagia to solids initially, but progressively complained of dysphagia to liquids as well. She reported a 10-pound weight loss since the initiation of her symptoms. She was admitted and was made nil per os (NPO) upon admission due to inability to tolerate both solids and liquids. The patient was taken for an esophagogastroduodenoscopy (EGD) the following day. The EGD revealed a nearly obstructing, friable, ulcerated mass extending from the proximal to mid-esophagus, approximately spanning from 27 to 34 centimeters down the length of the esophagus as seen in Figure 1 . The pathology report revealed that the mass was an infiltrating, poorly differentiated, SCC of the esophagus. The pathology report was also negative for HPV testing. Further evaluation with CT scans of the chest and abdomen and pelvis revealed no distant metastases. Subsequently, a decision was made to perform an EUS with concomitant esophageal stent placement for staging and palliative feeding purposes. The EUS revealed esophageal SCC classified as stage 3: Tumor-3 (T3) and Nodes-2 (N2) as seen in Figure 2 . A 120 millimeter x 23 millimeter self-expanding covered metal esophageal stent was deployed, with balloon dilation of the proximal opening as shown in Figure 3 . The patient tolerated the procedure well and departed the endoscopy suite in stable condition and returned back to the medical floors. During the remainder of her hospitalization, her diet was slowly advanced to clear liquids and was eventually transitioned to a low-residue diet. She was discharged from the hospital on day 6 of her admission and advised to follow-up with her new diagnosis as outpatient with gastroenterology.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50483

oa_package/89/48/PMC10788239.tar.gz

PMC10788240

38226078

Introduction The medical challenges faced by foreigners or immigrants in accessing healthcare in foreign countries, which result in health disparities and deterioration in health conditions, can be viewed as universal health inequities [ 1 - 3 ]. Indeed, foreign residents may need help utilizing medical services due to their unfamiliarity with the healthcare system [ 3 ]. In addition, these foreign residents may not be enrolled in health insurance or need more awareness of the available options [ 4 ]. The language barrier is also a particularly significant issue for foreign residents seeking to receive medical care in Japan. Communication difficulties can lead to anxiety and inconvenience, potentially making it challenging for healthcare professionals to understand the needs of their patients [ 5 ]. In this context, foreign residents could be a group heavily impacted by the novel coronavirus disease 2019 (COVID-19). The COVID-19 pandemic has been a significant factor causing substantial disruption to healthcare services globally. Particularly in Japan, the pandemic has led to social distancing and fear of transmission, causing a decrease in hospital visits and reports of individuals ignoring symptoms until their conditions worsen [ 6 , 7 ]. Therefore, healthcare professionals must understand foreign residents' communication needs and strive to provide language support as necessary [ 5 ]. In order to provide linguistic support for physicians in Japanese medical settings, a telephone-based human interpretation service is available. This service, called MediPhone, developed by a Japanese medical SaaS startup founded by former tech industry professionals, offers remote medical interpretation over the phone [ 8 ]. It acts as a bridge, connecting healthcare professionals with remote medical interpreters in real time, thereby facilitating communication between healthcare professionals and foreign patients who speak different languages. The service covers 31 languages and boasts a roster of over 300 registered medical interpreters. Users can access the service for free for up to 30 minutes per month, with additional charges incurred for extended usage. A distinct advantage of this platform is the employment of human interpreters. If there is any ambiguity in the patient's statements, the interpreter can engage in a clarifying dialogue with the patient before relaying the information to the medical staff, ensuring the patient's narrative is accurately conveyed. In previous studies, health inequity has been explored through four primary aspects: 1) race/ethnicity, 2) primary language, 3) gender, and 4) location [ 1 ]. Especially concerning primary language, Japanese is recognized as one of the most challenging languages, and the English proficiency of many Japanese medical professionals remains relatively low [ 5 , 9 ]. Given this context, human-mediated translation services in clinical practice are of paramount importance. However, it remains uncertain which specific patient groups most require these interpreter services and the circumstances driving this need. This study, therefore, was designed to highlight the patients who may face challenges accessing medical care due to language barriers in a single medical clinic.

Materials and methods Study site The study site was in Tokyo, where the number of foreign residents was reported to be approximately 540,000, accounting for about 20% of the foreign population living in Japan in 2022 [ 10 ]. Particularly in Tachikawa City, which has a population of over 180,000, there are around 4,650 foreign residents, and remarkably, despite the challenges of the pandemic, their numbers have been on the rise for the past seven years, indicating a latent demand for medical care tailored to foreigners [ 11 ]. Its strategic location and service offerings make it an ideal setting for examining healthcare service utilization among foreign residents in a metropolitan area. Settings and participants Navitas Clinic Tachikawa is an outpatient clinic (Tachikawa, Tokyo, Japan), consisting of internal medicine, pediatrics, and dermatology departments. The clinic also offers vaccinations and health check-ups. The clinic operates until 9 p.m. on weekdays and until 5 p.m. on Saturdays, providing a service format accessible to those working in central Tokyo who typically find it challenging to visit clinics during the daytime. The clinic is staffed with seven physicians, two pediatricians, and three dermatologists, all of whom are Japanese and speak Japanese as their first language. The clinic started implementing human-mediated translation services in November 2017. The target group for this study comprised foreign patients who utilized human-mediated translation services for consultations at our clinic between November 2017 and December 2021. This period was selected to capture a comprehensive dataset post-implementation of the translation services. Data collection and analysis We conducted a retrospective investigation of the medical records of foreign patients who used human-mediated translation services for consultations at our clinic from November 2017 to December 2021. The surveyed items included age, gender, the language used, department visited, diagnosis, whether any tests were conducted, whether health insurance was in place, methods of booking, and dates of visits. These variables were selected to provide a comprehensive overview of patient demographics and their interaction with the healthcare system. To ensure accuracy and comprehensiveness, data extraction was performed by a trained team of medical staff under strict confidentiality protocols. This approach aimed to identify patterns in service utilization and pinpoint specific needs of foreign residents within the healthcare system.

Results Table 1 presents the results of the visit list. During the study period, a total of 124 foreign patients had consultations using human-mediated translation services. Of these, 69 (56%) were male, and 55 (44%) were female. The median age was 35 years, ranging from 3 to 61 years. By age group, there were six patients (5%) aged 10 and under, no patients (0%) in their teens, 27 patients (22%) in their 20s, 82 patients (66%) in their 30s, five patients (4%) in their 40s, two patients (2%) in their 50s, and two patients (2%) in their 60s. By department, 107 patients (86%) visited internal medicine, nine patients (7%) visited dermatology, and six patients (5%) visited pediatrics. Two patients (2%) underwent COVID-19 antibody testing. In terms of reservation format, 12 patients (10%) made online reservations, and 112 patients (90%) registered at the reception desk. Of these, 30 males (52%) and 28 females (48%), a total of 58, were registered on the patient list. The characteristics of these individuals are summarized in Table 2 . The median age was 34 years. The languages used on human-mediated translation services were English for 34 patients (59%); Chinese for nine (16%); Spanish for four (7%); Hindi for three (5%); Nepalese for two (3%); and Russian, Thai, Vietnamese, French, Tagalog, and Bengali each for one patient (1%). Forty-seven patients (81%) were covered by the Japanese National Insurance, three patients (5%) did not have insurance, and eight patients (14%) were self-pay. Of the three patients without insurance, two were tourists, and one had forgotten to renew. There were eight patients (14%) who only had tests, 23 patients (40%) who had both tests and prescriptions; 22 patients (38%) who only had prescriptions; one patient (1%) who had a test, a prescription, and a referral; two patients (3%) who only had a consultation; one patient (1%) who only had a referral; and one patient (1%) who had a referral and was hospitalized. The most common condition was the common cold, affecting 29 patients (50%), and pre-return COVID-19 tests for border entry screening were performed on five patients (9%). There were also two patients (3%) each for HPV vaccination and hay fever, and one patient each (1%) for diabetes, acute gastritis, and urinary tract infection.

Discussion The results of this study demonstrate a clear need for human-mediated translation services, particularly for foreign residents visiting Navitas Clinic Tachikawa in Tokyo. During the study period, 124 foreign patients utilized human-mediated translation services, indicating a significant demand for language interpretation services in healthcare settings. The gender distribution was relatively balanced, and the median age of patients was 35 years. Regarding language needs, English was the most commonly used language in human-mediated translation services, followed by Chinese and Spanish. Notably, a large proportion of these patients were in their 30s, possibly reflecting the age distribution of foreign workers, with an average age reported to be 33.3 in Japan [ 12 ]. This information is especially useful for medical institutions planning resources and services for foreigners, such as in Osaka and Aichi, where there are large numbers of foreign workers [ 13 ]. In addition, 86% of the patients who visited our clinic had an internal medicine consultation, suggesting that they had a chronic disease or general health condition that required regular visits [ 14 ]. The fact that most patients had health insurance, and the majority of patients booked their appointments at the reception desk, might be indicative of a certain level of familiarity with the Japanese healthcare system among foreign residents. However, the three patients who did not have insurance highlight an important issue. Insurance enrollment is a key factor in ensuring access to healthcare services [ 15 ]. Healthcare professionals and policymakers need to address this issue to prevent health disparities among foreign residents. This study has some limitations. Firstly, human-mediated translation services are not always the go-to solution for straightforward English conversations. Other general digital translation tools, such as Google Translate and DeepL, or even non-verbal communication can influence the preference for or against using human-mediated translation services. Translation services might not be required for brief medical examinations conducted by physicians fluent in English. In addition, since some patients speak languages not covered by human-mediated translation services, or are seen with friends who speak Japanese, caution should be taken in interpreting this survey. Secondly, being a single-centre study, the findings may not fully represent all medical clinics in Japan. Thirdly, the study did not collect information about the patient's nationality, length of stay in Japan, or level of Japanese proficiency, which could influence the need for interpretation services. Consequently, future research should validate the efficacy of such human-mediated translation services by comparing consultation durations and the accuracy of comprehension. Despite these limitations, our study provides valuable insights into the utilization of medical interpretation services by foreign residents in Japan. Given the increasing globalization and diversity in society, the demand for such services will likely continue to grow. Human-mediated translation services play a crucial role in ensuring that foreign residents have equal access to healthcare services, ultimately contributing to the goal of health equity.

Conclusions In conclusion, this study underscores the vital importance of human-mediated translation services in enhancing healthcare accessibility for foreign residents in Japan, emphasizing their diverse linguistic needs and varied medical services they utilize. Future research should focus on expanding the scope of these studies to include multiple centers and diverse patient demographics, further exploring the impact of translation services on patient care quality and healthcare accessibility for foreign residents in Japan. This study, thus, serves as a foundational step in understanding and addressing the unique healthcare needs of Japan's increasingly diverse population.

Introduction Foreign residents in Japan often face challenges accessing healthcare due to language barriers, potentially leading to health inequities. This study aimed to assess the utilization and impact of human-mediated translation services in a specific medical setting in Tokyo. Methods A retrospective investigation was conducted on medical records of foreign patients who utilized human-mediated translation services at Navitas Clinic Tachikawa (Tachikawa, Tokyo, Japan) from November 2017 to December 2021. Data on age, gender, language used, department visited, diagnosis, insurance status, and booking methods were analyzed. Results Out of the 124 foreign patients who utilized the human-mediated translation services during the study period, 69 (56%) were male, and 55 (44%) were female. The median age was 35 years, with a range from 3 to 61 years. English was the predominant language used by 34 patients (59%), followed by Chinese for nine patients (16%) and Spanish for four patients (7%). The majority, 107 patients (86%) visited the internal medicine department, nine patients (7%) consulted dermatology, and six patients (5%) visited pediatrics. Regarding insurance status, 47 patients (81%) were insured, three patients (5%) were uninsured by the Japanese national health insurance system, and eight patients (14%) were self-pay. The primary mode of appointment booking was at the reception desk, with 112 patients (90%) using this method, while 12 patients (10%) made reservations online. Conclusions The findings of this study underscore the importance of human-mediated translation services for improving healthcare accessibility for foreign residents in Japan, emphasizing the need to address language barriers and promote health equity in clinical settings. Future studies should also explore challenges faced in patient-physician interactions from a linguistic perspective and potential technological solutions to enhance these services.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50493

oa_package/f6/e0/PMC10788240.tar.gz

PMC10788241

38226129

Introduction Heart failure (HF) is a complex clinical syndrome caused by a structural or functional cardiac disorder that results when the heart cannot provide the body with sufficient blood supply [ 1 - 3 ]. The most common cause of HF is when the left ventricular myocardial function is reduced, followed by the dysfunction of the pericardium, myocardium, endocardium, heart valves, or great vessels [ 4 , 5 ]. Further, HF is becoming a significant public health concern affecting an estimated 14 million Europeans [ 1 , 6 ]. According to the American Heart Association, over five million Americans over 20 have HF, and the estimated prevalence of HF in Saudi Arabia is 455,222 cases, with an estimated incidence of 32,200 cases annually [ 7 ]. Moreover, HF primarily results from injuries to the myocardium caused by a variety of factors, including ischemic heart disease, hypertension, and diabetes. Less common factors include cardiomyopathies, valvular disease, myocarditis, infections, systemic toxins, and cardiotoxic drugs [ 3 ]. A recent study in Saudi hospitals revealed that 20% of acute coronary syndrome admissions had HF [ 7 , 8 ], which can be classified based on the deficit site (as predominantly left ventricular, right ventricular, or biventricular) or according to the onset (as acute or chronic) [ 4 , 9 ]. Moreover, HF can be clinically classified based on the heart function status into two types: HF with preserved ejection fraction (EF) in those with a normal EF of over 50% and HF with a reduced EF of less than 40% [ 4 , 9 ]. The signs and symptoms of HF occur with inadequate cardiac output and poor blood supply. Primarily, the patients present with shortness of breath, coughing, wheezing, fatigue, weakness, lethargy, lower limb edema, and ascites [ 3 , 4 , 10 ]. For the diagnosis, various parameters are used with a physical examination to determine the presence of clinical signs and symptoms. Moreover, blood tests are employed, including a complete blood count, urinalysis, complete metabolic profile, blood urea nitrogen, serum creatinine, glucose, fasting lipid profile, liver function test, and thyroid stimulating hormone. Additionally, HF-specific laboratory tests are used, including brain natriuretic peptide (BNP) and N-terminal proBNP (which is more sensitive and less specific than BNP). Other diagnostic modalities include chest X-rays, transthoracic echocardiography, computerized tomography scans, and magnetic resonance imaging [ 4 , 5 ]. Other studies have demonstrated that over 40% of patients with HF die within the first year of their first hospitalization, indicating a poor prognosis for HF [ 6 , 11 ]. In this regard, public awareness of HF is critical for controlling the disease burden, recognizing disease severity, and determining its prognosis. However, studies worldwide on this topic are limited and primarily from Europe because there is a lack of information on public awareness of HF [ 12 ]. This study aims to assess the awareness and perception of HF among the population in Makkah City, Saudi Arabia.

Materials and methods This cross-sectional study was conducted between October 2022 and February 2023. The study includes participants over 18 who live in Makkah. Participants who refused to participate in the study were excluded. The sample size was calculated using Open Epi software v2.1 ( www.OpenEpi.com ) [ 13 ]. The minimum calculated sample size was 486, considering a confidence interval of 95% and a significance level of 5%. However, 1,053 participants were recruited in this study to increase the efficiency of generalizing the study findings. The Umm Al-Qura ethical committee granted this survey ethical approval in November 2022, with approval number HPAO-02-K-012-2022-11-1275, under the principles of the Helsinki Declaration. Participant names, phone numbers, and identity card numbers were not included to maintain anonymity and confidentiality. Before the survey, each respondent provided informed consent online and was made aware that the survey was voluntary, confidential, and only for academic purposes. The study tool, an online questionnaire translated independently by two bilingual translators, was adapted from a previously published study [ 12 ]. The Arabic version was established, and a backward translation was conducted to compare the English version with the original to assess the accuracy of the questionnaire. Finally, a pretest was conducted by distributing the survey to 10 participants and then collecting their data and feedback. The questionnaire is divided into three sections. The first section collects demographic information. The second section contains questions that measure the study population’s awareness of HF before defining it, and the third section contains questions after providing the definition. Statistical analysis was performed using Statistical Product and Service Solutions (SPSS, version 26) (IBM SPSS Statistics for Windows, Armonk, NY). The numerical data (including age only) are presented as the mean plus or minus the standard deviation. In contrast, the categorical data (including the rest of the variables) are presented as frequencies and percentages. A chi-square test was employed to assess the association between the sociodemographic characteristics of the participants and their knowledge of HF. Age was categorized as ≤30 or >30, and a p-value <0.05 was considered statistically significant.

Results A total of 1,053 participants who met the inclusion criteria completed the study questionnaire. The mean age of the participants was 33.5 (13.2) years, ranging from 18 to 70 years old. Additionally, 558 (53%) of the study participants were male, and 495 (47.0%) were female. The most predominant nationality of the participants was Saudi (n = 1006; 95.5%; Table 1 ). The first three questions evaluated the awareness of the symptoms of cardiovascular disease: Question (Q) 1 for angina/myocardial infarction, Q2 for stroke, and Q3 for HF. Less than half of the participants answered correctly regarding Q1 (44.0%), Q2 (32.5%), and Q3 (24.1%). Moreover, Q3 was correctly answered much less than the two other diseases. When respondents were asked if they ever heard of HF, 62.4% answered that they heard of it. When the participants were asked to choose the best representation of HF, 59.5% chose "the heart cannot pump enough blood around the body.” When asked about the severity of the following symptoms: “breathlessness, tiredness, or swollen ankles,” only 37.0% recognized it as a serious illness. When they were asked, “How soon will you go to the hospital if you feel breathlessness, tiredness, or swollen ankles,” most (69.2%) answered that they would visit the hospital in one to two days, followed by within a week (18.0%), one to three weeks (6.4%), and a month (2.7%), and some (3.7%) would skip visiting the hospital. About one-third of the participants’ families suffer from heart disease, 13.8% from angina or myocardial infarction, 13.5% from HF, 12.0% from arrhythmia, and 22.0% from valvular heart disease (Table 2 ). Regarding the etiology of HF, about half of the participants (50.6%) answered that HF is part of the normal aging process. Regarding the contributing factors for developing HF, most respondents selected hypertension (82.9%), while the rest chose lung disorder (17.1%). Regarding the perception of HF-related risk, when the participants were asked about the lifetime risk of developing HF, only 18.3% answered correctly: “20 in 100 people.” When the participants were asked which of the following diseases had the highest mortality after five years of diagnosis, 44.3% chose HF, 26.5% indicated myocardial infarction, 22.4% chose stroke, and 6.8% selected prostate or breast cancer. Regarding the risk of sudden cardiac death in HF patients, 69.4% of the participants answered that they were worried about sudden cardiac death, and 12.2% were not worried about it. When the participants were asked about the HF risk of mortality within the first year of discharge, only 10.3% correctly answered, “20 in 100 people might die.” When the participants were asked about the readmission rate within a year after discharge from HF, 15.4% correctly selected “20 in 100 people.” The participants were asked to choose which disease has the greatest influence on quality of life, and the majority (61.1%) indicated HF, 19.0% selected diabetes, 13.0% chose hypertension, and 6.9% indicated arthritis. As for HF treatment, most participants (48.5%) indicated that they would prefer a treatment that could improve their quality of life, and 19.2% preferred a treatment that would allow them to live longer. Their perceptions about HF medication were also investigated in the survey. About 51.6% agreed that HF medication can reduce death from HF. Further, 53.6% agreed that the current HF medications could improve the quality of life in patients with HF, and 35.1% agreed that current HF medications could prevent the occurrence of HF. When the participants were asked if they thought that HF patients should live quietly and reduce all physical activity, most (66.8%) answered “yes.” The preferred sources of HF-related information among the respondents were the internet (48.5%) and followed by primary healthcare (25.7%) and secondary or tertiary care clinics (23.6%; Table 3 ). Table 4 lists the association between the answers to Q3 (“What disease do you think of if someone has the following symptoms? Symptom: breathlessness with low-level activity, tiredness, and swollen ankles”) and the sociodemographic characteristics. Most participants younger than 30 years (32.0%) and over 30 years (41.1%) answered that these are related to the heart in general, with a P value <0.000. Additionally, male (36.7%) and female (36.2%) respondents answered “heart in general,” with a P value of 0.639. Further, Saudi (36.1%) and non-Saudi respondents (44.7%) answered that these symptoms were more related to the heart in general, with a P value of 0.133.

Discussion Although HF has a poor prognosis and that more than 40% of patients die within the first year of their first hospitalization, the awareness of HF in Europe remains unsatisfactory. Studies regarding the awareness among the Middle Eastern population are lacking as well (6,11,12). Therefore, this study assesses the awareness of HF among the population of Makkah City, Saudi Arabia. Accordingly, the findings revealed an acceptable level of awareness, whereas most participants have heard about HF (62.4%), and the majority defined it correctly (59.5%). Nevertheless, a substantial number of the respondents defined HF incorrectly. However, these results are significantly better than those in a Korean study, where only 47% could define HF correctly [ 12 ]. These results are consistent with the findings of the Europe (SHAPE) survey [ 6 ]. Regarding the etiology of the disease, around 50.6% indicated that it is related to the aging process. This misperception aligns with the opinions of one-third of the participants in the Korean study [ 12 ]. This finding indicates the need for more awareness programs to prevent the late diagnosis of the disease and the consequent poor outcomes. Moreover, 82.9% believed that HF is caused by hypertension. According to primary care records in the UK, the five-year mortality of HF has been increasing to 48.5% [ 14 ]. Furthermore, US Medicare data report that mortality is 75% [ 15 ]. Despite that, the majority of the respondents did not recognize that HF has the highest mortality rate in contrast to MI, stroke, prostate cancer, and breast cancer. Only 18.3% recognized the correct lifetime risk of developing HF, which the Framingham Heart Study reported as 20%, regardless of sex [ 16 ]. When HF was identified with the symptoms of heart conditions in general, the Korean population exhibited better awareness, where 62% correctly recognized HF [ 12 ]. This finding indicates the need for more educational programs and campaigns to improve the awareness of the population regarding HF. Conversely, most participants stated that HF was a serious illness. More than half of the participants agreed that the current HF medications could reduce death from HF and improve the quality of the patients’ lives. This state of awareness is beneficial for the acceptance and tolerance of medications and should be enhanced in the rest of the population. However, most respondents disagreed that HF medications could prevent the occurrence of HF. Two-thirds of the Korean population preferred secondary and tertiary care clinics as sources of information about HF [ 12 ]. In contrast, the study participants preferred the internet as a source of information regarding the disease. This preference demonstrates the necessity of effectively guiding the population to more reliable sources that provide evidence-based information about the disease. A few limitations exist in this study. A more randomized sampling technique should be used for more accurate results. The variations in the number of population categories affect the generalization accuracy of the study findings. Therefore, the variation should be minimized in further studies of the same design.

Conclusions The focus of this study was to assess the awareness and perception of HF among the population in Saudi Arabia. Most participants had heard of HF, and the majority correctly defined it. However, multiple flaws exist in their understanding of numerous aspects of the disease, requiring more educational programs and campaigns guided by specialized physicians to improve their knowledge comprehensively. The preferred source of HF information was the internet, highlighting the need to effectively guide the population to more reliable sources that provide factual and scientifically proven information about the disease.

Background Heart failure (HF), a major public health problem worldwide, is a complex clinical syndrome caused by structural or functional heart disorders occurring when the heart cannot supply sufficient blood to the body. The most common cause of HF is impairment of the left ventricle. Public awareness of HF is critical for controlling the disease burden, recognizing disease severity, and determining its prognosis. Therefore, this study assesses the awareness and perception of HF among the population in Makkah City, Saudi Arabia. Methods A cross-sectional study was conducted among 1,053 participants over 18 years of age who lived in Makkah City between October 2022 and February 2023. Participants were randomly selected and recruited using a validated online questionnaire. Results Of the participants, 62.4% had heard of HF, and the majority (59.5%) correctly identified it. Regarding the etiology of the disease, about 50.6% indicated that it was related to the aging process, and 82.9% indicated it was related to high blood pressure. Only 24.1% of participants correctly recognized HF symptoms; most defined the symptoms as general heart disease. Moreover, 51.6% of participants agreed that the current HF medications can reduce deaths from HF and improve quality of life. However, most respondents disagreed that HF drugs can prevent the onset of HF. Conclusion The findings emphasize the need for more awareness programs to raise the public awareness about HF and effectively guide the population to more reliable sources that provide evidence-based information about the disease.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50504

oa_package/36/cd/PMC10788241.tar.gz

PMC10788242

38226105

Introduction Bacillus Calmette-Guérin (BCG) therapy stands as a pivotal modality in the treatment arsenal for non-muscle invasive bladder cancer (NMIBC), offering a unique approach through immunotherapy [ 1 ]. The treatment involves the intravesical instillation of live attenuated Mycobacterium bovis, derived from the tuberculosis bacillus, with the aim of provoking a robust local immune response within the bladder [ 2 ]. It is a standard treatment for high-risk non-muscle invasive bladder cancer [ 1 ]. While BCG has demonstrated remarkable efficacy in reducing the recurrence and progression of superficial bladder tumors, its application is not devoid of challenges, including the potential for disseminated BCG infection (which occurs in less than 5% of patients) and infusion reactions [ 3 ]. Studies have reported a reduction in recurrence rates, and complete response rates ranging from approximately 30% to 70% for patients with NMIBC receiving BCG therapy after transurethral resection of bladder tumors (TURBTs) [ 3 ]. Understanding how BCG works is fundamental to appreciating its therapeutic potential. The live attenuated mycobacteria stimulate a local immune response by activating macrophages and T lymphocytes [ 4 , 5 ]. This immune cascade not only targets and destroys cancer cells within the bladder but also induces a systemic immune response [ 5 ]. The mechanism involves the release of various cytokines, such as interferon-gamma and tumor necrosis factor-alpha, orchestrating an immune-mediated assault on cancer cells [ 4 ]. BCG therapy, therefore, serves not only as a direct attack on existing tumor cells but also as a mechanism for bolstering the body's natural defenses against bladder cancer [ 6 ]. Despite its success, BCG therapy carries inherent risks. Disseminated BCG infection, though rare, remains a serious concern. This occurs when the instilled mycobacteria escape the confines of the bladder, disseminating through the bloodstream and potentially affecting distant organs. It often necessitates treatment with RIPE therapy (rifampin, isoniazid, pyrazinamide, and ethambutol), although there are some concerns about BCG treatment-resistant strains [ 7 ]. Concurrently, infusion reactions, including severe sepsis, can manifest as a result of the heightened immune response induced by BCG [ 3 , 8 , 9 ]. These complications necessitate a delicate balance in the management of patients undergoing BCG therapy.

Discussion BCG for bladder cancer is one of the oldest forms of immunotherapy [ 1 ]. A diagram depicting its mechanism of action can be seen in Figure 2 . While BCG is effective for both immunocompetent and immunocompromised patients, it can lead to systemic side effects in 30%-40% of patients and potentially life-threatening complications, including distant organ infections [ 10 ]. Management of these effects necessitates immediate intervention, continuous monitoring, and cautious medication use. A very small fraction of treated patients develop life-threatening sepsis [ 3 , 7 ]. BCG sepsis is treated with antimycobacterial antibiotics (RIPE) and glucocorticoids [ 3 , 6 ]. Known risk factors such as urothelial barrier disruption or concurrent urinary tract infections should be considered before BCG administration to minimize complications [ 10 ]. It is crucial to provide patient education and maintain vigilant follow-ups for potential long-term effects. Alternative approaches to non-muscular invasive bladder cancer are surgical [ 10 ]. TURBT is the mainstay therapy for non-muscle invasive disease, with subsequent chemotherapy instillations [ 10 ]. Additionally, en bloc bladder tumor removal (EBRT) has been proposed as a preferable technique, reducing cell spillage, and decreasing surgical risks. However, BCG immunotherapy stands as the sole conservative intervention proven to deter progression in high-risk NMIBC [ 10 ]. The future of bladder cancer treatment includes immunotherapy with checkpoint inhibition, targeted therapies, and antibody-drug conjugates [ 2 ]. Checkpoint inhibitors, which potentiate the body's immune response against cancer cells, are a key focus in ongoing trials for bladder cancer. Targeted therapies, such as FGFR inhibitors, offer a more precise treatment option for patients with specific genetic alterations. Additionally, antibody-drug conjugates like enfortumab-vedotin provide a combination of monoclonal antibodies and chemotherapy, minimizing damage to healthy tissues [ 1 ]. The evolving landscape suggests that the integration of these novel therapies with established methods will likely shape the standard of care, offering improved outcomes and more tailored treatment strategies for bladder cancer patients [ 10 ].

Conclusions In conclusion, BCG therapy remains a cornerstone in the management of NMIBC, exhibiting efficacy in reducing tumor recurrence. However, the presented case underscores the intricate challenges associated with BCG treatment, including chemical cystitis and severe sepsis and the differential diagnosis of the rare but serious complication of disseminated BCG infection. The future of bladder cancer treatment, as discussed, is evolving with the exploration of immunotherapy and targeted therapies, offering potential alternatives or complementary approaches. As we navigate this evolving landscape, the significance of prompt recognition, accurate diagnosis, and vigilant monitoring of BCG-induced side effects becomes paramount. While BCG therapy continues to be a valuable tool in bladder cancer treatment, ongoing research and a personalized approach are essential to optimize its benefits while minimizing potential complications, thus ensuring the best possible outcomes for patients facing this challenging disease.

This case report presents a 66-year-old male with a complex medical history, including testicular cancer, chronic obstructive pulmonary disease, obstructive sleep apnea, tobacco use disorder, erectile dysfunction, and obesity. The patient exhibited recurrent gross hematuria, leading to a comprehensive workup. Cystoscopy revealed a bladder tumor, prompting transurethral resection and mitomycin C instillation. Subsequent intravesical Bacillus Calmette-Guérin (BCG) therapy was initiated but resulted in severe sepsis during maintenance. Despite initial suspicion of BCG-induced sepsis, further evaluation suggested a reaction with chemical cystitis. Treatment involved brief antimicrobial therapy, and the patient's condition improved. This case highlights the challenges in managing BCG therapy complications, emphasizing the need for prompt intervention, careful monitoring, and consideration of risk factors. Patient education and vigilant follow-ups are crucial for addressing potential long-term effects.

Case presentation A 66-year-old male patient presented with a history of testicular cancer (managed with radical orchiectomy), chronic obstructive pulmonary disease, obstructive sleep apnea, tobacco use disorder, erectile dysfunction, and obesity. The patient presented with gross hematuria persisting for approximately one week, a symptom he had experienced one to two months prior, leading to treatment for a urinary tract infection despite being asymptomatic at the time. Additionally, he described nocturia and a slightly slower urine stream, with the effect of tamsulosin on these symptoms being uncertain. Despite these urinary symptoms, prostate screenings remained consistently unremarkable. Further evaluation revealed a post-void residual test (PVR) result of 8 and an overall symptom score of 3, indicating low burden. Given the patient's medical history and smoking status, a comprehensive workup for gross hematuria was initiated. Cystoscopy performed by the urology team revealed a bladder filling defect at the left ureteral orifice and a calcified left-wall bladder tumor on a stalk, raising suspicion for transitional cell carcinoma. Consequently, TURBT was performed, followed by the instillation of mitomycin C. After the initial treatment, the patient was scheduled for weekly intravesical BCG therapy for a duration of six weeks. Subsequent follow-up cystoscopies did not detect any recurrence. A CT of the abdomen and pelvis before and after treatment with tumor resection and mitomycin C can be seen in Figure 1A , 1B . During maintenance BCG therapy, the patient returned to the ER, following a three-week cycle of BCG therapy, presenting with shortness of breath and dysuria, but without hematuria. Upon admission, he met the criteria for severe sepsis, with fever reaching 40.6°C, tachycardia peaking at 110 bpm, respiratory rate at 32 breaths per minute, increased oxygen requirement, and lactic acidosis (2.80). Given an apparent urinary tract infection/sepsis, there was a high suspicion of BCG-induced sepsis and disseminated BCG infection. Upon consultation, the infectious disease team recommended the initiation of RIPE therapy and vancomycin. Acid-fast bacilli, urine, and blood cultures were consistently negative. As the patient's condition improved symptomatically and clinically, the initial clinical impression favored a BCG infusion reaction with chemical cystitis, given the timing of his presentation (on the day of the last instillation). Consequently, the RIPE therapy and antibiotics were discontinued after 48 hours. The patient reported a return to baseline status and has remained under follow-up care since.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50498

oa_package/c9/8a/PMC10788242.tar.gz

PMC10788243

38226080

Introduction Meningiomas are slow-growing neoplasms carrying a favorable prognosis, especially when entirely resected and in younger patients [ 1 ]. Accounting for 15-20% of all brain tumors, they rarely extend beyond the cranial boundaries and usually associate osteolytic changes in the skull [ 2 ]. Primary extracranial meningiomas are uncommon tumors that are often misdiagnosed and are located in a broad range of sites, with approximately 6-17% of all meningiomas being found in extranevraxial locations. Thus, recognizing this diagnosis in unexpected spots could prevent potential challenges [ 3 ]. Typically, meningiomas are grouped into three grades, based on the World Health Organization (WHO) Meningioma Classification as follows: grade I (benign), grade II (atypical), and grade III (malignant), with the benign ones being the vast majority of tumors met in this field [ 4 ]. Every case of extracranial meningiomas provides a description of the histological characteristics; they are divided into different subtypes, in consideration of the most predominant cellular morphology, including meningothelial (syncytial), fibrous (fibroblastic), psammomatous, angiomatous (angioblastic), and transitional (mixed) [ 3 ]. The risk of extracranial meningiomas increases with age, also being more susceptible in women than men. Regarding the therapeutical approaches, total surgical resection is the primary treatment of choice for extracranial meningiomas [ 5 ].

Discussion As stated before, histology defines five patterns of these tumors, of which syncytial (meningothelial) ones are most prevalent among the primary extracranial meningiomas [ 3 ]. A neoplasm composed of spindle-shaped cells with whorls, without any indication of atypia, suggests a meningothelial origin [ 6 ]. Along with this subtype, psammomatous meningioma, a densely calcified tumor characterized by the presence of numerous psammoma bodies, is most common in vertebral disorders. Fibrous meningiomas are identified by extended spindle-shaped tumor cells with narrow rod-shaped nuclei that are set in a collagenous or reticulum-rich environment. Compared to meningothelial subtypes, these have fewer whorls, and psammoma bodies are sporadically found [ 7 ]. Typically, transitional meningiomas display meningothelial cells organized in bundles of variable length, exhibiting a mix of specific fibroblastic appearance, syncytial arrangements, epithelioid cells, and consistently occurring whorls. Faint cytoplasmic boundaries and occasional intranuclear inclusion bodies are usually described [ 8 ]. A rare subtype of meningiomas, comprising only 2.1% of them, is represented by the angiomatous one. Such tumors are described by an abundance of blood vessels amidst meningothelial areas. They are classified as WHO grade I tumors without cellular atypia or anaplasia [ 9 , 10 ]. Overall, meningiomas are classified into three grades (1-3) based on their histological and molecular features. Grade 1 subtype displays less aggressive characteristics, accordingly, while second and third grade proves to be more concerning. Thus, grade 2 tumors exhibit increased mitotic figures, invasion of the brain tissue, and specific histological subtypes (choroid or clear cell kinds), while grade 3 surpasses the prior ones, describing hallmarks such as sarcoma, carcinoma, or melanoma-like appearances, along with telomerase reverse transcriptase (TERT) promoter mutation [ 11 ]. Among the various types of meningiomas that are being found within a mass, MEP is quite uncommon, and only a few reported cases have presented tumors that merge both the extradural and MEP types [ 12 ]. They display a ‘carpet-like’ infiltration of the adjacent bone, accompanied by extensive hyperostosis and dural thickening [ 13 ]. Typically, extradural spinal meningiomas arise from nerve roots, where the dura is thinner, thus making it easier to migrate into the extradural space. More often, the meningioma appears as a round type, while the en plaque type grows along the dura mater in sheet-like structures [ 11 ]. While studying the literature on extracranial meningiomas, we have delved into many studies that unveiled significant contributions to the field, many of them being elaborated in Table 1 . From the data presented, we have shown that there are multiple surgical strategies, such as wide local excision and supraorbital eyebrow approach, which is a minimally invasive technique that offers wide access to the anterior skull base region and parasellar area through a subfrontal corridor. The use of neuroendoscopy allows one to extend the approach further to the pituitary fossa, the anterior third ventricle, the interpeduncular cistern, the anterior and medial temporal lobe, and the middle fossa. The supraorbital approach involves a limited skin incision, with minimal soft-tissue dissection and a small craniotomy, thus carrying relatively low approach-related morbidity. Using this technique, the risk of recurrence was 0% [ 18 ]. Table 1 highlights the relationship between the degree of resection of the tumor and the risk of recurrence so that patients in whom total growth resection (GTR) was possible have a much lower risk of recurrence than those in whom STR or PR could be performed. In the cases presented, in patients in whom total tumor resection was possible, recurrence was 0%. Most recurrences occurred about five years after surgery [ 19 ]. From the cases presented in the table, we have highlighted the greater exposure that women experience, compared to men, in the case of this pathology, as women are more likely to develop this disease. Additionally, older ages, over 65 years, favor the appearance of this disease, as it is in our case. From the data presented, we have shown that there are multiple surgical strategies, highlighting the relationship between the degree of resection and the risk of recurrences so that patients in whom GTR was possible have a much lower risk than those in whom subtotal growth resection (STR) or partial resection (PR) could be performed. In the cases presented, where total tumor resection was possible, the recurrence rate was 0%, most of them occurring about five years after surgery. We also took into account the histological aspect, useful in the diagnosis of this disease [ 20 ].

Conclusions In wrapping up our discussion about this patient’s fascinating medical case, we have unraveled significant aspects regarding the pathology of meningiomas. Our female patient possesses all the factors that make insightful contributions to the condition she experienced as follows: her advanced age, her clinical picture through the neurological deficits she had, and the tumor’s location, along with the osteolytic changes. The surgical approach allowed the resection of the entire tumoral process, thus preventing any recurrence and ensuring the patient’s wellness.

The study reflects on a 69-year-old female patient with a history of cardio-respiratory disorders who was diagnosed with meningioma en plaque. Her clinical management entailed surgical resection of the tumor, which was followed by a complex postoperative course, including cardiorespiratory arrest and respiratory failure. Histologically, extracranial meningiomas are categorized into five subtypes based on predominant cellular morphology, with the meningothelial type being prevalent in this case. The report also examines the significance of complete tumor resection, noting a lower recurrence rate with gross total resection. Additionally, it discusses the increased susceptibility of extracranial meningiomas with advancing age and a higher incidence in females. Data from various studies underscore the importance of a surgical approach and extent of resection in predicting recurrence risk. The case report concludes by highlighting the critical aspects of the pathology of meningiomas and the surgical strategy that ensured the patient's recovery. The findings from this case contribute to the broader understanding of extracranial meningiomas, their diagnosis, and management.

Case presentation This is a case of a 69-year-old patient with a known history of cardio-respiratory disorders, such as atrial fibrillation with medium rhythm and bilateral bronchopneumonia. Neurological examination reveals left hemiparesis, intracranial hypertension syndrome, cerebral edema, and cerebrospinal fluid (CSF) fistula of the frontal lesion. Following morphopathological analysis, we suspect that the patient was experiencing a meningioma en plaque (MEP), which is an infiltration at the level of the dura and sometimes invades the bone with intraosseous tumor growth, leading to significant hyperostosis. The patient also shows osteolytic change of the skull, an element found in MEP (Figures 1 - 2 ). After surgical intervention, achieving gross total resection (Figure 3 ), the patient suffered cardiorespiratory arrest of cardiac cause, followed by acute respiratory failure and ventilator prosthesis. The patient required intubation and mechanical ventilation over 96 hours, presenting NYHA II heart failure, acute hemodynamic failure, acid-base balance disorders, and upper gastrointestinal hemorrhage.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50490

oa_package/65/b0/PMC10788243.tar.gz

PMC10788244

38226081

Introduction Advances in genetic technologies have made genetic testing more accessible than ever before, with whole exome sequencing (WES) gaining popularity. However, the actual need for daily genetic testing more often revolves around targeted gene analysis rather than comprehensive examination. In the case of rare genetic diseases, it is often necessary to launch new testing systems. When the gene of interest has an extensive number of exons, performing Sanger sequencing on each exon or setting up capture probes for each exon in next-generation sequencing (NGS) becomes challenging. Previously, we used CEL nuclease-mediated heteroduplex incision with polyacrylamide gel electrophoresis and silver staining (CHIPS) technology for mutation screening to streamline the Sanger sequencing process [ 1 , 2 ]. However, the current landscape has shifted towards targeted NGS sequencing [ 3 , 4 ]. There is a growing need for an NGS method that allows easy setup and on-demand analysis of individual genes. Long-range polymerase chain reaction (PCR)-based NGS provides a solution to this challenge. With improved DNA polymerase and high molecular weight DNA extraction methods, we can now reliably amplify approximately 20 kb of specific gene regions from genomic DNA, and library preparation from long PCR products is straightforward. Given that the median size of a human gene on the genome is around 26 Kb [ 5 ], in many genes, one or two primer sets can cover the entire gene including the promoter region. We call this method very long amplicon sequencing (vLAS). This approach has particular advantages for genes characterized by numerous exons within a limited region. To illustrate its effectiveness, we present here an analysis of common inherited bone diseases. Osteogenesis imperfecta (OI) is a genetic connective tissue disorder characterized by bone fragility, with its primary feature being bone fragility. The majority of OI cases result from pathogenic variants in either the COL1A1 or COL1A2 gene, which encode for collagen type I alpha chains [ 6 , 7 ]. OI is classified into four types based on clinical severity and radiographic findings: classic non-deforming OI with blue sclerae (Type I), perinatally lethal OI (Type II), progressively deforming OI (Type III), and common variable OI with normal sclerae (Type IV) [ 8 ]. As genetic diagnosis advances, it plays a crucial role in predicting the severity of the phenotype based on the genotype [ 6 ]. Mutations in the COL2A1 gene cause a variety of autosomal dominant skeletal dysplasias. The most severe phenotypes, such as achondrogenesis type II, hypochondrogenesis, and Torrance type platyspondylic dysplasia, are associated with neonatal death. Moderate severity phenotypes include spondyloepiphyseal dysplasia congenita (SEDC), Strudwick type spondyloepimetaphyseal dysplasia (SEMD), Kniest dysplasia, spondyloperipheral dysplasia, and Czech dysplasia. Milder forms include early-onset osteoarthritis and Stickler syndrome type I (SSI), the most common type II collagenopathy (1/10,000). Early diagnosis is crucial for appropriate patient care, follow-up, and genetic counseling for affected families [ 9 ]. COL1A1 spans a genomic region of approximately 17.5 kb and contains 51 exons. Similarly, COL1A2 and COL2A1 are approximately 36.7 kb and 31.5 kb and contain 52 and 54 exons, respectively. This paper presents vLAS-analyzed data for patients with OI and SSI.

Materials and methods This study was performed at the Center for Clinical Genomics, Kanazawa Medical University Hospital, Uchinada, Japan. Since 2013, we have been contracting out genetic testing for various genetic diseases from in-hospital and out-of-hospital facilities in Japan. A total of seven patients were enrolled, comprising five with OI and two with SSI. Among them, three cases were previously diagnosed using CHIPS screening and Sanger sequencing, while four were newly identified cases (Table 1 ). Patients six and seven had previously reported CHIPS results [ 1 , 2 ]. All patients were clinically diagnosed by an experts' panel including clinical geneticists, radiologists, orthopedic surgeons, and pediatricians, based on clinical symptoms and bone radiographs. For all participants, genetic testing was conducted following genetic counseling, and written informed consent was obtained from the participating patients or their parents in the study after explanation by the primary physician. This study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board of Kanazawa Medical University (G161 approved on August 29, 2022). vLAS was performed as previously reported [ 10 - 12 ]. Briefly, 20 ng of peripheral blood DNA was amplified to approximately 20 kb using the 0.15 micromolar PCR primers listed in Table 1 and KOD Multi&Epi (TOYOBO, Osaka, Japan) to cover the target genes. PCR products were purified using an AMPure XP (Beckman Coulter Life Sciences, San Jose, CA). An NGS library was prepared using an Illumina DNA Prep with Enrichment Kit (Illumina, San Diego, CA), and a 12.5 pM library was loaded on an Illumina MiSeq (Illumina, Inc., San Diego, California, United States) system using a Reagent Nano Kit v2 (500 cycles) (Illumina, Inc., San Diego, California, United States) according to the manufacturer's recommended protocol. Variant calling was performed using GATK's HaplotypeCaller (Version 4.0.6.0) (Broad Institute, Cambridge, Massachusetts, United States) [ 13 ], and functional classification of variants was performed using SnpEff (version 4.3t) [ 14 ]. The Database of Short Genetic Variations dbSNP (version 151) and ClinVar were used for variant annotation [ 15 , 16 ]. The Integrative Genomic Viewer (IGV, version 2.4.13) (Broad Institute, Cambridge, Massachusetts, United States) was used for visualization [ 17 ]. Classification of the pathogenicity of variants was followed by the American College of Medical Genetics and Genomics/American Association of Molecular Pathology (ACMG/AMP) standard guidelines [ 18 ]. In patients two and three, novel missense variants were detected in COL1A1. These variants are absent in the general Japanese population according to the Japanese Multi Omics Reference Panel, jMorp ( https://jmorp.megabank.tohoku.ac.jp / last accessed November 20, 2023), and the Genome Aggregation Database, gnomAD ( https://gnomad.broadinstitute.org / last accessed November 20, 2023). Also, this variant has not been registered in ClinVar ( https://www.ncbi.nlm.nih.gov/clinvar / last accessed November 20, 2023). Different missense changes, p.Gly1001Cys, p.Gly731Ala, and p.Gly731Val, of the same amino acid residue, have been reported as pathogenic on ClinVar. Multiple lines of computational evidence support a deleterious effect, and the patient’s phenotypes are highly specific for a disease. These variants are judged to be likely pathogenic according to ACMG/AMP Guidelines. All detected pathogenic variants in this study were registered in ClinVar, with the submission ID SCV004098662-004098668.

Results As a result, vLAS was implemented extremely efficiently. Long-range PCR amplification products of COL1A1, COL1A2, and COL2A1 were able to be stably obtained from all patient samples. The IGV image of sequencing results showed that the genomic region of each gene was covered extremely uniformly including all over 50 exons (Figure 1a ). Pathogenic variants were detected in all samples, two of which were novel variants (Figure 1b , Table 2 ).

Discussion With advances in DNA sequencing technology, WES and whole genome sequencing (WGS) have become increasingly available. However, in clinical genetic practice, there is often a need for genetic testing focused on individual diseases rather than a comprehensive analysis. Nevertheless, there are limited applications that address this issue, leading to a preference for WES in certain scenarios, including when the target gene has a large number of exons, there are multiple target genes, or the target disease is rare and no specific test is available. The declining cost of WES has further fueled this trend. While capture sequencing methods targeting coding exons are widely used to test individual genes by NGS, the process of setting up new probes is expensive and requires some effort. In addition, the construction of diagnostic panels for specific disease categories makes it daunting to make changes or add new genes as needed. vLAS efficiently addresses these difficulties especially when the number of genes to be tested is limited, and also provides several advantages. First, when analyzing a new gene, genetic testing can be freely performed by simply designing a long PCR primer set for any gene. Second, unlike the capture probe method, there is no hybridization process, so library preparation is easy, there is no off-target generation, and comprehensive analysis including intron regions can be performed [ 12 ]. Additionally, vLAS enables the detection of the breakpoint sequences caused by large intragenic deletion [ 10 ]. In cases where the gene exhibits mRNA expression in blood, splicing abnormalities due to deep intron variants can also be detected using long-range reverse transcribed PCR (RT-PCR) libraries [ 10 , 11 ]. Nano Kit v2 (500 cycles) reagent is sufficient to run 25 samples with different genes at 100 kb per sample. The average sequence depth is more than 200 and the total cost per sample is less than US$25 including DNA extraction and library preparation [ 12 ]. Although vLAS is particularly useful for genes that have a large number of exons in a small region, it can be applied to any gene. In actual clinical practice, testing for specific genes is often required based on clinical diagnosis rather than comprehensive genetic analysis. In such cases, vLAS is a promising option if no test system has been established.

Conclusions Streamlining the genetic diagnosis of type I and type II collagenopathies has been achieved with vLAS. vLAS is extremely useful for genetic diagnosis when the target genes can be narrowed down from the clinical diagnosis. It is easy to set up a new test system, and it is possible to detect a wider range of mutations at a low cost. Because of its versatility, vLAS can also be used to diagnose other genetic diseases.

In the practice of clinical genetics, gene testing is usually guided by clinical diagnosis. When dealing with rare diseases, it is often necessary to create new test systems. The handling of a gene with a substantial number of exons poses a challenge both in sequential Sanger sequencing for each exon, and in the setup of capture probes to each exon for next-generation sequencing (NGS). We present very long amplicon sequencing (vLAS), an optimized long-range polymerase chain reaction (PCR)-based NGS method that overcomes this challenge. By utilizing approximately 20 Kb long PCR products and short-read NGS, vLAS is emerging as a highly adaptable and effective solution, especially for genes with numerous exons concentrated in a limited genomic region. Here, we demonstrate vLAS in the analysis of five patients with type I and two with type II collagenopathies. The integration of user-friendly NGS methods into genetic diagnosis enhances the practicality of clinical genetics.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50482

oa_package/f0/b1/PMC10788244.tar.gz

PMC10788245

38226076

Introduction Being overweight and obesity are defined as abnormal or excessive fat accumulation that presents a serious health risk. A body mass index (BMI) of 25 or more is considered overweight, while 30 or more is obese [ 1 ]. Within the past several decades, there has been a significant increase in the prevalence of obesity [ 1 , 2 ]. Obesity is a major public health concern that is associated with numerous health problems. Further, it has been proven not to be attributable to a sole factor, but rather to complex interactions among multiple factors [ 3 , 4 ]. As it is a major concern that affects the world’s population, there is no doubt that techniques for every individual to manage it should be investigated. Particularly, these techniques can be divided into two major categories: non-surgical (natural and medicinal) and surgical [ 2 , 5 , 6 ]. There are multiple methods in the non-surgical approach, natural (diet and/or exercise), or medicinal [ 2 , 5 , 6 ]. While these methods are less invasive, a major role in their success depends upon the willingness and determination of the individuals who opt for them [ 2 ]. As diets require either reducing caloric intake or portion sizes primarily and other diets focus on a particular nutrient (i.e., keto), they do not necessarily lead to sustained weight loss and require close follow-ups [ 2 , 5 , 6 ]. With respect to exercise, physical activity in general has multiple positive effects on the body and mental health of those who practice it. An eight-year weight loss study, the Look AHEAD study, was performed to assess the effects of intentional weight loss in adults with type 2 diabetes. Its results showed that, of those who engaged in intense lifestyle modifications, 34.5% had achieved a weight loss of 10% or more in the first year, and of these participants, 39.3% had maintained the 10% or more loss at year eight [ 7 ]. With respect to the medicinal approach, anti-obesity medications have been recommended in recent years for patients who are classified as obese or have had obesity-related complications. However, the weight loss is sustained only for as long as the medication is used [ 2 , 8 ]. Another method, the surgical approach, is also a viable option. Ongoing studies and published research have demonstrated that it is a safer procedure overall, with better effects, which has made surgeons more comfortable with recommending this method of treating obesity [ 2 , 5 ]. However, from a medical standpoint, although a number of studies have proven this surgery’s efficacy, patients who undergo this life-changing procedure may face the stigma surrounding it [ 5 , 9 , 10 ]. This causes those who may benefit more from this surgery, and perhaps need it, to avoid the surgery altogether, and causes those who have undergone it to feel bad after they face such situations post-surgery. The goal of this study is to explore the stigma associated with bariatric surgery in the Al-Qassim Region because it is crucial to understand the prevalence of these stigmas among the public and the effect it has on both those who have undergone the surgery and those who are considering it as an option. Further, this can help break down barriers and increase access to life-changing treatments for those who need them.

Materials and methods The study employed a cross-sectional design, utilizing a self-administered questionnaire for data collection. The research was conducted in the Al-Qassim Region over a period of 12 months. To determine the sample size, Yamane’s Formula was applied, considering a target population of 1.52 million persons and a 5% margin of error, resulting in a required sample size of 399. Convenience sampling was chosen as the sampling technique, allowing for the quick and efficient collection of data from participants meeting the inclusion criteria: adults aged 18 or older residing in the Al-Qassim Region. Exclusion criteria comprised individuals under 18 years old and non-residents in the Al-Qassim Region. The data collection involved the distribution of a self-administered questionnaire through online platforms, such as social media and forums. The questionnaire comprises multiple sections addressing demographics, perceptions of obesity, and attitudes toward bariatric surgery. Participants are queried about their beliefs regarding the causes of obesity, societal views on individuals who undergo bariatric surgery, and their personal experiences after the operation. The questionnaire further explores concerns about public opinion influencing the decision to undergo surgery and opinions on the effectiveness of bariatric surgery compared to other weight loss methods. Additionally, participants are asked to share their thoughts on societal support and resources for individuals who have undergone bariatric surgery. A pilot study was conducted with 20 adult participants from the Al-Qassim Region to assess the feasibility of the study design and data collection tools. The questionnaire, comprising demographic information and participants’ perceptions of weight loss surgery, was deemed clear and concise. No issues arose during data collection, affirming the feasibility of the study design. Based on the pilot study's results, minor modifications were implemented before the main study to enhance accuracy and reliability. The results of the pilot study were omitted from the study sample. In the main study, data analysis was performed using Statistical Product and Service Solutions (SPSS, version 26) (IBM SPSS Statistics for Windows, Armonk, NY). Categorical variables were presented as frequencies and percentages, and a chi-square test was employed to compare participants' perceptions regarding weight loss surgeries and demographic data. A p-value of < 0.05 indicated statistical significance in the analysis of participants' thoughts about weight loss surgeries and related demographic factors and then further displayed as tables or pie charts, accordingly.

Results The participants’ sociodemographic data A total of 988 participants agreed to participate in the study, and their sociodemographic characteristics are presented in Table 1 . Specifically, 61.20% (n=605) of the participants were females and had a university education 69.9% (n=691). With respect to age groups (missing 65 values), the largest proportion (43.7%, n=403) of the participants were within the age group of 18-24. With respect to BMI (missing 36 values), the most common category (43.5%, n=414) was 18.5-24.9 (Figure 1 ). Knowledge of bariatric surgery Knowledge about bariatric surgery is presented in Table 2 . The majority of the participants (57.30%, n=566) agreed strongly that obesity is a disease, while 30.50% (n=301) agreed, and 9.30% (n=92) remained neutral. With respect to genetic factors’ role in causing obesity, a substantial proportion of them (38.80%, n=383) strongly agreed, and 43.90% (n=434) agreed that genetic factors are one of the causes of obesity. When asked about the factors that increase the risk of obesity, the options selected most commonly were “idle and lazy life” 76.50% (n=756) and “eating too much” 75.60% (n=747). A significant proportion of the participants also indicated other factors that were associated with obesity, such as genetic factors (68.40%, n=676), stress and tension (34.60%, n=342), and psychological diseases (39.00%, n=385). With respect to perceptions about obesity, a notable proportion of them disagreed or strongly disagreed that most obese people are lazy (24.80%, n=245) and that excessive eating is the primary reason that most individuals suffer from obesity (14.00%, n=138). With respect to blame for obesity, a majority of the participants disagreed (28.00%, n=277) and strongly disagreed (17.70%, n=175) that people who are overweight or obese should be blamed for their condition. The participants were also asked to choose the tips that they believe should be offered to individuals who are overweight or obese. The options selected most commonly were “reducing the amounts of food” (69.80%, n=690), having the motivations and will (67.10, n=663), and “prevent sugars” (52.90%, n=523) (Table 2 ). Importantly, less than half of them (44.43%, n=439), responded that they had never thought about treating obesity through surgical operations, while 9.62% (n=95) had considered it, and 3.74% (n=37) had actually undergone the operation (Figure 2 ). The participants’ experience after weight loss surgeries among those who underwent the procedure The respondents who underwent weight loss surgeries (3.74%, n=37) were asked about their experiences after they underwent the operations, and 43.20% (n=16) of them reported that they faced critical comments or poor treatment from the community because of the bariatric surgery they had. With respect to feeling ashamed or embarrassed to disclose their surgery, 35.10% (n=13) of the participants answered “yes”, while the majority 64.90% (n=24) answered “no.” In addition, 37.80% (n=14) of them reported that they avoid social situations or events because of critical comments or poor treatment received from the community, while 62.20% (n=23) stated that they did not avoid such situations. They were also asked about specific comments that they had received about their surgery. The comments reported most commonly were “You have taken the easy way out instead of adopting a healthy lifestyle,” and “Why didn’t you try to go on a diet?” (51.40%, n=19) of the respondents reported both. Other comments mentioned frequently included “Why didn’t you try exercising instead of surgery?” (43.20%, n=16) and “After the operation, a lot of loose skin will result” (48.60%, n=18). It is worth noting that 21.60% (n=8) of the participants reported that they had not received any critical comments. With respect to the support and resources that would be helpful for individuals who have undergone bariatric surgery, the majority of them 48.60% (n=18) mentioned the importance of receiving advice and access to community educational sources or resources. (45.90%, n=17) of the respondents also considered that support groups that consisted of individuals who underwent the same experience were beneficial. Volunteer campaigns and “other” resources were mentioned by smaller proportions of the participants (32.40%, n=12, and 2.70%, n=1, respectively). Further, they were asked about their opinions on the effect of increasing public awareness and education about obesity treatment and its benefits in reducing misconceptions. A significant majority (62.20%, n=23) strongly agreed that increasing public awareness and education can help reduce inaccurate impressions, while 21.60% (n=8) agreed, and 10.80% (n=4) remained neutral (Table 3 ). Intentions to undergo weight loss surgery The participants’ intentions to undergo weight loss surgery and their concerns related to public opinion and community treatment were assessed (Table 4 ). A significant proportion of the participants disagreed (30.1%, n=28), and slightly fewer strongly agreed (25.80%, n=24) that concerns about public opinion or community treatment could affect their decision to undergo surgery. Moreover, 41.10% (n=39) of the participants disagreed, while 17.90% (n=17) strongly agreed that society holds a negative attitude toward individuals who have undergone obesity treatment. When asked whether they had ever avoided telling people that they were considering the surgery because of potential critical reactions, 42.10% (n=40) of them responded affirmatively, while the majority (57.90%, n=55) indicated that they had not avoided disclosing this information. The participants were also asked whether those around them had motivated them to have the surgery. The majority (36.80%, n=35) responded neutrally, while 27.40% (n=26) agreed, and 9.50% (n=9) strongly agreed that those around them had motivated them to undergo weight loss surgery. With respect to public opinion’s effect on their decision to undergo surgery, 29.50% (n=28) of the participants agreed, and 21.10% (n=20) strongly agreed that public opinion can make the decision about having the operation to treat obesity difficult. The study also investigated the way that public opinion about bariatric surgery affects individuals who are considering it. The effects reported most commonly were fear of others’ critical reactions (46.30%, n=44) and low self-esteem and self-confidence (31.60%, n=30). The majority of the participants answered neutral (36.80%, n=35) with respect to whether those around them were motivating them to have the operation, followed by agreement with this statement (27.40%, n=26) (Table 4 ). Perceptions with respect to operations to treat obesity The participants’ perceptions of operations to treat obesity were assessed through various statements. With respect to the belief that most individuals who underwent the operation as a treatment for obesity failed to follow a healthy diet and exercise thereafter, the majority of the participants (32.20%, n=318) responded neutrally, followed by 30.20% (n=298) who agreed with the statement. When asked about the perception that most bariatric surgeries are limited only to lazy individuals, a significant proportion of the participants (38.70%, n=382) disagreed, while 12.00% (n=119) strongly agreed with the statement. With respect to the belief that people who had bariatric surgery had other weight loss options available to them, 42.30% (n=418) agreed, and 21.00% (n=207) strongly agreed. With respect to the perception that bariatric surgery is the easiest way to lose weight compared to making efforts to follow a diet and exercise, 29.00% (n=287) agreed, while 21.60% (n=213) strongly agreed. When asked about the appropriateness for young men and young women to undergo obesity surgery, the majority of the participants (79.00%, n=781) believed that it was inappropriate except in severe cases, while 7.90% (n=78) thought that it was unsuitable in all cases. A majority of the participants (47.40%, n=468) agreed, and 24.10% (n=238) strongly agreed and acknowledged the importance of addressing public opinion of bariatric surgery, while a smaller proportion disagreed or strongly disagreed (5.50%, n=54). The participants’ views on whether bariatric surgery is cosmetic surgery varied, with 33.90% (n=335) disagreeing, 26.90% (n=266) responding neutrally, and 20.30% (n=201) agreeing with the statement. With respect to the perception of the quality of life of obese people who underwent bariatric surgery, a similar proportion of the participants agreed (32.90%, n=325) and answered neutral (32.60%, n=322), while 17.60% (n=174) strongly agreed with the statement. The majority of them (37.10%, n=367) agreed that those who underwent bariatric surgery should be congratulated. With respect to the belief that health insurance should cover the cost of the operation to facilitate access, 35.30% (n=349) agreed, 30.20% (n=298) strongly agreed, and 9.20% (n=91) disagreed or strongly disagreed (Table 5 ). Factors associated with the participants’ attitudes and practice toward weight-loss surgeries Table 6 presents the association between the participants’ demographic characteristics and their attitudes toward weight-loss surgeries. Among the variables examined, only two showed a significant association with the participants’ attitudes. Age was found to be associated significantly with the participants’ thoughts about treating obesity through surgical operations (p<0.001). Specifically, participants aged 46-65 years (7.1%, n=11) and 25-35 years (5.5%, n=10) underwent weight loss surgeries significantly more often compared to other age categories (2.7%, 2.2%, and 0.0% (n=11, 4, 0) among those aged 18-24, 36-45 years, and > 66 years, respectively. Significantly more participants with higher BMI levels, in which 5.2% (n=12) and 5.1% (n=12) were overweight and obese, respectively, had undergone surgery compared to other categories (3.1%, n=13; among a normal BMI and 0.0% among underweight; p<0.0001; Table 6 ). The association between the participants’ perception of the operation to treat obesity and their attitudes toward weight-loss surgeries Table 7 presents the association between the participants’ perception of the operation to treat obesity and their attitudes toward weight-loss surgeries. The highest percentage (20.8%, n=5) of the participants strongly disagreed with the statement, “Most of those who underwent the operation as a treatment for obesity failed to follow a healthy diet and exercise” had undergone the operation before (p<0.001). Similarly, the highest percentage (4.7%, n=5) of the participants who strongly disagreed with the statement “Most bariatric surgeries are only limited to lazy individuals” had undergone the operation before (p<0.001). In addition, the highest percentage (14.0%, n=18) of the participants who disagreed with the statement “It is wrong for young men/young women to undergo obesity surgery” had undergone the operation before (p<0.001). Moreover, the highest percentage (5.8%, n=6) of the participants who strongly disagreed with the statement “Bariatric surgery is a cosmetic surgery” had undergone the operation before (p<0.001). In contrast, the lowest percentage of the participants (0.0%, n=0) who strongly disagreed with the statement “Bariatric surgery is the easiest way to lose weight instead of making efforts to follow a diet and exercise” had undergone the operation before (p<0.001). Similarly, the lowest percentage (0.0%, n=0) of the participants who strongly disagreed with the statement “It is important to address public opinion with respect to bariatric surgery” had undergone the operation before (p<0.001) and the lowest percentage (0.0%, n=0) of the participants who strongly disagreed with the statement “The quality of life of obese people who underwent bariatric surgery is likely to be better than that of obese people who did not undergo the operation” had undergone the operation before (p<0.001, Table 7 ). The association between the participants’ perception of the operation to treat obesity and sociodemographic data Tables 8 - 9 present the association between the participants’ perception of the operation to treat obesity and sociodemographic data. A statistically significant difference was observed in gender (p<0.001). The median value of the parameter for males was found to be 32.0 (IQR: 29.0-35.0), which was higher than for females. The participants were categorized based on their educational level, and the median values were higher in middle school (31.0, IQR: 29.0-35.0), university (31.0, IQR: 27.0-34.0), and post-graduate (31.0, IQR: 29.0-35.0). A statistically significant difference was observed among the educational levels (p=0.016). According to age groups (p<0.001), the median value for the age group ≥ 66 was 38.0 (IQR: 34.0-43.5) and was the highest median among the age groups. With respect to BMI categories, the median value for category 30 or more was 33.0 (IQR: 28.0-35.0), and there was a statistically significant difference among the BMI categories (p<0.001). The effect of the male participants’ perception of the operations to treat obesity was significantly stronger compared to that of females (beta=1.910, 95% CI: 1.290-2.531, p<0.001). The participants in the age groups 18-24 (beta=-7.563, 95% CI: -11.985 to -3.141, p=0.001), 25-35 (beta=-5.005, 95% CI: -9.442 to -0.569, p=0.027), 36-45 (beta=-5.646, 95% CI: -10.079 to -1.212, p=0.013), and 46-65 (beta=-5.642, 95% CI: -10.075 to -1.209, p=0.013) showed a significantly weaker effect on the perception of the operations to treat obesity compared to the reference group (≥66 years). The participants with a BMI of ≤18.49 (beta=-2.139, 95% CI: -3.513 to -0.765, p=0.002), 18.5-24.9 (beta=-1.142, 95% CI: -1.937 to -0.346, p=0.005), and 25-29.9 (beta=-1.437, 95% CI: -2.262 to -0.613, p=0.001) exhibited a significantly lower effect on the parameter compared to those with a BMI of 30 or more (Tables 8 - 9 ).

Discussion Obesity is a major public health concern that is associated with numerous health problems, including diabetes, heart disease, and certain types of cancer [ 11 ]. Locally, obesity in Saudi Arabia is sitting at 24%; however, when compared to previous years, this number is lower [ 12 ]. As for why this is happening, there is no justified answer as of yet, but there are some speculations that it may be as a result of policies, which encourage a healthier lifestyle [ 12 ]. Although the use of bariatric surgery is only one of the options available to manage obesity and associated illnesses, it remains the most effective [ 8 ]. However, the pervasive stigma associated with bariatric surgery may limit the candidate’s willingness to undergo the operation and have an effect on patients’ well-being both before and after the surgery. Therefore, it is crucial to carry out a study on these stigma’s nature and effects. To our knowledge, no single study in Saudi Arabia has been conducted to assess the effect of public stigmatization related to bariatric surgeries on both those who have undergone the surgery and those who are considering it as an option. Further, in our study, we assessed the public’s knowledge, attitude, and prevalence of obesity among males/females in the Al-Qassim Region. The participants’ sociodemographic data In this study of Saudi adults in the Al-Qassim Region, the BMI of nearly 43.50% (n=414) of the respondents was within the normal range, while 49.1% (n=467) were classified as overweight or obese. This prevalence is consistent with that of another study in the region that found that the prevalence of obesity and being overweight was 53.5% [ 13 ]. Further, it is consistently compared to other studies that have shown that the prevalence of obesity and being overweight were 57.7%, 40%, and 54%, respectively, among adults across Saudi Arabia [ 14 - 16 ]. Knowledge about bariatric surgery In this study, we observed that the participants had an appropriate level of knowledge about obesity. For example, 87.8% (n=867) of them strongly agreed or agreed that obesity is a disease, and 82.7% (n=817) believed that genetic factors may play a large role in obesity. Additionally, 76.50% (n=756) of them believe that the factors of an idle and lazy life, eating too much (75.60%, n=747), disease of all kinds (41.70%, n=412), stress and tension (34.60%, n=342), and psychological diseases (39.00%, n=385) can increase the risk of obesity. In comparison, our finding is consistent with those in local studies in Riyadh City published in January 2018 and in the Al-Qassim Region published in February 2019, in which the authors found that most of the respondents have a high level of knowledge about obesity’s causes and risk factors [ 13 , 16 ]. Another national study showed that the stereotype perceived most commonly was that obese individuals are lazy, which 62% of their sample endorsed [ 17 ]. In our study, with respect to blame for obesity, a majority of the participants disagreed (28.00%, n=277) and strongly disagreed (17.70%, n=175) that people who are overweight or obese should be blamed for their condition. In contrast to another study in which the participants were asked whether others had ever blamed them for their weight problems, the majority (66.4%) responded that they had not by any means [ 18 ]. Importantly, less than half of the participants (44.43%, n=439) responded that they had never considered treating obesity through surgical operations, 9.62% (n=95) had considered it, and 3.74% (n=37) had actually undergone the operation. The participants’ experience after weight-loss surgeries In this study, the patients who underwent bariatric surgery (3.74%, n=37) were asked about the weight stigma and their experiences after the surgery. Moreover, 43.20% (n=16) stated that they were exposed to unpleasant comments or poor treatment from the community, and 35.1% (n=13) felt ashamed or embarrassed to disclose their surgery. “You have taken the easy way out instead of adopting a healthy lifestyle” and “Why didn’t you try to go on a diet?” were the most common comments that they received from the community (51.4%, n=19). This result is consistent with that of another study conducted in Brazil [ 9 ]. Further, in this study, most of the participants agreed about the importance of receiving advice and access to community educational sources or resources (48.6%, n=18) to support those who had undergone the surgery. This is consistent with the evidence in a study conducted to investigate social support’s effect on mental health and eating disorders after the surgery was performed and showed that greater social support was associated with lower depression [ 19 ]. Intentions to undergo weight-loss surgery When we investigated what may affect a patient’s decision, their intention whether or not to opt for weight-loss surgery, and their concerns about the matter pre- or post-surgery, we found in our demographic survey that public concern with respect to the surgery had an overall mixed opinion, indicating that it may not be necessary to consider it. In addition, it was seen that society’s view toward those who did undergo the surgery does not seem to be critical overall. As per the other points assessed as well, such as whether some had avoided telling others that they underwent the surgery, or whether those around them played a part in their decision to have it or not, nearly equal numbers tended to be neutral. However, the majority of the participants agreed that the public’s perception of the surgery may make it more difficult for someone to undergo the surgery, albeit they said earlier that public opinion was not much of an issue. When considering if public opinion had affected an individual’s opinion regarding undergoing the surgery, "fear of negative reactions from others" had been predominantly chosen, followed by "low self-esteem and self-confidence," similarly pointed out in a systemic review, in a qualitative study, that people may try to hide their weight loss or the fact that they have undergone surgery as a means of avoiding the stigma and negative comments, which may be directed towards them [ 20 ]. Perceptions of surgery to treat obesity In our study, we found that the majority of the participants tended to believe that individuals who had undergone bariatric surgery to treat obesity had failed to follow a healthy diet and exercise subsequently. This was confirmed further when, again, the majority of them said that those who underwent the procedure had other options available to lose weight and that surgery was the easiest way, which was consistent with the results of other studies [ 10 , 21 , 22 ]. While there were mixed opinions in our study when the participants were asked whether bariatric surgery was considered cosmetic, rather than therapeutic, one of the largest studies related to bariatric surgery’s effect on obesity and its outcomes, the Swedish Obesity Study, has concluded that "bariatric surgery for severe obesity is associated with long-term weight loss and decreased overall mortality," which emphasizes the therapeutic side of the surgery. The study compared groups of patients who had undergone the surgical procedure to control their weight change/loss. Moreover, the group who had undergone the procedure for weight loss had a far more positive outcome and/or change in weight loss and any obesity-related outcomes, with the added bonus that it even helped those who had diabetes control their disease better [ 23 ]. Nevertheless, our sample also showed that many agreed that the quality of life of those who do undergo the surgery is more likely to be better than those who have not and deemed that health insurance should still cover the procedure’s cost. Factors associated with the participants’ attitudes and practice toward weight-loss surgeries It was also noticed that, when we investigated whether there were any factors that affected the way that the surgery was viewed, two variables had a noticeable effect on their attitude and practice toward bariatric surgery. Those variables were age and BMI and showed that individuals who were older or had a higher BMI were more likely to have undergone a weight-loss procedure already. Unexpectedly, gender did not cause a significant change or effect. The association between the participants’ perception of the operations to treat obesity and their attitudes toward weight-loss surgeries We examined the association between the participants’ perceptions of operations to treat obesity and their attitudes toward these surgeries in Saudi Arabia. The participants who strongly disagreed with negative statements about bariatric surgery (e.g., being lazy, taking the easy way out, or your quality life would be better if you did not undergo the surgery) had a significantly higher number of prior surgical experiences, indicating a positive correlation between favorable attitudes toward surgery and prior surgical history. These findings are consistent with those in previous research conducted among the Saudi adult population, in which a subset of individuals showed more positive attitudes toward bariatric surgery although the majority preferred a proper diet for weight loss [ 13 ]. This suggests that individuals who reject the prevailing negative perceptions and stereotypes associated with weight-loss surgeries may be more inclined to consider bariatric surgery as a viable option to address morbid obesity. These results emphasize the potential effect of efforts to reduce stigma in promoting informed decision-making and increasing eligible candidates’ uptake of bariatric surgery. These findings are reinforced when the strong desire of individuals who have undergone the surgery to increase public awareness and address the stigma associated with this procedure, which had the lowest disagreement 0.0% (n=0), among them (p<0.001), is taken into consideration. In addition, our study revealed that the participants who undergo the operation face surgery-related stigma, which is supported by a study conducted in a bariatric clinic in the American Southwest that found that people who have bariatric surgery for weight loss may trade one type of stigma for another [ 24 ]. Thus, individuals who qualify for bariatric surgery based on weight alone may be reluctant to explore the surgery as a viable option. Further, these findings are consistent with those in another study that emphasized the importance of addressing weight- and surgery-related stigma at a larger level, stating that societal perceptions’ effect extends beyond individuals who are considering bariatric surgery; it also affects patients’ experiences post-surgery [ 25 ]. Our findings supported this further by showing that the participants with positive attitudes toward bariatric surgery were more likely to have undergone the operation [ 13 , 25 ]. This highlights the need for healthcare providers to create a supportive and non-stigmatizing environment to enhance post-surgical outcomes. The association between the participants’ perception of the operations to treat obesity and sociodemographic data The significant association between the participants’ perception of obesity treatment operations and various sociodemographic factors, particularly gender, sheds light on the potential gender-specific disparities in attitudes toward weight-loss surgeries. Our study revealed that the male participants’ attitudes had a greater effect on their perception compared to females, suggesting a distinct perspective or awareness with respect to such surgical interventions. However, Aly et al.’s previous study, which explored demographic and quality of life factors that influence patients’ consideration of bariatric surgery reported that a higher proportion of women (40%) had considered weight-loss surgery compared to men (22%), and women’s physicians were more likely to recommend the surgery (22% vs. 14%) [ 26 ]. These gender-based disparities indicate that gender-specific societal norms and expectations may influence perceptions and motivations related to weight-loss surgeries. Understanding such nuances is crucial for healthcare providers to develop tailored approaches to support patients’ decision-making processes and address potential barriers to seeking weight-loss surgeries. By recognizing the effect of sociodemographic factors, including gender, on attitudes toward bariatric surgery, healthcare professionals can promote more equitable access to, and use of, this effective treatment option for obesity. Our findings revealed a significant association between the participants’ perceptions of obesity treatment operations and their educational level. Notably, the participants with higher education, such as university and post-graduate, exhibited similar median values in their perceptions (31.0), suggesting that individuals with more education may possess a better understanding of weight-loss surgeries. Education’s influence on attitudes toward weight-loss surgeries was supported by a separate investigation on laparoscopic sleeve gastrectomy (LSG) treatment. The study implemented three education sessions for one group of patients after LSG, while another control group received only written recommendations in the form of a guidebook. The results demonstrated that receiving education sessions had a significant effect on the study group’s weight loss and adherence to lifestyle recommendations compared to the control group (p<0.001). Moreover, the control group exhibited less motivation to comply with recommendations after a year of observation [ 27 ]. These findings underscore education’s importance in promoting successful weight loss and lifestyle changes after bariatric surgery and suggest that integrating education as a permanent element of the LSG procedure can enhance its effectiveness in treating morbid obesity. Moreover, the findings related to BMI categories showed a statistically significant difference (p<0.001). The participants with a BMI of 30 or more had a higher median value (33.0, IQR: 28.0-35.0) in their perception of obesity treatment operations. This suggests that individuals with a higher BMI may have a different and more positive attitude or opinion toward weight-loss surgeries compared to those with a lower BMI. Overall, the study highlighted gender, age, educational level, and BMI’s influence on the participants’ perception of obesity treatment operations. These sociodemographic factors should be taken into account when designing educational interventions and public awareness campaigns to improve the population’s understanding and acceptance of weight-loss surgeries. Limitations Despite the valuable insights gained from this study, it is important to acknowledge its limitations. Firstly, this research was conducted in the Al-Qassim Region, and, as such, the findings may not be representative of the broader population in Saudi Arabia or other Regions. In addition, the research relied primarily on self-reported data from the participants, which can be subject to recall and social desirability biases. The study’s cross-sectional design provides only a snapshot of attitudes and perceptions at a specific point in time, which makes it impossible to establish causal relationships or track changes over time. Finally, the research focused primarily on the general public’s perspective and did not delve into healthcare professionals’ experiences and perspectives, which could provide a more comprehensive understanding of the issue. These limitations should be taken into account when interpreting the findings and can serve as avenues for further research in the future.

Conclusions In conclusion, this study underscored the pressing need to address the stigma and misconceptions associated with bariatric weight loss surgery in the Al-Qassim Region. Societal attitudes’ effect on individuals who have undergone these procedures is evident and leads to feelings of shame and hesitation among those considering such treatments. Several recommendations emerged from this research to improve these individuals’ quality of life and ensure equitable access to life-changing bariatric surgery. Firstly, comprehensive public awareness campaigns should be launched to dispel myths and misunderstandings about bariatric surgery and educate the public about its benefits and medical necessity for many individuals. Secondly, healthcare professionals must play a pivotal role in this endeavor by engaging in proactive patient education and advocacy, challenging stigmatizing beliefs, and providing resources for those who have undergone the surgery. Moreover, the development of support networks, including support groups and community educational resources, is essential to help individuals navigate the post-surgery journey and combat societal bias’ adverse consequences. By implementing these recommendations, it is possible to create a more informed, empathetic, and supportive environment for individuals considering or recovering from bariatric surgery in the Al-Qassim Region and ultimately improve their physical and mental well-being.

Background Obesity is defined as abnormal or excessive fat accumulation that presents a serious health risk and is a major public health concern. Obesity prevention and management require evidence-based strategies that emphasize diet and physical activity. Bariatric surgery is also a life-changing procedure that can improve physical and mental health, but the stigma associated with it can prevent people from seeking treatment and affect their lives adversely. Studies have shown that bariatric surgery patients face discrimination from the public and healthcare professionals, which can lead to adverse psychological outcomes and hinder access to quality care. Goals and methods This study intends to explore the stigma related to bariatric surgery in Al-Qassim Region, Saudi Arabia, because it is crucial to understand its prevalence among the public and the influence it has on both those who have undergone the surgery and those who are considering it as an option. The participants had to complete an online questionnaire, comprised a general section and other sections based on whether or not the individual has, has not, or is considering bariatric surgery. Results A total of 988 individuals, 605 of whom were female (61.2%), agreed to participate in the study. The most common body mass index (BMI) category was 18.5-24.9 (43.5%, n=414). The majority of the participants had either agreed or strongly agreed that obesity is a disease (87.8%, n=867) and that genetic factors play a role in causing it (38.8%, n=383). The factors selected most commonly that increase the risk of obesity were “idle and lazy life” (76.5%, n=756) and “eating too much” (75.6%, n=747). Fewer than half of the participants (44.43%, n=439) reported that they had never thought about treating obesity through surgical operations, 9.62% (n=95) had considered it, and 3.74% (n=37) had actually undergone the surgery. Among those who underwent weight loss surgery (n=37), 43.20% (n=16) reported that they received critical comments or poor treatment from the community, 35.10% (n=13) felt ashamed or embarrassed to disclose their surgery, and 37.80% (n=14) avoided social situations or events because of those comments or poor treatment. The comments reported most often were “You have taken the easy way out instead of adopting a healthy lifestyle” (51.40%, n=19) and “Why didn’t you try to go on a diet?” (51.40%, n=19). Among those who have intentions to undergo weight loss surgery (n=95), a significant proportion of the participants (43%, n=40) agreed or strongly agreed that concerns about public opinion or community treatment could affect their decision to undergo weight loss surgery. Moreover, 32.6% (n=31) of them agreed or strongly agreed that society has a negative attitude toward individuals who have undergone obesity treatment. When asked whether they had ever avoided telling people that they were considering surgery because of potential adverse reactions, 42.10% (n=40) of the participants responded that they had. Conclusion This study helped bring attention to, and prove, the stigma related to bariatric surgery in Al-Qassim Region. Such stigma has prevented patients from seeking or undergoing a surgical option to manage their weight, even if it is the option recommended for them. As such, public education and awareness campaigns are encouraged to help reduce the stigma, as well as improve access to bariatric surgery for those who need it.

Appendices Questionnaire Demographic Characteristics Age: _________ Gender: o Male o Female Weight (kg):_____________ Height (cm):_____________ Education Level: o Elementary o Middle School o High School o College (Non-health Major) o College (Health Major) o Post-graduate o Other:_____ 1 - Obesity Obesity is considered a disease. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Genetic factors are considered one of the causes of obesity. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Which of the following increases the risk of obesity? (Select all that apply). ○ Hereditary ○ Inactive/Sedentary Lifestyle ○ Eating too much ○ Illness (All kinds) ○ Stress ○ Psychological Diseases ○ Other: ______ Most obese people are lazy. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Excessive eating is the main cause of most people suffering from obesity. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree People who are overweight or obese should be blamed for their condition. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Choose the advice that you think should be given to people who are overweight or obese. ○ Reducing amount of food ○ Abstain from sugary food ○ Surgery to get rid of obesity ○ Through ONLY exercise, you will lose weight ○ Take diet pills ○ Reduce sleep, since excess sleep increases weight ○ You must have motivation and will ○ I wouldn’t suggest any of these tips ○ Other:______ Have you ever thought about treating obesity through surgical operations? ○ Yes ○ No ○ No, but I know someone who did ○ I’ve already undergone the procedure 2 - Experience after the operation Have you been exposed to negative comments or poor treatment from society because of the obesity surgery you underwent? ○ Yes ○ No If yes, what was your experience? (Optional) ______________________________________ Did you feel ashamed or embarrassed to tell others that you underwent obesity surgery? ○ Yes ○ No Have you ever avoided social situations or events because of negative comments or bad treatment from the society that was directed at you because of your obesity surgery? ○ Yes ○ No Have you been subjected to comments, such as: ○ I took the easy way out instead of adopting a healthy lifestyle. ○ Why don't you go on a diet? ○ Why didn't you try exercising instead of surgery? ○ After the procedure, a lot of sagging skin will result. ○ I have not been subjected to any negative comments. ○ You can live with obesity without resorting to operations. ○ Certainly there were other solutions to obesity, such as (exercise, extreme dieting, etc.). ○ Other: ______ People around you motivated you to have the operation. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree What type of support or resources do you think would be helpful for individuals who have undergone bariatric surgery to deal with any negative experiences or treatment from society? ○ Volunteer campaigns ○ Advice ○ Resources or educational resources for the community ○ Support groups (for those who have undergone the same experience) ○ Other: _________ Do you think that increasing public awareness and education about the procedure to treat obesity and its benefits can reduce the wrong impression? ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree 3 - Thinking of undergoing the surgery Concerns about public opinion or community treatment affect your decision to undergo surgery as a treatment for obesity. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Society takes a negative attitude towards individuals who have undergone obesity treatment surgery. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Have you ever avoided telling people that you are considering surgery because of potential negative reactions? ○ Yes ○ No Those around you were motivating you to undergo the operation. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Public opinion can make the decision about whether to undergo surgery to treat obesity difficult. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree How does public opinion about obesity surgery affect individuals considering it? ○ Fear of negative reactions from others ○ Fear of discrimination and prejudice ○ Low self-esteem and self-confidence ○ Other: ____________ 4 - Bariatric Surgery Most of those who had the procedure as a treatment for obesity failed to follow a healthy diet and exercise. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Most bariatric surgeries are limited only to lazy individuals. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree People who had bariatric surgery had other options available to them to lose weight. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Surgery to treat obesity is the easiest way to lose weight instead of putting in the effort by following a diet and exercising. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree It is wrong for young men/women to undergo surgery to treat obesity. ○ Yes (in all situations) ○ Yes (In some cases) ○ No It is important to address public opinion regarding bariatric surgery. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Bariatric surgery is a cosmetic surgery. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree The quality of life of people who are obese and have undergone bariatric surgery is likely to be better than that of people who are obese and have not had the procedure. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Those who underwent the surgery to treat obesity should be congratulated. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree Health insurance should cover the cost of the operation to facilitate access to it. ○ Strongly Disagree ○ Disagree ○ Neutral ○ Agree ○ Strongly Agree

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50477

oa_package/36/9b/PMC10788245.tar.gz

PMC10788246

38226073

Introduction Currarino syndrome (CS) is a rare disease with an incidence of approximately one in 100,000 [ 1 ], characterized by three clinical features: presacral mass, anorectal malformation, and sacral bone deformity. Some variants of this “Currarino triad” require not only pediatric surgery but also neurosurgery. The most common type of anorectal malformation associated with CS is anorectal stenosis [ 2 ], whose severity and extent vary. There are currently no established guidelines on CS treatment owing to the paucity of comprehensive studies of the disease [ 1 , 3 ]. We herein reviewed the clinical course, surgical methods, and outcomes of CS cases treated at our hospital and retrospectively suggested strategies for the treatment of CS associated with anorectal stenosis because there are a few previous reports [ 1 , 3 ] that clearly summarize the order of surgical treatment and determined surgical methods for anorectal stenosis according to clinical symptoms.

Materials and methods The present, retrospective review of the records of patients with Currarino syndrome (CS) who underwent surgery for the disease was approved by the Institutional Review Board of Tokyo Metropolitan Children’s Medical Center. The present study was approved by the Institutional Review Board (Research Ethics Review Board No. 2020b-21). Seven patients with CS with anorectal stenosis who were seen at our hospital between 1998 and 2021 were reviewed for the clinical detail, primary symptoms, histological findings of any presacral mass, medical history of intra-tumoral infection, presence of a fistula between the presacral mass and anorectum, level of sacral bone deformity, associated malformations, family history of CS, surgical method, order for treatments for anorectal stenosis, presacral mass, and spinal tethered cord, pathological findings of anorectal stenosis, and postoperative defecation function. The diagnosis of these patients with CS was performed based on a triad of sacral malformation, presacral tumor, and anorectal stenosis. All patients were diagnosed using ultrasonography (US) and magnetic resonance imaging (MRI). Computed tomography (CT) and contrasted enema were additionally performed for some patients for the evaluation of anorectal stenosis. Familial history included even in the cases of incomplete CS. Histological evaluation was evaluated with hematoxylin and eosin (H-E) staining and azan staining using the standard protocol. The presence of a fistula between the presacral mass and anorectum was evaluated from the histopathological findings in the presacral mass and specimens of anorectal stenosis.

Results Patient background All seven patients with CS with associated anorectal stenosis treated at our hospital were subadult females. Their median age at CS diagnosis was eight months (range: 1-18 months). The primary symptoms were abdominal distension in three patients, constipation in two patients, sacral skin depression and urinary tract infection in one patient, and anuresis in one patient. Pathological analysis of the presacral masses found a mature teratoma in six patients; a meningocele in three patients; and a lipoma, including duplication, in one patient. Three patients had a fistula communicating between the tumor and anorectum, and two of these three patients had a past medical history of intratumoral infection. All cases of sacral bone deformity were scimitar-type sacral bone defects, with three and four cases of S3- and S4-level sacral bone defects, respectively. Five cases of the tethered spinal cord were observed. Other, associated, congenital malformations were scoliosis or a giant ureter in one patient and an anterior anus in two patients. Table 1 summarizes the details of each patient. Neurosurgery for CS The order of neurosurgery for CS varied by patients, as seen in Table 1 . All five patients with tethered spinal cords underwent untethering surgery simultaneously or before surgery for anorectal stenosis and presacral mass resection. All three patients with a presacral meningocele also had an associated teratoma. Duraplasty was required in these patients to reduce the risk of meningitis because the presacral mass was complicated with a fistula communicating the intraspinal cavity with the anorectum. In two of the three patients, the operation was performed simultaneously with untethering surgery. Surgical treatment for the anorectal stenosis and presacral mass Only one patient (Case 1) required a colostomy for severe anorectal stenosis. Surgery was performed for anorectal stenosis in six patients, except the patient in Case 4, who underwent dilation with a metal bougie. The surgical method for anorectal stenosis treatment was end-to-end anastomosis with a partial resection, anorectal strictureplasty, pull-through, posterior sagittal anorectoplasty (PSARP), and cutback after presacral mass resection. In four patients, the surgery for anorectal stenosis was performed simultaneously with excision of the presacral mass. Surgical resection of the presacral mass was performed in all six patients with a mature teratoma. In Case 2, PSARP was performed because the anterior anus had a moderate degree of stenosis, and the presacral mass was diagnosed as a lipoma based on radiological findings and, therefore, did not require removal. In Case 5, stenosis of the anterior anus was initially diagnosed as mild stenosis not requiring surgery; however, a cutback procedure involving transanal excision of scar tissue was performed to relieve the stenosis after severe fecal embolization requiring disimpaction occurred after resection of the infected presacral mass. In Case 7, a pull-through procedure was performed for the anorectal stenosis according to the Soave-Denda method, which is typically performed for Hirschsprung’s disease. In Cases 6 and 7, a covering ileostomy was created simultaneously with the surgery for anorectal stenosis to reduce the risk of anastomotic leakage. Postoperative course and defecation/urinary function In terms of complications in the early postoperative period, two patients in Cases 5 and 6 had a deep surgical site infection; both had a history of a preoperative intratumoral infection, which was treated with lavage. The current, median age of the patients is seven years (range: 2-23 years), and the median follow-up period from the last surgery is five years (range: 1-21 years). In terms of long-term, postoperative defecation function, two patients required hospitalization for fecal impaction, and a cutback was performed additionally for the anorectal stenosis in Case 5, as described above. Except for Case 4, which was treated only with anal dilation with a metal bougie, glycerin enemas were required for defecation management owing to constipation, but none of the patients experienced incontinence. To restore postoperative urinary function, clean intermittent catheterization was begun for neurogenic bladder in the patient in Case 1 at age 15 years. The other patients did not experience any urinary issues. Pathological findings of anorectal stenosis in resected specimens Resected specimens of the anorectal stenosis were obtained from five patients (Cases 1, 2, 5, 6, and 7). Pathological analysis revealed disorganized and rough smooth muscle fibers and replacement of the stroma by an increased quantity of collagen fibers (Figures 1 -a, 1-b, 1-d, 1-e). Degeneration of the intestinal smooth muscle and increased fibrotic tissue were considered to be the cause of the anorectal stenosis in all the cases. Three patients (Cases 5, 6, 7) had a presacral mass with the fistula communicating the anorectum with the tumor, which histopathological analysis revealed to be lined with columnar or squamous epithelium surrounded by fibrous scar tissue (Figures 1 -c, 1-f).

Discussion CS is a secondary neural tube dysplasia in which the anterior wall of the spine is formed in the fourth week of embryonic development and the gastrointestinal tract and neural tube are separated. Currarino et al. [ 2 ] and several other authors [ 4 , 5 ] suggested that a neuro-enteric fistula is formed in the intestinal and neural tubes, resulting in a presacral mass with a fistula caused by the abnormal splitting or deviation of the notochord during the stage. Gupta et al. [ 6 ] also suggested that the failure of some epiblast cells to migrate from their primitive node may leave remnants in the primitive streak, which may develop into a teratoma in the sacrococcygeal region. A mutation in HLXB9 or MNX1 (formerly HLXB9) in the 7q36 gene has been suggested as the cause of CS. Ross et al. reported that HLXB9 is expressed in the anterior horn of the spinal cord and the lid plate during neural tube development. These abnormalities comprise the so-called “Currarino triad” [ 7 ]. However, Dworschak et al. [ 8 ] reported that other genes or regulatory regions may contribute to CS, and several cytogenetic studies have implicated further loci in the etiology of CS besides MNX1, based on the finding that mutation analysis detected likely MNX1 variants in only 57.4% of all CS patients. The question of CS etiology, therefore, remains moot. Because patients with CS present with various, associated malformations and complicated clinical symptoms, an individualized, multifaceted treatment approach is needed [ 9 ]. AbouZeid et al. reported 17 cases of CS and described an individualized treatment strategy to address the wide variation in complications [ 3 ]. In addition, Martucciello et al. created a diagnostic/therapeutic protocol based on their experience with six patients, but the treatment flow is diverse and complicated [ 1 ]. Anorectal stenosis is the most common type of anorectal malformation in CS, accounting for about 75% of complications [ 10 ]. The degree of anorectal stenosis varies from severe, which requires a colostomy in the neonatal period, to mild, which can be relieved by dilation using a metal bougie. Based on our experiences in our seven cases, we created a treatment flow chart for CS with anorectal stenosis (Figure 2 ). First, the severity of the anorectal stenosis should be assessed. If fecal impaction caused by severe anorectal stenosis is present in the neonatal period, a colostomy should be performed. Defecation in mild-to-moderate cases can be managed by dilation with a metal bougie, glycerin enemas, or a laxative. Next, malformations associated with CS, such as tethered spinal cord and presacral mass, should be assessed using radiological methods, such as magnetic resonance imaging (MRI), computed tomography (CT), or ultrasonography (US). If the presacral mass is infected owing to a fistula communicating with the anorectum, transanal or percutaneous drainage and antibacterial treatment should be performed simultaneously. Contrast enema examination and colonoscopy are also useful to evaluate the degree and extent of anorectal stenosis and may aid in the detection of any fistula communicating the presacral tumor with the anorectum (Figure 3 ). Reports of CS complicated with Hirschsprung's disease suggest that anorectal manometry and a rectal mucosal biopsy should also be performed if possible [ 11 , 12 ]. Further, as urinary function can deteriorate in the tethered spinal cord or through neurosurgical complications, a bladder function test should be performed pre- and postoperatively. In Case 1, she was diagnosed by urodynamics, and the intravesical pressure gradually increased from adolescence. A neurogenic bladder can be caused by tethering of the spinal cord, sacral agenesis, recurrence of a presacral mass, or injury to parasympathetic nerves during surgery; however, it was not clear to detect the reason for the neurogenic bladder in our case. If the presacral mass is diagnosed as a teratoma based on radiological findings, a complete surgical resection should be performed to prevent the development of a malignancy or abscess [ 13 ]. Determining the surgical indications for anorectal stenosis prior to the removal of the presacral mass is necessary because the surgery for anorectal stenosis can be performed simultaneously with presacral mass resection via the same posterior sagittal incision (Figure 4 ). Based on pathological analysis of the associated lesion, tumor resection may be insufficient to treat the anorectal stenosis, and surgery at least to the level of the intestinal muscular layer may be needed. Therefore, the procedures for presacral mass excision and stenosis correction should be performed simultaneously if physiological and radiological findings suggest that the stenosis is likely to affect defecation. If the presacral mass is a meningocele, duraplasty and untethering of the tethered spinal cord should be performed first to reduce the risk of meningitis, as the surgical field can become contaminated when the intrarectal lumen is opened to relieve the stenosis in certain procedures. Whether the surgical procedures for anorectal stenosis and neurosurgery should be sequenced or performed simultaneously is debatable. Given the potential for infection and issues in postoperative management, sequential treatment may be preferable [ 14 ]. However, some studies have reported that both procedures can safely be performed simultaneously [ 15 , 16 ]. Various surgical procedures for anorectal stenosis have been reported, and no standard surgical procedure has yet been established. AbouZeid et al. described PSARP for anorectal stenosis in CS to mobilize the rectum in the entire circumference by dividing the muscle complex [ 2 ]. On the other hand, Hamrick et al. reported a surgical technique for preserving the anterior anorectal wall by limiting mobilization to the posterior wall half circumference of the anorectal stenosis. This procedure is advantageous for postoperative defecation management because it preserves the sensory nerves near the dentate line [ 17 ]. Additional surgery should be considered in patients who experience difficulty with defecation management even after the presacral mass has been removed via the posterior sagittal incision. Given the possibility of postoperative adhesion, using the transanal approach, such as the cutback procedure used in Case 5, which can improve the stenosis without using a posterior sagittal incision, may be preferable. Laparoscopically assisted pull-through surgery has been applied to surgery for anorectal stenosis and is suitable for patients who have previously undergone presacral mass removal [ 18 ]. Some patients requiring surgery for segmental dilation of the colon resulting from chronic constipation associated with CS in the distant postoperative period have also been reported [ 19 ]. The present case series contained three cases of presacral mass complicated with a fistula between the mass and the anorectum. The histopathological findings of these cases revealed that the location of the anorectal stenosis and the fistula was approximately the same, suggesting that the formation of the presacral mass may be closely related to the anorectal stenosis. Additionally, the anorectal stenosis is more likely to be a congenital, rather than an acquired, abnormality because inflammatory findings were not present in all the cases. The intestinal smooth muscle was disorganized and rough, and the stroma was replaced by an increased quantity of collagen fibers. The fibrous scar tissue may thus lead to anorectal stenosis in CS. The present case series has several limitations, including its retrospective design and small sample size. Accumulating additional data and undertaking future, multicentric studies with a larger population and a higher level of evidence are required. The surgical benefits for postoperative defecation function were difficult to evaluate fully because three of the patients were too young for evaluation of the postoperative outcomes. In addition, several factors besides anorectal stenosis affect defecation function in CS, such as neurological problems, complications of pelvic surgery, hypoplasia of the anal sphincter and levator ani muscles, and colonic dilation resulting from chronic constipation.

Conclusions We reviewed seven patients with CS who underwent treatment in our hospital and proposed a protocol to optimize the treatment flowchart. Since CS has a wide spectrum of clinical presentations, there is no consensus on the therapeutic protocol. Consequently, there is a chance of missing the diagnosis at the initial workup, and optimal treatment could not be given at the appropriate time in some cases. Establishing the treatment strategy could reduce the risk of delayed diagnosis, and it would be possible to improve treatment outcomes based on a full understanding of the complicated pathophysiology of CS.

Purpose: The present study aimed to review the treatment experience and outcomes of Currarino syndrome (CS) complicated with anorectal stenosis to evaluate the current treatment strategies. Methods: Seven cases of CS complicated with anorectal stenosis, treated at our hospital between 1998 and 2021, were retrospectively investigated. This is a case series article from a single institution. Results: In six and three cases and one case, the presacral mass was a mature teratoma, meningocele, and lipoma, respectively. Resection of the lesion was performed in all six cases of mature teratoma, and duraplasty was performed before resection in all three cases of meningocele. Moreover, surgery for anorectal stenosis was performed simultaneously in four patients. Surgery was performed for six cases of anorectal stenosis, with the remaining case relieved by dilation using a metal bougie. The surgical methods used were a partial resection with end-to-end anastomosis, anorectal strictureplasty, pull-through, posterior sagittal anorectoplasty, and cutback after mass resection. Pathological analysis of the anorectal stenoses revealed disorganized and rough smooth muscle fibers and the replacement of the stroma by an increased quantity of collagen fibers. Conclusions: The clinical outcomes of CS can be improved by establishing a treatment flow chart and understanding the complicated pathophysiology of the disease.

We wish to acknowledge Dr. Satoshi Ihara, a pediatric neurosurgeon at Tokyo Metropolitan Children’s Medical Center, for his help in interpreting the neurosurgical findings, and thank Mr. James Robert Valera for his assistance with editing this manuscript.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50512

oa_package/b4/5e/PMC10788246.tar.gz

PMC10788247

38226115

Introduction and background Vitamin A is a fat-soluble vitamin essential in the normal nonpathogenic functioning of many human body processes. It plays an important role in eye development, prevention of blindness, bone development, skin and mucosa protection, and immunity, and aids in the development of epithelial tissue, such as teeth and hair. In addition to these important functions, it is also essential for standard embryo development [ 1 ]. During pregnancy, a mother’s nutrient requirement increases to maintain proper fetal development. This requirement is especially important during the third trimester when the fetus undergoes rapid growth and needs the developmental benefits of vitamin A [ 2 - 4 ]. Many developing countries have mothers who experience vitamin A deficiency (VAD), leading to abnormal development of the fetus. This deficiency can be due to socioeconomic factors since most deficiencies are of higher prevalence in poor, underserved, countries with inequalities in income, education, housing, and access to medical care [ 2 ]. If an expectant mother has an excess intake of vitamin A during the first trimester, central nervous system derangements, cardiovascular system abnormalities, or spontaneous abortions may result [ 2 - 4 ]. This is because high serum levels of retinoic acid, one of the three forms of vitamin A, interfere with genes essential for the development of the fetus. A pregnant woman should watch the levels of vitamin A that she consumes because any extra can result in teratogenicity in the fetus, this can occur through education about what foods or supplements contain vitamin A, and monitoring their intake of such foods. The World Health Organization (WHO) recommends a maximum dose of 10,000 IU daily during the first 60 days of fetal development when crucial and susceptible development occurs for all mothers [ 2 - 4 ]. Because of the dangers of both vitamin A excess and deficiency, it is important that levels are monitored during pregnancy to ensure the mother is maintaining an appropriate level for her fetus.

Conclusions The pregnant mother must weigh the need for an adequate supply of vitamin A to avoid deficiency against excess consumption, potentially leading to teratogenicity. The unregulated use and lack of education on the teratogenic effects of vitamin A multivitamins and supplements in developed countries make it much easier to consume doses greater than the recommended quantity by the WHO. Mothers who consumed more than 25,000 IU/day of vitamin A had children with an increased risk of urinary tract malformations. However, the recommendation is no more than 10,000 IU daily before 60 days of gestation and no more than 25,000 IU weekly to prevent the teratogenicity risk. Educating expectant mothers on the risks associated with both deficient and excessive levels of vitamin A can prevent deformities related to vitamin A malnutrition in the child.

Vitamin A deficiency (VAD) or excess in expectant mothers can result in fetal abnormalities such as night blindness, bone anomalies, or epithelial cell problems. In contrast, excessive vitamin A in pregnancy can precipitate fetal central nervous system deformities. During pregnancy, a pregnant woman should monitor her vitamin A intake ensuring she gets the recommended dosage, but also ensuring she doesn't exceed the recommended dosage, because either one can result in teratogenicity in the fetus. The widespread and unregulated use of multivitamins and supplements makes consuming doses greater than the recommended quantity more common in developed countries. While vitamin A excess is more common in developed countries, deficiency is most prevalent in developing countries. With proper maintenance, regulation, and education about VAD and excess, a pregnant mother can diminish potential harm to her fetus and potential teratogenic risks.

Review Vitamin A overview Vitamin A is a fat-soluble vitamin found naturally in many foods, including fruits, vegetables, and animal-based products. It is crucial for the normal functioning of many distinct functions of the human body, such as visual adaptation, immunity, and epithelial cell differentiation [ 5 ]. Once ingested, it must be metabolized into its biologically active forms, which include retinol, retinal, and retinoic acid, with the most prominent forms being 11-cis-retinal and all-trans-retinoic acid (ATRA) [ 5 ]. Each form has different roles in the human body [ 5 ]. While it may be administered topically or as an injectable, vitamin A is predominantly taken orally and absorbed by enterocytes after metabolizing it into retinol. Once absorbed, it can be transported to other cells bound to cellular retinol-binding protein. Its main target is the liver, where it is stored as retinyl palmitate [ 5 , 6 ], and its absorption is increased with increased fat intake through the diet [ 5 ]. The kidneys eliminate vitamin A derivatives through urine or the liver through bile [ 5 ]. Because it is fat-soluble and stored in the liver, it can take months before deficiency becomes evident. On the other hand, toxicity can occur easily with excess intake or altered metabolism. Intake varies depending on sex and status of pregnancy and/or lactation and is measured in retinol activity equivalent (RAE), where one RAE equals retinol 1 microgram [ 6 ]. The recommended daily allowance is 900 micrograms per day for males, 700 micrograms per day for non-pregnant and non-lactating females, 750-770 micrograms per day for pregnant females over the age of 18, and 1300 micrograms per day for lactating females over the age of 18; regardless of sex or pregnancy and lactation status, intake should not exceed 3000 micrograms per day [ 6 ]. The serum concentration of vitamin A in adults ranges from 300 to 700 ng/mL with a peak plasma time of four to five hours in oil solution or 304 hours when water-miscible [ 6 ]. Vitamin A and its levels can be impacted by estrogens and oral contraceptives, increasing retinol-binding protein levels [ 5 ]. In addition, alcohol damages the liver over time and decreases the amount of stored vitamin A [ 5 ]. Current uses of vitamin A Vitamin A is found naturally in many food items, including liver, butter, other dairy products, eggs, chicken, beef, fish, and certain vegetables such as carrots and sweet potatoes. In addition to its natural sources, vitamin A and its derivatives can also be found in over-the-counter supplements and are prescribed as a treatment for certain medical conditions [ 7 ]. If a pregnant mother was educated on which foods contain vitamin A, she could easily monitor her intake. Vitamin A supplementation is crucial in individuals who do not receive adequate intake through diet alone, which is uncommon in developed countries. Sufficient intake is necessary to prevent symptoms of VAD, such as night blindness, dry skin, keratomalacia, and immunosuppression. In addition to its benefits for development and metabolic processes, vitamin A has important pharmaceutical benefits. One commonly prescribed medication is isotretinoin, a vitamin A-derived medication administered orally over months to treat severe cystic acne [ 8 ]. Because vitamin A also plays a role in adaptive immunity, its supplementation in children is part of the supportive treatment plan for measles [ 9 ]. Similarly, supplementation of vitamin A may be an option for treating severe acute respiratory syndrome coronavirus 2 [ 10 ]. Treatment modalities in these cases include a topical nasal application for anosmia and oral supplements in combination with steroids for patients with severe active infection. Individuals recovering from the virus may also benefit from oral vitamin A supplementation [ 10 ]. Teratogen effects Studies have examined the relationship between vitamin A use in pregnancy, specific organ outcomes, and developmental anomalies. One meta-analysis of six studies examined the association between vitamin A during pregnancy and nonsyndromic orofacial clefts. More specifically, the study explored vitamin A use during the periconceptional period, or three months before conception, and the first trimester. Results demonstrated a significant protective effect of periconceptional vitamin A on the nonsyndromic cleft lip with or without cleft palate (NSCL/P) but a nonsignificant protective effect for nonsyndromic cleft palate only. It has been posited that excess or deficient levels of vitamin A during pregnancy can affect the development and growth of the kidneys and urinary tract in offspring. Previous studies exhibited smaller kidneys in the offspring of women with deficient levels of vitamin A during pregnancy [ 11 , 12 ]. Ozisik et al. found a significant shared commonality in vitamin A target genes and genes known to cause congenital anomalies of the kidney and urinary tract [ 11 , 12 ]. While this finding suggests that vitamin A plays a role in cell signaling, growth, and development of the kidneys and urinary tract, further research is necessary to study specifics, including periods of development that may be more vulnerable to variations in vitamin A and doses [ 11 , 12 ]. VAD is most common in developing countries. In contrast, excess vitamin A consumption, possibly teratogenic, is more prevalent in developed countries [ 2 ]. The widespread and unregulated use of multivitamins and supplements in developed countries, for example, vitamin A supplements, makes consuming doses greater than the recommended quantity much easier. This is partially due to supplements not being held to the same regulations as pharmaceuticals in the United States, and patients should always consult a physician before beginning any supplement regimen, especially during pregnancy [ 4 ]. The WHO defines VAD as <0.70 μmol/L serum retinol levels [ 2 ]. During the third trimester, VAD becomes a concern when maternal blood volume and fetal development rate increase [ 2 ]. One study evaluated the effects of chemically induced maternal VAD in mice and its effect on the development of congenital diaphragmatic hernias (CDH). Abnormalities in the retinoid signaling pathway are one mechanism proposed to contribute to CDH. Results showed that low levels of maternal vitamin A increased the incidence of teratogen-induced CDH [ 13 ]. VAD is known to have dangerous effects on the immune system, ocular health, and growth and development of children. VAD, low serum retinol, and low insulin-like growth factor 1 (IGF-1) are thought to contribute to a poor immune system and poor growth and development in children with Down syndrome (DS). A cross-sectional study on 47 children ages 24-72 months with DS revealed a prevalence of VAD of 25.5% (n = 12; 95%CI: 13.9-40.3) [ 14 ]. The relative dose-response test determined VAD results that suggested low vitamin A stores in the liver. About 74.5% of children had serum retinol levels <0.70 μmol/L and there was no significant association between VAD and IGF-1 levels, there was a positive correlation between retinol levels and IGF-1 levels [ 14 ]. While vitamin A was shown to be protective against NSCL/P [ 11 ], another study using mouse embryonic palatal mesenchymal cells investigated the effects of the vitamin A derivative ATRA on their differentiation and mineralization. The results illustrated that ATRA inhibited Wnt signaling, therefore impeding bone formation. This demonstrates a possible connection between ATRA and cleft lips/palates, though further animal and human models are necessary to continue exploring this relationship [ 2 ]. Vitamin A plays a role in the development of pharyngeal arch arteries, cranial nerves, and hindbrain, which all function in swallowing. Excess maternal vitamin A in mouse models of 22q11.2 deletion syndrome resulted in pups with exacerbated abnormalities of the fourth pharyngeal arch artery, cranial nerve V, and expression of Cyp26b1 in the hindbrain that contributed to swallowing difficulties as well as increased lung inflammation, a sign of aspiration [ 15 , 16 ]. Doses over 10,000 IU a day are thought to have the potential for teratogenic risks, especially if excessive intake occurs during the first quarter of the pregnancy [ 2 ]. Studies have shown that doses exceeding this amount affect the development of neural crest tissues, urinary tract, and cardiac structures [ 2 ]. Vitamin A's current recommendations Although it is still unclear whether specific amounts of these metabolites are generated with each vitamin A supplement, 13-cis-retinoic acid is the main teratogen of Isotretinoin, a common treatment for cystic acne [ 2 ]. Because of the body's variability in metabolism and lack of clinical examples, there is a paucity of data on doses of vitamin A that establish a threshold of teratogenicity. It is generally assumed that when doses rise above 10,000 IU per day of vitamin A in a pregnant woman with baseline normal vitamin A levels, there is a potential risk of teratogenicity, and reports suggest that fetuses of pregnant mothers taking doses greater than 25,000 IU/day had urinary tract malformations [ 17 , 18 ]. According to the WHO, supplementation is not recommended in developed countries with a nutritionally adequate diet, and pregnant mothers should consume a dietary allowance of 2670 IU of vitamin A. Currently, guidelines indicate a maximum of up to 10,000 IU daily before 60 days of gestation and no more than 25,000 IU weekly after 60 days. This difference is due to the higher risk of early pregnancy effects which can be seen in Table 1 [ 19 - 24 ]. Suppose a pregnant woman's intake of retinol exceeds 10,000 IU per day. In that case, the American Heart Association recommends that fetal echocardiography be obtained during the prenatal period due to the risk for cardiomyopathy with a minimal absolute risk between 1% and 2% [ 3 ]. Clinical trials related to vitamin A toxicity More research is required to better understand vitamin A’s mechanisms and effects, especially during the prenatal period. For example, factors causing increased CYP2D6, an enzyme important for drug metabolism, and its activity during pregnancy, as well as the role of undernutrition on thymulin in neuroendocrine regulation of inflammation, require additional study as seen in Table 1 [ 19 , 24 ]. Other studies attempting to ascertain the effects of vitamin A yielded contrasting results. Notably, in Ding et al., daily oral vitamin A supplementation helped improve vitamin A status in many pregnant women who are deficient [ 20 ], while in Haskell et al., small-quantity lipid-based nutrient supplements did not affect vitamin A levels [ 21 ]. Regardless, in these two studies, along with Nga et al., various versions of vitamin A supplementation did not affect maternal or infant outcomes as seen in Table 1 [ 22 ]. While this research provides some insight into the role of vitamin A in the prenatal period, it also demonstrates the necessity of further exploration into these relationships. Discussion According to current evidence, it is essential that pregnant mothers maintain recommended levels of vitamin A as recommended by WHO throughout pregnancy to ensure their health and the health of their baby. Vitamin A is a fat-soluble vitamin found in many foods, fruits, vegetables, and animal-based products. It is important for visual acuity, immunity, bone growth, and epithelial cell differentiation and therefore very important in the development of fetuses [ 5 , 25 , 26 ]. In developed countries, vitamin A supplementation is critical in patients who do not receive satisfactory consumption through their diet. Appropriate levels of vitamin A are necessary to prevent symptoms of VAD, such as night blindness, dry skin, keratomalacia, and immunosuppression. One of the most common uses of prescribed vitamin A isotretinoin is a vitamin A-derived medication administered orally over months for treating severe cystic acne; however, it has a teratogenic association [ 8 , 27 ]. Many individuals may start medication for their acne without realizing the teratogenic effects that it will have on their future child. Results from trials assessing the teratogenicity of vitamin A have demonstrated that it affects gene signaling and inhibits bone formation, resulting in cleft lips or palate [ 11 , 28 ]. It has also been shown that excess can lead to abnormalities in pharyngeal arches, which results in impaired swallowing and aspiration [ 16 ]. However, deformities in the fetus can also occur in deficient levels of vitamin A, which is why individuals often take vitamin A supplementation. Vitamin A levels should be monitored very carefully to prevent teratogenic effects. Pregnancy increases the body’s nutritional requirements, including micronutrients such as vitamin A. There is a higher incidence of excessive vitamin A consumption in developed countries, whereas, in developing countries, VAD is more common [ 2 , 29 ]. However, both the excessive and deficient levels of vitamin A result in a negative effect on the fetus. With proper education to patients, vitamin A teratogenicity can be avoided and therefore result in a healthier baby.

The authors wish to acknowledge the Paolo Procacci Foundation for its generous support in the publication process.

CC BY

no

2024-01-16 23:41:56

Cureus.; 15(12):e50513

oa_package/3f/ac/PMC10788247.tar.gz

PMC10788254

38226357

Background The fabric of a rock or sediment is related to the spatial and geometric configuration of the particles that compose it. In sedimentary rocks, fabric is a property that depends on the depositional environment and can provide information about the directions of the currents at the time of deposition. For this information to be obtainable, some of the particles composing the flowing granular material (crystals, clasts, but also bubbles and voids) must be elongated. In this case we use the term shape-fabric. Shape-fabric refers to the tridimensional orientation and the degree of clast iso-orientation that is recorded in a deposit [1] , [2] , [3] . The first applications of this textural property to paleo-flow direction go back to the study of turbidity deposits with the purpose of understanding the origin area of these sedimentary materials (e.g., [4] , [5] , [6] ). Later, the same property has been used successfully in volcanology to identify flow directions, vent location, and other characteristics of pyroclastic density currents [ 1 , 3 , 7 , [8] , [9] , [10] , [11] ] and lahars [12] . Applications of fabric analysis using intercept methods were described by Launeau and Robin [13] . There are various properties that can be studied using shape-fabric, such as determining the polarity of the flow movement, recognizing fluctuations in the flow direction during deposition, and the relationship between the degree of clast iso-orientation and the rheological characteristics of the flow [3] . Shape-fabric can be measured in different ways [14] ; those that measure the physical bulk properties of a sample as a whole (bulk methods), and those that measure the orientation of the particles directly, in other words, particle by particle. To the first category belong those methods that measure physical properties such as anisotropy of magnetic susceptivity (AMS) [15] , [16] , [17] , [18] , [19] , [20] , [21] , [22] , [23] , permeability [24] , and dielectric intensity [25] . The second category includes the quantitative textural analysis (QTA) method, which is the method described in this paper. QTA applied to a particle's constituent rock or sediment provide geometric data that can be directly related to the paleocurrents, assuming a model for the iso-orientation mechanism ( Fig. 1 ). In practical applications focused on granular flow directions in a sedimentary environment, clast shape-fabric can be used in different ways, and a variety of different information can be obtained. Data is obtained by analyzing the particles (within the oriented sample) in a horizontal plane or parallel to the bedding planes. For flow polarity it is necessary to measure clast imbrication (by analyzing particles in a vertical plane, oriented parallel to the previously determined flow direction). In some cases, an apparent fabric can also be determined by directly measuring the particles on a vertical wall of an outcrop [26] . The information obtained in this case depends on the outcrop's orientation with respect to the flow, and it can help measure flow polarity (analyzing the imbrication in the outcrop) or to show the presence of erosive channels, lenses, swirls during deposition. Clast shape-fabric is easy to use, accurate, and low-cost. It consists of taking oriented samples in the field and analyzing them by circular statistics. Here we describe the protocol and methods applied in Cerca-Ruiz [27] and Cerca-Ruiz et al. (in progress), where the method has been used to measure the clast shape-fabric direction and iso-orientation degree, information that allow to construct flow trajectory maps of the pyroclastic deposits of the Joya Honda Maar in San Luis Potosí, Mexico. The same methodology can also be applied to other igneous, sedimentary, and metamorphic rocks. It should be noted that previous versions of this methodology have been applied in the works of Hernandez-Rivas et al. [12] and Zrelak et al. [3] , yielding excellent results. The following sections focus on the methods of sampling and sample processing in the laboratory and describe the protocols of image analysis and data processing.

Method details Background The fabric of a rock or sediment is related to the spatial and geometric configuration of the particles that compose it. In sedimentary rocks, fabric is a property that depends on the depositional environment and can provide information about the directions of the currents at the time of deposition. For this information to be obtainable, some of the particles composing the flowing granular material (crystals, clasts, but also bubbles and voids) must be elongated. In this case we use the term shape-fabric. Shape-fabric refers to the tridimensional orientation and the degree of clast iso-orientation that is recorded in a deposit [1] , [2] , [3] . The first applications of this textural property to paleo-flow direction go back to the study of turbidity deposits with the purpose of understanding the origin area of these sedimentary materials (e.g., [4] , [5] , [6] ). Later, the same property has been used successfully in volcanology to identify flow directions, vent location, and other characteristics of pyroclastic density currents [ 1 , 3 , 7 , [8] , [9] , [10] , [11] ] and lahars [12] . Applications of fabric analysis using intercept methods were described by Launeau and Robin [13] . There are various properties that can be studied using shape-fabric, such as determining the polarity of the flow movement, recognizing fluctuations in the flow direction during deposition, and the relationship between the degree of clast iso-orientation and the rheological characteristics of the flow [3] . Shape-fabric can be measured in different ways [14] ; those that measure the physical bulk properties of a sample as a whole (bulk methods), and those that measure the orientation of the particles directly, in other words, particle by particle. To the first category belong those methods that measure physical properties such as anisotropy of magnetic susceptivity (AMS) [15] , [16] , [17] , [18] , [19] , [20] , [21] , [22] , [23] , permeability [24] , and dielectric intensity [25] . The second category includes the quantitative textural analysis (QTA) method, which is the method described in this paper. QTA applied to a particle's constituent rock or sediment provide geometric data that can be directly related to the paleocurrents, assuming a model for the iso-orientation mechanism ( Fig. 1 ). In practical applications focused on granular flow directions in a sedimentary environment, clast shape-fabric can be used in different ways, and a variety of different information can be obtained. Data is obtained by analyzing the particles (within the oriented sample) in a horizontal plane or parallel to the bedding planes. For flow polarity it is necessary to measure clast imbrication (by analyzing particles in a vertical plane, oriented parallel to the previously determined flow direction). In some cases, an apparent fabric can also be determined by directly measuring the particles on a vertical wall of an outcrop [26] . The information obtained in this case depends on the outcrop's orientation with respect to the flow, and it can help measure flow polarity (analyzing the imbrication in the outcrop) or to show the presence of erosive channels, lenses, swirls during deposition. Clast shape-fabric is easy to use, accurate, and low-cost. It consists of taking oriented samples in the field and analyzing them by circular statistics. Here we describe the protocol and methods applied in Cerca-Ruiz [27] and Cerca-Ruiz et al. (in progress), where the method has been used to measure the clast shape-fabric direction and iso-orientation degree, information that allow to construct flow trajectory maps of the pyroclastic deposits of the Joya Honda Maar in San Luis Potosí, Mexico. The same methodology can also be applied to other igneous, sedimentary, and metamorphic rocks. It should be noted that previous versions of this methodology have been applied in the works of Hernandez-Rivas et al. [12] and Zrelak et al. [3] , yielding excellent results. The following sections focus on the methods of sampling and sample processing in the laboratory and describe the protocols of image analysis and data processing.

Fabric analysis is essential for understanding the evolution of volcaniclastic deposits. Here we present a comprehensive and efficient methodology, called “Clast shape-fabric analysis,” which is part of the Quantitative Textural Analysis (QTA). This methodology combines high-resolution image analysis techniques with geospatial data processing tools. The fabric of a deposit refers to the three-dimensional orientation of the particles with respect to space, where the degree of iso-orientation of the major axes of the particles is taken into account. The process begins with the collection of oriented samples in the field. Then, in the laboratory, the samples are processed to obtain high-resolution images. The final stage involves the analysis of these images using the FabricS program, which combines image processing techniques and circular statistics. An application of the method was made at the Joya Honda Maar in Mexico, where shape-fabric analysis was used to identify the emission centers of pyroclastic materials. In summary, the “Clast shape-fabric analysis” is a reliable, low-cost and high-potential methodology that can be applied in several geoscientific disciplines and other areas of scientific research. • New Methodology for shape-fabric analysis is presented. • The methodology involves field work, laboratory work and image analysis. • Identification of particle orientations in volcaniclastic deposits. Graphical abstract Keywords Method name

Specifications table Method details In order to accurately measure the shape-fabric of a sample from a volcanoclastic deposit, and obtain reliable and reproducible data, it is necessary to follow a specific procedure ( Fig. 2 ). Sampling Location and orientation of samples are among the most important aspects of the process because the accuracy of the data obtained in clast shape-fabric analysis depends on these data. The following workflow summarizes the procedure for obtaining oriented samples in the field ( Fig. 3 ): 1. The unit to be sampled is identified in the outcrop. It is important to note that if more than one stratigraphic unit forms the outcrop, sampling should be carried out in each of them individually depending on the precision required. It is recommended to look for parts of the outcrop surface that are as vertical as possible. 2. With the help of a shovel, hammer and chisel, drill, or any tool that can be used for digging and shaping, a cuboid of approximately 15 × 15 × 15 cm is sculpted in the outcrop. This dimension depends on clast size; the larger the clasts, the larger the sample. If the deposit is unconsolidated, the cuboid should be constantly moistened with a solution of water (75 %) and sodium silicate (25 %) to prevent it from disintegrating while it is being carved out. If the deposit is well consolidated, simply sculpt the cuboid. 3. If the sample is poorly consolidated, it is necessary to stabilize it by means of gauze moistened in a mixture of plaster and water. Moisten the gauze in this mixture and place it on the cuboid until it is covered. Then add more plaster on the cuboid and use a spatula to try to smooth each of its faces as much as possible. Let it dry (approximately 8 hrs). 4. Once the plaster cast has hardened, proceed to orient the sample. With the help of a level, draw a horizontal line on the front face of the cuboid. Draw another line perpendicular to the horizontal line near the bottom of the sample. This line will indicate the direction of the basal part of the sample. Measure the azimuth of the sample by placing a compass on the horizontal line (making sure that the compass is perfectly horizontal, with the help of the level contained in the compass). Next, measure the inclination of the front face of the cube by placing the compass parallel to the vertical line or measuring the inclination with an inclinometer. Finally (if possible), mark an arrow pointing north on the upper or lower face of the cuboid. 5. If the sample is consolidated, draw the corresponding lines on the faces of the cuboid with a marker. 6. Carefully remove the cuboid by detaching the part that is still in contact with the outcrop. This can be done using a hammer and a long chisel or a drill with a long bit. This procedure must be slow and precise to avoid breaking the cuboid. 7. In the case of unconsolidated samples, adding a mixture of sodium silicate (40 %) and water (60 %) to the oriented sample can help to pre-consolidate it (if necessary) to avoid any deformation of the particles inside the cuboid. Processing the samples in laboratory Once the samples have been oriented and collected in the field, they are carefully transported to the laboratory. The objective of the laboratory procedure is to obtain high-resolution images of horizontal slices of the samples for subsequent image analysis ( Fig. 4 ). The processing of the samples in the laboratory is shown in the following workflow: 1. If necessary, the samples are placed in a vacuum chamber for as long as needed until they are completely impregnated. The time will depend on the type of sample, its degree of consolidation and porosity. It is important to note that with poorly consolidated samples it is recommended not to exceed −0.8 PSI to avoid disintegration. 2. Once consolidated, the sample can be sliced with a saw. A first cut can divide the sample in half in a plane parallel to the vertical. This is so that one half can be processed to analyze the clast shape-fabric and the other half to analyze imbrication of the clasts. Next, a series of cuts are made parallel to the horizontal plane to obtain slices approximately 10 to 20 mm thick. 3. The slices are then polished using sandpaper and abrasives (between 80 and 600 sandpaper grit size) to obtain a smooth finish in which the particles are visible with the greatest possible clarity. On each slice, it is important to mark an arrow indicating the direction of north. 4. Once the slices are polished, they are photographed, placing the north arrow parallel to the vertical side of the photograph. The slice should be placed on a horizontal table and the photograph should be perpendicular to the slice. 5. Once the shape-fabric analysis for flow orientation is finished, the second half of the sample is cut vertically parallel to the flow orientation in order to measure the imbrication. The slab thickness is the same as in Step 2. Image analysis and data processing To measure particle orientation on the photographs taken in the previous phase, a series of procedures needs to be followed. These processes include segmentation of the photograph and subsequent image analysis using the FabricS program [2] . The purpose of the segmentation phase is to obtain an image in which the particles are filled with white while the matrix (particles that are not visible in the photograph) is black. For the analysis of segmented images, the use of the FabricS program is highly recommended. FabricS [2] is a new easy-to-use free software by which it is simple to calculate the shape-fabric of volcano-sedimentary particles. FabricS is the first software dedicated to shape-fabric analysis that combines image analysis processing and circular statistics algorithms. The data obtained by FabricS are the orientation, iso-orientation degree, P-value and rose diagrams of the studied outcrop, verified by rigorous statistical tests such as the Rayleigh test. Both processes (segmentation and image analysis) are detailed in the following workflow: Segmentation The segmentation process consists of separating the matrix and the particles using image editing software. The particles are colored in white and the matrix is colored black in order to obtain binary images. It is recommended that this process be carried out manually in order to minimize the errors and increase the accuracy of the data. The process of segmentation in Adobe Photoshop is detailed in the following work list; however, there are other free software similar to Photoshop (Ink Scape, Image J, Paint.net) that can be used instead: 1. Input the photograph into Photoshop. 2. Using the “Quick Selection” option, select the sample. 3. In the top menu select the option selection → invert. 4. Using the “quick selection” tool, mark the outline of each of the particles in the photograph. For a better visualization of the finest particles, it is recommended to zoom in. Select all visible particles. 5. From the menu bar choose Image→ Adjustments → Hue/Saturation. In the Hue/Saturation window activate the “Colorize” option and set the following values: Hue= 0, Saturation= 25 and Lightness= 100. This will assign the color white to the particles. 6. To color the matrix black, in the menu bar select the option selection → Invert. This selects only the matrix. Repeat Step 5, but changing the “Lightness” value to −100. This differentiates the matrix from the particles. Any editing software that have at your disposal can be used, as long as it can produce an image where the matrix and particles are differentiated from each other. Image analysis by FabricS The main purpose of the overall methodology described in this paper is to obtain the preferential orientation of the particles that make up the analyzed samples. In order to achieve this, we suggest the use of the FabricS program, developed at the LAIMA laboratory of the Universidad Autonoma de San Luis Potosi. FabricS was developed in a MATLAB environment. It is not open source software and currently runs on Windows 7 or higher operating systems. FabricS combines image analysis procedures and circular statistics algorithms. It has an easy-to use graphical interface and basic functions for image processing. The software can process a single image or a set of images. It also allows some parameters to be adjusted, such as the minimum particle size to be considered in the analysis and the elongation of the particles. For more information, we suggest reading the work of Moreno-Chavez et al., [2] . The FabricS user interface contains four sections: file, process, fabric, and export, briefly described below: File The file section allows the user to load binary or color images. This button will open a dialog box where the user can load the binary image first and then the original color image. The process is similar when multiple images need to be loaded. Process The process section has two main tools for processing the image. The first is the complement section, which allows the user to set the background of the image to black and the particles to white if needed. The second tool is the ROI button which enables the user to select a region of interest to analyze instead of the whole image. In addition, the program allows two important parameters to be adjusted, the size threshold and the particle eccentricity threshold. The size threshold can be used to select the minimum size (in pixels) of the particles to be considered in the analysis. The particle eccentricity threshold determines the degree of ellipticity of the particles ranging from 0 to 1, where zero is an indicator of circular particles and 1 indicates very elongated particles. It is important to note that the more circular the shape of the particle, the less orientation information it will yield. Therefore, the eccentricity threshold eliminates all particles that do not provide accurate information due to their round shape. For recommended parameters, it is suggested to see Moreno-Chavez et al., [2] . Once these thresholds have been optimally configured, click on the DO button to perform the analysis. Fabric In the fabric section, the user can define the orientation (north) of the image(s) using the reference option. After that, the user can select the desired statistical parameters (mean, median, or mode). The user must then click on the “stats” button so that the program displays the selected statistics on the image in vector form. The fabric section graphically visualizes the data obtained from the analysis. If the user wishes to export the data, the user must go to the export tab. Export In this tab, the user can export the data obtained from the analysis, such as the statistics and the statistical tests that indicate the certainty of the results. The processed images can also be exported. The program may take a few minutes, depending on the processed image(s). Another important product that the program can export is the rose diagram of the data, which graphically shows the iso-orientation of the analyzed particles ( Fig. 5 ). Method validation The versatility and potential of the shape-fabric analysis can be demonstrated on a study carried out in Joya Honda Maar (work in progress) located 40 km north of the city of San Luis Potosí in central Mexico. According to previous work, the eruption that created the Joya Honda maar occurred approximately 311 ± 19 ka ago along a fissure and formed a crater 1.3 × 0.9 km and 270 m deep [28] . Five eruptive phases were documented on Saucedo et al. [28] , and the emission center was estimated based only by means of stratigraphy. In order to locate the possible center or centers of emission, 77 oriented samples were taken around the crater in the work of Cerca-Ruiz [27] and in Cerca-Ruiz (work in progress) ( Fig. 6 ). All five units were sampled. However, for practical purposes and to expose the effectiveness of the method, only the results obtained from the Unit III are presented below. The complete research work will be presented later in a scientific article. A total of 38 oriented samples were collected from Unit III following the methodology described in this work. At the same time, a geophysical study (magnetometry) were performed in the maar crater and the surroundings and a ballistic impact study were carried out in the same Unit III. As a result of the shape-fabric study, a trajectory map was constructed ( Fig. 6 a) evidencing a possible emission zone in the northern sector of the crater. However, according to the results of particle orientations in all five units another emission center was located at the southern sector of the crater. An interesting aspect of this study was that the other methodologies (magnetometry and ballistic analysis) yielded similar results. Through magnetometry an anomaly related to a possible conduct was detected at North of the crater and ballistic analysis provide evidence that an emission center probabli was located in the North of the crater. This is in agreement with the results reported by Loera et al. [29] and Saucedo et al. [28] . This case study confirmed that shape-fabric analysis is a reliable methodology that can be used on other similar volcanoes to identify emission centers or source areas of pyroclastic materials or sediments. The potential of this method is vast, as it can be applied in other geosciences such as structural geology ( [30] , [31] , [32] ), geophysics [ 33 , 34 ], or sedimentology [ 35 , 36 ], but also in other fields of science such as planetary science [ 37 , 38 ], biostatistics [ 17 , 39 ], and more, due to its simplicity and low cost. CRediT authorship contribution statement L.A. Rodríguez-Sedano: Conceptualization, Supervision, Investigation, Validation, Data curation, Writing – original draft, Writing – review & editing. D. Sarocchi: Conceptualization, Supervision, Investigation, Writing – original draft, Writing – review & editing. F. Castillo Rivera: Software, Validation, Data curation. G. Moreno-Chávez: Conceptualization, Software, Validation, Writing – review & editing. M.F. Cerca-Ruiz: Validation, Data curation. J.A. Montenegro-Ríos: Software, Visualization, Writing – review & editing. Declaration of Competing Interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Data availability No data was used for the research described in the article. Acknowledgment CONAHCYT grants 1075397, 210817, FAI Project UASLP C19-FAI-05-83.83, British Royal Society Project NAF-R2-180833, Instituto de Geología UASLP and Maestría en Ciencias del Procesamiento de la Información, UAZ.

CC BY

no

2024-01-16 23:41:57

MethodsX. 2023 Dec 13; 12:102519

oa_package/a9/02/PMC10788254.tar.gz

PMC10788261

38226283

Introduction Infectious diseases were responsible for billions of deaths before 1980 [ 1 ]. Many infectious diseases have been brought under control, and their associated mortality rates have reduced significantly thanks to developments in medical treatment, public health interventions, and countermeasures (diagnostics, therapeutics, and vaccines) [ 2 ]. However, emerging infectious diseases (EIDs), caused by unknown pathogens with high case-fatality ratios and contagiousness that carry the risk of becoming pandemic, is a primary concern for global health systems. Moreover, there is a limited scope for preventing, treating or controlling EIDs [ 3 , 4 ]. The 21st century has witnessed the emergence of several landmark events in the field of EIDs including the SARS (2003), Marburg (2004), H1N1 influenza (2009), MERS (2012), Ebola (2013), Zika (2016), Lassa fever (2018) and COVID-19 (2019) which have caused widespread morbidity and mortality, health care disruptions, economic and societal costs, and human suffering [ 5 , 6 ]. Many studies have demonstrated that the most effective approach to combat EIDs is the use of immunotherapy, which can be categorized to active and passive immunotherapy. Active immunotherapy involves artificial immunization with specific antigens by vaccination to stimulate the immune response, whereas passive immunotherapy confers immunity through the adoptive transfer of pathogen-specific antibodies [ 7 , 8 ]. While vaccination can provide durable and protective immunity against EIDs, human vaccine approval is a lengthy process [ 9 ]. For instance, the first vaccine was approved for emergency use in the midst of the COVID-19 pandemic in December 2020, whereas the pandemic had begun about a year prior; during this time, the pandemic had tragically resulted in the loss of 8.1 million lives, according to WHO statistics [ 10 , 11 ]. On the other hand, since begun of pandemic incidence until first vaccine approval when other therapeutic options are unavailable, passive immunotherapy is the best method for treating critically ill patients because it can induce immediate immunity in individuals [ 12 ]. Passive immunotherapy utilizes polyclonal and monoclonal antibodies. The monoclonal antibodies are produced in laboratories, while polyclonal antibodies are derived from immunized humans and animals. Despite the widespread use of monoclonal antibodies in a variety of infectious and non-infectious diseases, the process of producing and approving these antibodies is time-consuming, making them inappropriate for use in EIDs when the pathogen is unknown or when the pathogen's neutralizing epitopes have not been identified. In contrast, the use of polyclonal antibodies is preferred in new EIDs due to their low cost, large scale, rapid production, and recognition of a broad spectrum of epitopes, which reduces the pathogen's chances of escaping mutation [ [13] , [14] , [15] ]. Human-derived polyclonal antibodies have been used in post-exposure prophylaxis of rabies, hepatitis B, and tetanus; removal of the foreign agent prior to activation of the individual's immune system (as in using anti-Rh antibodies in Rh-negative women who have given birth to Rh-positive infants); and immunodeficient patients (such as those with Bruton agammaglobulinemia) [ [16] , [17] , [18] ]. Animal-derived polyclonal antibodies have been utilized to neutralize snake and scorpion venoms, as well as diphtheria and botulinum toxins, and to prevent acute rejection after organ transplantation through immunosuppressive treatment using anti-thymocyte globulin [ [19] , [20] , [21] ]. Plasma or purified IgG from convalescent individuals or purified IgG derived from animals can be used to prepare polyclonal antibodies for passive immunotherapy. Also, when the vaccine becomes available to the public, the use of plasma and purified IgG from vaccinated individuals can be an alternative option to treat critically ill patients and immunocompromised individuals. Despite the widespread use of various passive immunotherapy approaches, the optimal treatment for EIDs has yet to be determined. In this study, we will investigate the neutralization potency of plasma and purified IgG from convalescent and vaccinated individuals, as well as purified IgG from rabbits, against SARS-CoV-2.

Material and methods Collection of human sera and plasma In a cross-sectional study conducted between March 2020 and July 2021, 100 volunteers without underlying diseases provided serum samples with negative SARS-CoV-2 RT-PCR results. The volunteers were screened and classified into three groups of healthy, convalescent, and vaccinated, using the enzyme-linked immunosorbent assay (ELISA) for anti-nucleocapsid (anti-N) (Pishtazteb, Tehran, Iran) and anti-spike (anti-S) (Sinabiotech, Tehran, Iran) antibodies. Human samples collected prior to the outbreak of COVID-19 were also used as negative controls (healthy group). Four individuals were selected from each group based on high immunogenicity criteria (two doses of the ChAdOx1 nCoV-19 vaccine in the vaccinated group) and age and gender matching (in all three groups). Subsequently, 5 mL of citrate anticoagulant-containing blood and 5 mL of non-anticoagulant-containing blood were collected from these individuals. Serum and plasma samples were pooled separately and stored at −80 °C until use. Prior to enrolling participants in the study, it was ensured that each individual provided their consent by signing an informed consent form. Each participant was thoroughly informed about the aims of the study, the potential advantages and risks that might be involved. The Ethics Committee of Mazandaran University of Medical Sciences, Iran, approved the research project (IR.MAZUMS.REC.1399.106). Purification of human polyclonal IgG and assessment of the reactivity of collected plasma and purified IgG with SARS-CoV-2 spike protein In order to purify human polyclonal IgG, selected individuals' sera were diluted 1/20 in PBS and filtered through a 0.45-μm filter. The diluted sera were then injected into a Hi-Trap protein A column (GE Healthcare Life Sciences, Uppsala, Sweden). The concentration of purified IgG was measured using the extinction coefficient following dialysis in PBS. Before the reactivities of plasma and purified IgG were compared with SARS-CoV-2 spike protein, the total IgG content of plasma samples was determined. All samples, including plasma and purified IgG, were normalized by diluting them to 1 mg/ml of total IgG in PBS. After normalization of human plasma and purified IgG in different groups, endpoint titration ELISA was used to evaluate the antibody's reactivity against SARS-CoV-2 spike protein. Briefly, 5 μg/ml of spike protein was used to coat 96-well plates (SPL, Life Science, Korea) and kept at 4 °C overnight. Unbound antigens were removed by washing three times with PBST (phosphate-buffered saline containing Tween detergent). Then, coated plates were blocked with 250 μL of blocking buffer (5 % BSA in PBST) at room temperature for an hour, washed 3 times with washing buffer. The samples were subsequently titrated against the coated antigen. An HRP-conjugated goat anti-human IgG (Abcam, ab6858) was added and incubated at 37 °C for 60 min. Plates were subsequently washed, and tetramethylbenzidine (TMB) substrate (Pishtazteb, Tehran, Iran) was added. The reaction was terminated by adding 1 M H2SO4, and the optical density (OD) was measured at 450/630 nm using an ELISA reader (BioTek, Winooski, VT, USA). The positive endpoint titration for human plasma and purified IgG samples was determined by samples with an OD value above the cutoff point (mean plus twofold standard deviation of healthy human plasma and purified IgG as a negative control). Preparation of purified IgG from hyperimmunized rabbits and assessment of the reactivity of purified IgG with SARS-CoV-2 spike protein As described in our previous study [ 22 ], we intramuscularly immunized two female New Zealand white rabbits (Pasteur Institute of Tehran, Iran) with 20 μg recombinant S1-FC fusion protein accompanied by 10 μg of CpG adjuvant (Bioneer, Republic of Korea) four times with two administration per week. Two weeks after the fourth immunization rabbit sera and also sera of two non-immunized rabbits was collected. Immunized and non-immunized rabbit IgG (IR-IgG and NR-IgG, respectively) were purified using a protein A chromatography column following the dilution of sera in PBS. After the concentration of purified IgG was measured, the antibody's reactivity was calculated by endpoint ELISA titration according to the previous ELISA section, except that HRP-conjugated mouse anti-rabbit IgG (Abcam, ab99697) was used for secondary antibody detection. Surrogate virus neutralization test (sVNT) The SARS-CoV-2 sVNT kit (Pishtazteb, Tehran, Iran) was used in accordance with the manufacturer's instructions to determine the neutralizing antibodies potency of human and rabbit samples. Briefly, normalized samples were serially diluted with sample dilution buffers from 1000 to 0.01 μg/ml of total IgG and subsequently mixed with an equal volume of HRP-conjugated ACE2. Afterward, the solution was added to the pre-coated RBD wells and incubated at 37 °C for 30 min. Unbound HRP-conjugated ACE2 was washed away after incubation. Upon the addition of TMB, the OD value was measured at 450/630 nm using an ELISA reader. The percentage of inhibition was calculated using the following formula: (1 − sample OD value/average negative control OD value) × 100. In addition, the IC50 value was determined to represent the neutralizing antibodies potency. Graphing and statistical analysis GraphPad Prism (Version 9, San Diego, California, USA) was utilized for raw data analysis and graphing. The correlation between the neutralization IC50 value and positive endpoint concentration of spike-specific antibodies was analyzed by Spearman correlation coefficients. There were at least three repetitions of each experiment. Data were presented as means ± standard deviations. Differences were deemed statistically significant with *P < 0.05 and **P < 0.01, or when results exceeded the cutoff point (mean plus twofold standard deviation of the negative control sample).

Results Screening and categorization of healthy, convalescent, and vaccinated individuals One hundred volunteers with negative results for SARS-CoV-2 RT-PCR were tested with an ELISA to determine their antibody levels against the S and N proteins. Positive samples for anti-S and anti-N were deemed convalescent, whereas positive samples for anti-S but negative for anti-N were deemed vaccinated. In addition, samples with positive results for anti-S and anti-N that were vaccinated against SARS-CoV-2 according to a questionnaire were excluded from the study. Likewise excluded were samples with negative results for anti-S and anti-N that were collected after the COVID-19 pandemic. Human samples collected before the COVID-19 pandemic were also used as a negative control (healthy group) ( Fig. 1 A). On the basis of the questionnaire and ELISA results, four individuals from each group were selected for further analysis; their characteristics are depicted in Fig. 1 B. Evaluation of endpoint titration of spike-specific polyclonal IgG from human and rabbits samples For each group's spike-specific endpoint titration, an indirect ELISA was conducted with sample dilutions ranging from 10,000–4 ng/ml. As illustrated in Fig. 2 A, all vaccinated human plasma (VH-plasma), vaccinated human purified IgG (VH-IgG), and convalescent human plasma (CH-plasma) groups had OD values greater than the cutoff point. The cutoff point was determined using human control groups, healthy human plasma (HH-plasma), and healthy human purified IgG (HH-IgG). Positive endpoint titrations for VH-plasma, VH-IgG, CH-plasma, and CH-IgG were 81.7, 130, 979, and 1164 ng/ml, respectively. The results indicate that vaccinated individuals have higher levels of spike-specific antibodies in their plasma and purified IgG than convalescent individuals. In addition, endpoint titration was performed on purified rabbit antibodies. As depicted in Fig. 2 B, the OD values of IR-IgG were greater than the cutoff (with 27.3 ng/ml positive endpoint titration). Comparison of viral neutralization potency in human and rabbit samples The sVNT test was performed to compare the levels of neutralizing antibodies in human and animal samples. The response of neutralizing antibodies was greater in hyperimmunized rabbits and vaccinated humans than in all other groups ( Fig. 3 A). In addition, the neutralization IC50 values for the VH-plasma, VH-IgG, CH-plasma, CH-IgG, and IR-IgG groups were 11.48, 18.84, 122.6, 153.5, and 2.08 μg/ml, respectively. As expected, no neutralizing antibody response was observed in the control groups ( Fig. 3 B). Moreover, a two-tailed Spearman correlation test was conducted between the positive endpoint concentration of spike-specific antibodies and the IC50 values of each group, which reveals a direct and significant correlation between the spike-specific antibody level and that of neutralizing antibody in each group (r = 0.99; p < 0.001; Fig. 3 C).

Discussion Multiple methods can be used to prepare passive immunotherapy from human and animal sources. The utilization of convalescent or vaccinated human plasma is a common passive immunotherapy method [ 14 ]. According to a study by Hung et al., the administration of convalescent plasma to influenza A (H1N1) patients resulted in a 34 % decrease in mortality compared to the conventional treatment group [ 23 ]. Using purified antibodies derived from humans is another method of passive immunotherapy. Huygens et al. demonstrated that human-derived purified antibodies reduce by 68 % the risk of severe COVID-19 infection in immunocompromised patients compared to the control group [ 24 ]. Nonetheless, these two approaches require collecting large quantities of pathogen-specific antibodies from human volunteers, whose neutralizing antibody titers may vary. While another passive immunotherapy method employing animal-derived purified antibodies can rapidly produce pathogen-specific antibodies on a large scale [ 14 ]. Today, there is a tendency to use animal-derived purified antibodies because they can be produced on a large scale without the need for human volunteers [ 14 ]. Despite the widespread interest in passive immunotherapy, no study has yet compared the neutralization potency of different passive immunotherapy methods. In this study, we collected plasma and IgG from healthy, convalescent, and vaccinated individuals ( Fig. 1 ), as well as IgG from non-immunized and hyperimmunized rabbits against SARS-CoV-2 spike protein. We used an endpoint ELISA titration to evaluate the spike-specific antibody levels. The results showed that in comparison to the control groups, both artificial (VH-plasma, VH-IgG, and IR-IgG) and natural (CH-plasma and CH-IgG) immunization approaches induced a significant spike-specific antibody response ( Fig. 2 A and B). These antibodies were significantly higher in hyperimmunized animals and vaccinated individuals than in convalescent individuals, attributable to the injection of booster doses and potent adjuvants. Moreover, despite identical total IgG concentrations in all samples (as described in section 2.2 , plasma and purified IgG were normalized to 1 mg/ml of total IgG), the results demonstrated that plasma has a slightly higher reactivity to spike protein than purified IgG. This difference may be attributable to antibody destruction resulting from structural changes, aggregation, and pH variation during the purification procedure [ [25] , [26] , [27] ]. The sVNT test was conducted to determine the level of neutralizing antibodies among different groups. The results demonstrated that the neutralizing antibody response of hyperimmunized rabbits is significantly greater than other groups ( Fig. 3 A and B). This distinction may stem from the presence of a plateau level of neutralizing antibodies in hyperimmunized rabbits resulting from a higher number of immunizations and potent adjuvants, compared to the variable levels of neutralizing antibodies in the vaccinated human group, which have not yet reached a plateau [ 22 , 28 ]. In addition, a significant positive correlation was observed between the neutralization IC50 value and the positive endpoint concentration of spike-specific antibodies, suggesting that an increase in the level of spike-specific antibodies corresponds to increased neutralizing antibodies ( Fig. 3 C). In line with our findings, Grunau et al. demonstrated a linear correlation between viral neutralizing antibody titers and anti-spike antibody levels in COVID-19-vaccinated individuals [ 29 ]. As illustrated in Fig. 4 , animal-derived purified antibodies can be employed to combat new EIDs. These antibodies can be produced rapidly on a large scale and contain the highest concentration of neutralizing antibodies. By increasing the neutralization potency of antibodies, the cost of treatment is reduced and open up alternative routes of administration possible that extend antibody effective half-life [ 30 ]. However, it is more challenging to prepare antibodies with high neutralization potency in vaccinated individuals due to the lengthy process of human vaccine approval. Therefore, animal-derived purified antibodies appear more appropriate than other passive immunotherapy methods for EIDs [ 14 ]. One further outcome of this research is the possibility of employing purified antibodies instead of plasma therapy. Multiple studies have demonstrated that plasma contains over a thousand proteins, of which only 20 % are antibodies and less than 2 % (∼200 μg/ml) of these antibodies are pathogen-specific [ 15 , 31 , 32 ]. The overabundance of proteins in plasma is associated with known risks, such as allergic reactions (ranging from mild to life-threatening anaphylaxis) and transfusion-related acute lung injury (TRALI), as well as a transfusion-associated circulatory overload (TACO). Purified IgG, on the other hand, has a purity of over 96 %, allowing for smaller injection volumes and easier storage and transport [ [33] , [34] , [35] ]. Notably, unlike plasma therapy, purified antibodies do not require blood group matching, and the risk of transmitting blood-borne diseases is minimized [ 35 , 36 ]. Currently, several animal-purified polyclonal antibodies are used therapeutically in patients for such applications as preventing acute organ rejection after transplantation (Atgam® and Thymoglobulin®), removing toxic levels of digoxin (Digifab®), neutralizing crotalid snake venoms (CroFab®), and treating symptomatic botulism (BAT®). These products have shown a promising safety and efficacy record, particularly in critical situations where the benefits of treatment far exceed the potential risks [ [37] , [38] , [39] , [40] ]. To mitigate the risk of hypersensitivity reactions to xenogenic antibodies, one approach is to use enzyme digestion to generate Fab fragments instead of using whole IgGs. However, a more appropriate method is to utilize transchromosomic animals. This technology allows for the production of fully human antibodies against pathogens [ 14 ]. There are a few potential limitations to this study. First, affinity-purified specific antibodies were not used in this study, which is deemed essential for future research. Second, because our laboratory did not have access to biosafety level 3, we were unable to use the gold standard conventional virus neutralization test [ 41 ]. Our research indicates that the purified antibodies from hyperimmunized animals may be a promising strategy for treating critically ill patients in new EIDs than other passive immunotherapy methods. We hope that using a more appropriate passive immunotherapy method, we will be able to control EIDs during pandemic outbreaks and reducing their associated mortality.

The use of passive immunotherapy, either as plasma or purified antibodies, has been recommended to treat the emerging infectious diseases (EIDs) in the absence of alternative therapeutic options. Here, we compare the neutralization potency of various passive immunotherapy approaches designed to provide the immediate neutralizing antibodies as potential EID treatments. To prepare human plasma and purified IgG, we screened and classified individuals into healthy, convalescent, and vaccinated groups against SARS-CoV-2 using qRT-PCR, anti-nucleocapsid, and anti-spike tests. Moreover, we prepared purified IgG from non-immunized and hyperimmunized rabbits against SARS-CoV-2 spike protein. Human and rabbit samples were used to evaluate the neutralization potency by sVNT. All vaccinated and convalescent human plasma and purified IgG groups, as well as purified IgG from hyperimmunized rabbits, had significantly greater levels of spike-specific antibodies than the control groups. Furthermore, when compared to the other groups, the purified IgG from hyperimmunized rabbits exhibited superior levels of neutralizing antibodies, with an IC50 value of 2.08 μg/ml. Additionally, our results indicated a statistically significant positive correlation between the neutralization IC50 value and the positive endpoint concentration of spike-specific antibodies. In conclusion, our study revealed that purified IgG from hyperimmunized animals has greater neutralization potency than other passive immunotherapy methods and may be the most suitable treatment of critically ill patients in EIDs. Graphical abstract Keywords

Funding Mazandaran University of Medical Sciences provided the funds for this research project (Grant No. 7306). Data availability statement Data will be made available on request. No data associated with this study has been deposited into a publicly available repository. Ethics declarations This study was reviewed and approved by the Research Ethics Committee of Mazandaran University of Medical Sciences (approval ID: IR. MAZUMS.REC.1399.106). All participants provided informed consent to participate in the study. Additional information No additional information is available for this paper. CRediT authorship contribution statement Hossein Ranjbaran: Writing – review & editing, Methodology, Investigation. Yahya Ehteshaminia: Writing – original draft, Methodology, Investigation. Mohammadreza Nadernezhad: Writing – original draft, Methodology, Investigation. Seyedeh Farzaneh Jalali: Methodology, Investigation. Farhad Jadidi-Niaragh: Supervision, Formal analysis, Data curation, Conceptualization. Abdol Sattar Pagheh: Supervision, Formal analysis, Data curation, Conceptualization. Seyed Ehsan Enderami: Supervision, Formal analysis, Data curation, Conceptualization. Saeid Abedian Kenari: Supervision, Formal analysis, Data curation, Conceptualization. Hadi Hassannia: Writing – review & editing, Supervision, Project administration, Methodology, Investigation, Funding acquisition, Data curation, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements Figures were created using BioRender software ( https://biorender.com ).

CC BY

no

2024-01-16 23:41:57

Heliyon. 2023 Dec 16; 10(1):e23478

oa_package/1e/aa/PMC10788261.tar.gz

PMC10788266

37903929

Introduction Cell and gene therapies (CGTs) are in the vanguard of modern medicine. There are more than 3500 CGTs currently in development targeting a wide array of diseases, including cancer, amyotrophic lateral sclerosis, sickle cell disease, and acquired immune deficiency syndrome, as well as many rare genetic diseases [ 1 ]. In the context of this study, we define gene therapy as the therapeutic delivery of nucleic acid into a patient’s cells to treat disease, and cell therapy as the administration of living cells into a patient’s body to treat or cure disease. These broad definitions encompass treatments for a range of diseases that are responsive to CGTs. By 2030, it is projected that up to 60 new CGTs could be launched that would affect more than 350,000 patients, making such therapies one of the most impactful areas of research and investment in medicine [ 2 – 4 ]. The adoption of such therapies is accompanied by a multitude of challenges, and the stakeholders, processes, and outputs involved in CGTs are quite different than traditional pharmaceuticals, necessitating updated approaches for planning, manufacturing, and delivery [ 5 ]. A key challenge common to all highly specialized care, including CGTs, is to deliver quality health services to a specific, often small, number of geographically dispersed patients. This is particularly relevant with CGTs, given the associated infrastructural and therapeutic costs from the demand and supply sides. The conventional approach of concentrating clinical, logistical, and infrastructural expertise and capacities in a few large centers of care is not conducive for optimized delivery of CGTs. In low- and middle-income countries (LMICs) in particular, multiple barriers exist to the access to and delivery of CGTs, including inequitable healthcare access, lack of resources, funding shortages, the prohibitive cost of CGTs, and complex regulatory systems [ 6 , 7 ]. Furthermore, CGT manufacturing, utilization, and access is concentrated in high-income countries. As of 2021, only three of the 20 approved gene therapies worldwide were approved in LMICs [ 1 , 8 ]. This inequitable landscape of CGTs is worrying, especially given that LMICs carry approximately 90% of the global burden of disease [ 9 ]. Concerted efforts are required to expand the delivery of CGTs in LMICs; otherwise, the divide between the disease burden and therapeutic availability will keep growing. Although it is true that many of the challenges faced by the administration of CGTs are common to other advanced treatments such as bone marrow transplants, CGTs have unique aspects that set them apart. These therapies are often highly personalized, requiring the modification of the patient’s own cells or the design of a therapeutic genetic sequence specific to the patient’s condition. This personalized nature poses unique manufacturing and logistical challenges. Moreover, many CGTs hold the promise of being one-time treatments that can provide lasting benefits, which brings new considerations for cost-effectiveness analyses and reimbursement models. Importantly, CGTs hold substantial potential benefits for LMICs, particularly for diseases prevalent in these areas. For example, gene therapy for β-thalassemia, a prevalent inherited disorder in LMICs, has demonstrated promising results in restoring hemoglobin production and reducing the need for blood transfusions [ 10 ]. These unique aspects necessitate specific considerations in the planning, delivery, and post-administration management of CGTs, especially in LMICs. The need to adapt traditional approaches or consider and adopt novel approaches in delivering such therapies is great. One such approach is the adoption of the hub and spoke healthcare model for delivering CGTs. In a hub and spoke model for healthcare delivery, a main health facility (hub), which receives the most resources and delivers the most intensive services, is complemented with less complex health facilities (spokes), which offer a more limited array of services [ 5 , 11 ]. Some hub and spoke models also include a partner spoke, a smaller health facility aimed at expanding access to the health services provided by hubs and spokes, especially in rural areas [ 12 ]. This model is scalable, efficient, and most importantly, adaptable based on needs and context [ 13 ]. The model has been implemented in cancer care to distribute services and improve patient outcomes, in pain management to promote telehealth programs and reach underserved areas, and in acute stroke care to enable remote consultation and timely treatment [ 12 , 14 – 17 ]. In the context of CGTs, traditional delivery models have focused on manufacturing of therapeutic products without full consideration of delivery and access [ 18 , 19 ]. Recently, calls for delivery of gene therapy, specifically adeno-associated virus–based gene therapy, through hub and spoke models have been observed [ 20 – 22 ]. We propose the application of a hub and spoke model for CGT delivery in LMICs, accompanied with a developed framework for core CGT stakeholders. This proposed hub and spoke model is intended to address a wide spectrum of these diseases, which includes inherited conditions, such as certain metabolic disorders, hemophilia, and muscular dystrophy, in which mutations in a single gene can be targeted. It also includes diseases such as cancer, in which gene therapy can be used to modify immune cells to recognize and attack cancer cells, and viral infections, in which gene therapy can potentially be used to target and eliminate the viral genome. In the context of cell therapies, our model may address conditions including cancers in which modified T cells (like CAR-T cells) can be used and autoimmune diseases in which regulatory T cells can potentially restore immune tolerance. This model is then simulated in two distinct scenarios in LMICs: a within-country scenario in Brazil, and a cross-country scenario in the Middle East and North Africa (MENA). There are a number of emerging treatments that could benefit from our proposed CGT delivery model. Advances in gene editing technologies, such as CRISPR-Cas9, are promising for addressing genetic diseases at their source. Other gene therapies, such as those targeting hemophilia, have already received approval in the Unites States [ 23 ], and clinical trials are also being conducted in Brazil. Similarly, CAR-T cell therapies are revolutionizing the treatment of certain cancers and may soon become more prevalent [ 24 – 26 ]. Our proposed model could serve as a vehicle to ensure these treatments reach patients in LMICs, further underscoring the relevance and potential of our study. CGT delivery in Brazil According to the American Society of Gene & Cell Therapy (ASCGT), 21 CGT clinical trials, nine of which are in phase 3, are underway in Brazil, which is an LMIC as defined by the World Bank [ 1 , 27 ]. The first CGT-related clinical trial in Brazil was conducted in the early 2000s, after the government developed and implemented more than 10 cellular therapy research centers in recognition of the potential of CGTs [ 28 ]. Later developments in the field eventually led to the recent resolution by the National Health Surveillance Agency (Agência Nacional de Vigilância Sanitária, ANVISA) aimed at regulating the research, development, clinical application, and registration of advanced therapeutic products in Brazil. More recently, ANVISA, in collaboration with the United Nations Development Program, called for the development of the National Network of Specialists in Advanced Therapies (Rede Nacional de Especialistas em Terapias Avançadas, RENETA), a network of experts that support the evaluation of clinical trials, product registration, and monitoring processes of advanced therapeutic products. RENETA also assists in building human capacity at ANVISA to ensure proper regulation over advanced therapeutic products [ 29 ]. Thus, Brazil represents a country that is actively engaging with the CGT space. The government’s proactive approach to the development of cellular therapy research centers and its recent regulatory resolutions demonstrates a strong commitment to facilitating the adoption of CGTs. CGT delivery in the MENA region CGT delivery is still in its infancy in the MENA region. The first record of gene therapy administration in this region dates to 2018, where the first dose of nusinersen, a gene therapy product for patients with spinal muscular atrophy, was administered at the East Jeddah Hospital in Saudi Arabia [ 30 ]; the drug is now approved in Kuwait, Qatar, and the United Arab Emirates (UAE) [ 31 ]. In 2019, voretigene neparvovec, a gene therapy treatment for inherited retinal blindness, was approved by the Ministry of Health and Prevention and the Saudi Food and Drug Authority in the UAE and Saudi Arabia, respectively [ 32 , 33 ]. Another gene therapy product, onasemnogene abeparvovec, was used to treat children with spinal muscular atrophy in Egypt, Saudi Arabia, Kuwait, Qatar, and the UAE [ 34 – 39 ]. Outside of North America, the Hamad General Hospital in Qatar was the first to administer this drug. For cell therapy products, tisagenlecleucel, a chimeric antigen receptor T-cell therapy product, was approved by the Saudi Food and Drug Authority in 2022 [ 40 ]. This reflects the variability in resources and capacities across the MENA region. Indeed, according to the World Bank, the MENA region includes high-income (Bahrain, Kuwait, Oman, Qatar, Saudi Arabia, UAE) and low- and middle-income countries (Lebanon, Djibouti, Egypt, Iran, Iraq, Morocco, Tunisia, Jordan, Libya, Morocco, Syria, Tunisia, West Bank and Gaza, Yemen) [ 27 ].

Materials and Methods Stakeholder framework and mapping The core stakeholder framework and mapping for CGT delivery was informed by the existing literature in addition to the websites and online documentation of relevant pharmaceutical, governmental, academic, and information technology sectors. A major source of information, BioPhorum, a leading business-to-business membership organization with more than 6000 subject matter experts, has developed more than 30 documents on CGT. These documents include a detailed analysis and description of CGT-related challenges [ 41 ], considerations [ 42 ], and more importantly, process maps [ 43 , 44 ] and actors [ 45 ]. Another example is the joint partnership between the European Association for Hemophilia and Allied Disorders and the European Hemophilia Consortium to develop a framework for a hub and spoke model for delivery of hemophilia CGT [ 46 ]. In Brazil, facilities for cell processing were extracted from a data set on cell processing centers [ 47 ] and the National Cell Therapy Network (Rede Nacional de Terapia Celular, RNTC) website [ 48 ]. Facilities involved in CGT clinical trials were extracted from the literature, clinicaltrials.gov [ 2 ], the RENETA website [ 29 ], and the ASCGT website [ 1 ]. Information on potential CGT manufacturers in Brazil was extracted from CGT clinical trial sponsors, which may or may not be the CGT manufacturer, in addition to the literature and governmental or pharmaceutical websites in Brazil [ 49 – 51 ]. To decide whether the health facilities identified were academic centers, the corresponding website and SUS (Sistema Único de Saúde) information page of the site were reviewed. Accreditation of health facilities was reviewed using the Foundation for the Accreditation of Cellular Therapy and the Association for the Advancement of Blood & Biotherapies/Associação Brasileira de Hematologia, Hemoterapia e Terapia Celular accreditation websites [ 52 , 53 ]. Health facilities able to perform hematopoietic stem cell transplantation were extracted from DATASUS data set [ 54 ]. In the MENA region, cell processing centers were extracted from data sets, websites, and publications on cord blood banks and facilities in the Middle East [ 55 – 59 ]. It is important to note that cord blood banks may or may not have the capacity to process cells for cell therapies. Facilities involved in CGT clinical trials were extracted from the literature as well as the clinicaltrials.gov website [ 2 ] and the ASCGT website [ 1 ]. Information on potential CGT manufacturers in the MENA region was extracted from CGT literature and governmental/pharmaceutical websites in the region. Only two pharmaceutical companies in Saudi Arabia were identified, Saudi Vax and Sudair Pharma [ 60 , 61 ]. The Abu Dhabi Stem Cell Center in the UAE is also advertised to locally engineer chimeric antigen receptor T-cells [ 62 ]. The sponsors identified in CGT clinical trials in the MENA region are multinational companies with offices in the Middle East, but information on whether such companies manufacture CGT products in the region was not available. Shortlisting hubs, spokes, and partner spokes for simulation Given that CGT delivery is still in its infancy in Brazil, CGT clinical trials are a rich source of information for simulating the CGT hub and spoke model. Sites that can perform and lead clinical trials have the capacity to administer CGTs. Thus, the health facilities involved in CGT clinical trials were shortlisted as hubs. One of the main criteria suggested for a CGT spoke is having the clinical capacity to administer at least one type of CGT therapy and host processing facilities. Bone marrow transplantation requires collection, processing, cryopreservation, and reinfusion, which are processes similar to CGT delivery. Thus, to identify potential CGT spokes, a list of hospitals that provide bone marrow transplantation services in Brazil were extracted. Those that provide both allogenic and autogenic bone marrow transplantation were shortlisted, and those that were listed as hubs were filtered out. Potential CGT partner spokes were extracted from the DATASUS data set, particularly health facilities that have hematological and/or hemotherapy care units. This would satisfy two main criteria of CGT partner spokes: the capacity to perform apheresis and the presence of qualified care professionals. Importantly, many hematological and/or hemotherapy care units are hosted within basic health units, which are highly embedded within communities in Brazil. In the MENA region, consolidated data sets on types of facilities and associated services were absent, and fragmented data from international and governmental websites were core sources. In addition to health facilities involved in CGT clinical trials, health facilities in the MENA region that already provide CGT services were extracted through literature search and news, as well as governmental and nongovernmental websites [ 62 – 65 ]. These were also included in the list of potential hubs. For potential spokes, a partial list of bone marrow transplantation centers in the MENA region was retrieved from the European Society for Blood and Marrow Transplantation and the Eastern Mediterranean Blood and Marrow Transplantation membership documentation [ 66 , 67 ]. Other centers were extracted through literature search, websites for health facilities, and governmental and social media sites [ 68 – 75 ]. The final list of 37 potential spokes was produced after filtering out hospitals that were already shortlisted as hubs. Because of an absence of data sets on services provided by health facilities in the MENA region, potential CGT partner spokes were considered to be the rest of the hospitals that were not shortlisted as hubs or spokes. The rationale is that these hospitals, many of which are municipal, are well embedded into communities in the MENA region and would have qualified personnel for potential apheresis. Importantly, these facilities would also have the capacity to screen patients and refer them to spokes. For both Brazil and the MENA region, shortlisted hub, spoke, and partner spoke criteria and respective health facilities were coded when producing the matrices to allow for a more streamlined assessment. Geographic mapping All GIS mapping was done using QGIS, a free and open-source cross-platform desktop geographic information system application that supports viewing, editing, printing, and analysis of geospatial data [ 76 ]. In the simulation models, a distance to point function was used in QGIS to measure the closest distance between the clinical trial sites and manufacturer. Sites that were within 20 kilometers of a manufacturer were considered physically close.

Results Developing a framework for core CGT stakeholders for CGT delivery The evolution of the current linear and centralized delivery model of CGT into a nonlinear hub and spoke model requires a detailed understanding and mapping of the current and potential roles of relevant stakeholders. The stakeholders involved in the different categories of CGT (i.e., cell therapies, gene modified cell therapies, gene therapies, and tissue-engineered products [ 10 ]) are variable. It is therefore important to draw on all of these to formulate a common framework of stakeholders with their core roles. The proposed framework includes four direct and two indirect CGT delivery stakeholder types (Fig. 1 , Supplementary Material). Core CGT stakeholders include beneficiaries (patients or donors), therapeutic centers, treatment coordination actors, manufacturing stakeholders, payers, and regulatory agencies. A therapeutic center collects specimens and performs pretesting on patients to inform eligibility, initiates the supply chain, prepares the patient for therapy, and administers the therapy. Treatment coordination involves planning, patient operations, accounting, an orchestration platform, and courier services to ensure proper scheduling, quality, and risk assessment of the supply process. Manufacturing analyzes specimens, prepares cells, and develops and tests therapeutic products using sequencing labs, cell preprocessing facilities, and vector manufacturing facilities. Coordination between one or more payers initiates the CGT therapeutic process, and regulatory agencies authorize and monitor the CGT pipeline. Process flow mapping for the various types of CGTs is presented in Supplementary Fig. 1 (stock genetic vector), Supplementary Fig. 2 (personalized gene therapy), Supplementary Fig. 3 (allogenic cell therapy), and Supplementary Fig. 4 (autologous cell therapy). The hub and spoke model for CGT delivery The proposed hub and spoke model for CGT delivery comprises three main interconnected components: hub, spoke, and partner spoke. Briefly, the CGT hub is a leading academic medical center that is experienced in both comprehensive care and delivering CGT. A spoke is a healthcare center that has minimal CGT experience but will serve as the home center for patients. A partner spoke is a supporting facility that is not necessarily a health center but facilitates the function of spokes within the system (Supplementary Material). The CGT hub is an advanced medical center specializing in CGT research and practice, with the therapeutic center having the capacity for screening, diagnosing, and treating all categories of CGT (Fig. 2A ). The hub also possesses cryopreservation and apheresis facilities, as well as co-located manufacturing facilities, and works in collaboration with biotech and pharmaceutical companies. It should have an established logistics and supply chain, with compliant manufacturing and an existing CGT registry, and offer training to HCPs. Hub and spoke personnel should collaborate to ensure optimal patient outcomes, and the orchestration platform should include a visibility and monitoring unit and an information technology harmonization unit. The CGT spoke is responsible for administering at least one type of CGT, with apheresis centers housed in spokes. A spoke should have a cell/tissue processing facility, optimize donor starting material, and provide referral routes to hubs. The spokes also act as donation centers and should standardize treatment protocols and process flows in coordination with the hub. A treatment coordinator should be stationed at spokes, and HCPs should oversee patient information sharing (Fig. 2B ). The CGT partner spoke operates in small communities without full accreditation, serving as screening, referral, and collection centers to optimize cell transit time and simplify logistics (Fig. 2C ). Outpatient centers and clinics aid in increasing referrals to spokes for CGTs and increasing access to clinical trial participants. Partner spokes should have qualified personnel, and collection centers at partner spokes may coordinate with spokes for proper shipping and initiation of the manufacturing process. A summary of the capacities of core stakeholders required at the hubs, spokes, and partner spokes, including beneficiaries, therapeutic centers, treatment coordination, and manufacturing, is provided in Supplementary Tables 1 – 4 . Simulation of the hub and spoke model in Brazil and the MENA region The theoretical clinical utility and cost effectiveness of a hub and spoke model for delivering CGT does not guarantee its successful adoption. It is thus imperative to simulate institutional readiness for potential adoption of such models. To pressure test the hub and spoke institutional criteria developed, we simulated the adoption of a within-country hub and spoke model of CGT delivery in Brazil and a cross-country model of CGT delivery in the MENA region. First, we mapped the distribution of key facilities that may be involved in a potential hub and spoke model for CGT delivery in Brazil. These included cell processing centers (Supplementary Table 5 ), health facilities involved in CGT clinical trials (Supplementary Table 6 ), and potential manufacturers (Supplementary Table 7 ). The distribution of facilities reflects the health inequality in Brazil, whereby CGT-related facilities were found to be concentrated in major cities in Brazil, especially in São Paulo (Fig. 3 ). Next, shortlisted hubs (Supplementary Tables 8 , 9 , and Supplementary Fig. 5 ), spokes (Supplementary Fig. 6A ), and partner spokes (Supplementary Fig. 6B ) were assessed against the respective matrices developed previously (Figs. 2 , 3 ). Finally, the locations of the shortlisted hubs, spokes, and partner spokes were populated, and the shortest distance between hub and spoke/spoke and partner spoke was visualized. Two hub and spoke models were simulated in Brazil. The first simulation included all clinical trial sites as hubs (Fig. 4A ). This scenario allows for a better population reach; however, designating 35 sites as hubs may be challenging. The second simulation included only clinical trial sites that performed both gene and cell therapy trials as hubs (Fig. 4B ). Importantly, in both models, the hubs, spokes, and partner spokes are concentrated in the southeast region of Brazil, where health access and development are most advanced. Similar to the simulation in Brazil, an extensive mapping exercise was undertaken to identify the potential health facilities that may serve as hubs, spokes, and partner spokes in the MENA region. The identity, location, and geographic distribution of cell processing centers (Supplementary Table 10 ), potential manufacturers, and health facilities involved in CGT clinical trials (Supplementary Table 11 ) were populated. Potential CGT facilities were concentrated in the Gulf Cooperation Council countries, like the UAE and Saudi Arabia, where heavy investments in health infrastructure were made during the past decade (Fig. 5 ). Next, health facilities that were shortlisted as hubs (Supplementary Tables 8 and 12 , and Supplementary Fig. 7 ), spokes (Supplementary Fig. 8A ), and partner spokes (Supplementary Fig. 8B ) were assessed against the developed matrices. Finally, the location of the shortlisted hubs, spokes, and partner spokes was used to simulate the hub and spoke model in the MENA region (Fig. 6 ).

Discussion Access to CGTs varies massively, with LMICs less likely to benefit, especially given the prohibitive cost associated with such therapies. The preliminary, adaptable roadmap for more equitable CGT delivery in LMICs we propose aims to increase access to CGTs in these countries, but the hub and spoke model is also associated with significant infrastructural, clinical, administrative, and policy requirements. Such requirements are likely to be lacking in most LMICs, and thus the model should be adapted to fit each contextual factor of each country or region. The simulation of the models in Brazil and the MENA region also revealed many opportunities and challenges for future consideration. The model requires an existing infrastructure that is conducive for expanded CGT services. Brazil has a well-established infrastructure for enabling CGT services and was identified as one of three LMICs to have activity across five advanced therapies/related categories: dedicated government funding, goods and services, pharmaceutical and non-pharmaceutical firms, publications, and academic groups [ 28 ]. Further, new centers for CGT manufacturing are under development in Brazil, bolstering local manufacturing of such therapies, and more than 20 CGT clinical trials are underway, setting the stage for increased adoption of CGTs [ 49 ]. The proposed model is expected to enhance patient catchment, improve market coverage, and expand patient access to CGTs. Indeed, many of the health centers that were shortlisted as hubs in Brazil are academic medical centers with previous experience in CGT, since they are involved in CGT clinical trials and located near manufacturing facilities in major cities in Brazil. Shortlisted spokes, which have the capacity to collect, process, cryopreserve, and reinfuse, would serve as the treatment home for patients. Shortlisted partner spokes in Brazil, which have apheresis capabilities, are well embedded within communities throughout the country. The simulation in Brazil has also uncovered challenges that should be considered in future development or potential implementation of the model. First, much of North and West Brazil do not include any hub in the proposed model, which vastly increases distances between hubs and spokes, creating logistical and infrastructural challenges. Second, the model was based on data sets/information from government, nongovernment, pharmaceutical, and literature sources and websites. Many government data sets have not been updated since 2016, and most information needed for full designation of hubs/spokes/partner spokes could not be identified. This may have hindered the identification of potential hubs in North and West Brazil. Engaging the different stakeholders may better inform the potential designation of hubs. Third, the simulation adopts the shortest distance between hub/spoke and spoke/partner spoke. An optimal approach would incorporate the geographic proximity of hubs/spokes/partner spokes, potential clients, distance decay, availability of services relative to demand, road network travel time, and patient catchment into the model. However, building such a model requires extensive data collection across Brazil, and these data sets are currently not available [ 77 – 80 ]. The developed simulation highlights at its core the inequitable capacity and resources for healthcare in the MENA region. The simulated model is polarized: while potential hubs were mostly identified in high-income countries in the Gulf region, especially in Saudi Arabia, not a single spoke was identified in Yemen. Regional health coordination mechanisms are absent in this region, except for the Gulf Cooperation Council countries through the Gulf Health Council, which is hosted in Saudi Arabia and serves six countries. These coordination mechanisms will be essential in planning, executing, and maintaining a hub and spoke model for CGT delivery in the MENA region. Capacity in the MENA region to provide CGTs is severely hindered by a multitude of challenges, especially in countries impacted by political conflict. For example, the total number of clinical trials in Yemen is four, and no bone marrow transplantation facilities could be identified. In Palestine, no clinical trials are registered in international databases [ 81 ]. Still, opportunities for implementing the hub and spoke model for CGT delivery in the MENA region are numerous. First, expanding CGT delivery in the MENA region is already underway. In Saudi Arabia, the Ministry of Investment signed a Memorandum of Understanding with Novartis to expand local activities for CGT, including transfer of technology, research and development, and building local capacities [ 82 ]. This will ensure better clinical translation and local manufacturing of such therapies. Second, calls for implementing a hub and spoke model for healthcare delivery have been raised, specifically in the Gulf Cooperation Council area [ 83 ]. A Saudi BioPharma company named Saudi Vax has signed a letter of intent to establish a facility that will become a CGT hub in the region [ 60 ]. The role of regulatory bodies in the administration of CGTs is pivotal. From the approval of clinical trials to the authorization for market access, regulatory authorities play a critical part in the CGT landscape. Within the hub and spoke model, regulatory approval plays a crucial role in defining the responsibilities of different centers, the administration of CGTs, and patient management. The hubs, which are the main healthcare facilities, would need stringent regulatory compliance and accreditation because of the complex nature of the services they offer. Spokes and partner spokes, however, would manage less complex procedures but still need to adhere to the standards set by regulatory bodies. During clinical trials, the regulatory focus is on safety and efficacy, which would require close monitoring of patients and rigorous data collection at all centers. Post-approval, the focus shifts toward broader patient accessibility, long-term safety, and real-world effectiveness. This might require changes in the roles of hubs, spokes, and partner spokes in the model, and also in the processes for patient management and therapy administration. In addition to establishing robust regulatory frameworks, the implementation of the hub and spoke model necessitates developed infrastructure, with the hubs requiring the most resources and delivering the most complex services. Indeed, many low-income and lower-middle-income countries currently lack the necessary infrastructure, including transplantation centers and apheresis facilities, for the widespread implementation of CGTs. This infrastructural gap presents a substantial challenge for these countries and, without targeted efforts to build capacity and develop necessary infrastructure, could exacerbate existing health inequities. In addition, capacity building is also essential, as healthcare personnel need specialized training to administer CGTs. Partnerships with pharmaceutical companies, academic institutions, and international organizations can help with both capacity building and financing. Finally, innovative financing strategies are needed to manage the high costs associated with CGTs, ensuring that these life-changing therapies are accessible to all who need them. Although the hub and spoke model can enhance access to CGTs in LMICs, it is essential to acknowledge the disadvantages. One disadvantage is the potential to further exacerbate existing health inequities. As noted in the simulations in Brazil and the MENA region, medical centers that are equipped to handle CGTs are often located in urban or more affluent areas or countries. This can create a disparity in access to these advanced treatments, further advantaging the already privileged population. It is crucial to address this issue and ensure that access to CGTs is equitable and reaches underserved populations. Moreover, investing in CGT in LMICs may have substantial population health implications. While these therapies hold promise for treating specific diseases, diverting financial resources toward their development and implementation could potentially hinder other essential healthcare initiatives. Many LMICs face challenges in providing basic healthcare services, such as newborn screening, cancer screening, and other preventive measures that have a broader population impact. Prioritizing CGTs without addressing these fundamental healthcare needs could perpetuate inequities and neglect the broader health needs of the population. This study provides insight into the long road ahead to ensure expanded access to CGTs in LMICs. The inclusion of pharmaceutical, clinical, and policy stakeholders from LMICs in the planning, development, and expansion of CGT services is a must. Finally, further investigation into the practical adoption of such a model, focusing on the development of documentation required for health institutions to evaluate readiness for and eventual adoption of the proposed hub and spoke model, is needed.

In the rapidly evolving landscape of biotechnologies, cell and gene therapies are being developed and adopted at an unprecedented pace. However, their access and adoption remain limited, particularly in low- and middle-income countries (LMICs). This study aims to address this critical gap by exploring the potential of applying a hub and spoke model for cell and gene therapy delivery in LMICs. We establish the identity and roles of relevant stakeholders, propose a hub and spoke model for cell and gene therapy delivery, and simulate its application in Brazil and the Middle East and North Africa. The development and simulation of this model were informed by a comprehensive review of academic articles, grey literature, relevant websites, and publicly available data sets. The proposed hub and spoke model is expected to expand availability of and access to cell and gene therapy in LMICs and presents a comprehensive framework for the roles of core stakeholders, laying the groundwork for more equitable access to these lifesaving therapies. More research is needed to explore the practical adoption and implications of this model. Subject terms

Supplementary information

Supplementary information The online version contains supplementary material available at 10.1038/s41434-023-00425-x. Acknowledgements Editorial support was provided by Jennifer Weintraub of Kay Square Scientific, Newtown Square, PA, USA. This support was funded by Novartis Gene Therapies, Inc. Author contributions Conceptualization: SS, OD, MT, SDS. Data Collection: SS, OD, AP, SDS. Data Analysis: All authors. Writing of the First Draft: SS. All authors have reviewed and approved the manuscript for submission. Funding This study was funded by Novartis Gene Therapies, Inc. Data availability Data generated for this study are included in the supplementary material. Additional data are available from the corresponding author on reasonable request. Competing interests OD, DA, DT, MD, AP, and BKT are employees of Novartis Gene Therapies, Inc., and own stock/other equities. SS and MT have received consulting fees from Novartis Gene Therapies, Inc., for this research. SDS has received research grants and consulting fees from Novartis Gene Therapies, Inc., and served as consultant/on an advisory board for Bayer US. ST has received honoraria from Novartis Gene Therapies, Inc. He has also received payment for participation on a BioMarin advisory board and participated in a scientific advisory board meeting for UCB Pharma.

CC BY

no

2024-01-16 23:41:57

Gene Ther. 2024 Oct 30; 31(1-2):1-11

oa_package/81/23/PMC10788266.tar.gz

PMC10788267

38225934

Introduction Glycosylphosphatidylinositol (GPI) anchors various proteins to the plasma membrane. There are over 150 types of GPI-anchored proteins (GPI-APs) in mammalian cells, and they play various roles in fertilization, development, and immune responses as enzymes, adhesion molecules, receptors, and complement regulatory proteins. 1 Thirty genes are involved in the biosynthesis and modification of GPI-APs, and variations in these genes can cause inherited GPI deficiency (IGD). Complete deficiency of GPI is lethal because the absence of over 150 GPI-APs on the cell surface is not viable; therefore, most IGD patients have a partial GPI deficiency. No effective treatment for IGD is available at present. To elucidate IGD pathology and to develop treatments, we and others previously generated IGD model mice, in which one of the genes required for GPI biosynthesis, phosphatidylinositol (PI) glycan anchor biosynthesis class O ( PIGO ) and PIGV gene, respectively, was partially deficient. 2 , 3 We reported the effectiveness of adeno-associated virus (AAV)–based gene therapy in Pigo -deficient model mice using a gene editing strategy, homology independent targeted integration assisted with a low level of transgene expression. 2 Whereas PIGO and PIGV function in the middle of the GPI biosynthesis pathway, PIGA is involved in the first step of GPI biosynthesis, the transfer of N -acetylglucosamine (GlcNAc) from UDP-GlcNAc to PI to generate GlcNAcPI. This step is mediated by the GlcNAc transferase complex consisting of PIGA, PIGH, PIGC, PIGP, PIGQ, PIGY, and DPM2. 1 PIGA is a catalytic component and is essential for this reaction. PIGA is X-linked; therefore, only males who receive a variant allele from their mothers are affected. The main symptoms of PIGA deficiency are neurological abnormalities, such as developmental delay, intellectual disability, and seizures. To further investigate the effectiveness of AAV-based gene therapy of IGD, we generated CNS-specific Piga knockout (KO) mice by crossing CNS-specific Cre recombinase expressing Nestin-Cre mice with Piga -floxed mice. 4 Nestin is expressed at approximately embryonic day (E) 7.5 during neuronal development. 5 In the hemizygous Piga KO male mice, GPI-APs would be lost from neurons, astrocytes, and oligodendrocytes if Cre-mediated recombination occurs, whereas in the heterozygous Piga KO female mice, GPI-APs would be lost from half of those CNS cells, in which Cre-mediated recombination occurs because of X-inactivation. Similar to the previous report, 6 male Piga KO mice die by approximately postnatal day (P) 10 after birth and have severely decreased levels of GPI-APs in the brain, whereas female Piga KO mice have a severe defect in myelination and die by approximately P25. Using these model mice as an evaluation system, we developed AAV-based gene therapy for PIGA deficiency. Here, we show that AAV-based gene replacement therapy is effective for improving some of the phenotypes of CNS-specific Piga KO mice. However, liver cancer developed in all three treated mice after 1 year. Although the occurrence of liver cancer has not been reported for AAV-based gene therapy in humans, careful consideration is needed in the dose, route of administration, and selection of suitable promoters for AAV-based gene therapy of IGD.

Materials and methods Generation and genotyping of mice Piga -floxed mice were generated in our laboratory. 4 CNS-specific Cre expressing transgenic mice, B6.Cg-Tg( Nestin-Cre ) RBRC02412, were provided by the RIKEN BioResource Research Center through the National BioResource Project of the Ministry of Education, Culture, Sports, Science, and Technology, Japan. Homozygous Piga -floxed female mice were crossed with Nestin-Cre mice. Piga is X-linked; therefore, male mice with the Cre transgene were CNS-specific Piga KO mice, whereas female mice with the Cre transgene were mosaic for Piga expression because of X-inactivation. CNS-specific Piga KO mice were sacrificed at the indicated age and whole brains, hearts, and skeletal muscles were taken out. gDNA was isolated after homogenization. Primers for wild-type and Piga -floxed alleles were primer 1, 5′-ACCTCCAAAGACTGAGCTGTTG-3′, and primer 2, 5′-CCTGCCTTAGTCTTCCCAGTAC-3′ (fragment sizes 420 and 250 bp, respectively); primers for the targeted allele were primer 1 and primer 3, 5′-TGTGGGTTTCAGTTCATTTCAGA-3′ (fragment size 550 bp) ( Figures 1 A and 1B); those for the Cre transgene were primer 4, 5′-AGGTTCGTTCACTCATGGA-3′, and primer 5, 5′-TCGACCAGTTTAGTTACCC-3′ (fragment size 235 bp). Mice were maintained in a specific pathogen-free animal facility at the Research Institute for Microbial Diseases, Osaka University, Japan. Animals Mice were maintained under a 12-h light/12-h dark cycle in a temperature-controlled environment, with food and water provided ad libitum . All of the animal procedures were approved by the Animal Care and Use Committee of the Research Institute for Microbial Diseases, Osaka University, and were carried out in accordance with the approved guidelines. Nestin-Cre transgenic mice were maintained by mating with wild-type C57BL/6 mice and Piga -floxed mice were maintained by mating female homozygous Piga -floxed mice with hemizygous Piga -floxed male mice. Histological analysis of the mouse brain and liver Mice were anesthetized and fixed by cardiac perfusion with 4% paraformaldehyde in 0.1 mol/L phosphate buffer (pH 7.2). Brains were removed from the mice and further immersed in the same fixative overnight at 4°C. Samples processed for paraffin embedding were cut into 5-μm sections with a semimotorized rotary microtome (RM2245; Leica, Nussloch, Germany) and placed on silane-coated glass slides. Samples for cryosections were embedded in optimal cutting temperature compound (Sakura Finetek, Tokyo, Japan) after cryoprotection in 10% and 20% sucrose in 0.1 mol/L phosphate buffer (pH 7.2) and sectioned at 10 μm with a cryostat (CM3050; Leica). The sections were placed on silane-coated glass slides and stored at −80°C until used. Meyer’s H&E staining was performed on paraffin-embedded sections. For MBP immunohistochemistry, deparaffinized sections were stained with rat anti-MBP IgG (no. MCA409S) overnight at 4°C, further incubated with biotinylated goat anti-rat IgG for 1 h, and finally with peroxidase-conjugated streptavidin (Vector Laboratories, Newark, CA) for 1 h at room temperature. Staining for peroxidase was performed using 0.0125% 3,3′-diaminobenzidine tetrahydrochloride and 0.002% H 2 O 2 in 0.05 mol/L Tris-HCl buffer (pH 7.6) for 10 min. For HA immunohistochemistry, cryosections were incubated with rabbit anti-HA IgG (Cell Signaling Technology, Danvers, MA) and further incubated with fluorescein isothiocyanate (FITC)-conjugated donkey anti-rabbit IgG. Samples were analyzed using a BZ-X800 microscope (Keyence, Osaka, Japan). Electron microscopy Mice were fixed by cardiac perfusion with 2% paraformaldehyde and 2% glutaraldehyde in 0.1 M phosphate buffer (pH 7.2). Brains were removed and 1-mm-thick brain slices were postfixed with 2% paraformaldehyde and 2% glutaraldehyde in 0.1 M phosphate buffer (pH 7.4) overnight followed by postfixation with 1% OsO 4 , dehydration with a graded ethanol series and embedding in Epon812 (Oken Shoji, Tokyo, Japan). Ultrathin sections were cut with an ultramicrotome UC6 (Leica Microsystems), stained with uranyl acetate and lead citrate, and examined with a transmission EM HT7700 microscope (Hitachi, Tokyo, Japan). Western blotting After perfusion with saline, mouse brains were homogenized and solubilized in 60 mM n -octyl-β- d -glucoside-containing lysis buffer, followed by centrifugation to remove debris. After bicinchoninic acid assay measurement of protein content, lysates were processed by SDS-PAGE followed by western blotting. The primary antibodies used were rat anti-MBP (MCA409S, Bio-Rad, Hercules, CA) and mouse anti-glyceraldehyde 3-phosphate dehydrogenase (GAPDH) (AM4300, Thermo Fisher Scientific, Waltham, MA). Secondary antibodies used were horseradish peroxidase–conjugated anti-rabbit, anti-rat, or anti-mouse IgG. Generation of AAV pscAAV- CAG-GFP was purchased from Addgene (no. 83279). pscAAV- CAG-2HA-PIGA was generated from pscAAV- CAG-GFP by the replacement of an AgeI/NotI fragment containing GFP with PCR amplified 2HA-human PIGA . AAV was packaged by the lipofection of pscAAV- CAG-2HA-hPIGA with PHPeB capsid 35 and pAd5 helper plasmid (Addgene) into AAVpro 293T cells (Takara Bio, Kusatsu, Japan), which were purified by polyethylene glycol precipitation followed by iodixanol gradient ultracentrifugation. 36 Viral titer was determined by qPCR using Taq-Man technology (Thermo Fisher Scientific). Intravenous AAV injection into newborn Piga -deficient mice Newborn (P1–P2) Nestin-Cre / Piga -floxed mice were subjected to intravenous AAV injection as described previously. 37 Before the procedure, pups were anesthetized by placing on ice for 1 min. They were subsequently injected via the temporal vein using a 30G insulin syringe with needle (Lo-Dose Insulin Syringe with needle, 30G, 1/2 mL, Becton Dickinson, Franklin Lakes, NJ). Pups were then allowed 2–3 min to rewarm and recover and were then returned to their cage. MRI We conducted in vivo and ex vivo MRI of mice using an 11.7-T vertical bore scanner (AVANCE II 500WB; Bruker BioSpin, Ettlingen, Germany). In vivo T 2 weighted brain MRI images of Piga +/− mice at 21 days of age and their wild-type littermates was compared ( Figure 3 A). Mouse anesthesia was initially induced with 2% isoflurane and maintained with 1.6% isoflurane during MRI. Body temperatures of mice were maintained at 37°C with circulating warm water. In vivo T 2 weighted images were obtained by the rapid acquisition with relaxation enhancement technique. 38 The acquisition parameters were field of view = 15 × 15 mm, matrix size = 256 × 256, in-plane resolution = 59 μm, slice thickness = 300 μm, repetition time = 5000 ms, echo time = 39.5 ms, number of averages = 16, and acquisition time = 21 min. Ex vivo diffusion weighted MRI of mice was performed to compare the brain regions and brain lengths. AAV-treated male Piga −/ mice and their wild-type littermates were imaged at 17 days old. Ex vivo images of AAV-treated female Piga +/− mice and their wild-type littermates were obtained at 19 days old ( Figure S2 ). The acquisition parameters for ex vivo diffusion weighted MRI were field of view = 15 × 15 mm, matrix size = 256 × 256, in-plane resolution = 59 μm, slice thickness = 300 μm, repetition time = 5,000 ms, echo time = 19.5 ms, b-value = 3,000 s/mm 2 , number of averages = 6, and acquisition time = 2 h 8 min. Ex vivo T 2 weighted imaging of AAV-treated Piga +/− mice and their wild-type littermates at 19 and 54 days old was performed ( Figure S3 ). The acquisition parameters for ex vivo T 2 weighted imaging were field of view = 15 × 15 mm, matrix size = 256 × 256, in-plane resolution = 59 μm, slice thickness = 300 μm, repetition time = 5,000 ms, echo time = 50 ms, number of averages = 40, and acquisition time = 53 min. Video EEG recordings and analysis Adult mice at 4 months of age or older were used for EEG recordings as previously described. 2 In brief, mice were anesthetized with isoflurane and implanted with EEG electrodes (no. 8201: 2 EEG/1 EMG Mouse Headmount, Pinnacle Technology, Parsippany, NJ) according to the manufacturer’s instructions. The electrodes were attached to thin cables linked to a computer running software that allowed the visualization of EEG activity with simultaneous video recording (Vital Recorder, Kissei Comtec, Nagano, Japan). Video EEGs were recorded overnight, and the files were reviewed for background activity, epileptic discharge, and seizure activity (SleepSign, Kissei Comtec, Matsumoto, Japan). In addition, low-dose pentylenetetrazole (20 mg/kg) was administered intraperitoneally, and pentylenetetrazole-induced seizure susceptibility was evaluated using the modified Racine scale. 39 Frequency and amplitude in 8 h of EEG data (from 8:00 to 4:00) were calculated by fast Fourier transform power spectral analysis. 40 Data were analyzed in 10-s epochs using SleepSign software. The EEG signal was separated into five regions per epoch. Each region was fast Fourier transform calculated using 256 datum points (2 s) before the 5 spectra were averaged. The spectrum had a resolution of 0.5 Hz. Animal behavioral analysis Muscle weakness and coordination deficit were measured by the four-limb hanging test. The latency to fall was recorded with a 3-min cutoff time. The performance of AAV-treated female variants was compared with that of wild-type controls. Working memory and exploratory activity were measured using a Y-maze apparatus (arm length: 40 cm, arm bottom width: 3 cm, arm upper width: 13 cm, height of wall: 15 cm; BrainScience Idea, Osaka, Japan). Each mouse was put in the Y-maze, one arm of which was blocked, for 5 min. One hour later, each mouse was placed in the bottom area of the Y-maze with both arms open. The number of entries into the arms were recorded for 5 min. Working memory and activity were calculated as the number of correct alterations/number of total new arm entries, as previously described. 41 Proteomic analysis of the mouse brain Six-day-old brains of Nestin-Cre/ Piga -floxed male mice, AAV-treated mice, and wild-type mice were homogenized and lysed in 500 μL of lysis buffer (10 mM Tris-HCl, pH 7.4, 150 mM NaCl, 5 mM EDTA, protease inhibitor cocktail) containing 2% Triton X-114 (Nakalai Tesque, Kyoto, Japan) for 30 min on ice. After centrifugation at 21,900 × g at 4°C for 15 min, the supernatant was incubated for 10 min at 37°C and aqueous and detergent phase separation was performed by centrifugation at 5,600 × g at 37°C for 7 min. After recovering the aqueous phase, 350 μL of lysis buffer was added to the detergent phase and further incubated with phosphatidylinositol-specific phospholipase (1 U/mL) at 16°C for 2 h. Phase separation was performed and the aqueous phase was combined with the previously recovered aqueous phase. Four volumes of acetone were added to the combined aqueous and detergent phases, and protein precipitation was performed at −80°C for 1 h. Protein precipitate was pelleted by centrifugation at 13,400 × g for 30 min at 4°C and then air dried. The air-dried pellet was dissolved in 20 μL 0.1% RapiGest (Waters, Milford, MA) and reduced with 10 mM DTT, followed by alkylation with 55 mM iodoacetamide, digestion with trypsin, and purification with a C18 tip (AMR, Tokyo, Japan). The purified peptides were subjected to nanocapillary reversed-phase liquid chromatography with tandem mass spectrometry (LC-MS/MS) analysis using a C18 column, Nikkyo NTCC-360 (75 μm × 150 mm, 3.0 μm, Nikkyo Technos, Tokyo, Japan) in a nanoLC system (Bruker Daltonics, Billerica, MA) connected to a timsTOF Pro mass spectrometer (Bruker Daltonics) and a modified nano-electrospray ionization source (CaptiveSpray; Bruker Daltonics). The mobile phase consisted of water containing 0.1% formic acid (solvent A) and acetonitrile containing 0.1% formic acid (solvent B). Linear gradient elution was carried out from 2% to 35% solvent B for 22 min at a flow rate of 400 nL/min. The ion spray voltage was set at 1.6 kV in the positive ion mode. Ions were collected in the trapped ion mobility spectrometry device over 100 ms and MS and MS/MS data were collected over an m/z range of 100–2,000. During the collection of MS/MS data, the trapped ion mobility spectrometry cycle was adjusted to 0.53 s and included 1 MS plus 4 parallel accumulation serial fragmentation-MS/MS scans, each containing on average 12 MS/MS spectra (>100 Hz). 42 , 43 Nitrogen gas was used as collision gas. The resulting data were processed using DataAnalysis version 5.2 (Bruker Daltonics), and proteins were identified using MASCOT version 2.7.0 (Matrix Science, London, UK) against the SwissProt database. Quantitative values were calculated with Scaffold 5 (Proteome Software, Portland, OR) for MS/MS-based proteomic studies. 44 Values were calculated from the sum of all of the spectra from a specific protein referred to the total spectrum counts. Each sample was normalized with the total spectrum counts. Measurement of virus-derived PIGA expression in various tissues and endogenous Rtl1 in liver cancers At 1 year of age, AAV-treated mice and wild-type littermates (n = 3) were euthanized and tissues were dissected. Total RNA was isolated with an RNAeasy kit (Qiagen, Hilden, Germany) after homogenization and gDNA was removed. Total RNA was reverse transcribed using a Superscript VILO kit (Thermo Fisher Scientific) and qPCR was performed with the cDNA using SYBR Green PCR mater mix (Thermo Fisher Scientific) and primers on the StepOnePlus Real-Time PCR System (Thermo Fisher Scientific) with the comparative Ct method. (The primers used for qPCR were, for mPiga: 5′- GTGAAGTCGGGGACATTGCC and 5′- GCAAACATGTAGCCCGTCAC; for hPIGA: 5′- GGGACATTGCCAGCTCCAGA and 5′- TCTGTCCAGTCGTTTGTCCATTGG; for mPigo: 5′-GCAGTAACTTTGCCAGCCATGC and 5′- TAGAGGTGTTCCAAGATGCCG; for mRtl1: 5′- TACTGCTCTTGGTGAGAGTGGACCC and 5′- GGAGCCACTTCATGCCTAAGACGA; for mGpc3: 5′- CAGCCCGGACTCAAATGGG and 5′- AGCCGTGCTGTTAGTTGGTATTTTTC; for endogenous control: TATA box binding protein; and for mTbp: 5′-TATGACCCCTATCACTCCTG and 5′-TTCTTCACTCTTGGT.). As for comparison of AAV-derived hPIGA expression level with endogenous mPiga expression level, copy number standard curves were generated by the dilution of hPIGA and mPiga expression plasmids. The copy numbers of hPIGA and mPiga mRNA of AAV-treated mice were analyzed and normalized with Tbp expression. As for the time course analysis, 3 pregnant mice were purchased from CLEA Japan, and 16 pups were intravenously injected with 10 11 vg/mouse of AAV PHPeB hPIGA. The pups were divided into 4 groups and 4 mice each were sacrificed at days 4, 10, 15, and 25 after injection for qPCR analysis of hPIGA expression with the comparative Ct method mentioned above.

Results Generation and phenotypes of CNS-specific Piga KO mice Nestin-Cre transgenic male mice were crossed with homozygous Piga -floxed female mice to induce Cre-mediated deletion of Piga exon 6 in neurons ( Figure 1 A). Percentages of exon 6-depleted alleles relative to Piga -floxed and wild-type alleles were calculated from the band intensities of the PCR products generated using three primers ( Figure 1 A and 1B). In male mice, the percentage of deleted ( Piga − ) alleles in Piga -floxed alleles was approximately 65% in various brain regions (hereafter, these mice are called Piga −/ ), suggesting that 35% of Piga expression levels remained. In female mice, the percentage of depleted alleles in Piga -floxed alleles was approximately 70% in various regions (hereafter, these mice are called Piga +/− ) ( Figure 1 C). It has been reported that nestin is expressed not only in nerve cells but also in heart and skeletal muscle during embryogenesis. 7 , 8 However, exon 6–depleted alleles were not detected in the heart and skeletal muscle of newborn Piga −/ and Piga +/− mice ( Figure S1 ). Considering the random inactivation of the X chromosome, Piga expression in the brains of Piga +/− mice can be expected to be 65% (50% + 15%) of that of wild-type mice. qRT-PCR analysis revealed that the average Piga expression in the brains of Piga +/− mice was approximately 56% of that in wild-type mice being in good agreement with the expectation, indicating that normal CNS cells did not proliferate dominantly relative to Piga -deleted cells during development in Piga +/− mice ( Figure 1 D). Piga +/− and Piga −/ mice showed severely defective growth. Piga +/− mice started limb clasping and ataxic gait from approximately P10, and these symptoms were progressive ( Figures 2 A and B). Survival time was drastically shortened; the Piga +/− mice died by approximately P25 and the Piga −/ mice died by approximately P10 ( Figure 2 C). Brain MRI of Piga +/− mice at P21 showed defective myelination; while myelination progressed in the corpus callosum and anterior commissure of wild-type brains, as indicated by the appearance of low-intensity regions ( Figure 3 A, red and yellow arrows, respectively), corresponding regions in Piga +/− brain maintained high intensities ( Figure 3 A). Myelination defect in Piga +/− mice was confirmed by (1) sparse staining of the major myelin protein, myelin basic protein (MBP), in the corpus callosum and cingulum compared to wild-type brains ( Figure 3 B), (2) decreased numbers of myelinated nerve fibers bearing electron-dense myelin sheath ( Figure 3 C), and (3) decreased levels of MBP in Piga +/− cerebrum and cerebellum (C) compared to wild-type tissues (W) as determined by western blotting ( Figure 3 D). AAV-PHPeB-mediated expression of human PIGA (hPIGA) cDNA prolonged survival time of CNS-specific Piga KO mice Piga +/− and Piga −/ mice were intravenously administered 10 11 virus genomes (vg)/mouse of AAV-PHPeB CAG - HA-hPIGA (AAV- hPIGA ) at P1 or P2. The growth defect was only partially restored in Piga +/− and Piga −/ mice ( Figure 4 A), but survival time was significantly extended for both genotypes. Of the Piga +/− mice, 37% survived until approximately 1 year of age and 50% of Piga - / mice survived until P21 ( Figure 4 B), indicating that self-complementary (sc) AAV treatment was effective as early as 1 week after administration because most of the nontreated Piga − / mice died by P10. The time course analysis of AAV-derived hPIGA expression in the wild-type mice revealed that it was expressed at the level almost similar to the endogenous level at day 4 after injection of AAV ( Figure S2 ). Neurological amelioration of AAV- hPIGA -treated mice In the hanging test, AAV- hPIGA -treated Piga +/− mice showed an ability comparable to that of wild-type mice ( Figure 4 C). The Y-maze test measured total entries, special working memory, and time spent in the new arm ( Figure 4 D). Of these, AAV- hPIGA -treated Piga +/− mice showed a significantly higher value only for the total number of entries, and the other two did not differ from controls, suggesting that AAV- hPIGA -treated Piga +/− mice were restless and hyperactive ( Figure 4 D, left). Electroencephalogram (EEG) analysis of AAV- hPIGA -treated Piga +/− mice showed that high voltage with slightly slower waves was prominent in the dark phase ( Figures 4 E and S3 ). They showed no spontaneous seizures, but tonic-clonic seizures (score 4, Table S1 ) were induced in one out of four mice that received a low dose of pentylenetetrazole (20 mg/kg), indicating mild susceptibility to seizures ( Figure S3 ). We could not test the untreated Piga +/− mice for comparison in these neurological tests because their neurological abnormalities were too severe, and they do not survive past P25. Myelination of AAV- hPIGA -treated Piga +/− mice was delayed at P19, as in nontreated mice ( Figure S4 A). Myelination of AAV- hPIGA -treated Piga +/− mice had progressed at P54 but to a lesser extent than in wild-type mice ( Figure S4 B). Brain MRI showed that in AAV- hPIGA -treated Piga −/ mice at P17, the length of the corpus callosum was shorter and the cerebellum was smaller than in their wild-type littermates ( Figures S5 A and S5B), suggesting that the structural abnormalities of the brain could not be improved by the gene replacement after birth. To note, none of the untreated Piga −/ mice were available for comparison at P17 because they did not survive past P10. Proteomic analysis of Piga −/ mouse brains revealed that Contactin 1–6 (CNTN1–6) levels, except for CNTN6, were severely decreased at P6 and not rescued by AAV- hPIGA administration at P6 or P17 ( Figures 5 A and B). CNTN1–6 are GPI-APs and occur in membrane-bound and soluble forms. CNTN1 is indispensable for paranodal junction formation. It also plays an important role in myelination. Therefore, decreased levels of CNTN1 may have contributed to the phenotype of Piga +/− mice. As for other GPI-APs, levels of voltage-gated calcium channel alpha2/delta subunit 1–3 (CACNA2D1–3) were severely decreased in Piga −/ mice at P6 and were partially rescued by AAV- hPIGA treatment at P17. Levels of glial cell line–derived neurotrophic factor (GDNF) family receptor alpha 1,2 and Reticulon 4 receptor-like 2 were also decreased but not rescued by AAV- hPIGA treatment. Levels of immunoglobulin (Ig)-like cell adhesion (family members, such as NTRM, LSAMP, NEGR1, and IgLON5, and Netrin-G1 and G2 were not or were mildly decreased in Piga −/ mice at P6 and were partially rescued by AAV- hPIGA at P17 ( Figures 5 A and B). Evaluation of hPIGA/Piga levels in AAV- hPIGA -treated mice AAV-PHPeB-derived hPIGA expression in various tissues from three Piga +/− mice of approximately 1 year of age was analyzed. qRT-PCR analysis revealed that AAV- hPIGA treatment resulted in widespread and robust expression of hPIGA in the brains of treated Piga +/− mice, whereas in the periphery, hPIGA expression was highest in skeletal muscle and was low but significant in the kidney ( Figure 6 A). Surprisingly, the expression of endogenous Piga was significantly decreased in the brains of AAV- hPIGA -treated Piga +/− mice compared with the levels in nontreated Piga +/− mice (33% in Figure 6 B versus 56% in Figure 1 D), suggesting potential endogenous feedback regulation of Piga expression. hPIGA was N-terminally tagged with hemagglutinin (HA) and anti-HA immunohistochemistry of brain tissues predominantly stained nerve cells ( Figure S6 ). Copy numbers of AAV-derived hPIGA mRNA were compared with those of endogenous Piga mRNA in these mice. Two out of three mice showed hPIGA expression above the endogenous mPiga expression level even after 10 months ( Figure 6 C). Hepatocellular carcinoma (HCC) after AAV-PHPeB gene delivery in Piga +/− mice Although AAV treatment has been largely reported as safe and well tolerated in rodents and larger animals and even in humans, there are reports that described the development of HCC in mice after the systemic delivery of AAV gene therapy vectors. 9 , 10 In our study, liver tumors were found in all three Piga +/− mice that were euthanized for gene expression analysis at approximately 1 year of age. Figure 7 A shows representative tumor images of one mouse. It was proven to be pathologically cancerous by H&E staining ( Figure 7 B). Anti-HA staining showed that the hPIGA transgene was not overexpressed in the tumor ( Figures 7 C, 7D, and S7 ). Because HCC development was attributed to AAV integration into the RNA imprinted and accumulated in nucleus ( Rian ) locus and the resulting overexpression of proximal microRNAs and retrotransposon-like 1( Rtl1 ), 10 liver tissues of normal appearance from three AAV treated Piga +/− and one Piga +/+ mice and the liver tumor from one of the AAV treated Piga +/− mice were analyzed by qRT-PCR for the expression of Rtl1 and hPIGA ( Figures 8 A and 8B). Expression of Rtl1 was drastically increased in the tumor (approximately 30,000 times that of wild-type liver) and other liver tissues (6–40 times that of wild-type liver), suggesting AAV integration into the Rian locus ( Figure 8 A). These findings were consistent with the liver tissues of pathological appearance in fact being cancerous. In contrast, the expression of hPIGA was decreased in the tumor tissue ( Figure 8 B), which is consistent with the result of the immunohistochemical staining for HA-hPIGA ( Figures 7 C, 7D, and S7 ). Unexpectedly, the expression of endogenous Piga was increased in the tumor ( Figure 8 C); however, the expression of Pigo, another GPI biosynthesis gene, or Glypican3 , the highly expressed GPI-AP in the liver cancer, was not drastically increased in the tumor ( Figure S8 ).

Discussion PIGA is a catalytic subunit of the enzyme complex involved in the first step of GPI biosynthesis. Because PIGA is an X-linked gene, males who receive the maternal pathogenic allele develop the disease. Therefore, the frequency of PIGA deficiencies is higher than that of other IGDs, with approximately 100 patients reported worldwide up to now. 11 , 12 , 13 Human females who are heterozygous for pathogenic PIGA alleles are predicted to be mosaic for GPI-AP expression because of X-inactivation. These females develop normally and become healthy carriers of PIGA pathogenic alleles. 12 Fluorescence-activated cell sorting analysis of granulocytes revealed that one carrier showed two peaks (normal and decreased) of CD16 expression, whereas others showed normal expression. 12 X-inactivation occurs randomly early in development, but GPI + cells become dominant, especially in tissues in which GPI-APs are critical during development, such as neuronal tissues. However, this does not seem to be true in mice. Human cytomegalovirus (CMV)-Cre mice crossed with Piga -floxed female mice generate hemizygous total-KO Piga −/ male and heterozygous total-KO Piga +/− female mice, both of which are lethal. Total-KO Piga −/ male embryos die at E9 and Het-KO Piga +/− female embryos die at E13, with severe malformation. 14 We attempted several times to generate Piga knockin mice bearing a patient’s variant allele either by direct injection of Cas9 with gRNA and donor oligo into fertilized eggs or by standard ES cell methodology, but our attempts failed. We were unable to generate heterozygous females (data not shown). Instead, all of the reported Piga KO models have been based on CNS-specific KO mice. 6 , 15 To determine whether a gene therapy approach is useful for PIGA -IGD, we treated CNS-specific KO mice with an AAV-based gene therapy. Nestin-Cre is active as early as E7.5; significant recombination is detected in the developing brain as well as in the neural tube at E10.5–11.5, and recombination in the CNS is almost complete by E15.5. 16 In Nestin-Cre/ Piga -floxed mice, neurons, astrocytes, and oligodendrocytes were defective in Piga . The Piga gene was deleted in approximately 70% of the brain DNA in the floxed allele of Piga +/− females ( Figure 1 C), which indicated that Piga expression would be 65% of that in wild-type mice if random inactivation of X chromosome took place. Actually, endogenous Piga expression was approximately 56% of that in wild-type mice ( Figure 1 D), indicating that normal CNS cells did not proliferate dominantly against Piga -deleted cells in these mice. Although Piga −/ males and Piga +/− females were born alive, they showed severely defective growth and defective myelination and most of them died by P10 and P25, respectively ( Figures 2 and 3 ). Intravenous injection of scAAV-PHPeB- CAG - hPIGA at P1–P2 was very effective at extending their survival; half of the Piga −/ males lived up to 3 weeks and 40% of the Piga +/− females lived more than 1 year ( Figure 4 ). The growth defect was not completely rescued, but significant improvements in neurological phenotypes, such as muscle weakness and limb clasping, were observed ( Figure 4 ). Brain MRI showed improved myelination in AAV-hPIGA-treated Piga +/− females at P54, and no spontaneous seizures were observed, indicating no prominent neurological abnormality. Proteomic analysis of the brains of Piga −/ male mice revealed that levels of CNTN1–6 were severely decreased at day 6 and not rescued by AAV- hPIGA administration at P6 or P17 ( Figures 5 A and 5B). CNTN1 is indispensable for early interactions between axons and glia. CNTN1 clusters at the paranodal junction, establishing a complex with contactin-associated proteins (CNTNAPs, Casprs) and interacts with glial neurofascin-155 to establish axon glial contacts for the insulating function of myelin. 17 CNTN1 is essential for the proper localization of potassium Kv1.2 channels at juxtaparanodal regions, indicating that it is needed for correct action potential repolarization during action potential conduction. 17 The defect in nerve conduction and excitability of the target muscles may cause muscle atrophy, which is consistent with hypotonia in human patients and hindlimb weakness in Piga −/ , Piga +/− , and Cntn1 KO mice. Cntn1 KO mice are severely growth restricted and have a myelination defect and cerebellar dysfunction and die at approximately 3 weeks of age; 18 these phenotypes are similar to those of Piga +/− mice. A human infant with a homozygous variation in the CNTN1 gene has been reported. Their phenotype was lethal myopathy because of decreased levels of CNTN1 at neuromuscular junctions, leading to disrupted communication between muscles and nerves. 19 CNTN1 also plays an important role in myelination. The maturation and differentiation of oligodendrocytes is controlled by CNTN1-dependent signal transduction through its interaction with PTPRZ, NOTCH, PTPα, and FYN on oligodendrocytes and their precursor cells. 20 Therefore, the decreased expression of CNTN1 would contribute to the phenotype of Piga −/ and Piga +/− mice and of human IGD cases. The levels of CACNA2D1–3 were also severely decreased at day 6 of Piga −/ mice and were partially rescued by AAV- hPIGA treatment at P17. CaVs express as a heteromeric proteins on the plasma membrane of skeletal muscles and neurons, in which the α1 subunit is associated with two auxiliary subunits, the intracellular β subunit, and the α2δ subunits; the latter are encoded by four genes, CACNA2D1–4, and are reported to be GPI-APs. 21 They play important roles in the trafficking and function of the CaV channel complexes. There are several reports that the defect in CACNA2D1 or CACNA2D2 causes developmental and epileptic encephalopathies, hypotonia, and severe cerebellar ataxia, 22 symptoms of which can be also observed in IGDs, including PIGA deficiencies. 12 Therefore, the decreased expression of CACNA2D1–3 ( Figure 5 ) may contribute to the mouse phenotypes such as muscle weakness and ataxic gate. Other important GPI-APs include the GDNF receptor family of proteins, GFRα-1,2, which are involved in various signaling pathways for neuronal cell survival and migration through activation of RET tyrosine kinase receptor. RET and GFRα-1 are the responsible genes for Hirschsprung disease, 23 from which severe cases of IGD suffer. 24 In the mouse models, we did not find any organ abnormalities. Tissue nonspecific alkaline phosphatase (TNAP), a GPI-AP, was mildly decreased at day 6 of Piga −/ mice. Decreased surface expression of TNAP is responsible for seizures in IGD cases. TNAP dephosphorylates pyridoxal phosphate (PLP) to PL, a membrane permeable form of vitamin B 6 , which is converted to PLP intracellularly and functions as a cofactor for GABA synthase. TNAP KO mice developed seizures due to decreased GABA levels in the brain, which was rescued by PL treatment. 25 Likewise, the administration of pyridoxin is very effective in controlling seizures in some IGD cases. 24 TNAP is also involved in the uptake of vitamins B 1 and B 2 , the latter of which requires another GPI-AP, CD73, which also converts flavin adenine dinucleotide to flavin mononucleotide. 26 We could not detect folate receptor 1 (FOLR1) in the mouse brain; however, it is known that this GPI-AP expresses on the choroid plexus epithelium being involved in the transcytosis of 5-methyl tetrahydrofolate (5MTHF) from blood to cerebrospinal fluid (CSF) and decreased expression of FOLR1 causes cerebral folate deficiency. 27 Low concentration of 5MTHF in CSF is often found in IGD cases, 28 suggesting that FOLR1 is one of the genes responsible for psychomotor retardation, cerebellar ataxia, and seizures in IGDs. Despite the effectiveness of AAV-based gene therapy in Piga +/− mice, treated mice developed HCC 1 year after treatment. AAV is regarded as nonpathogenic and is considered to be a promising vector for gene delivery. However, recent reports have questioned the safety of AAV. A subset of studies shows that in AAV-treated mice, AAV preferentially integrates into the Rian locus, resulting in the overexpression of proximal microRNAs and Rtl1 , which can lead to carcinogenesis. 10 The induction of carcinogenesis depends on the AAV dose, enhancer/promoter selection, and the timing of gene delivery. Chicken β-actin (CBA) or liver-specific thyroxine-binding globulin (TBG) promoters plus the CMV enhancer have been hypothesized to promote increased transcription (transactivation) of genes proximal to Rian that drive the formation of HCC. The Rian locus is highly expressed in neonates and is therefore susceptible to AAV integration. However, the overexpression of microRNAs did not occur when AAV was driven by promoters other than CBA or TBG, indicating that vector-encoded cis -regulatory sequences were responsible. 10 In humans, the upregulation of delta-like homolog 1-deiodinase type 3, the orthologous locus to the mouse Rian locus, has been associated with poor survival in patients with hepatic carcinoma. 29 RTL1 overexpression activates the Wnt pathway by increasing the levels of DOCK4 and MACF1, both of which enhance the release of β-catenin from the destruction complex and increase the stability of β-catenin in melanoma cells. 30 Fortunately, no human cases treated with an AAV transgene driven by CBA have developed liver cancer. 31 It is not known whether the AAV integration preference is different between humans and mice. Unexpectedly, the expression of endogenous Piga was increased in the tumor but not the vector-derived hPIGA ( Figure 8 C). This is probably not the cause of carcinogenesis but the result from the Wnt signal activation caused by Rtl1 overexpression. Consistent with this, the expression of Pigo, another GPI biosynthesis gene, or Glypican3 , the highly expressed GPI-AP in liver cancer, were not drastically increased in the tumor ( Figure S8 ). Zolgensma (onasemnogene abeparvovec), AAV9- CAG-SMN1 , has been approved for the treatment of spinal muscular atrophy (SMA) in various countries, including Japan. Most of the treated patients developed liver dysfunction, thrombocytopenia, and thrombotic microangiopathy after AAV administration, and the severity of these adverse effects was correlated with the dose of AAV and its expression level. 32 , 33 , 34 This is known to be caused by the immunological reaction to AAV. However, there is no evidence of AAV integration and subsequent HCC in these patients, with the long-term follow-up data suggesting the safety and tolerability of AAV9. Zolgensma treatment in SMA patients is followed up to 7.5 years postdosing. Nonetheless, there are many parameters, such as administration routes, amount of virus, timing of administration, and the promoters, to be considered for the use of AAV to select the safest and most effective method for the gene therapy of IGD. One limitation of our work is that this mouse model is not an accurate disease model. We could not establish the knockin mouse bearing the same mutation of the affected individual. Individuals with PIGA deficiency are always partial deficiencies because complete deficiency is lethal. In these individuals, PIGA expression is decreased not only in the CNS but also in whole bodies and expressions of various GPI-APs are decreased to various degrees. AAV-based gene therapy after birth could not completely rescue the mouse phenotypes because Cre recombinase driven by Nestin promoter completely depletes the Piga gene as early as approximately E10 in the CNS, and the defect in brain development due to loss of various GPI-APs at this stage was not reversible. In utero administration of AAVPIGA would be required to overcome the defects in embryogenesis. As for the gene therapy for partial PIGA deficiency, we believe that most of the symptoms are reversible based on the fact that the PIGO -deficient mouse model bearing the same mutation of the affected individual was successfully treated with AAV-based gene therapy. 2

Thirty genes are involved in the biosynthesis and modification of glycosylphosphatidylinositol (GPI)-anchored proteins, and defects in these genes cause inherited GPI deficiency (IGD). PIGA is X-linked and involved in the first step of GPI biosynthesis, and only males are affected by variations in this gene. The main symptoms of IGD are neurological abnormalities, such as developmental delay and seizures. There is no effective treatment at present. We crossed Nestin - Cre mice with Piga -floxed mice to generate CNS-specific Piga knockout (KO) mice. Hemizygous KO male mice died by P10 with severely defective growth. Heterozygous Piga KO female mice are mosaic for Piga expression and showed severe defects in growth and myelination and died by P25. Using these mouse models, we evaluated the effect of gene replacement therapy with adeno-associated virus (AAV). It expressed efficacy within 6 days, and the survival of male mice was extended to up to 3 weeks, whereas 40% of female mice survived for approximately 1 year and the growth defect was improved. However, liver cancer developed in all three treated female mice at 1 year of age, which was probably caused by the AAV vector bearing a strong CAG promoter. Graphical abstract Neurological abnormalities due to defective expression of various GPI-anchored proteins in CNS caused by the Nestin-Cre-mediated Piga deletion during mouse embryogenesis is ameliorated by intravenous administration of scAAVPIGA immediately after birth. This suggests that neurological symptoms in patients with inherited GPI deficiency are reversible if treated early after birth. Keywords

Data and code availability All of the data are available from the corresponding author on reasonable request.

Supplemental information Acknowledgments We thank Andrew Kwon, PhD, for connecting us to Steven and Ann Nguyen, the parents of the PIGA-CDG patient Emmett Nguyen. We thank Junji Takeda and Takahiro Kodama (Osaka University) for discussion, Masahito Ikawa (Osaka University) and Gen Kondoh (Kyoto University) for helping us to generate Piga knockin mice, and Keiko Kinoshita (Osaka University) for technical help. We also thank Edanz ( https://jp.edanz.com/ac for editing the English text of a draft of this manuscript. This work was supported by funds raised by Steven and Ann Nguyen. This publication has been made with the parents’ permission. This work was also supported by the 10.13039/501100001691 Japan Society for the Promotion of Science and 10.13039/501100001700 Ministry of Education, Culture, Sports, Science, and Technology KAKENHI grants (21H02415 for T. Kinoshita), a grant from the 10.13039/501100003478 Ministry of Health, Labour, and Welfare , and a grant from the Practical Research Project for Rare/Intractable Diseases from the 10.13039/100009619 Japan Agency for Medical Research and Development (AMED) (23FC1033, JP22ek0109614, and JP23bm1223019 to Y.M.). Author contributions Y.M., T. Kinoshita, and K.C.M. designed the study. S.U., K.I., and Y.M. acquired the data and conducted the experiments. S.L. made the AAV. Y.Y. performed the MRI analysis. M.K. performed the electron microscopy analysis and T.S. performed the histological analysis. A.N. performed the proteomics. Y.M. and T. Kinoshita wrote the paper. Declaration of interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:57

Mol Ther Methods Clin Dev. 2023 Dec 14; 32(1):101176

oa_package/f4/ad/PMC10788267.tar.gz

PMC10788271

37542151

Introduction Parkinson’s disease (PD) is the second most prevalent neurodegenerative disease, and its incidence increases with aging [ 1 ]. In idiopathic PD, the vulnerability and dysfunction of the dopaminergic neurons that causes their death produce an imbalance between the direct and the indirect basal ganglia pathways [ 2 – 4 ], causing motor symptomatology such as tremors, bradykinesia, rigidity, and postural instability that are characteristics of PD, and they occur when more than 75% of the dopaminergic neurons die [ 1 , 5 ]. The pathophysiology of PD has been described as the death of the dopaminergic neurons of the substantia nigra pars compacta (SNpc); which produce the dopamine (DA) necessary for the correct motor control exerted by the basal ganglia (BG) [ 5 ]. Because of this, the gold standard for treating PD is the synthesis replacement of DA by the administration of its precursor, Levodopa (L-DOPA) [ 6 , 7 ]. However, this treatment has a limited useful time of around five years. Furthermore, the use of L-DOPA does not stop the progressive death of the dopaminergic neurons and induces several problems such as dyskinesias (abnormal involuntary motor movements). The disease is associated with deficits in neurotransmitters other than DA that can result in non-motor symptoms, such as cognitive, psychiatric and sensorial symptoms (dementia, depression, anosmia, etc.) and gastrointestinal complications (constipation, drooling, dysphagia, and nausea) that occur as the disease progresses [ 8 – 10 ]. Current pharmacological, surgical and alternative PD treatments preserve the use of L-DOPA [ 5 , 11 ]. Diverse promising gene and cell therapy strategies have been tested for PD. These include transplants of neurons derived from induced pluripotent stem cells (iPSc cells) [ 12 ]. Also, the expression of disease-modifying transgenes like neurotrophic factors (glial cell-line derived neurotrophic factor (GDNF), neurturin (NRTN), artemin (ARTN), and persephin (PSPN) or of non-disease modifying transgenes, including tyrosine hydroxylase (Th), and glutamic acid decarboxylase (GAD) have been tried with interesting results in experimental models, and some in early clinical protocols [ 13 – 16 ]. However, using viral vectors in some of these protocols may cause the integration of transgenes into the host genome, which may induce mutagenesis or activate the host immune response limiting the effectivity [ 17 , 18 ]. Astrocytes are essential for the immunological response of the brain in PD, and their participation in that response has been observed in both the substantia nigra and the striatum in animal models and postmortem studies in PD patients [ 19 ]. There is an important reactive astrogliosis in PD, which indicates the relevant role of astrocytes in the response to the disease [ 20 ]. Astrocytes are resistant to insults and participate in the dopamine (DA) metabolism because they express both the amino acid transporter (LAT) and the DA transporter (DAT), allowing the uptake of L-DOPA and DA observed in the striatum [ 20 , 21 ]. Moreover, astrocytes express aromatic L-amino acid decarboxylase (AADC), an enzyme that converts L-DOPA into DA. In cell culture studies, the capacity of astrocytes to convert L-DOPA to DA in a concentration-dependent manner has been demonstrated [ 22 ]. Thus, Th is the only enzyme that astrocytes require to produce DA. We previously showed that either implanted astrocytes or endogenous astrocytes expressing transgenic Th produce DA in the striatum causing behavioral recovery in 6-OHDA-lesioned rats and in a non-human primate MPTP model [ 16 , 23 , 24 ]. Clustered Regularly Interspaced Short Palindromic Repeats associated with Cas (CRISPR-Cas) has already changed the gene therapy field since it represents a highly efficient gene editing tool [ 25 , 26 ]. Synergistic Activation Mediator (SAM) is a second-generation CRISPR system capable of activating gene expression using an enzymatically inactive Cas9 (dCas9) and co-transcriptional activators. This CRISPR-based system has demonstrated be a robust and efficient method to induce gene activation [ 27 – 29 ]. Therefore, in the present work, we performed a proof-of-concept experiment to test the use of SAM in gene therapy for the treatment of PD. This work aims to evaluate the effectiveness of the endogenous activation of genes to induce the synthesis of DA in cells of glial origin and the effects of its implantation in a murine model of PD. We propose that the endogenous expression of th from rat astrocytes induced by SAM will allow these cells to produce DA and cause behavioral improvement in the 6-OHDA hemiparkinsonian rat model. The therapy proposed here will avoid fluctuations due to dosing, absorption and penetration into the brain, and simultaneously spare multiple organ systems from exposure to L-DOPA and its metabolites. In the brain, it will spare large areas from the resulting conversion to DA, and spare specific neurons, such as serotoninergic neurons from the competition with tryptophan, both for transport from plasma and for decarboxylation. The alternative offered here presents a novel method to achieve this targeted therapy that will increment the drug-free period of treatment and can combine with other therapies in more advanced cases.

Materials and methods Sequence and cloning The lentiSAMv2 and lentiMPH v2 plasmids (a gift from Feng Zhang Addgene #75112 and #89308) were used for CRISPR-Cas9 activation. We analyzed the promoter region of the rat th gene, 1000–1200 bp upstream of the transcription start site, searching for PAM regions. We obtained ~20 nucleotides sequences that were evaluated using BLAST analysis to check for mismatched sites on the rat ( Rattus Norvegicus ) genome. In addition, the th promoter region was also assessed with E-CRISP, CRISPRESSO2, CHOPCHOP, and Benchling software to select the candidate sequences by specificity score, annotation score, efficiency score, mismatched score, and the number of hits. Finally, we chose 13 sgRNA candidates that were evaluated for mismatches using the BLAST tool ( https://blast.ncbi.nlm.nih.gov/Blast.cgi ). The single-guide RNA (sgRNA) sequences used in this study are listed in Supplementary Table 1 , and were cloned in the lentiSAM v2 vector, employing the Golden Gate protocol using BsmB I (Thermo Fisher Scientific cat. no. IVGN0136) for plasmid restriction, t4 PNK (NEB, cat.no. M0201) for oligo annealing and t7 ligase (NEB, cat.no. M0318) for cloning [ 27 ]. The cloning sgRNA region was amplified by PCR using specific primers. The amplified PCR products were purified using the Sap-Exo kit. (Jena Bioscience, Cat. No. PP-218L, Germany) following the manufacturer ́s instructions. The purified products were sequenced (forward and reverse) by automatic sequencing on an ABI PRISM 3100 with the Kit of BigDye Terminator v3.1 cycle sequencing (Applied Cat No. 4337458, Biosystems, USA). We show in Supplementary Fig. 1B the sequence of the chosen th sgRNA. Cell culture HEK293T/17 cells (ATCC CRL-11268) were maintained in DMEM High Glucose (Gibco, cat. no. 12800-017) and pyruvate supplemented with 10% fetal bovine serum (FBS) (Gibco, cat. no. 16000-044) and 1% penicillin/streptomycin (Thermo Fisher Scientific cat. no. 15140122). Cells were incubated at 37 °C, 5% CO 2 and 95% humidity. Cells were passaged every other day at a ratio of 1:6 or 1:5 using Trypsin 1x (Sigma, cat. no. 59428C). The C6 rat glioblastoma cell line (ATCC CCL-107) was maintained in F12 DMEM (Gibco, cat. no. 12500-062) with L-glutamine and pyruvate supplemented with 10% fetal bovine serum (Gibco, cat. no. 16000-044) and 1% penicillin/streptomycin (Thermo Fisher Scientific, cat. no. 15140122), incubated at 37 °C, 5% of CO 2 , and 95% humidity, passaged every other day at a ratio of 1:6 or 1:5 using Trypsin 1x (Sigma, cat. no. 59428C). CTX-TNA2 cells, a primary astrocyte non-tumorigenic line obtained from rat cerebral cortex (ATCC CRL-2006), was maintained in high-glucose DMEM (Gibco, cat. no. 12800-017) with pyruvate supplemented with 10% fetal bovine serum (Gibco, cat. no. 16000-044) and 1% penicillin/streptomycin (Thermo Fisher Scientific, cat. no. 15140122). Cells were incubated at 37 °C, 5% of CO 2 and 95% humidity , passaged every other day at a ratio of 1:6 or 1:5 using Trypsin 1x (Sigma, cat. no. 59428C). Transfection of C6 cells C6 cells were transfected with Lipofectamine 3000 (Thermo Fisher Scientific, cat.no. L3000-15). The SAM system with each sgRNA was tested in C6 cells seeded in 60 mm Petri dishes; for each sequence cells were seeded at ~40% confluence (~1.2 ×10 6 cells) the day before and 0.5 μgrams of plasmid DNA was transfected overnight with 1.5 ml of medium without serum and medium changed the following day with fresh medium with 10% FBS. Lentivirus production and cell infection One day before transfection, HEK293T cells were seeded at ~50% confluency (~4.4 ×10 6 cells) in 100 mm Petri dishes. Cells were transfected the next day at ~70–80% confluence (~6.6 ×10 6 cells). For each petri dish, 3 μg of each of the following plasmids were transfected using 24 μL of Lipofectamine 3000 (Thermo Fisher Scientific, cat.no. L3000-15) and 24 μL of P3000 Enhancer (Therm Fisher Scientific, cat.no. L3000-15): pMD2.G (Addgene 12259); pMDLg/pRRE (Addgene 12251) and pRSV-Rev (Addgene 12253). Cells were transfected with 8 mL of high-glucose DMEM (Gibco 12800-017) without fetal bovine serum, and 12 h after the transfection, the medium was changed for complete maintenance medium. The virus supernatant was harvested 48 h post-transfection and centrifuged at 1200 rpm to eliminate debris and dead cells. The virus supernatant was then aliquoted and stored at −80 °C. Different cell lines were infected with a 0.3 viral titer on 1.5 ml of pseudolentivirus with the SAM system, and each sgRNA and incubated overnight; then, the medium was removed, and fresh medium was applied; after 48 h, cells were processed for RNA and protein extraction. Resistance curves for blasticidin S HCl (10 μg/ml, Thermo Fisher Scientific, cat.no. A1113903) and hygromycin B (300 μg/ml, Sigma cat.no. H3274) were constructed, and cells were selected by their resistance to antibiotics. After 24 h of infection, fresh media containing the antibiotics were applied, and cells were maintained under antibiotic selection. RT-PCR Cells were seeded in 60 mm plates and grown to 90% confluency (~2.8 ×10 6 cells) before RT-PCR. Total RNA was extracted using the Trizol reagent (Invitrogen, cat. no. 15596026). RNA quantification and purity were assessed by spectrophotometric analysis using the Nanodrop instrument (Thermo Fisher Scientific, cat. no. 13-400-525). DNAse I (Sigma, cat.no. AMPD1) was used to degrade DNA; cDNA was synthesized using M-MLV (Invitrogen, cat. no. 28025-013), dNTPs 100 mM (Invitrogen, cat. no. 10297-018), and oligo dT (T4 Oligo Oligo dT 18-mer). RT-PCR was performed using Taq Polymerase Kappa (Sigma, cat.no. BK1004), dNTPs (Invitrogen, cat. no. 10297-018), UltraPureTM DNase/RNase Free Distilled Water (Invitrogen, cat. no. 10977015) and the following primers were used: for th (FW ‘GGAGAGCTCCTGCACTCC’ REV ‘GGCATAGTTCCTGAGCTTG’), for β-actin (FW ‘TCACGCACGATTTCCCTCTCAG’ REV ‘TGGCACCACACCTTCTACA’), for dCas9 (FW ‘GCACATACCACGATCTGCTG’ REV ‘CGCTTCAGCTGCTTCATCAC’) and for ms2 (FW ‘GGGATGTGACAGTGGCTCC’ REV ‘GGACCTCCACCTTGATGGTATAC’). (Sigma-Aldrich). We used the following temperatures for the PCR assays: 94° for denaturation, 57 °C for annealing, and 72° for the extension step using a T100 Bio-Rad thermal cycler. Western blot Protein extraction from cells or tissue was performed using a lysis buffer that contained a protease inhibitor cocktail (Complete, Roche Diagnostics, cat. no. 04574834001). Total protein (25 μg) was loaded for each sample for separation by SDS–polyacrylamide gel electrophoresis (8%) and transferred onto nitrocellulose membranes (BioRad; cat. no. 1620115). Membranes were blocked with 5% non-fat milk diluted in Tris-buffered saline containing 0.1% Tween-20 (TBST) for one h and incubated in the presence of anti-Th (Abcam ab112 monoclonal anti-rabbit 1:1000), or anti-Th F-11 (sc-25269 monoclonal anti-mouse, 1:1000), or anti-Th Millipore ab152 (monoclonal anti-rabbit 1:1000), anti-Glial Fibrillary Acidic Protein (DAKO Z0334 polyclonal anti-rabbit 1:2000), anti-Beta-actin-peroxidase (Sigma A3854 monoclonal anti-mouse 1:25000). Proteins were revealed using secondary peroxidase-coupled anti-rabbit and anti-mouse antibodies (Jackson ImmunoResearch), using the Western Lightning Plus-ECL Kit (PerkinElmer; Waltham, MA). Images were digitally captured using the FUSION SOLO S instrument (Vilbert Smart imagining). Autoradiograms were analyzed by densitometry using the Image J ® software (NIH). Immunofluorescence C6 and astrocytes were seeded on coverslips (~80% of confluence in a 60 mm petri dish, ~2.5 ×10 6 cells), fixed with 4% paraformaldehyde and/or cold methanol, washed twice with PBS, and incubated using 0.2% Triton X-100/PBS/1% BSA for 30 min at room temperature. Cells were incubated using a primary mouse monoclonal antibody against anti-Th (Abcam ab112 monoclonal anti-rabbit) in C6 cells and anti-Glial Fibrillary Acidic Protein (DAKO Z0334 polyclonal anti-rabbit) and anti-Th F-11 (sc-25269 monoclonal anti-mouse) in astrocytes for 12 h. Coverslips were washed with 0.2% Triton X-100/PBS and incubated using the appropriate secondary antibodies, Alexa Fluor 594 donkey anti-mouse IgG (H + L) (Invitrogen A21203), and Alexa Fluor 488 donkey anti-rabbit IgG (H + L) (Invitrogen A21206.) Nuclei were counterstained using DAPI with Vectashield (Vector Laboratories). Experimental subjects Male Wistar rats weighing ~200 g at the start of the experiment were used. Rats were housed on a 12-h light/dark cycle with free access to food and water at ~25 °C, with a relative humidity of 40–60%. All animal studies were performed according to the Guide for the Care and Use of Laboratory Animals [ 30 ], as adopted by the US National Institutes of Health and the Mexican Regulation of Animal Care and Maintenance (NOM-062-ZOO-1999). Rats were maintained and handled according to the guidelines of the CINVESTAV Animal Care Committee, making all efforts to minimize suffering and the number of animals used. The principles of the 3Rs (Replacement, Reduction, and Refinement) were followed to guarantee a humane endpoint. Brain tissue immunofluorescence Rats were euthanized ten weeks after the transplant of astrocytes or the sham treatment with an overdose of sodium pentobarbital (200 mg/kg i.p.) and transcardially perfused with saline followed by 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS). Brains were dissected and post-fixed in PFA at 4 °C for 24 h and then cryoprotected in 10% sucrose/PBS for 24 h, 20% sucrose/PBS for 24 h, and 30% sucrose/PBS for an additional 24 h at 4 °C. Frozen coronal 35-μm-thick sections were cut in a sliding microtome (Leica Microsystems), collected in 2% PFA/PBS, and stored at 4 °C until processing. Brain sections chosen for the analysis of the striatum were incubated in free-floating conditions in 0.2% Triton X-100/PBS for 30 min and blocked for 30 min in 1% BSA/0.2% Triton X-100/PBS at room temperature. Then, sections were incubated overnight with the primary rabbit antibody against the glial fibrillary acidic protein (Dako Z0334 polyclonal antirabbit GFAP, 1:500) and anti-Th (Sigma T1299 monoclonal anti-mouse, 1:200) diluted in 0.02% Triton X-100/0.1% BSA/PBS. For primary antibody detection, sections were rinsed with 0.02% Triton X-100/PBS and incubated for 1 h at room temperature with the secondary antibody Alexa Fluor 594 donkey anti-mouse IgG (H + L) A21203 and Alexa Fluor 488 donkey anti-rabbit IgG (H + L) A21206 diluted in the same solution as the primary antibody. Nuclei were counterstained using 4′,6-diamino-2-phenyldole (DAPI) with Vectashield (Vector Laboratories). Brain tissue immunohistochemistry Immunohistochemistry was performed using a rabbit monoclonal anti-tyrosine hydroxylase Th antibody (1:400, Abcam ab112), a rabbit antibody against the glial fibrillary acidic protein (Dako Z0334 polyclonal antirabbit GFAP, 1:500) and a biotinylated anti-rabbit IgG (H + L) (1:100; Vector Laboratories, Burlingame, CA, USA). The immunohistochemical staining was developed using the avidin–biotin–peroxidase complex (1:10; ABC Kit; Vector Laboratories) and DAB (Sigma). Immunohistochemical labeling was observed with an Olympus BX53 microscope and images were digitized. Confocal microscopy Triple-labeled images were obtained using a confocal laser-scanning microscope (Leica TCS-SP8) in the XYZ (Z-stacks) mode using a 63X (oil immersion) objective. The following excitation lasers/emission filters settings were used for the various chromophores: an argon laser was used for the Alexa Fluor 488, with a peak excitation at 490 nm and emission in the 505–530 nm range; a He-Ne laser was used for the Alexa Fluor 594 with a peak excitation at 543 nm and emission in the 568–615 nm range; and a UV laser was used to reveal DAPI with a peak excitation at 456 nm and emission in the 410–480 nm range, using the sequential acquisition of separate wavelength channels to reduce interference between channels. The Z-stacks (3–4 optical slices) were then converted into a three-dimensional projection image using the Leica LAS AF lite software. High-performance liquid chromatography (HPLC) Brain slices, 100 μm thick, were cut using a Leica microtome (RM2125 RTS) in ice-cold Krebs-Henseleit solution. Immediately, left and right striata sections were dissected and collected in an amber 1.5 ml Eppendorf tube with sterile PBS solution at 4 °C (100 μl) and 50 μl of perchloric acid 0.1 N, the tissue was manually homogenized, sonicated and ultracentrifuged, the supernatant was filtered through a 0.22 μm filter (Millipore GSWP04700) to assay for DA and DOPAC [ 31 , 32 ]. The medium culture samples were collected in an amber 1.5 ml Eppendorf tube with 50 μl of perchloric acid 0.1 N per 950 μl of medium, samples were shaken with a vortex for a minute and ultracentrifuged, the supernatant was filtered through a 0.22 μm filter (Millipore GSWP04700) to assay for DA and DOPAC. Briefly, dopamine levels were determined using an HPLC method with electrochemical detection as previously described [ 33 , 34 ]. The separation of Dopamine and DOPAC was performed using a dC-18 microbore column (Atlantis, 2.1 ×150 mm, Waters Co.). The mobile phase was: buffer 97%, NaCl 135 mg/l; citric acid 10.5 mg/l; EDTA 20 mg/l; OSA 20 mg/l; methanol 3%, pH 2.9 adjusted with NaOH. The flow rate was 0.3 ml/min at 30 °C. The electrochemical detection system was an Intro; Waters Co. coupled to a glassy carbon electrode (VT-03, Antec Leyden). The oxidation potential was þ380 mV vs. silver/silver chloride reference electrode (ISSAC). 6-OHDA stereotaxic lesion surgery Rats were anesthetized with Ketamine 75 mg/kg and Xylazine 5 mg/kg i.p; placed on a David Kopf stereotaxic frame and injected unilaterally with 6-hydroxydopamine (6-OHDA by Sigma, cat. no. H4381); 16 μg/μl of saline containing 0.1% ascorbic acid) in the medial forebrain bundle at coordinates (AP −1.8; ML 2.4; DV −7 mm) relative to Bregma according to the rat brain atlas of Paxinos and Watson [ 35 ], and 5 μl were administered at a rate of 1 μl per minute. To prevent noradrenergic neuron damage, rats were pre-treated with desipramine (10 mg/kg i.p.) 40 min before the surgery. Twelve days after the 6-OHDA lesion, only rats showing ten or more ipsilateral turns/min 30 min after amphetamine injection were included in the study [ 36 ]. A group of sham-lesioned rats with the surgical procedure but no stereotaxic lesion or administration of 6-OHDA was used as a control. We call lesioned side the right side receiving 6-OHDA and as the intact side the left side that did not receive stereotaxic lesion and 6-OHDA administration. Implantation of astrocytes Hemi-parkinsonian rats (~300 g) were anesthetized with Ketamine/Xylazine (75/5 mg/kg i.p.) and placed on a David Kopf stereotaxic frame and received 20,000 astrocytes in each of the two sites in the lesioned striatum : anterior (AP 1.9; ML 2.2; DV −5) and posterior (AP 0.9; ML 3.0; DV −4) relative to Bregma according to the brain atlas of Paxinos and Watson [ 35 ]. Astrocytes (AST) and astrocytes expressing Th (AST-TH) were trypsinized from 90% confluent 60 mm Petri dishes, counted, and 20,000 cells were immediately concentrated in 3–4 μl of culture medium without serum to be used for the implant procedure [ 37 ]. Sham-lesioned rats did not receive implants. Behavioral tests Rats were trained for two weeks before the 6-OHDA lesion. Two weeks after the 6-OHDA lesion, astrocytes and AST-TH were implanted in the striata of experimental subjects. Implanted rats were evaluated for the following nine weeks. Experimental and control rats were sacrificed ten weeks after the lesion to obtain striatal tissue for HPLC, immunofluorescence, and western blot assays. Amphetamine-induced rotations test Rats were challenged with amphetamine (8 mg/kg i.p. Sigma, cat. no. NMID420D) and tested for ipsilateral circling behavior; the number of turns per minute was measured and recorded for 30 min, 15 min after the peritoneal administration of amphetamine [ 38 ]. Cylinder test The test was performed as follows: Rats were placed individually inside an acrylic cylinder (diameter, 22 cm; height, 26 cm) with a mirror behind it at a 45° angle to allow 360° vision. Rats were video recorded for 5 min after they first touched the cylinder walls with either the impaired or unimpaired forelimb or both simultaneously. Scores were calculated by the asymmetry ratio: right-left/ (right + left + both). Scores on the forelimb asymmetry ratio range from −1 to 1 [ 39 , 40 ]. Inclined beam balance test Animals were trained for three days to walk along 2 m long beams from the starting platform to a cage on the upper part; animals were presented with a negative stimulus in the base of the beam as a yellow light lamp and a sugar cookie reward in the dark cage. The beams were inclined at 15°; two different beams with 24- and 18-mm widths were used. The animals were tested for the basal measure four days after the start of the training before the 6-OHDA lesion. The test recorded the time it took the animals to walk from the base of the beam to the cage at the end of the shaft. Only animals that completed the test in less than 120 s during the basal determination were included in the experiment [ 41 ]. Distribution of implanted astrocytes Three 6-OHDA-lesioned Wistar rats were transplanted with AST-TH and ten weeks after implantation were euthanized, and striatal coronal sections obtained (~30 μm) from AP +1.60 mm to −0.92 mm. Sections were stained for GFAP, Th and counterstained with DAPI to reveal the nuclei. The estimated coordinates were established using the Paxinos and Watson rat brain atlas [ 42 ]. We determined the number of GFAP + /TH + cells of three independent fields of each striatal slice from three rats. Statistical analysis Results are expressed as mean ± SD ( n = 4). The sample size was calculated with G*Power software (Faul, Erdfelder, Lang and Buchner, 2007); a large effect size was used. We used the GraphPad Software version 8.0 to analyze the data. We evaluated data distribution with the Shapiro–Wilk test. Dopamine HPLC determination was analyzed with Brown–Forsythe ANOVA and Welch ́s ANOVA test with a two-stage linear step-up procedure of Benjamini, Krieger, and Yekutieli test as post hoc . DOPAC/DA ratio and Th densitometry comparison were analyzed by One-Way ANOVA and Tukey ́s multiple comparations test as post hoc. P < 0.05 was considered statistically significant. For the behavioral tests we used four animals per group, after applied inclusion criteria based on the results of the test training, animals were randomized place in each group. The study was blinded during the data analysis. Amphetamine-induced rotation test data were analyzed with a two-way ANOVA with Sidak ́s multiple comparisons post hoc test. Forelimb placement asymmetry, and the inclined beam motor balance test (24 and 18 mm) data were analyzed with Repeated Measures ANOVA and Tukey ́s multiple comparison post-hoc test. P < 0.05 was considered statistically significant, variance between the compared groups was estimated, and it was similar.

Results CRISPR activation screening for Th expression We analyzed the promoter region of the rat genome using the DNA sequence from the Gene library of NIH (Gene ID: 25085, Norway rat chromosome 1) to determine potential th sgRNAs, as described in the methodology. We chose sequences with the highest specificity score, annotation score, efficiency score, and a high number of hits but non-genome mismatches (other genome sites where the oligo can align differently to the specific region). These sgRNAs were analyzed with the BLAST tool to check for mismatches in other rat genome regions ( Rattus Norvegicus ) and sgRNAs that did not align with other areas of the rat genome were selected (Fig. 1A ). Thirteen sgRNAs were chosen to activate the rat th gene (Supplementary Table 1 ) and were individually cloned in The SAM v2 vector (Addgene #75112) using The Golden Gate protocol [ 27 ]. To test the selected sequences, C6 cells were transfected with the SAM system with each Th sgRNA, protein expression induced by each sgRNA transfected, are shown in Supplementary Fig. 1A . The 13 th sgRNA sequences were tested, but we only show the results from TH4 sgRNA since this sequence achieved the highest levels of Th protein expression. The selected th sgRNA was sequenced to verify its sequence (Supplementary Fig. 1B ). RT-PCR was used to evaluate the levels of the th mRNA in C6 cells transfected with the selected th sgRNAs and, as a positive control, the rat striatum was employed. We did not find th mRNA expression in C6 control cells or those transfected with SAM system without th sgRNA (Fig. 1B , lane 3), th expression was only observed in C6 cells transfected with the complete SAM and the th sgRNA (Fig. 1B , lane 4). Moreover, we also determined the expression of the Th protein in the C6 transfected cells using two monoclonal antibodies (Abcam 112 and Santa Cruz Technologies F-11). In Fig. 1C , we show the blot of C6 cells transfected with the SAM system together with TH4 sgRNA, and it can be observed that only these cells express Th (Fig. 1C , lanes 2 to 4), in contrast we found no expression of Th in control (lane 1) or C6 cells transfected without th sgRNA (lane 5). As a positive control for the expression of Th, we used rat striatum (lane 6). In C6 transfected cells selected with blasticidin and hygromycin, DA production was detected. The stable C6 Th-expressing cell line released dopamine (DA) to the culture medium but DA could not be detected in the medium from control C6 cells (Fig. 1D ). The double antibiotic selection was applied to ensure that all elements of the SAM system and the selected th sgRNA were stably expressed. Then we showed the expression of Th by immunofluorescence in C6 cells with the SAM system transfected with or without the th sgRNA (Fig. 1E, F ). As can be seen in cells that express th mRNA, a positive stain for Th was observed, in contrast to C6 cells that do not express the messenger. These data demonstrate that the SAM-TH system can activate the expression of the endogenous th gene in cells of glial origin and induce the synthesis of an active protein that can produce and release DA to the extracellular medium. Expression of Tyrosine hydroxylase in cultures of rat astrocytes A cortical astrocyte cell line was infected with the p-Lenti SAM TH and the MPHv2 complex, cells selected with blasticidin and hygromycin for 14 days and RNA and proteins were extracted from the cells. Figure 2A shows the expression of the th gene and of dCas9, and ms2 mRNA in infected astrocytes as assessed by RT-PCR. We did not detect th , dCas9 or ms2 mRNAs in astrocytes cells (Fig. 2A , lane 2), but they were expressed in astrocytes infected with the complete system (Fig. 2A , lane 3); from now on we will refer this cell line as AST-TH. We also determined the expression of the Th protein in infected astrocytes using a monoclonal antibody against Th (Abcam 112). In Fig. 2B , we show the blot of the Th protein of AST-TH cells, and it can be observed that these cells express Th (Fig. 2B , lane 2). As expected, no signal for Th was detected in control astrocytes (lane 1). As a positive control, we used the same amount of protein from the rat striatum (25 μg, lane 3). Moreover, in cultures of infected astrocytes, we determined the production of DA, and observed that the stable AST-TH-expressing cell line released DA to the culture medium (Fig. 2C ), yet no detectable DA was observed in the medium of the culture of cortical control astrocytes. We then showed the expression of Th by immunofluorescence in AST-TH cells and in astrocytes (Fig. 2D, E ). It can be seen that AST-TH cells stain for Th; in contrast, control astrocytes do not express the protein. The viral titer used to infect astrocytes was 3.0 with an infection efficiency of 26.8 ± 3% ( n = 3). Transplanted Th-producing astrocytes (AST-TH) induce motor recovery in unilaterally 6-OHDA-lesioned rats We tested whether AST-TH could induce motor control improvement in hemiparkinsonian rats. We used the amphetamine-induced circling, cylinder, and inclined balance beam tests to evaluate motor recovery. Figure 3A shows the timeline of the training and behavioral tests applied to the three experimental groups, which were followed for 13 weeks. A sham group was included in the experimental design to compare the magnitude of recovery. Lesioned rats were separated into two transplanted groups; one received AST-TH cells, and the other only primary astrocytes. Astrocytes were implanted in two different sites in the lesioned striatum . Nine weeks after cell implantation, the rotation test in response to amphetamine was repeated; results are shown in Fig. 3B . We found a significant increment in the circling behavior of rats receiving unmodified astrocytes compared with those receiving AST-TH cells. Sham-lesioned rats showed negligible turning behavior, and it did not change in the course of the experiment (Fig. 3B ). The cylinder test was applied to the three groups before the lesion to determine the basal asymmetry index. We observed that before the lesion or simulated surgery (sham), the rats did not show significant differences in forelimb placement asymmetry (FPA, Fig. 3C ); in fact, values in the three groups were lower than 0.5, indicating low or no asymmetry in forelimb use. Then, we repeated the test 4 and 8 weeks after transplants, as previously reported [ 39 , 40 ]. Four weeks after the transplants, rats treated with control astrocytes showed an increment of FPA over the values of the sham and AST-TH transplanted rats. This value increases significantly over 0.5 by the eighth week, whereas it remains under 0.5 in the other two groups, indicating that the AST-TH cell transplant prevents increments in FPA (Fig. 3C ). The training for the inclined beam balance started ten days before the 6-OHDA lesion; after it, a basal line was established for the three groups. Figure 3D, E show the results of the tests performed every week after transplantation (week 1–8). In Fig. 3D , the AST-TH transplanted group shows similar time climbing the beam compared with the sham group. In contrast, the primary astrocytes transplanted group increased the time needed to climb. Then, we used an 18 mm thick beam allowed us to measure fine motor control and balance. We observed no differences in the time it took the rats to climb the beam for the sham and the AST-TH group. However, the rats that received only astrocytes increased time scores and were statistically different from the other two groups (Fig. 3E ). These data indicate that the implant of AST-TH cells decreases the motor imbalance and asymmetry of 6-OHDA-lesioned rats. Dopamine metabolism determined in implanted rats We evaluated DA content and the DOPAC/DA turnover ratio in the different experimental groups, both in the lesioned and the intact striatum . Figure 4A shows that there were no DA content differences in the intact side of any of the groups; however, in the lesioned side (Fig. 4B ), we found that DA cannot be detected in the lesioned-side of rats that received just astrocytes, whereas there is a clearly detectable level of DA in the striatum implanted with AST-TH. Then, we evaluated the turnover of DA in the striatum . In the left, control striatum , turnover values were similar in all groups. In contrast, no turnover can be determined in the control astrocytes transplanted group on the lesioned side. However, the turnover in the AST-TH group shows values that are not different from the sham group or intact striatum , suggesting an accelerated DA metabolism in the AST-TH implanted striatum . Besides, we evaluated whether AST-TH cells still expressed Th protein in the transplanted striatum . Figure 4E shows the expression of Th as determined by western blot analysis from the homogenates used to determine DA content in the three groups of animals. It can be observed that homogenates from the striata of all the rats express the Th protein in the intact (left) side and that only in the sham and AST-TH groups, Th was detected in the lesioned (right) side. Densitometric analysis of Th expression using ab152 antibody indicates that in the intact side of control astrocytes and AST-TH groups, an increment of Th expression occurs (Fig. 4F ); also, in the lesioned side of the AST-TH group, there is a recovery of Th expression comparable to the sham group. We can observe in the intact side of the transplanted groups a significant increase of Th expression in comparison with the Sham group; this effect has been attributed to a cross hemispheric nigrostriatal branching that occurs in the unilateral 6-OHDA lesion on the MFB, so dopaminergic fibers migrate to the contralateral striatum [ 43 , 44 ]. Expression of Th and GFAP in the lesioned striata These series of experiments were designed to show the expression of both Th and GFAP in the implanted astrocytes. Figure 5 shows the expression of Th and GFAP in the intact (row A) and in the lesioned striata (row B) of rats receiving control astrocytes. Astrocytes can be observed in A and B, however, the TH signal can only be seen in the intact side, but not overlapping with the GFAP stain (see merge). Rows C and D show the same markers in rats implanted with AST-TH in the striatum . Astrocytes were observed on both sides. However, unlike the control astrocytes treated group (row D), Th stain in the lesioned side co-localizes with GFAP. In contrast, in the left control side, staining is like in the control astrocytes-treated group (compared to row A). Row E shows the Z-reconstruction of an astrocyte expressing both GFAP and Th from D row, indicating that the co-localization of Th and GFAP occurs in astrocytes. We also show DAB immunohistochemistry images of Th and GFAP to show the overall Th and GFAP staining on the striatum (Supplementary Figs. 2 , 3 ). Supplementary Fig. 3A, B show a magnification of immunohistochemistry (IH) Th images of right and left striata , respectively. Distribution of AST-TH in the implanted striatum Finally, we analyzed the distribution of AST-TH cells within the striatum . To accomplish this goal, we obtained coronal serial sections of the striatum from AP + 1.60 mm to −0.92 mm, stained for Th, and counterstained with DAPI to reveal the nuclei. Figure 6A shows a series of progressive anteroposterior drawings of coronal sections of the left striatum level in which color squares indicate the approximate site of AST-TH cell injection. Mean of Th positive cells from the three subjects, along the stereotaxic coordinates, is shown in Fig. 6B . In Fig. 6C , it can be observed that all Th-positive cells in the lesioned and implanted striatum have an astrocyte-like morphology (Supplementary Fig. 5 ) and that they can be found widely distributed within the striatum , although with a higher concentration closer to sites of implantation (Fig. 6B, C ). Moreover, to compare astrogliosis between the intact side and the lesioned side that received AST-TH, immunofluorescence against Th and GFAP was performed in different sections (Supplementary Fig. 6 ). And we can observe strong astrogliosis in the lesioned side (as well in Supplementary Fig. 4A , we can also observe an increase of the number of astrocytes on the lesioned striatum on DAB IH images at 4x, 10x and 20x), and AST-TH astrocytes expressing Th, but there are also GFAP positive cells, not expressing Th, indicating they are endogenous astrocytes. Finally, no Th positive astrocytes were observed on the intact side, although TH positive terminals are detected in the intact side.

Discussion This work is a proof-of-concept experiment using a CRISPR-Cas9 system to treat a neurodegenerative disease. The strategy of this experiment is based on the fact that the expression of Th induces the synthesis of DA in astrocytes since it is the only enzyme absent in astrocytes to complete the biosynthetic pathway to produce DA [ 23 , 24 ]. We had previously demonstrated that the transfer of the th gene into astrocytes causes the expression of the enzyme, and motor recovery in both rats and in a non-human primate model of PD [ 16 , 23 , 24 ], thus demonstrating that astrocytes expressing Th can induce behavioral recovery in different animal models of PD. In the present work, we used a new and different approach: activating the endogenous astrocyte th gene. SAM induces higher transcriptional levels than other transgene transfer systems (retroviral and lentiviral methods), and even when compared to other CRISPR activation systems; it is highly specific; it does not disrupt the genome or cause double-strand breaks (DSBs), and the epigenomic modification is reversible [ 28 , 29 ]. Moreover, third-generation pseudo-lentivirus (p-lenti) are an efficient and safe gene delivery system to obtain stable cell lines expressing SAM. Furthermore, the lentiviral system allows the delivery of multiple sgRNAs potentially obtaining the regulation of several genes within the same cell. Finally, these viruses are self-inactivating and less immunogenic than previous methods of viral gene transfer, and thus it is a safer method to be used in gene therapy protocols [ 45 , 46 ]. Employing the SAM co-transcriptional complex, we activated the expression of the endogenous th gene in cells of glial origin, C6 glioma cells and rat astrocytes, which synthesized the protein, produced, and released DA into the medium. Furthermore, when the DA-producing SAM-engineered primary astrocytes were implanted in a rat PD model, they could reverse motor behaviors induced by the 6-OHDA lesion (Fig. 3 ). The motor improvement is attributable to the DA produced by astrocytes expressing their endogenous th gene, since no behavioral motor improvement was observed when astrocytes that did not activate the gene were implanted. We also demonstrated the expression of the th gene in astrocytes by double labeling with the glial marker GFAP and Th. Furthermore, tyrosine hydroxylase protein was only detected in the lesioned striatum of rats implanted with Th-producing astrocytes, which also contained significant amounts of dopamine, that was absent in lesioned rats that received control astrocytes and did not express the th gene. All these data demonstrate that it is possible to induce the expression of the endogenous th gene of astrocytes with a CRISPR activation system and induce behavioral improvement in an PD animal model. The levels of DA in the lesioned striata of rats receiving AST-TH were lower than those of control non-lesioned animals. There is a wide variation in the literature regarding the amount of DA necessary to induce behavioral recovery after striatal lesions, which can be 11% of those of normal rats [ 47 , 48 ]. Thus, the levels shown in this paper (14.42%) are within the range previously reported to induce behavioral recovery. The effect induced by lower DA concentrations is probably due to the increased extracellular lifetime of DA likely caused by poorer DA reuptake caused by a reduction of DA transporters on the lesioned side, and by the hypersensitivity of DA receptors [ 49 – 51 ]. We consider that DA production could be increased by implanting more DA-producing astrocytes in the lesioned rats. We chose the first week after the 6-OHDA lesion to select the animals based on their turning behavior, since one week after the lesion the dopaminergic degeneration is sufficient to induce motor impairment, yet, the complete damage is observed between the third- and fourth-weeks post-lesion [ 52 – 54 ]. Several arguments support the use of astrocytes as a platform for producing dopamine and other molecules that may impact the development and treatment of PD, such as BDNF and GDNF, that can help preserve the few dopaminergic neurons remaining and improve the dopamine metabolism in the striatum [ 55 , 56 ]. Previous studies demonstrated that astrogliosis is present in the site of the lesion in the 6-OHDA hemiparkinsonian model, and other animal models and postmortem PD studies have found inflammation on the affected striatum , specifically astrogliosis [ 20 , 57 ]. In Fig. 5 , on the lesioned sites of both groups, we can observe astrogliosis in the striatum but not in the intact side, this indicates that the implant of astrocytes in both the lesioned and transplanted groups have the same glial response. Astrogliosis and increases of GFAP expression have been reported in previous works after both and short and long periods after the lesion [ 58 , 59 ]. Astrocytes can synthesize DA from L-DOPA and release it into the medium. Thus, the expression of Th in astrocytes is sufficient for the production and release DA from activated cells [ 22 , 60 ]. Moreover, astrocytes possess DA reuptake mechanisms to regulate the levels of extracellular DA, express DAT (which recycles the dopamine back to the cell), and COMT/MAO (which metabolizes dopamine) [ 21 ]. For all these reasons, astrocytes are an excellent option to express Th and regulate DA levels in the basal ganglia. We chose the CTX-TNA2 rat astrocyte line because the double antibiotic selection necessary to create a stable cell line that expresses the SAM complex requires several cell passages. This astrocyte line is non-tumorigenic, and thus it is not rejected by the host. We observed (Fig. 6 ) the distribution of cells in the lesioned striatum nine weeks after implantation. Cells appeared healthy and with an extensive distribution in the tissue, which seems to follow a gradient from the injection sites. The 6-OHDA hemiparkinsonian model is a late-stage PD, since it causes the unilateral death of the dopaminergic neurons, thus one hemisphere is depleted of DA. The lesioned rats present imbalance and motor asymmetry, as the lack of DA induces a loss of motor control from the lesioned side [ 61 ]. In our study, we wanted to determine whether DA released from AST-TH cells could correct the imbalance and asymmetry induced by the 6-OHDA lesion. To determine the motor alterations induced by the implant, we used a pharmacological test, the amphetamine-induced rotation sensitive to the release of dopamine from the nigrostriatal terminals and the sensitivity of the postsynaptic DA receptors. The amphetamine test allows us to evaluate the motor asymmetry improvement induced by the transplant, as previously shown [ 62 – 64 ]. To evaluate other aspects of motor behavior that the transplants can improve, we decided to perform a spontaneous test (cylinder test) and a test that requires training (inclined beam test). The cylinder test evaluates the asymmetry and the sensory-motor function that present the lesioned rats, and the beam test is a balance, sensory-motor, and motor performance evaluation. Rats that received AST-TH showed improvement in all tests, thus showing that the SAM-modified astrocytes, capable of expressing their endogenous th gene, are functional in an in vivo model of PD. We chose the SAM system as a first approach of a proof-of-concept experiment because it is robust, stable and has a high capacity to activate endogenous genes in mammalian cells [ 65 ], multiple gene transcriptional activation [ 27 ], and the generation of iPSC from human fibroblast [ 66 ]. Moreover, cells modified using SAM have been used in a mouse model of metabolic syndrome with good results [ 67 ]. This indicates that SAM can be used to reprogram other mammalian cells, besides astrocytes, and perhaps also directly transfer the SAM system into the endogenous brain astrocytes, thus without the need of implanting foreign cells. Double-stranded DNA breaks (DSB) occur when using CRISPR, causing mutations that could transform the cells and alter gene expression [ 25 , 68 – 71 ]. However, the SAM system does not cause DSB and thus does not present this kind of problem. Host immune response to Cas proteins can be a concern [ 72 ], but as other studies have shown, the improvement of less immunogenic, higher fidelity CRISPR-Cas systems, together with new, more specific and safer delivery systems may overcome the aforementioned problems associated with the use of CRISPR-Cas for therapy [ 73 – 75 ]. The therapeutic approach used can be improved by activating the expression of neurotrophic factors such as BDNF and GDNF that can increase DA metabolism in the striatum [ 56 , 76 , 77 ], or the use pharmacological inhibitors of COMT and MAO-B to increase the levels of DA in the extracellular space. In summary, we showed a proof-of-concept experiment that demonstrates the functional capacity of astrocytes expressing their endogenous th gene to produce DA and induce behavioral motor improvement in an experimental rat model of PD. This same approach can be improved by achieving the expression of other proteins that can protect the DA neurons and using different cell types like stem cells or induced pluripotent stem cells to better integrate the cells within the basal ganglia neuronal circuits.

Parkinson`s disease (PD) is the second most prevalent neurodegenerative disease, and different gene therapy strategies have been used as experimental treatments. As a proof-of-concept for the treatment of PD, we used SAM, a CRISPR gene activation system, to activate the endogenous tyrosine hydroxylase gene ( th ) of astrocytes to produce dopamine (DA) in the striatum of 6-OHDA-lesioned rats. Potential sgRNAs within the rat th promoter region were tested, and the expression of the Th protein was determined in the C6 glial cell line. Employing pseudo-lentivirus, the SAM complex and the selected sgRNA were transferred into cultures of rat astrocytes, and gene expression and Th protein synthesis were ascertained; furthermore, DA release into the culture medium was determined by HPLC. The DA-producing astrocytes were implanted into the striatum of 6-OHDA hemiparkinsonian rats. We observed motor behavior improvement in the lesioned rats that received DA-astrocytes compared to lesioned rats receiving astrocytes that did not produce DA. Our data indicate that the SAM-induced expression of the astrocyte ́s endogenous th gene can generate DA-producing astrocytes that effectively reduce the motor asymmetry induced by the lesion. Subject terms

Supplementary information

Supplementary information The online version contains supplementary material available at 10.1038/s41434-023-00414-0. Acknowledgements We want to thank Araceli Navarrete and Paula Vergara for invaluable technical assistance; Ivan José Galván for his technical support in the acquisition of the images and the LaNSE confocal microscopy unit, Cinvestav-Zacatenco; and Dr. Dulce María Delgadillo for plasmid sequencing and to The Genomics, Proteomics and Metabolic Unit of The National Laboratory of Experimental Services, Cinvestav - Zacatenco. Author contributions LFNP: designing research studies, conducting experiments, acquiring data, analyzing data, writing the manuscript’s first draft, and manuscript corrections. FPB: conducting experiments, HPLC. JAAF: conducting experiments, stereotaxic surgery. AC-R: reviewing the manuscript. BF: providing reagents, HPLC, stereotaxic surgery and reviewing the manuscript. JS: providing reagents, designing research studies, analyzing data, reviewing the manuscript and corresponding, funding acquisition. Funding This work received financial support from Pfizer Through its independent medical grants program (JS). Data availability Data analyzed in this study are available from the corresponding author upon reasonable request. Competing interests The authors declare no competing interests. Ethics approval All animal studies were performed according to The Guide for The Care and Use of Laboratory Animals [30], as adopted by the US National Institutes of Health and the Mexican Regulation of Animal Care and Maintenance (NOM-062-ZOO-1999). Rats were maintained and handled according to the guidelines of the CINVESTAV Animal Care Committee, making all efforts to minimize suffering and the number of animals used. The principles of the 3Rs (Replacement, Reduction, and Refinement) were followed to guarantee a humane endpoint.

CC BY

no

2024-01-16 23:41:57

Gene Ther. 2024 Aug 4; 31(1-2):31-44

oa_package/e2/d1/PMC10788271.tar.gz

PMC10788274

38225952

Background Allergic rhinitis (AR)/rhinoconjunctivitis (ARC) is a disorder of the nose and eyes affecting about one-fifth of the general population primarily driven by an immunoglobulin E (IgE)-mediated type I hypersensitivity response, due to an allergen exposure. 1 , 2 In Europe, the prevalence of diagnosed AR was recently estimated at around 25%. 2 AR is often characterized by nasal congestion, nasal secretion, post nasal drip, nasal and throat itching, and sneezing, with impact on social activities, social life, school performance, labor productivity, and quality of life, and accompanied by comorbidities such as conjunctivitis or asthma. 2 , 3 , 4 , 5 , 6 , 7 House dust mite (HDM) allergy is a major cause of allergic respiratory disease and a large proportion of patients with AR are sensitized to HDM, predominantly Dermatophagoides pteronyssinus and Dermatophagoides farinae . 8 , 9 , 10 , 11 Medical treatment of AR aiming at eliminating both symptoms and inflammatory reactions is mainly based on antihistamines, nasal steroids, or leukotriene receptor antagonists, but symptom relief does not extend beyond the end of treatment. 12 Allergen immunotherapy (AIT) is currently the only disease-modifying option available for AR, that also controls symptoms, and reduces medication use. 1 , 3 , 6 , 12 , 13 , 14 , 15 It interferes with the basic pathophysiological mechanisms modulating the allergic immune response and should be used as early as possible in the course of the illness to avoid progression of the disease and new sensitizations. 1 , 3 , 14 , 15 , 16 , 17 Various allergen formulations administered either subcutaneously or sublingually (liquid or tablet) have been approved for the treatment of moderate to severe AR with or without controlled allergic asthma induced by HDMs. 12 , 18 , 19 , 20 , 21 , 22 , 23 The 300 index of reactivity (IR) sublingual immunotherapy (SLIT) tablet of D. pteronyssinus and D. farinae extracts (total allergenic activity ratio 1:1) has demonstrated efficacy and safety in adults, adolescents, and children with HDM-induced AR (irrespective of mono- or poly-sensitization status or the concomitant presence of mild asthma) in multinational, double-blind, randomized, controlled trials (RCTs), and has been approved for commercialization in the Asia-Pacific region and Europe. 18 , 22 , 24 , 25 , 26 , 27 , 28 Of these, the most recent, international, phase 3 pivotal RCT confirmed the benefits of a 12-month course of treatment with the 300 IR HDM SLIT tablet versus placebo in a large population of adults and adolescents. 25 In this trial, the primary endpoint was the average total combined score (TCS, scale 0–15), defined as the sum of the patient's daily rhinitis total symptom score (RTSS, scale 0–12) and daily rescue medication score (RMS, scale 0–3), during 4 weeks at the end of the treatment period. The primary endpoint was met with a difference in total combined score (TCS) of −16.9% between the 300 IR and placebo groups. The pre-specified secondary endpoints consistently confirmed this result demonstrating the robustness of outcomes. The aim of this post hoc analysis was the evaluation of the European Union (EU) patients that participated in this international trial 25 to analyze the possible impact of differences in clinical management and/or in patients’ monitoring between regions in addition to possible intrinsic differences between patients from various regions. Moreover, the treatment effect of the 300 IR HDM SLIT tablet in this EU subpopulation primarily assessed on the imbalanced TCS 0-15 was compared with that on 2 combined scores equally weighing symptom and medication scores as recommended by Health Authorities and international experts: 29 , 30 i) the European Academy of Allergy and Clinical Immunology (EAACI)-proposed total combined symptom and medication score (CSMS 0-6 , a pre-defined endpoint) and ii) the total combined rhinitis score (TCRS 0-24 , a post hoc endpoint). Further insight into how this treatment effect can be translated into a clinically relevant improvement perceivable by the patients is also provided.

Methods Study design and materials The study design, set up, and results of the large, international, double-blind, placebo-controlled, phase 3 trial (EudraCT 2014-004223- 46 and ClinicalTrials.gov NCT 02443805) have already been published. 25 Here we report the results observed in the 91 centers from 9 EU countries involved in the trial: Belgium, Bulgaria, Czech Republic, France, Germany, Italy, Poland, Slovakia, and Spain. Briefly, the screening phase lasted 6 weeks to 6 months (including a 5-week, single-blind, placebo run-in period), the double-blind treatment phase for approximately 12 months, and the follow-up period around 2 weeks. Patients with an average TCS of 5 or more out of 15 over the last 4 weeks of the 5-week run-in period were centrally randomized 1:1 (computer-generated randomization with stratification by study site per participating country) to receive HDM AIT (a sublingual tablet formulation of standardized, purified, freeze-dried, sieved D. pteronyssinus and D. farinae extracts for daily administration; allergen content per 300 IR formulation: 14–17 μg Der p 1 and 53–68 μg Der f 1; Stallergenes Greer, Antony, France) or placebo. During the dose escalation phase, patients received a 100 IR tablet on Day 1 (under medical supervision for 30 min), 2 100 IR tablets on Day 2, and a 300 IR tablet on Day 3, or matching placebo tablets. The maintenance treatment consisted of 300 IR tablet or placebo once daily. Trial population The main inclusion criteria were reported previously. 25 Briefly, the population ranged from 12 to 65 years, had physician-diagnosed HDM-induced AR with or without concomitant asthma, and proven sensitization to D. pteronyssinus and/or D. farinae (skin prick test ≥5 mm than negative control and specific serum IgE ≥3.5 kU/L). The main exclusion criteria followed the guidelines, notably confounding allergies during the 4-week primary evaluation period, partly controlled or uncontrolled asthma according to Global Initiative for Asthma (GINA) 2014 classification, 31 previous AIT and the standard contraindications to AIT. Ethics approval and consent to participate This double-blind, placebo-controlled, randomized clinical trial was performed in accordance with good clinical practice defined by the International Council for Harmonization and the principles that have their origin in the Declaration of Helsinki and local laws and regulations. All participants or parents or legal representatives (for participants 17 years or younger) gave their written consent to participation, after having been informed of the trial objectives and procedures. Endpoints As previously described, the primary endpoint of the main study was the average of the daily TCS during the 4-week primary evaluation period. The daily TCS 0-15 was the sum of the patient's daily RTSS 0-12 and the daily RMS 0-3 . One of the key pre-defined secondary endpoints of this RCT was the CSMS 0-6 calculated as RTSS 0-12 /4 + RMS 0-3 , thus equally weighing the symptom severity and rescue medication use as recommended by the European Medicines Agency (EMA) 30 and the EAACI. 29 It is worth noting that the RMS component of both TCS and CSMS used a stepwise approach endorsed by both Health Authorities and international experts, scoring 0 for no treatment, 1 for an oral H1 antihistamine, 2 for intranasal corticosteroid, and 3 for oral corticosteroid. 29 , 30 Based on previously published scores, 32 an alternate daily medication score (DMS) was analyzed post hoc imputing a score of 4 per tablet of oral antihistamine (maximum daily score = 4), a score of 2 per puff of intranasal corticosteroid (maximum daily score = 8, ie, 2 puffs per nostril) and, if applicable, a score of 12 on the day of oral corticosteroid intake. The DMS ranged from 0 to 12. A second balanced combined score, TCRS calculated as RTSS 0-12 + DMS 0-12 and thus ranging from 0 to 24, was further assessed as complementary post hoc efficacy endpoint. Both TCRS 0-24 and CSMS 0-6 were analyzed over the primary period in the subgroup of EU patients and compared with the imbalanced TCS 0-15 and the results in the overall population. Other clinical endpoints of the main trial were also analyzed post hoc in the EU subpopulation: RTSS, RMS, rhinoconjunctivitis total symptom score (RCTSS), 6 individual rhinoconjunctivitis symptom scores (ISSs), total ocular symptom score (TOSS), rhinoconjunctivitis quality of life questionnaire (RQLQ) score, and safety variables. Clinical relevance Based on the CSMS 0-6 , the treatment effect (ie, the reduction in symptom and medication score with the 300 IR HDM tablet versus placebo over the primary period) was translated into a clinically relevant improvement from the patients' perspective by considering either component of the combined score: RTSS 0-12 or RMS 0-3 . The reduction in RTSS 0-12 was correlated to a decrease in symptom severity while the reduction in RMS 0-3 was correlated to a decrease in the number of days with less therapy for a patient taking antihistamines or nasal corticosteroids daily over the time period. Statistical analysis Data were analyzed with the same methodology used in the main study. 25 By definition, the modified full analysis set (mFAS) comprised all participants who received at least one dose of the investigated product (IP) and had at least one evaluation of the primary variable during the primary evaluation period. The baseline period was defined as the last 4 weeks of the 5-week run-in period and the primary evaluation period as the 4 weeks preceding the last record within a time window of last IP ± 28 days for patients treated at least 300 days. The ANCOVA model for appropriate assessment scores (TCS, CSMS, TCRS, RTSS, RMS, RCTSS, ISSs, and TOSS) was executed on the square root of the average score during the primary evaluation period with treatment group as main effect, and the pooled center, square root of the baseline average score, gender, age, asthma and sensitization status as covariates; p-values were two-sided. The back-transformed least squares (LS) means for each treatment group were used to assess the absolute (point estimate) and relative LS mean differences, with the relative LS mean difference (%) = 100 x [(LS mean 300 IR – LS mean placebo)/LS mean placebo]. For the quality-of-life endpoint analysis (RQLQ score), the ANCOVA model did not used the square root transformation. Additional models for a subgroup analysis of CSMS and TCRS have been performed by adding interaction between the treatment group and the subgroup of patients with “at least one rescue medication use at baseline”. A Mixed Model with Repeated Measures has been used to assess the treatment effect at study timepoints (3 months, 6 months, 9 months, 12 months) by adding timepoint effect and the interaction between treatment group and timepoint. The safety analysis was conducted on the safety set comprising all randomized patients having received at least 1 dose of the IP. Statistical analyses were performed with SAS software (version 9.4, SAS Institute, Inc, Cary, NC).

Results Study patients Of the 1607 adults and adolescents randomized in this trial, 992 were recruited in European centers and treated either with the 300 IR HDM tablet (N = 498) or with placebo (N = 494). The modified full analyisis set (mFAS) comprised 818 EU patients (384 in the 300 IR group and 434 in the placebo group). There were no differences between the 300 IR and placebo groups regarding baseline sociodemographic variables, clinical characteristics, and sensitization status of this subpopulation ( Table 1 ). Baseline symptom and medication scores were similar in both treatment groups ( Supplemental Table 1 ) with mean RTSS = 6.82 ± 1.98 (300 IR) and 6.74 ± 1.92 (placebo), mean RMS = 1.01 ± 0.773 (300 IR) and 0.94 ± 0.773 (placebo), mean TCS 0-15 = 7.83 ± 2.145 (300 IR) and 7.67 ± 2.023 (placebo), mean CSMS 0-6 = 2.72 ± 0.93 (300 IR) and 2.62 ± 0.88 (placebo), mean DMS = 4.26 ± 3.743 (300 IR) and 3.99 ± 3.755 (placebo), and mean TCRS 0-24 = 11.07 ± 4.368 (300 IR) and 10.73 ± 4.176 (placebo). The mean overall RQLQ score was 2.40 ± 1.06 in the 300 IR group and 2.47 ± 1.04 in the placebo group. In the safety set, the median overall exposure to the study treatment was 364 days in the 300 IR and the placebo groups. Compliance was good throughout the treatment period (>90% per group). Combined symptom and medication scores: TCS 0-15 , CSMS 0-6 and TCRS 0-24 Table 2 shows the results of the 3 combined scores in the overall population and in the EU subpopulation during the primary evaluation period. The LS mean average TCS 0-15 was significantly lower in the 300 IR group than in the placebo group (point estimates = −0.74, 95%CI [−1.08; −0.38] and −0.90, 95%CI [−1.34; −0.48] in overall and EU evaluable patients, respectively, p < 0.0001 for both), with relative LS mean differences from placebo of −16.9% and −20.3%, respectively. With regard to the balanced symptom and medication scores, average CSMS 0-6 and TCRS 0-24 , statistically significant LS mean differences between the 300 IR and the placebo groups were also observed. Point estimates were −0.26, 95%CI [−0.38; −0.14] and −1.07, 95%CI [−1.35; −0.79], respectively, in the overall population and −0.32, 95%CI [−0.46; −0.17] and −1.28, 95%CI [−1.63; −0.94], in the European subpopulation (p < 0.0001 for all). The corresponding relative differences of about −18% in all patients and −21% in EU patients were consistent with that of the TCS 0-15 in each population, respectively. In the EU subpopulation, the components of the combined scores: RTSS 0-12 and RMS 0-3 for the TCS 0-15 and the CSMS 0-6 , RTSS 0-12 and DMS 0-12 for the TCRS 0-24 also showed significant and consistent results ( Table 2 ). Point estimates and relative differences from placebo were −0.79, 95%CI [−1.18; −0.41], p < 0.0001, −20.6% for RTSS 0-12 ; −0.10, 95%CI [−0.17; −0.03], p = 0.0034, −29.0% for RMS 0-3 ; −0.41, 95%CI [−0.56; −0.26], p < 0.0001, −30.0% for DMS 0-12 . The clinical relevance of the treatment effect of the 300 IR HDM tablet for an EU patient is illustrated in Fig. 1 based on the results of the EAACI-recommended CSMS 0-6 . In calculating the CSMS 0-6 , the RTSS 0-12 component was divided by 4 to give equal importance to the RMS 0-3 . Thus, proportionately, for each of the 4 rhinitis symptoms initially scored from 0 to 3, a change of one severity level corresponded to a score of 0.25 with a maximum score of 0.75. A reduction of −0.32 in the CSMS 0-6 may reflect a decrease in one severity class in 1 symptom, for instance blocked nose, for the whole year, assuming the 3 other symptoms remain stable in severity and the use of rescue medication remains also stable. From the score calculation, it is possible to translate the severity of blocked nose into days with mild, moderate or severe symptom or even days with no symptom over the time period. The example in Fig. 1 A considers a patient treated with placebo with an average CSMS 0-6 of 1.52 and suffering from mild blocked nose over 41% of days, moderate blocked nose over 22% of days and severe blocked nose over 15% of days. With a CSMS 0-6 reduction of −0.32, a patient treated with 300 IR HDM tablet scored 1.20 can expect to be free from moderate to severe blocked nose over 1 year. Alternately, looking at the RMS 0-3 component, it is also possible to translate the medication score into days with use of antihistamines, nasal or oral corticosteroids, assuming the 4 rhinitis symptoms remain stable in severity. The example in Fig. 1 B considers a patient treated with placebo (CSMS 0-6 = 1.52) needing oral antihistamines for over 40% of days and nasal corticosteroids for over 7% of days. With a CSMS 0-6 reduction of −0.32, a patient treated with 300 IR HDM tablet (CSMS 0-6 = 1.20) can expect to be free from nasal corticosteroids and to reduce oral antihistamine intake over 1 year. Additional analyses with CSMS 0-6 and TCRS 0-24 In the EU subpopulation, ANCOVA analyses of the average CSMS 0-6 and TCRS 0-24 adding interaction between treatment group and the use of any rescue medication at baseline consistently showed that the treatment effect was more pronounced among patients with rescue medication use: CSMS 0-6 point estimate −0.34, p < 0.0001, relative difference −22.5%; TCRS 0-24 point estimate −1.38, p < 0.0001, relative difference −22.6%. However, due to the low number of patients who did not use a rescue medication at baseline (n = 43 in 300 IR group, n = 67 in placebo group), the interaction between treatment group and use of rescue medication did not show statistical significance (CSMS 0-6 and TCRS 0-24 point estimates −0.14 and −0.62, relative differences −9.7% and −10.9%). On the other hand, the treatment effect was assessed over time on CSMS 0-6 and TCRS 0-24 using a repeated measures model. A significant difference between the 300 IR and placebo groups was observed from 3 months of treatment onward whichever the score (all p ≤ 0.001, Fig. 2 ). Relative differences at Month 3, Month 6, Month 9, and Month 12 were −13.6%, −15.0%, −18.1%, −21.9% for the CSMS 0-6 and −13.8%, −15.0%, −20.0%, −22.1% for the TCRS 0-24 . Other endpoints During the primary evaluation period, all the clinical endpoints assessed in the 10.13039/100006939 EU subpopulation collectively demonstrated the 300 IR HDM tablet was significantly better than placebo ( Table 3 ), supporting the results of the combined symptom and medication scores and their components. In addition, the RCTSS, the TOSS, and 5 ISSs: nasal pruritus, sneezing, nasal congestion, ocular pruritus, and tearing improved compared to placebo for 20% or more for each score ( Table 3 ). The 300 IR group and the placebo group differed significantly. With regards to the quality of life, at the end of the treatment period, significant differences were observed in the overall RQLQ score and all the 7 domain scores between the 300 IR and placebo groups ( Fig. 3 ). The point estimate of the overall RQLQ between the 300 IR group and the placebo group at the end of treatment was −0.29 (95%CI [−0.42; −0.15], p < 0.0001) with a relative difference of −17.8%. For the 7 individual domains, the relative differences from placebo ranged from −15.7% for practical problems to −19.1% for nasal symptoms ( Fig. 3 ). Safety Overall, the safety profile in the EU subpopulation of this RCT was consistent with that in the overall population, 25 in line with the known safety profile of SLIT. No deaths occurred during the study, and there were no reports of anaphylaxis. No EU participants treated with 300 IR HDM tablet received epinephrine. During the 12-month treatment period, 324 (65.1%) patients in the 300 IR group and 254 (51.4%) patients in the placebo group reported at least one treatment-emergent adverse event (TEAE). The most common TEAEs related to the 300 IR HDM tablet were mild or moderate application-site reactions such as oral pruritus, throat irritation, ear pruritus, and mouth edema. One patient experienced a serious TEAE related to the 300 IR HDM tablet. This was an adult who experienced a severe pharyngeal reaction on the second day of treatment immediately after taking the IP at home. He took cetirizine tablets immediately, and the event resolved 1 h later. The patient discontinued the study on the same day. Forty-one patients (8.2%) in the 300 IR group and 2 patients (0.4%) in the placebo group prematurely discontinued the study due to TEAEs suspected to be drug-related, mostly application-site reactions.

Discussion The present study aimed to evaluate post hoc the results of the EU population that participated in the recently published international, randomized, placebo-controlled, phase 3 clinical trial with the 300 IR HDM sublingual tablet, 25 with a focus on 2 balanced symptom and medication scores following recommendations from Health Authorities and international experts, 29 , 30 the CSMS 0-6 and TCRS 0-24 . A large patient population of adults and adolescents with HDM-induced AR was recruited in the primary trial and two thirds (n = 818) of the 1262 patients evaluable for efficacy (ie, included in the mFAS) were European. In this subpopulation as in all patients, treatment with the 300 IR HDM tablet was associated with a significant and consistent reduction in the three assessed combined symptom and medication scores (imbalanced and balanced). Absolute and relative differences from placebo were even greater in EU patients with −0.90 and −20.3%, respectively for the average TCS 0-15 , −0.32 and −20.9% for the average CSMS 0-6 , and −1.28 and −21.2% for the average TCRS 0-24 . Noteworthy, this treatment effect is similar or even better than that observed in a RCT in EU patients with HDM-induced AR with another HDM SLIT tablet (12 SQ-HDM, ALK-Abelló, Hørsholm, Denmark) and using a balanced combined score close to the present TCRS 0-24 . 32 In that trial, absolute and relative differences in TCRS from placebo were −1.22 and −18%, respectively. For all 3 combined scores, the relative differences between the 300 IR HDM tablet and placebo in this EU subpopulation were greater than the threshold for difference in the primary outcome relative to placebo recommended by the United States Food and Drug Administration (FDA) to achieve clinically meaningful difference (ie, at least −15% with a 95%CI upper limit of at least −10%). 33 These differences also exceeded the −20% difference stated by the World Allergy Organization (WAO) 34 and the Global Allergy and Asthma European network (GA2LEN) of the Allergy and its Impact on Asthma (ARIA) initiative. 35 However, there is international consensus that a formal validation for the minimal clinically important difference (MCID) of a balanced symptom and medication score as primary endpoint in future trials in AIT is needed for defining the cut-off for a “clinical relevant” treatment effect-size. 29 , 36 From the patient perspective, the results of the EAACI-recommended CSMS 0-6 supported this clinically relevant effect in that they could be translated into a perceivable reduction in symptom severity, for instance blocked nose (reported as the most bothersome), and/or a decrease or cessation of symptom-relieving medication, notably corticosteroids, over a time period. This is a clinically meaningful result obtained already in the first year of treatment of a perennial disease affecting patients continuously exposed to allergens and having difficulty to control the daily symptoms and reduce their need for rescue medication. The positive results of the other clinical scores assessing the symptoms and medication use in the EU subpopulation like the components of the combined scores, RTSS, RMS, and DMS as well as the RCTSS and the individual nasal or ocular ISSs, collectively strengthen those of the combined scores, supporting the relevance of the 300 IR HDM tablet. In addition, the overall and the seven RQLQ domain scores were significantly lower in the 300 IR group than in the placebo group. Such an improvement in quality of life, which is known to be greatly impaired in patients with HDM-induced AR, contributes to the meaningfulness of this treatment for the patients. The 300 IR HDM SLIT tablet is currently marketed in Australia, Japan, New Zealand, South Korea, and more recently Europe. In seeking approval in Europe in 2021, the treatment effect of the product was specifically assessed in the EU population that participated in the primary international trial, which was considered pivotal. 25 Pivotal trials are normally performed in different countries worldwide to cover different healthcare systems, daily practices, and patient populations. It is remarkable that all the endpoint results in the EU subpopulation consistently exceeded those observed in the overall population. 25 The lower results in the overall population were essentially driven by differences between active treatment and placebo of lesser extent in other regions and notably in North America. Such a variation between regions may be explained by differences in clinical trial management and/or patients’ monitoring. Indeed, it has been acknowledged that in multi-regional clinical trials, treatment effects of drugs may be impacted by varying factors (including recruitment, compliance, subject retention, medical practice, environmental, cultural, socio-economic factors, etc) across regions or subpopulations. 37 The EMA ICH guideline E17 on “general principles for planning and design of multi-regional clinical trials” has been recently developed to address these issues but came into effect after the primary international trial was conducted. In addition, differences in patients' clinical characteristics and variations in allergen exposure across regions might have interfered with the trial results. 38 A post hoc analysis of the patients’ baseline characteristics by region ( Supplemental Table 2 ) revealed that, compared with the EU subpopulation, the North American (NA) was less exposed to HDM, had lower mean levels of HDM-specific IgE, included a higher proportion of polysensitized patients (about two-thirds in North America versus less than 40% in the European Union) with more patients exhibiting confounding sensitizations, therefore more likely to present intercurrent factors that might have affected the efficacy results. In addition, NA participants presented with a longer duration of HDM-AR suggesting they were likely more used to live with their allergy, which could have impacted their way of scoring. Though no notable differences in the combined score results were observed at baseline, the RMS component results were about 2-fold lower in NA versus EU patients and associated with a slightly higher RTSS ( Supplemental Table 3 ). A lower intake of rescue medications in the former might explain the lower decrease in combined score seen in the active group at the end of treatment. Furthermore, the proportion of dropouts was 1.5–1.8-fold higher within the NA versus the EU subpopulation, the primary reason being due to adverse events (AE) in the active group and withdrawal by the subject in the placebo group ( Supplemental Table 2 ). Assuming that the patients who prematurely withdrew due to AEs (mostly reactions to active treatment) were those who might have benefited most from the active treatment if they have continued, this suggests that remaining NA patients were likely less affected so that the difference in scores versus placebo was less marked. Interestingly, the clinical efficacy of the 300 IR HDM tablet was more pronounced in EU patients with severe enough symptoms to require symptom-relieving medication during the baseline period. This reinforces the value of this treatment during the periods with troublesome symptoms, also allowing the patients to reduce their consumption of rescue medications, particularly nasal corticosteroids. Finally, as expected, the safety profile of the 300 IR HDM tablet in the EU subpopulation was found as acceptable as in the overall population. 25 No severe anaphylactic reactions were reported. The strengths of this analysis were the large size of the subpopulation studied consisting of about two-thirds of the primary trial population and the consistency of results whichever the scores evaluated. The main limitation was the treatment duration of 12 months. However, the increased effect of this tablet over the 12-month treatment period, as observed in the overall trial population, is important to highlight given that EAACI guidelines recommend a minimum of 3 years of AIT to achieve long-term efficacy. 1 Other clinical trials with the 300 IR HDM tablet demonstrated an early onset of efficacy after an interval of only 8–12 weeks, indicating that benefits were apparent also in the first year of treatment. 24 , 27 In addition, one of these trials conducted in Europe showed that efficacy was maintained during a treatment-free follow-up year after 1 year of therapy. 24 These findings support the potential of AIT for further improvement beyond 12 months and that continuing AIT will ensure to “consolidate” clinically meaningful benefits which can last after treatment cessation. 39

Conclusion The post hoc data from the international, randomized, placebo-controlled clinical trial focusing on the European HDM-AR adults and adolescents confirm the efficacy and safety of the 300 IR HDM SLIT tablet during the first year of treatment. The results were even better than those observed in the overall population. Moreover, when applying the recommended balanced symptom and medication scores CSMS 0-6 or TCRS 0-24 , a clinically relevant efficacy above the required 20% improvement was shown and translated into clinically meaningful benefits from the patients' perspective. The significant improvement in the overall RQLQ score as well as in all seven domains of the RQLQ unpin the meaningfulness of this treatment for European patients.

Background House dust mite (HDM)-induced allergic rhinitis (AR) is a major cause of allergic respiratory disease. The efficacy and safety of the 300 IR HDM sublingual immunotherapy (SLIT) tablet in patients with moderate-to-severe HDM-AR was confirmed in a large, international, phase 3 randomized controlled trials (RCTs). Here, we analyzed the results in the European population. Methods Data from 91 European centers that participated in the international, double-blind, RCT (EudraCT 2014-004223-46, NCT02443805) with the 300 IR HDM SLIT tablet versus placebo over 12 months were analyzed post hoc. The treatment effect in European adults and adolescents was notably assessed through the European Academy of Allergy and Clinical Immunology (EAACI)-recommended combined symptom and medication score (CSMS 0-6 , pre-defined endpoint) and the total combined rhinitis score (TCRS 0-24 , post hoc endpoint, also balanced) during the primary evaluation period (4 weeks at the end of treatment period) using analysis of covariance (ANCOVA). Results There were 818 patients who comprised the modified full analysis set in Europe. Over the primary period, the differences in CSMS 0-6 and TCRS 0-24 between the 300 IR and placebo groups were statistically significant (p < 0.0001): −0.32 (95%CI [-0.46; −0.17]) and −1.28 (95%CI [-1.63; −0.94]), respectively, with relative differences of −20.9% and −21.2%. All post hoc and the rhinoconjunctivitis quality of life endpoints were significantly improved with 300 IR versus placebo. The 300 IR HDM tablet was generally well tolerated. Conclusion This RCT sub-analysis confirmed the 300 IR HDM SLIT tablet is an effective and safe treatment for European adults and adolescents with HDM-AR with clinically meaningful benefits from the patients' perspective. Trial registration NCT02443805. Registered on April 29, 2015./EudraCT 2014-004223-46. Registered on September 16, 2015. Keywords

Abbreviations AIT, Allergen Immunotherapy; ANCOVA, Analysis of Covariance; AR, Allergic Rhinitis; ARC, Allergic Rhinoconjunctivitis; ARIA, Allergy and its Impact on Asthma Initiative; CI, Confidence Interval; CSMS, Combined Symptom and Medication Score; DMS, Daily Medication Score; EAACI, European Academy of Allergy and Clinical Immunology; EMA, European Medicines Agency; FAS, Full analysis set; FDA, Food and Drug Administration; GA2LEN, Global Allergy and Asthma European network; GINA, Global Initiative for Asthma; HDM, House Dust Mite; Ig, Immunoglobulin; IP, Investigational product; IR, Index of Reactivity; ISS, Individual Rhinoconjunctivitis Symptom Score; kU/L, Kilounits per Liter; LS, Least Squares; mFAS, Modified Full Analysis Set; MCID, Minimal Clinically Important Difference; N, n, Number of patients; RCT, Randomized Controlled Trial; RCTSS, Rhinoconjunctivitis Total Symptom Score; RMS, Rescue Medication Score; RQLQ, Rhinoconjunctivitis Quality of Life Questionnaire; RTSS, Rhinitis Total Symptom Score; SLIT, Sublingual immunotherapy; TEAE, Treatment Emergent Adverse Event; TCS, Total Combined Score; TCRS, Total Combined Rhinitis Score; TOSS, Total Ocular Symptom Score; WAO, World Allergy Organization Funding The study as well as the medical writing assistance was sponsored and funded by Stallergenes Greer (Antony, France). Availability of data and materials The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. The data of this post hoc analysis have been presented as abstract at the German Allergy Congress “DEUTSCHER ALLERGIE KONGRESS” (2021), published as “Pfaar O, Kleine Tebbe J, Demoly P, and Bahbah F. Clinical relevance of treatment with 300IR house dust mite SLIT tablet.” Allergo J Int. 2021; 30:215. Authors' contributions and consent for publication All authors participated in the double-blind placebo-controlled clinical trial as country-principal investigators and/or were involved in the discussion of the post hoc analysis, contributed to the manuscript from draft stage and approved the final manuscript for submission. Ethics approval and consent to participate This double-blind, placebo-controlled, randomized clinical trial was performed in accordance with good clinical practice defined by the International Council for Harmonization and the principles that have their origin in the Declaration of Helsinki and local laws and regulations. All participants or parents or legal representatives (for participants 17 years or younger) gave their written consent to participation, after having been informed of the trial objectives and procedures. Data used from RCT (EudraCT 2014-004223-46, NCT02443805). Declaration of competing interest O. Pfaar reports personal fees from Stallergenes Greer, during the conduct of the study; grants and/or personal fees from ALK-Abelló, Allergopharma, Stallergenes Greer, HAL Allergy Holding B.V./HAL Allergie GmbH, Bencard Allergie GmbH/Allergy Therapeutics, Lofarma, ASIT Biotech Tools S.A., Laboratorios LETI/LETI Pharma, GlaxoSmithKline, from ROXALL Medizin, Novartis, Sanofi-Aventis and Sanofi-Genzyme, Med Update Europe GmbH, streamedup! GmbH, Pohl-Boskamp, Inmunotek S.L., John Wiley and Sons, AS, Paul-Martini-Stiftung (PMS), Regeneron Pharmaceuticals Inc., RG Aerztefortbildung, Institut für Disease Management, Springer GmbH, AstraZeneca, IQVIA Commercial, Ingress Health, Wort&Bild Verlag, Verlag ME, Procter&Gamble, ALTAMIRA, Meinhardt Congress GmbH, Deutsche Forschungsgemeinschaft, Thieme, Deutsche AllergieLiga e.V., AeDA, Alfried-Krupp Krankenhaus, Red Maple Trials Inc., Königlich Dänisches Generalkonsulat, Medizinische Hochschule Hannover, ECM Expro&Conference Management, Technische Universität Dresden, Lilly, Paul Ehrlich Institut (PEI), Japanese Society of Allergy, Forum für Medizinische Fortbildung (FomF), all outside the submitted work and within the last 36 months; and he is member of EAACI Excom, member of ext. board of directors DGAKI; coordinator, main- or co-author of different position papers and guidelines in rhinology, allergology and allergen-immunotherapy. F. de Blay reports financial interests from Stallergenes Greer for participation in advisory boards. G.W. Canonica reports having received research grants as well as being lecturer or having received advisory board fees from: Allergy Therapeutics, HAL Allergy, Stallergenes-Greer. T.B. Casale reports on-financial support from ThermoFisher Scientific, outside the submitted work. P. Gevaert reports personal fees and non-financial support from Stallergenes Greer, during the conduct of the study. P. Hellings reports grants and personal fees from ALK-Abelló, Astra Zeneca, GlaxoSmithKline, Novartis, Sanofi/Regeneron, Stallergenes Greer and Viatris, outside the submitted work. K. Kowal reports personal fees and/or honoraria for lectures, or royalties from ALK-Abelló, AstraZeneca, Aurovitas Pharma, Berlin Chemie, Meda Pharma, Stallergenes Greer and UpToDate, outside the submitted work. G. Passalacqua reports consulting fees, honoraria for lectures, and/or research funding from ALK-Abelló, Allergopharma, Lofarma and Stallergenes Greer. M. Tortajada-Girbés declares he has no conflicts of interests to disclose regarding this work. C. Vidal reports lecture fees from ALK-Abelló, AstraZeneca, Novartis, Sanofi, Leti, Mundipharma, Chiesi and Stallergenes Greer, outside the submitted work. M. Worm reports honoraria and/or consultation fees from Abbvie, Aimmune Therapeutics, ALK-Abelló, Amgen, AstraZeneca, Boehringer Ingelheim Pharma, DBV Technologies, GlaxoSmithKline, Kymab Limited, Leo Pharma, Lilly, Mylan, Novartis Pharma, Pfizer Pharma, Regeneron Pharmaceuticals, Sanofi-Aventis, Stallergenes Greer. F. Bahbah was a former employee of Stallergenes Greer . P. Demoly reports grants from: ALK-Abelló, AstraZeneca, GlaxoSmithKline, Menarini, Puressentiel, Stallergenes Greer, ThermoFisher Scientific, Viatris, Zambon, outside the submitted work.

Supplementary data The following is the Supplementary data to this article: Acknowledgments We thank all the investigators, nurses, and participants who made this study possible, and all members of the SL75.14 clinical trial for their commitment to the study. Post hoc statistical analyses have been provided by Catherine Gentil, Axiodis (Toulouse, France). Medical writing and editorial assistance were provided by Dr. A. Narkus, MC Narkus GmbH Medical Consulting & Services supported by Dr. Silvia Scurati, Dr. Wadad Hobeika and Dr Josiane Cognet-Sicé, Stallergenes Greer Global Medical Affairs (Antony, France).

CC BY

no

2024-01-16 23:41:57

World Allergy Organ J. 2023 Dec 22; 17(1):100849

oa_package/94/98/PMC10788274.tar.gz

PMC10788280

38154447

INTRODUCTION Over the years, the poultry industry has undergone changes to meet increasing demand for meat from the global population and minimize economic losses. Furthermore, in commercially-raised birds, such as broilers and laying hens, understanding body composition is valuable for various purposes, including breeding and selection programs, nutritional recommendations, and poultry management. Birds with fast growth have been successfully selected to achieve target body weight earlier and for feed efficiency ( Havenstein et al., 2003 ). However, these selection programs have led to several indirect consequences, such as metabolic disorders, leg problems, and increased fat deposition ( Shim et al., 2012 ; Schallier et al., 2019 ). In order to detect these inherent problems and help breeders take appropriate actions, methods for measuring chickens' body compositions have been employed. Carcass chemical analysis is the most commonly used method to assess body composition and is considered the gold standard for this process. However, this method is time-consuming, complex, invasive, and requires the animal to be sacrificed to establish the content of protein, fat, water, and ash ( Kasper et al., 2021 ; Martinez et al., 2022b ). Although carcass chemical analysis produces reliable measurements, it does not allow for repeated measurements on the same animal throughout its growth, which is feasible with live animal methods. Furthermore, live animal methods meet the ethical demands of certain groups of society for animal care and use, by preventing slaughter of several individuals; moreover, this method can reduce costs and environmental pollution associated with chemical analysis ( Gonçalves et al., 2018 ). Dual-energy X-ray absorptiometry ( DEXA ) is a noninvasive method of measuring body composition that was originally designed for humans, but in recent decades, has been used as a tool in livestock research ( Scholz et al., 2015 ), and in the food industry for meat quality inspection ( Pipe, 2023 ). The equipment software uses the reduction of the dual X-ray beam caused by distinct absorbing materials to estimate the values of total tissue, soft lean tissue ( SLT ), fat tissue, bone mineral content ( BMC ), and fat percentage in the projected figures ( Schallier et al., 2019 ). DEXA has been proven useful in animal science; its application varies between prediction of intramuscular fat in beef steak ( Nunes et al., 2023 ), body composition of swine ( Soladoye et al., 2016 ; Kasper et al., 2021 ), body composition and phosphorus retention in rainbow trout ( Ndiaye et al., 2020 ), mineral density and content in laying hens ( Hester et al., 2004 ; Schreiweis et al., 2004 ), broiler bone resistance ( Shim et al., 2012 ), and growth performance ( Castro et al., 2019 ; Meyer et al., 2019 ). Furthermore, DEXA is a promising noninvasive method for longitudinal studies to determine fast heating production, and consequently net energy requirements of chicken growth ( Martinez et al., 2022a ; Suesuttajit et al., 2022 ), and longitudinal studies about potential body growth of broiler and pullet strains ( Alves et al., 2019 ; Gonçalves et al., 2020 ). Despite these advantages, DEXA needs to be calibrated to estimate the body composition of birds because the technique is not capable of directly estimating protein and water contents. Some works have shown that DEXA is effective at estimating body composition of chickens by comparing DEXA results with chemical analysis ( Mitchell et al., 1997 ; Swennen et al., 2004 ; Salas et al., 2012 ; Gonçalves et al., 2018 ; Schallier et al., 2019 ). According to Mitchell et al. (2011) , chicken feathers are not detected by DEXA. In this sense, the DEXA results must be related to chemical composition of the feather-free body ( FFB ) of broilers ( Alves et al., 2019 ). To the best of our knowledge, Schallier et al. (2019) is the most recent study to use DEXA to predict body composition of chickens, for which the authors developed reliable prediction regressions; however, in that study, the birds’ ages ranged from 21 to 56 d. Nonetheless, there are different brands of DEXA devices and they are all equipped with specific software versions, necessitating the development of individual prediction regression models for all DEXA devices and software combinations ( Swennen et al., 2004 ; Kasper et al., 2021 ). For this reason, 2 trials were carried out in the present study. Trial 1 aimed to develop appropriate prediction equations for broiler body composition using a Lunar Prodigy Advance DEXA fan beam scanner (software enCORE v18SP2) and chemical analysis, with broilers of different body compositions ranging from 1 to 42 d of age. Trial 2 aimed to externally validate the models developed to predict the body composition of broilers obtained in Trial 1.

MATERIALS AND METHODS Ethics Experiments were conducted to create regression equations (Trial 1) and for equation validation (Trial 2). All procedures adopted for both trials were previously approved by the Ethics Committee in the Use of Farming Animals at the Universidade Federal de Viçosa ( CEUAP-UFV ) under the accession number 029/2020 for Trial 1 and the accession number 057/2022 for Trial 2. DEXA Scanning Birds’ body composition was measured using the standard mode of the small animal module of the DEXA scanner model Lunar Prodigy Advance (PRODIGY, GE HEALTHCARE, Madison, WI) equipped with enCORE v18SP2 software in Laboratory of Body Composition and Densitometry of the Department of Animal Science of the Universidade Federal de Viçosa ( UFV ). In each day prior to the scanner analyses of the birds, a quality assurance ( QA ) program was performed using a phantom standard to ensure an accurate calibration of the equipment. Next, the birds were placed in the dorsal position on the scanner table with spread wings and stretched legs to avoid overlap of body parts. Birds were scanned from the cranial to caudal direction. After the scanning, the “region of interest” ( ROI ) was selected and personalized by drawing a custom rectangular area around the whole body of the bird image, as shown in Figure 1 . The following DEXA traits were recorded: SLT (g), fat tissue (g), BMC (g), and fat percentage (%). The sum of soft lean and fat tissue with the BMC of the DEXA measurements was assumed to be the total body weight (g). SLT (%) and BMC (%) were calculated using the respective gram values as a ratio of total body weight. Trial 1: Development of Linear Regression Equations For regression equation development a total of 300 Cobb500 male broilers were raised from 1 to 42 d of age. Diets were formulated for starter (1–14 d of age), grower (14–28 d of age), and finisher (28–42 d of age) feeding phases according to Rostagno et al. (2017) (as shown in Supplemental Table S1 ), with various levels of methionine inclusion. The application of these ages and diet variations were important for achieving a wide range of body weights and compositions, which is essential for establishing accurate regression equations. During the experimental period, birds were reared on floor pens under standard temperature and light conditions according to genetics guidelines, with diets and water provided ad libitum. Nine 1-day-old birds and 97 birds for each age group of 14, 28, and 42 d of age were selected, respectively, for a fasting period of 12 h, and then euthanized by cervical dislocation and defeathered. As DEXA does not detect feathers, we chose to pluck and freeze carcasses prior to scanning to facilitate future processing for chemical analysis, and to allow scanning of several number of birds at the same age. The whole carcasses FFB (including organs, heads, and limbs) were weighted, then frozen (−20°C) and later defrosted for DEXA scanning analysis according to the aforementioned method. After scanning, each of the FFB carcasses were frosted again and were further cut with a band saw (SI-282HD, SKYMSEN, Volta Grande, SC, Brazil), grounded twice with industrial grinder (PBM081, BECCARO, Rio Claro, SP, Brazil), homogenized, and a sample was collected to be lyophilized for chemical analysis conducted at the Laboratory of Animal Nutrition of the Department of Animal Science of the UFV. One-day-old birds were pooled before lyophilization to obtain 3 representative samples. Lyophilized samples were ground in a ball mill and used for analysis of dry matter ( INCT-CA, 2021 ), crude protein Kjeldahl N × 6.25 ( INCT-CA, 2021 ), ether extract ( AOCS, Am 5-04, 2009 ), and ash ( INCT-CA, 2021 ). All chemical values obtained were converted into natural matter to express the quantitative amount of total ash, ether extract, and the protein of total FFB composition (with water) in order to be compared with DEXA scan results. Regarding the 1-day-old birds DEXA scan results, the mean of 3 chicks used for the pool was compared with the wet chemical counterpart (total n = 294). Due to the fact that DEXA SLT mass considers the amount of protein and water of tissue, the chemical SLT was considered the sum of crude protein and water. Trial 2: Linear Regression Equations Validation A second trial was conducted to evaluate the accuracy of the regression equations developed in Trial 1. A total of 395 Cobb500 male broilers were raised from 1 to 42 d of age on floor pens. During the rearing period, the animals received corn and soybean-meal based diets in order to meet or exceed the requirements established by Rostagno et al. (2017) from 1 to 21 and from 22 to 42 d of age, with variations in mineral sources and quantities (shown in Supplemental Table S2 ). Birds received diet and water ad libitum and the ambient temperature and the light program were controlled following the genetics guidelines. At the beginning of Trial 2, a group of 10 one-day-old birds were selected, fasted 12 h, and euthanized by cervical dislocation. In order to acquire sufficient data from chickens with different ages for validation of the prediction equations developed in Trial 1, at the end of each 14, 21, 28, 35 and 42 d of rearing, a total of 77 birds were selected, fasted 12 h, euthanized, defeathered, and frozen, respectively. Afterward, the FFB broiler carcasses were defrosted and the FFB carcasses were scanned using the DEXA method mentioned previously, and then all FFB carcasses were frozen for further chemical analysis procedures like those described in Trial 1. For all body composition traits, 77 chickens per age group (14, 21, 28, 35, and 42 d of age) were used for chemical analyses of protein, water, and ash (total n = 385). However, 55 chickens for each age groups, were used for fat traits (total n = 275). Ten 1-day-old birds were not processed for chemical analysis, a decision that is explained below. Statistical Analysis DEXA measurements and wet chemical data were organized into the following variables: weight, fat, ash, SLT, water, and protein of the FFB broiler expressed as weight (g) and content (%) ( Table 1 ). All statistical procedures were performed using R software ( R Core Team, 2022 ), the script of which is available in the supplemental material ( Table S3 ). In all statistical tests described below, results were considered significant when P < 0.05. The first step ( n = 294) for the development of regression equations for each variable was to verify the confidence interval ( CI ) and significance of the least-squares linear regression parameters: intercept ( β 0 ) and slope ( β 1 ). Afterward, a Pearson correlation ( ρ ) analysis between DEXA measurements and wet chemical data was performed. The second step was to select, based on β 0 , β 1 , and ρ significances, what body composition variables were able for development of the linear regressions. Therefore, to verify that the model was not overfitted, a random split was performed on the Trial 1 dataset for a 5k-fold cross-validation analysis; 70% of dataset ( n = 206) was used for training to create the linear regression equations, and the remaining 30% ( n = 88) was used to test the prediction of linear regressions. Additionally, the mean absolute error ( MAE ) and the root-mean-squared error ( RMSE ) was estimated for each variable, and together with the coefficient of determination ( R 2 ), were used as indicators of precision and accuracy. The third step was to validate the linear regression equations created and tested with Trial 1 dataset using the Trial 2 dataset. DEXA measurements of Trial 2 were used in linear regression equations to obtain the predicted body composition values. The 1-day-old birds from Trial 2 data were not used for validation because the prediction results were not reliable (e.g., negative values observed for protein mass) ( Table 1 ). The predicted values were then compared to the wet chemical counterparts with a regression analysis (predicted vs. observed) and R 2 , MAE, RMSE, and P were assessed as precision and accuracy indices ( n = 385, except fat = 275).

RESULTS Variation in Body Composition of Broiler Chickens Variation in body composition of broiler chickens of different ages for Trial 1 and Trial 2 was achieved; the means with the coefficient of variation are shown in Table 1 . Regardless of birds’ ages or analysis method (DEXA or chemical), and in both trials, body measurements from highest to lowest variation were: fat, ash, protein, weight, SLT, and water. Trial 1: Development of Linear Regression Equations The linear regression parameters of variables measured by DEXA scanner and chemical carcass analysis are presented in Table 2 . β 0 indicates the intercept and β 1 the slope of linear regression, and their respective CI are also exhibited. The degree of correlation is indicated by ρ as the correlation coefficient and the R 2 are also shown in this table. The total weight measured by DEXA scanner was the sum of SLT, fat, and BMC tissues and was compared with the carcasses scale weight. For weight mass linear regression parameters verification, both β 0 and β 1 were significant ( P < 0.001) and a high positive correlation ( ρ = 1) and R 2 value (0.999) between DEXA weight and scale weight was observed ( Table 2 ). Based on these results, a linear regression equation was developed to estimate the broiler carcass weight (R 2 = 0.999, MAE = 25.12, RMSE = 38.99) and is shown in Figure 2A . The test of regression equations of the weight mass with Trial 1 data split is shown in Table 3 , and a high R 2 (0.999) value was observed ( P < 0.001, MAE = 18.20, RMSE = 25.22). For both fat content and fat mass, β 0 and β 1 were significant ( P < 0.001) and a high positive correlation was observed, with the R 2 value of parameter verification for fat content at 0.833 and for Fat mass 0.982 ( Table 2 ). The linear regression equations developed to estimate broiler fat content (R 2 = 0.855, MAE = 0.81, RMSE = 1.05) and fat mass (R 2 = 0.981, MAE = 13.87, RMSE = 21.28) are shown in Figures 3A and 2B , respectively. The test of the regression equations of fat content (R 2 = 0.781, P < 0.001, MAE = 0.93, RMSE = 1.21) and fat mass (R 2 = 0.984, P < 0.001, MAE = 13.07, RMSE = 18.33) with the Trial 1 data split is shown in Table 3 . The ash content's β 0 and β 1 were significant ( P < 0.001); nevertheless, the R 2 value was low (0.070) and a negative correlation ( ρ = −0.27) was observed ( Table 2 ). Based on these results, a linear regression equation for ash content was not developed. For ash mass, the β 0 and β 1 were also significant ( P < 0.001), and a high positive correlation ( ρ = 0.98) and R 2 value (0.953) was observed ( Table 2 ). The linear regression equation developed for ash mass (R 2 = 0.956, MAE = 3.98, RMSE = 5.61) is also shown in Figure 2C . The test of linear regression equations using the Trial 1 data split for ash mass (R 2 = 0.947, P < 0.001, MAE = 3.91, RMSE = 5.60) is shown in Table 3 . The SLT content and SLT mass were compared to the sum of water and protein obtained through humidity (via dry matter) and wet crude protein chemical analysis. β 0 and β 1 were significant ( P < 0.001) for both SLT content and SLT mass; therefore, a positive correlation ( ρ = 0.79) was observed for SLT content (R 2 = 0.628) and a high positive correlation ( ρ = 1) was observed for SLT mass (R 2 = 0.997) ( Table 2 ). A linear regression equation was developed for SLT content (R 2 = 0.658, MAE = 1.01, RMSE = 1.28) and for SLT mass (R 2 = 0.997, MAE = 35.73, RMSE = 52.45) and is shown in Figures 3B and 2D , respectively. The test of regression equations of SLT content (R 2 = 0.566, P < 0.001, MAE = 1.08, RMSE = 1.31) and SLT mass (R 2 = 0.998, P < 0.001, MAE = 29.66, RMSE = 40.46) with Trial 1 data split is shown in Table 3 . As SLT measured by the DEXA scanner represented the muscle tissue composed of protein plus water, the SLT was considered to be the counterpart of chemical-based water and protein data for the linear regression equations development. Considering this, for water content and water mass, both β 0 and β 1 were significant ( P < 0.001) and a positive correlation ( ρ = 0.80) for water content (R 2 = 0.635) and a high positive correlation ( ρ = 1) for water mass (R 2 = 0.997) was observed ( Table 2 ). A linear regression equation was also developed for water content (R 2 = 0.678, MAE = 0.99, RMSE = 1.27) and for water mass (R 2 = 0.997, MAE = 29.56, RMSE = 43.94) and is shown in Figures 3C and 2E , respectively. The test of regression equations of water content (R 2 = 0.532, P < 0.001, MAE = 1.15, RMSE = 1.42) and water mass (R 2 = 0.997, P < 0.001, MAE = 26.60, RMSE = 37.55) with Trial 1 data is shown in Table 3 . Regarding protein content, β 0 was significant ( P < 0.001) but β 1 was not significant ( P = 0.22) and no correlation between DEXA measurements and chemical data was observed ( Table 2 ). Based on these results, a linear regression equation for protein content was not developed. However, in relation to protein mass, both β 0 and β 1 were significant ( P < 0.001) and a high positive correlation ( ρ = 0.99) and R 2 value (0.989) was observed ( Table 2 ). The linear regression equation developed to estimate broiler protein mass (R 2 = 0.989, MAE = 12.94, RMSE = 19.05) is shown in Figure 2F . The test of linear regression equation using Trial 1 data for protein mass (R 2 = 0.991, P < 0.001, MAE = 10.95, RMSE = 15.32) is shown in Table 3 . Trial 2: Linear Regression Equations Validation The linear regression equations validation was performed by comparing the values estimated using the equations developed in Trial 1 (predicted values) against the values observed via chemical analysis through a regression. For all variables expressed in weight, high R 2 values were observed (>90%), whereas, for all variables expressed in content, the R 2 values observed were around 50%. The regressions of predicted vs. observed values for each variable expressed in weights and contents are shown in Figure 4 , Figure 5 , respectively.

DISCUSSION The first Trial aimed to compare the DEXA scanner estimates against chemical carcass analysis of broilers aged 1 to 42 d and to establish reliable predictive linear regression equations. Chemical analysis is still the gold standard for measuring chicken broiler composition, and consequently, DEXA accuracy can be determined by comparison ( Schallier et al., 2019 ). Furthermore, Trial 2 evaluated the validation of prediction equations developed in Trial 1 using external data. Validation is highly recommended for verifying whether a mathematical model is useful and accurate to a real system ( Inca et al., 2022 ). Independent of the method used to measure the FFB composition, DEXA and chemical analysis estimates showed similar tissue pattern of variation in the following (descending) order: fat, ash, protein, weight, SLT, and water. The diets and age conditions applied to the animals in Trials 1 and 2 successfully provided a diverse dataset for application of tests, development of prediction equations, and subsequent validation. The results of development of the linear regression equations (Trial 1) demonstrated that DEXA was an effective approach for predicting body carcass composition of broilers. The weight determined by DEXA as the sum of SLT mass, fat mass, and BMC showed an almost perfect correlation with scale weight in this study. The equation slope of 1.079 demonstrated that DEXA slightly underestimated the FFB weight of chickens, which may be due to underestimation of SLT mass and ash and the overestimation of fat, which is consistent with the observations of Schallier et al. (2019) . Nonetheless, other studies with chickens demonstrated an overestimation of DEXA weight even in the presence of a high positive correlation with scale weight (i.e., Mitchell et al., 1997 ; Swennen et al., 2004 ; Salas et al., 2012 ; Gonçalves et al., 2018 ; Martinez et al., 2022b ). This inconsistency among previous studies, although small, reinforces the importance of calibration of specific instruments, software versions, and ROI methodologies. The criteria for stating equation robustness were to observe high R 2 and low MAE or RMSE error metrics ( Martinez et al., 2022b ). Therefore, the linear equation for weight demonstrated a high accuracy when we observed the reduction of 6.92 g MAE (or, RMSE = 13.77) value of testing using Trial 1 split data. The high correlation of fat mass observed in this study is also in agreement with what was observed ( ρ = 0.98) by Schallier et al. (2019) . Furthermore, the R 2 (98%) of this study was higher and was corroborated by other values in the literature ranging from 89 to 96% ( Korine et al., 2004 ; Swennen et al., 2004 ; Salas et al., 2012 ; Gonçalves et al., 2018 ; Schallier et al., 2019 ). An exception to this were the results of Mitchell et al. (1997) , who observed a value of 62% for R 2 that the authors attributed to the DEXA scanning mode, program version, and to the weight of the birds, as birds weighing less than 2,000 g had more discrepancies in fat mass. However, Gonçalves et al. (2018) , even with a high R 2 (91%), also observed similar discrepancies in birds with low body weight (7 d of age), which had high fat mass; the contrary occurred in heavyweight birds (28 and 77 d of age). In the present study, DEXA overestimated fat mass ( β 1 = 0.756), which is in agreement with other studies ( Korine et al., 2004 ; Swennen et al., 2004 ; Salas et al., 2012 ; Schallier et al., 2019 ). This was previously attributed to variation in soft tissue hydration, as discussed by Pietrobelli et al. (1998) , although the authors concluded in the same work that the magnitude of DEXA fat estimation error related to soft tissue hydration is small under normal circumstances, and should not pose any substantial limitations to the accuracy of DEXA technique. Nonetheless, Korine et al. (2004) pointed out the following reasons for DEXA fat overestimation in their study: first, DEXA calibration was based on rat bone density (mouse phantom of Lunar PIXImus 2 model); second, the feathers were mistaken as fat by the DEXA scan; lastly, the ether petroleum used as a solvent for lipid extraction extracted mainly triglycerides (84% of total lipids), whereas the DEXA scanner estimates total lipids. In the present study, the first point can be disregarded due to this DEXA scanner model's and software differences. Although defeathered chickens were used in this study, it is important to note that, while Mitchell et al. (2011) stated that chicken feathers are not detected by DEXA, the study by Korine et al. (2004) focused on migratory birds that may have very different feather compositions than domestic broiler chickens. In this context, the last point about lipid extraction method may explain the fat mass overestimation observed in our study. Another topic discussed by the authors is that DEXA's X-ray attenuation coefficient distinguishes a pixel based on the coefficients of high (e.g., bone tissue) and low (e.g., soft tissue) attenuation energy levels; in areas close to the bones that may not contain adequate amounts of soft tissue, such as the wings, neck, head, and feet, the interpolation process is necessary to estimate the soft tissue composition. In such cases, the process is based on reconstruction of the soft tissue composition by applying the average body fat percentage of the entire animal; thus, if this percentage is greater than the fat percentage of these areas, the fat will be overestimated ( Mitchell et al., 1997 ; Nagy and Clair, 2000 ; Korine et al., 2004 ). Regarding fat content, there was also a positive correlation ( ρ = 0.91) between DEXA and chemical analysis results, while slightly less than that observed for fat mass. Although this decrease is in agreement with other works, the correlation found in our study was higher compared to Swennen et al. (2004) and Schallier et al. (2019) , who found correlation values 0.59 and 0.77, respectively. These differences may be related to evolution of DEXA software, which has been calibrated over the years; additionally, the scanner used in this study was more recently developed. The linear equation for fat mass proved to be robust in the split data Trial 1 test, as no changes in R 2 and a slight reduction of MAE and RMSE values were observed. However, the fat content prediction equation test showed a reduction of 7.43% in the R 2 value and an increase of 0.12% in MAE (or, RMSE = 0.16) value, indicating a possible inaccuracy of this prediction model. Ash content was compared using the DEXA BMC percentage as an independent variable, and as result, a negative correlation and a small R 2 value were observed in this study, thereby discarding the possibility of prediction model development and corroborating results observed in other studies ( Swennen et al., 2004 ; Schallier et al., 2019 ). A possible explanation for these results is that ash content comprises a small range of values among all age groups of chicken, reaching 0.3 to 1.6% for BMC (fold difference of 5.3) and 2 to 3% for chemical ash (fold difference of 1.5). On this basis, a linear regression model, contrasting ash chemical and BMC content, with a low coefficient of determination, was expected. However, a high correlation was observed for ash mass, as well as reported by Swennen et al. (2004) and Schallier et al. (2019) . A high R 2 was likewise observed by these authors, as well as by Salas et al. (2012) and Gonçalves et al. (2018) . Only 1 study observed a low coefficient of determination of 46% between ash mass and DEXA BMC, and the authors were unable to clarify the absence of correlation ( Mitchell et al., 1997 ). Observing the equation slope ( β 1 = 1.578), DEXA greatly underestimated the ash quantity of broiler bodies compared to chemical ash weight. This finding was also reported in other studies ( Swennen et al., 2004 ; Salas et al., 2012 ; Gonçalves et al., 2018 ; Schallier et al., 2019 ). The observed discrepancy between BMC and ash chemical weight may be attributed to the fact that DEXA exclusively measures the mineral content of bones; conversely, mineral chemical analysis, carried out by burning the sample in a muffle furnace, takes into account the entire broiler body, including bones and nonbone tissues, which also contain substantial mineral content (such as organs, muscle, and liver) ( Johnson et al., 2016 ; Gonçalves et al., 2018 ). Additionally, based on the results of the linear regression test using split data of Trial 1, DEXA BMC allowed a good estimation of total body ash mass of broilers when corrected with the corresponding linear regression. A slight reduction of 0.92% in R 2 was observed, whereas the MAE value was reduced to 0.07 g (or, RMSE = 0.01), indicating the robustness of the prediction model. Water comprises more than 70% of the whole bodyweight of broilers. Because bone and (hydrophobic) fat have less water content due to their chemical nature, it is assumed that SLT is composed of protein and a larger proportion of water ( Brommage, 2003 ). Hence, the comparison between DEXA SLT and water plus protein obtained chemically was accepted as “chemical soft lean tissue mass.” In this study, a high positive correlation between DEXA SLT mass and chemical SLT, water, or protein mass was consistent with previous works on chickens ( Mitchell et al., 1997 ; Swennen et al., 2004 ; Salas et al., 2012 ; Gonçalves et al., 2018 ; Schallier et al., 2019 ). In spite of this, an overestimation was observed by Swennen et al. (2004) in SLT determined by DEXA, and the same was found by Gonçalves et al. (2018) in chickens with heavy weights. On the other hand, similar to the present study ( β 1 = 1.157), Schallier et al. (2019) observed DEXA underestimated the SLT mass ( β 1 = 1.047), which the authors attribute to characteristics of every unique scanner and software, emphasizing the need for proper calibration. Furthermore, the linear regression developed with R 2 = 0.997 and MAE = 35.73 g (RMSE = 52.45) allowed a proper SLT prediction based on DEXA estimations; as when tested with Trial 1 split data, a reduction of 6.07 g in MAE (or, RMSE = 11.99) was observed. The same behavior was observed when linear equations for water and protein mass were tested using Trial 1 split data, as reductions in MAE of 2.96 g (or, RMSE = 6.39) and 1.99 g (or, RMSE = 3.73) were observed, respectively. Similarly to fat content, SLT ( ρ = 0.79) and water ( ρ = 0.80) content showed a positive correlation to weight but to a lesser extent ( ρ = 1, for both). In relation to protein content, no correlation was observed, making it impossible to develop a prediction equation based on DEXA SLT estimates. Although diets guaranteed a body protein content variation ( CV ∼5%) within the broiler ages, the protein content percentages among the different age groups were 15.55% (1 d), 15.56% (14 d), 15.95% (28 d), and 15.80 % (42 d), with practically no variations between the group ages. When compared to protein mass, we observed among the same different age groups, results of 6.91 g (1 d), 74.96 g (14 d), 243.5 g (28 d), and 479.4 g (42 d); this explains why no significance was observed for β 1 of protein content percentages but was observed for protein mass. These results were similarly observed in other studies involving chickens, where the correlation between DEXA values and the water content was better than that for the protein ( Mitchell et al., 1997 ; Schallier et al., 2019 ). These observations can be attributed to the high content of water relative to protein within the tissue when calculating their relationship with SLT. The equations developed for SLT and water content showed R 2 values of 66 and 68%, respectively, which is not high, and can cause inaccurate prediction. Therefore, the equations tests using Trial 1 split data demonstrate a reduction in R 2 of 9.22% for SLT content and 14.52% for water content, as well as an increase in MAE value of 0.07 and 0.16%, respectively. Consequently, accurately distinguishing minor variations in fat, SLT, and water content within a narrow range of tissue weights is more difficult ( Schallier et al., 2019 ). Thus, a possible explanation for the results observed for the variables, expressed as percentage of fat, water, SLT, and especially for ash and protein, can be attributed to the smaller range of values (0–100%, as a ratio of total body weight) compared to the wide range of absolute data points ( Swennen et al., 2004 ). Nevertheless, validation with external data is still necessary to ensure model efficiency and no overfitting ( Martinez et al., 2022c ). The method of validation using unavailable data when the model was developed is considered the most rigorous approach ( Harrell, 2015 ), and it has been demonstrated that this method can identify high prediction errors in models previously considered accurate at the development phase ( Inca et al., 2022 ; Martinez et al., 2022c ). In this sense, Trial 2 was conducted to obtain external data and use it for validation of the prediction equations developed in Trial 1. The criteria for validation were the same as those described by Martinez et al. (2022c) , which considered a conservative change for external validation if the ∆R 2 < 25% (∆R 2 = R 2 of prediction model ˗ R 2 of validation), when the R 2 is high enough. In this way, the equations for prediction of broiler FFB weight, fat mass, ash mass, SLT mass, water mass, and protein mass based on DEXA estimates were validated (all regressions showed ∆R 2 = 0, except for fat mass, for which ∆R 2 = 4%) and can be securely used as a tool to measure chicken body composition. An increase in the MAE and RMSE values were observed for all of these variables in validation, although the opposite was observed in the results of the Trial 1 split-data test. The reason for this is that splitting data may cause an overestimate of the efficiency of models, underlining the need for validation using external data ( Harrell, 2015 ; Martinez et al., 2022c ). In relation to fat content, the linear regression for prediction was not validated (∆R 2 = 35%), and the R 2 between the predicted and real observations was considered low for this purpose (R 2 = 51%). Additionally, SLT and water content showed ∆R 2 values of 13 and 12%, respectively. However, the coefficient of determination for both was low, with values of 53 and 56%, respectively. Based on these results, the prediction equations for any variable expressed as a percentage in this study were not validated, indicating that DEXA is only recommended to estimate the absolute weight of body tissues in broilers. Hence, different model structures to predict these tissue percentages may be explored. Nonetheless, other models validated herein can be used to obtain the percentage of the target tissue as ratio of total weight (e.g., ash content = ash mass/weight × 100). It should be noted that, when DEXA estimates of 1-day-old birds from Trial 2 were applied in the prediction equations, the body prediction results were not reliable, especially for the protein mass-negative values that were observed. Because of this, we decided not to use these records for validation. A possible explanation for these results could be one disadvantage of GE Prodigy DEXA, which is the lower weight limit of approximately 250 g ( Johnson et al., 2016 ). Despite this lower limit of DEXA, one aim of this study was to test the accuracy of measurements at this early stage, as well as to obtain a wide range of dependent variable for prediction equations, similar to Salas et al. (2012) . Another disadvantage mentioned in studies that use DEXA to evaluate the body growth rate of chickens, is that a parallel study using the slaughter technique must be carried out to evaluate the feather growth rate ( Gonçalves et al., 2018 ; Alves et al., 2019 ). Due to the inability of DEXA to detect chicken feathers, it is impossible to measure the composition of feathers growth of individuals over time ( Gonçalves et al., 2020 ). However, for the meat industry it would not be a problem, as one of the market's interests is the body composition of the FFB carcass, mainly protein and fat. In the current study, only male broilers were used for Trials 1 and 2. However, the prediction models obtained here can be reliably used for female birds as well, as no trait associated with sex influence on the DEXA measurements has been previously reported in the literature ( Salas et al., 2012 ; Soladoye et al., 2016 ; Gonçalves et al., 2018 ). The strain of bird also may not have an influence on DEXA results, as Swennen et al. (2004) developed the prediction equations with the Cobb strain and validated it with the Ross strain. Therefore, further studies evaluating the effect of different chicken genetic strains on DEXA-traits are suggested, as was observed for swine ( Soladoye et al., 2016 ). In the literature, there are several studies which could benefit from the equations developed here. Castro et al. (2019) , who studied the effect of L-arginine supplementation, used the DEXA scanner (Prodigy, GE Healthcare) method to evaluate the body composition of broiler chickens. When applying the equations of our study, in addition to properly calibrating the data, the authors would discover that increasing L-arginine could result in an increasing protein weight of broilers, which would require statistical procedures to verify significance. Wang et al. (2019) , investigating the efficacy of different levels of phytase and multicarbohydrase on growth performance and bone mineralization in male broilers, could take advantage of our prediction equations to evaluate the effect of body composition on the body weight gain, since the authors used DEXA scanner but did not used the DEXA results of SLT. In a study by Chen et al. (2020) , although the authors did not specify that they used GE Healthcare DEXA scanner and used the small animal module, the use of DEXA results in the prediction equation could contribute to enriching the understanding of the effect of 25-hydroxyvitamin D 3 supplementation for laying hens on the animals’ total body compositions, most notably on total body ash, in addition to bone structural development. For broiler breeders, caution should be exercised when using the prediction equations developed in our study. Studies employing DEXA to estimate body composition ( Aranibar et al., 2020 ; Avila et al., 2023 ), could benefit from these prediction equations for accurate calibration of body composition. Nevertheless, researchers must carefully consider poultry age and body traits before applying prediction equations to avoid extrapolating results. In summary, we demonstrate here that DEXA is a good method of measuring body carcass composition of broiler chickens. Furthermore, DEXA, in contrast to chemical analysis, is a nondestructive and noninvasive method that allows longitudinal studies to be carried out using a limited number of birds, as there is no need to kill the animals. It is also important to emphasize that chemical analysis, although it remains the gold standard, is associated with increased labor, time consumption, multiple reagents and machinery usage, higher costs, and the generation of pollution residuals. In the poultry industry, dual X-ray technology can be used for meat inspection, quality control, and to typify carcasses. Thus, predictive equations, such as those validated in the present study, are strongly recommended for proper calibration of DEXA estimates.

CONCLUSIONS The linear models for chicken body composition prediction validated in this study are possibly specific to particular equipment, software version, and procedures (such as bird positioning and ROI specifications). Nonetheless, the same equation structures can be adapted to new data using different DEXA scanner systems by following the same development and validation processes of this study to acquire new customized prediction models. In Trial 1, predictive linear models were developed for broiler weight, fat mass, ash mass, SLT mass, water mass, protein mass, fat content, SLT content, and water content. Therefore, Trial 2 was conducted for validation, and remarkably, all absolute tissues were validated, indicating a high accuracy prediction of these equations. Thus, the prediction equations validated, can be used to obtain the tissue content as ratio of total weight. The DEXA method is an effective approach to measuring the body composition of broilers by using regression equations to calibrate results. The equations developed in this study could greatly contribute to DEXA applicability in poultry research, genetic improvement, meat production, and the poultry industry.

Two trials were carried out to develop and validate linear regression equations for body composition prediction using Dual-energy X-ray absorptiometry ( DEXA ). In Trial 1, 300 Cobb500 male chickens raised from 1 to 42 d of age were scanned in DEXA to estimate total weight, fat mass, soft lean tissue ( SLT ) mass, bone mineral content ( BMC ), and fat percentage. DEXA estimates were compared to body ash, crude fat, SLT (sum of protein and water) and scale body weight. The dataset was split, with 70% used for prediction equations development and 30% for testing, and the 5k-fold cross-validation analysis was used to optimize the equations. The R 2 , mean absolute error ( MAE ), and root-mean-squared error ( RMSE ) were used as precision and accuracy indicators. A negative correlation ( ρ = ˗0.27) was observed for ash content, while no correlation was observed for protein content ( P > 0.05). Predictive linear equations were developed to assess broiler weight (R 2 = 0.999, MAE = 25.12, RMSE = 38.99), fat mass (R 2 = 0.981, MAE = 13.87, RMSE = 21.28), ash mass (R 2 = 0.956, MAE = 3.98, RMSE = 5.61), SLT mass (R 2 = 0.997, MAE = 35.73, RMSE = 52.45), water mass (R 2 = 0.997, MAE = 29.56, RMSE = 43.94), protein mass (R 2 = 0.989, MAE = 12.94, RMSE = 19.05), fat content (R 2 = 0.855, MAE = 0.81, RMSE = 1.05), SLT content (R 2 = 0.658, MAE = 1.01, RMSE = 1.28), and water content (R 2 = 0.678, MAE = 0.99, RMSE = 1.27). All equations passed the test. In Trial 2, 395 Cobb500 male chickens were raised from 1 to 42 d of age and used for validation of prediction equations. The equations developed for weight, fat mass, ash mass, SLT mass, water mass, and protein mass were validated. In conclusion, DEXA was found to be an effective approach for measuring the body composition of broilers when using predictive equations validated in this study for estimate calibration. Key words

Supplementary materials ACKNOWLEDGMENTS Gustavo A. C. C. de Aguiar is a doctoral researcher funded by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) (no 88887.510984/2020-00). Gabriel R. Braga is a undergraduate student funded by scientific initiation program of Fundação de Amparo à Pesquisa do Estado de Minas Gerais (FAPEMIG) (no 6.36/2021). The chemical analysis was financially supported by Instituto de Ciência e Tecnologia de Ciência Animal (INCT-CA) (no 465377/2014-9), CAPES (no 88887.844747/2023-0), and FAPEMIG (no 6.36/2021). Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) (no 465377/2014-9) for financial support. DISCLOSURES The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in the present study.

CC BY

no

2024-01-16 23:41:57

Poult Sci. 2023 Dec 9; 103(2):103363

oa_package/35/06/PMC10788280.tar.gz

PMC10788281

38226212

Introduction The climate crisis is one of the grand societal challenges. At all levels of governance, nation states must take far-reaching measures to counteract the already dramatic effects of the climate crisis. Austrian social service and healthcare (SSHC)-nonprofit-organizations (NPOs), on which the empirical part of the paper focuses, have to deal to an increasing extent with the detrimental effects of global warming in their fields of services. For example, the climate crisis has led to a significant increase in climate-induced migration, poverty, and social injustice. Not only in Austria, the climate crisis has led to an increase in heatwave days, which have negatively impacted the main beneficiary groups of SSHC-NPOs (children, the elderly and people with health issues). SSHC-NPOs are confronted with increasing ecological risks in their management activities and increasing expectations of resource-providing stakeholders to play a more proactive role in the green transformation processes. Tackling the climate crisis is an essential part of the societal mandate of SSHC-NPOs for a more just and fair society. In line with the United Nations (UN) Agenda 2030 and the 2015 UN Paris Agreement, the European Union (EU) has intensified its coercive pressures towards a green transformation. One of the most recent steps taken by the EU Parliament is the adoption of the Corporate Sustainability Reporting Directive (CSRD) in November 2022. From 2025 onwards, the CSRD will also apply to SSHC-NPOs that are private law corporations in the (legal) form of a (non-profit) limited liability company or a (non-profit) public limited company. They will be subject to the CSRD if they meet two out of the three following criteria: at least 250 employees, a turnover of at least EUR 20 million, or a balance sheet total of at least EUR 40 million [ 1 ]. The CSRD creates a uniform framework for the disclosure of ESG (environmental, social, governance) matters. ESG-reporting requires governance structures for sustainable matters, policies and objectives, as well as a description of the main risks for companies in terms of sustainable activities. CSRD's dual focus on advancing external non-financial reporting and internal strategic and operative sustainability management activities is directed at reducing greenwashing by focusing not only on the talk level, i.e., non-financial reporting but also on the action level, i.e., sustainability strategy and sustainability management control practices (SMCP) [ 2 ]. SMCP represent a set of mechanisms and instruments that organizations use to monitor, measure and regulate their environmental, social and economic impacts. These mechanisms and instruments are crucial for the implementation of sustainable and responsible business practices [ 3 , 2 ]. While there is an extensive body of literature focusing on non-financial reporting and SMCP of for-profit and state-owned enterprises [e.g., [4] , [5] , [6] , [7] , [8] , [9] , [10] ], the research on the integration of environmental matters into management control (MC) practices in SSHC-NPOs is at a very early stage (see section 2 ). Prior research on sustainability matters in SSHC-NPOs has mainly addressed social value creation [e.g., [11] , [12] , [13] , [14] , [15] ] and financial matters [e.g., [16] , [17] , [18] ]. This is not surprising as the social purpose of SSHC-NPOs plays a critical role in upholding organizational legitimacy. Financial matters have gained importance for SSHC-NPOs since the introduction of New Public Management (NPM) reforms in the late 1980s. For social service providing NPOs in corporatist welfare systems such as Austria, Germany, the Netherlands and the Benelux states [ 13 ], NPM reforms resulted in the loss of the privileged positions in welfare state arrangements between the service providing NPOs and the state. Reasons for this are the introduction of quasi-market elements, competitive tendering for short-term service provision contracts, public funding cuts and excessive reporting obligations towards public sector funders [ 13 ]. As a consequence of NPM, the financial risk for providing SSHC-services has been transferred from public funders to the service-providers. In corporatist, liberal and social welfare states alike, SSHC-NPOs have adopted business-style management practices to increase their managerial professionalism, and to become more entrepreneurial [ 19 , 13 , 20 ]. Service-providing NPOs often face the accusation of pursuing financial goals at the expense of their social missions when they adopt corporate management practices. The justification needs to deal with the accusations of mission drift towards legitimacy-creating stakeholders. This has increased significantly over the last ten years, even though no conclusive research (findings) on the performance reducing effects of mission drift has been established yet [ 19 , 20 ]. Compared to the academic body of literature on social and financial sustainability matters of SSHC-NPOs, the call for systematic integration of environmental aspects into the strategies, SMCP and non-financial reporting of SSHC-NPOs is much more recent. In this context, the paper addresses how large Austrian SSHC-NPOs are preparing for the increased integration of environmental matters in preparation for the mandatory implementation of the CSRD. The significance of environmental matters in SSHC-NPOs is unclear, as the organizational identity of SSHC-NPOs is linked to their social mission. This would justify the conclusion that environmental matters have a lower priority. The counter-argument is that SSHC-NPOs should actively contribute to reducing the negative impacts of climate change in line with their societal mandate. The main drivers for this are the increased expectations of stakeholders and the increasing demands on the services of SSHC-NPOs to address the negative impacts of climate change in times when public funding is decreasing and legislative pressures is increasing. So far, NPOs and non-governmental organizations (NGOs) are portrayed by some authors as slow implementers regarding voluntary sustainability reporting [e.g., [21] , [22] , [23] , [24] ]. How that changes with the CSRD is unclear. In order to avoid the accusation of symbolic actions or greenwashing, it is necessary to move beyond mere non-financial reporting by integrating sustainable matters into an NPO's strategy and daily business [ 25 ]. SMCP play a central role in the promotion and realization of sustainable and responsible business practices [ 3 , 2 ]. While there is a growing body of literature on SMCP in general [e.g., [2] , [25] , [26] ], SMCP in NPOs [e.g., [27] , [28] , [29] ] and environmental management practices in NPOs [e.g., [30] , [31] ] have not received sufficient attention. Against this background, the paper addresses the following research question: RQ: How well are Austrian SSHC-NPOs prepared to comply with the CSRD requirements regarding environmental matters? a) How is the environmental dimension already implemented in their strategy and management controls and what steps are taken to improve the integration of the environmental dimension? b) How are Austrian SSHC-NPOs preparing themselves to improve their reporting on environmental matters? The subsequent sections are organized as follows. To answer the RQs, section 2 provides a brief overview of the existing literature on SMCP and environmental sustainability in NPOs as well as the conceptual and theoretical background for our study. Theory-wise, the paper draws on authors who use (neo-)institutionalism. To assess which level of environmental integration SSHC-NPOs are at, the paper uses two dynamic models, namely Shabana et al.'s [ 32 ] three-stage model and Gond et al.'s [ 3 ] three forms of sustainability integration in MC practices. Section 3 focuses on the methodology applied and provides an overview of the 21 Austrian SSHC-NPOs. According to current statistical data, the Austrian NPO sector accounts for 6 % of all employees in Austria. Among the various fields of activities within the Austrian NPO sector, SSHC-NPOs are employers of 54,4 % of the paid workforce in the Austrian NPO sector and generate 58 % of the economic value added of Austrian NPOs [ 33 ]. Among Austrian NPOs, SSHC-NPOs are the ones that are impacted the most by the CSRD, primarily due to their legal forms and size. The empirical results regarding the significance of environmental matters and their integration into strategic and operative management practices are presented in section 4 . This is followed by a discussion of the results in section 5 which presents the answers to the research questions and an evaluation of the development stages of the integration of environmental matters in the focused SSHC-NPOs. The conclusions, limitations and areas for further research are presented in section 6 . With the focus of the paper being on the integration of environmental matters into the sustainability management practices of Austrian SSHC-NPOs, the paper contributes to the least addressed sustainability management dimension. The paper provides empirical insight regarding the assigned significance of environmental matters in comparison to social and financial matters. Furthermore, the paper investigates the extent to which the integration of environmental matters is implemented at the instrumental level, especially in non-financial reporting and SMCP. To the best of our knowledge, this paper is the first, to assess the readiness of SSHC-NPOs for the forthcoming EU CSRD concerning environmental matters in a corporatist welfare state system.

Methodology and sample A total of 22 expert interviews were conducted with 29 top management representatives from 21 Austrian SSHC-NPOs since autumn 2022. The interview guideline contained open-ended questions and covered the following thematic blocks: interviewees' understanding of sustainability, operational and strategic management of environmental matters, the status of SMCP based on Malmi and Brown [ 34 ] and Lueg and Radlach [ 2 ] and the steps towards a non-financial reporting in line with the CSRD requirements. The interviews lasted between 70 and 120 minutes and were conducted in Austria, partly in person and partly online via Zoom. The interviews were recorded, transcribed, and then deductively and inductively coded using the MAXQDA 2022 software. Table 3 provides an overview of the SSHC-NPOs included. The overview of the sample shows that our interview partners are important actors in Austrian SSHC-NPOs. The sample was selected to comprise secular SS-NPOs, church-affiliated SS-NPOs, and HC-NPOs. The HC-NPOs category is comprised of sizable religious HC-NPOs and state-owned hospitals. The latter were included because they derive their legitimacy from their public mission mandates, which prioritize non-profit objectives over for-profit ones. The methodology section inherently possesses certain limitations, particularly in relation to sample size, study period, and the geographical emphasis on Austrian SSHC-NPOs (refer to section 6 for further details). While we believe that the depth of our interviews reached a satisfactory saturation point, it is essential to acknowledge potential criticisms related to the selection of the interviewees, given that the empirical part of the paper is based only on insights from 21 NPOs. Size-wise the included SSHC-NPOs are important actors. The number of employees ranges from 660 to 64,000. Only five SSHC-NPOs in our sample have less than 1,000 employees. All are well above the CSRD threshold of 250 employees [ 1 ].

Results Strategic relevance of sustainability Understanding of sustainability In the interview section focusing on the organizational and personal understanding of sustainability, most of the interview time was dedicated by the interviewees to the environmental dimension of sustainability. Some interviewees provided examples of the efforts they have taken to reduce their personal CO 2 footprint and the role their children play in questioning the behavior of their parents’ generation. When asked about their understanding of sustainability and its relationship to their efforts to achieve the goals of the UN 2030 Agenda, a third of organizations reported their organizational commitment to social and climate justice (I1; I2; I3; I4; I7; I8; I14; I21; I22). The social dimension of sustainability was classified by the interviewees as their organizational core. In SSHC-NPOs, history and organisational values play an important role in presenting their social commitments. In all three types of SSHC-NPOs, the importance of the social dimension was emphasized. This was underlined by their current social purpose, their organizational self-image, and the fact that they are SSHC-NPOs whose primary objective is to provide a wide range of social and healthcare services to the public, rather than to generate profits. The interviewees also stressed that their organizations are committed to providing safe, high-quality client and patient care within their tight financial constraints. In particular, interviewees from HC-NPOs stressed their contribution to regional development. In all SSHC-NPOs, the financial dimension is also very important. The economic efficiency of service provision is given high priority (I2; I3; I5; I6; I7; I8; I9; I11; I13; I15; I16; I20). All top-level executives emphasized that their organizations function as businesses and, therefore, must adhere to business logic. As enterprises, they must generate sufficient revenue to prevent losses. This does not exclude SSHC-NPOs from cross-subsidizing some services internally, but the overall profit and loss statement must cover costs. Unlike private for-profit social and health service providers, profit maximization is not an objective of the sampled SSHC-NPOs. In comparison, the environmental dimension of sustainability has only gained priority in recent years [e.g., [ [58] , [59] ]] and is not yet on equal footing with the social and financial dimensions. Depending on the world-view, this is justified by the increasingly urgent obligation to use the earth's resources in a much more ecologically sustainable manner (I3; I9; I10; I11; I13; I15) and the responsibility for the planet or the responsibility for creation (I2; I3; I4; I13; I15; I16; I17; I18; I19; I20; I22). For example: "Sustainability, of course, means much more than just climate protection. When I think of biodiversity and many other issues, like the protection of the oceans [ ...], but my personal motivation has always been especially all that has to do with climate protection“ (I15). CEOs from Catholic hospitals and church-related social economy organizations stressed the importance of the papal encyclical Laudato Si for improving their environmental orientation. A recurring topic was the trade-offs between financial, environmental and social sustainability. Conflicts between the three dimensions of sustainability are resolved at the expense of the environmental dimension. The interviewees emphasized that the commitment to environmental sustainability must be profitable (I5; I7; I10; I18; I19; I20) which is shown pars pro toto in the following quote: “You have to be able to make a business out of it. We are not profit-oriented, but we cannot operate in areas that are permanently in deficit." (I7). Environmental engagement as part of the sustainability strategy Environmental sustainability is already included into the strategic orientation of nine SSHC-NPOs (I4; I5; I6; I7; I9; I13; I14; I15; I120). With varying prioritization, another eight organizations have already named environmental sustainability as a strategic field (I4; I6; I7; I9; I13; I14; I15; I120). Across the eight SSHC-NPOs, there are significant variations. Environmental and climate protection issues were mentioned by the interviewees as a strategic priority. The importance assigned to environmental and climate change actions varied in the eight SSHC-NPOs from one in four to one in 25 strategic priorities. A prominent strategic objective for a third of the SSHC-NPOs is to be climate neutral in 2030, as the following quote shows: „ [w]e have all collectively resolved, as an organisation itself, to be climate neutral or CO 2 neutral in 2030[ ...]" (I4). In three organizations, environmental sustainability is considered a cross-cutting issue that cannot be reduced to a strategic field of action (I1; I13; I16), as the following quote illustrates: "[s]ustainability plays the highest role in our strategy because it is an underlying inner attitude that is relevant in all decisions." (I13). In nine organizations, environmental sustainability is not yet an integral part of the corporate strategy (I2; I8; I10; I11; I12; I17; I18; I19; I22) but some interviewees stressed the importance of integrating environmental objectives in the near future, as the following quote illustrates: "[t]hat is our weak point, so to speak. I think we have a well-developed attitude on the subject in the hierarchy and an institutional opinion. And where it comes to actually formulating a sustainability strategy across all areas in our organization, I would say we have the headline and table of contents" (I19). Only one interviewee opposed the integration of ecological sustainability into the organizational strategy: “I want to be number one in care, I want to be number one in nursing and care services, I want to be the best employer in the social sector." (I10). With regard to strategic fields of action ( Fig. 1 ) for the environmental transformation, references were made to the reduction of the carbon footprint through an appropriate mobility strategy for employees (I1; I4; I5; I6; I8; I12; I14; I20), more ecological fleet management (I10; I12), the expansion of the use of green energy and green aesthetic gases (I1; I2; I4; I5; I7; I8; I10; I13; I14; I16; I17; I19; I20). Also mentioned were green buildings (I8; I11; I13; I14; I19), the intensification of green purchasing (I2; I4; I7; I8; I12; I14; I17; I19), the existence of appropriate waste management concepts and the already existing approaches to avoid food waste (I1; I4; I6; I8; I12; I13). Another topic was the aspect of "re-use", in which interviewees, especially those who operate hospitals and care facilities, addressed the limits of re-use. Some interviewees stressed that re-use starts with green procurement. Greening the organisation was described by numerous interviewees as a highly challenging topic. Management controls Cultural Controls: Concerning the five MC types by Malmi and Brown [ 34 ], the interviewees assigned the highest relevance to intensifying the environmental orientation in their cultural controls. Interviewees stressed the importance of their organizational culture for being a societal responsible organization, as the following quotes illustrate: " [c]orporate culture [ ....], leading by example and authenticity." (I16); "You don't need so much else if you have the appropriate organizational culture.“ (I11). The majority of interviewees stressed the importance of persons within their organizations who are leading by example when it comes to sustainability matters. Some interviewees linked that to a strong intrinsic motivation and highlighted the importance of their NPO's values. They are crucial for SSHC-NPOs as they guide decision-making, shape organizational culture, and align actions with their mission (I1; I2; I3; I4; I6; I7; I8; I12; I16; I20). Concerning environmental matters a few interviewees mentioned the importance of leading by example, because it establishes norms, inspires and motivates less committed employees, and reinforces desired behaviors (I7; I8; I14; I15). Interviewees are well aware of the essential role of an organization-wide communication strategy regarding environmental matters (I1; I2; I3; I4; I8; I11). Organization-wide awareness building, proactively communicating green transformation projects, and embedding the importance of green transformation in personnel development programs are all steps to an organization-wide roll-out of environmental matters. Planning: Regarding long-term planning, many church-related SS-NPOs and HC-NPOs with a recent EMAS certification stressed their ambitions for a net-zero footprint. About a third of the SSHC-NPOs aim to be climate-neutral by 2030 as the following quote illustrates: „We have all collectively resolved, as an organisation itself, to be climate neutral or CO 2 neutral in 2030 [ ...]" (I4). Several church-related NPOs indicated that if compensation for unavoidable CO 2 emissions is necessary, they intend to invest in their projects in the Global South because their own engagement in the Global South will be of more value than other compensation initiatives. Collecting donations for climate protection projects in the Global South is also a part of today's donor-fundraising of church-related SS-NPOs. The integration of environmental concerns into short-term planning appears to be in its infancy, as the connection between the two was generally discussed in a vague manner. A minority of interviews mentioned internal annual target agreements (I5; I6; I9; I14) with the middle and lower management level but in a more general way. Only in one EMAS-certificated HC-organization, environmental targets are systematically integrated into the yearly performance evaluations of division managers. Cybernetic Controls: The interviews stressed that SSHC-NPOs already have a broad set of financial, performance, and quality indicators. SSHC-NPOs are in an early stage when it comes to environmental KPI, as the following quote illustrates: „That I say I have a key figure that shows me whether I am in the green zone, whether I have to readjust something. Frankly we are not there yet. [...]“ (I11). In one-third of the SSHC-NPOs, selective KPI are implemented for measuring energy efficiency (I3; I8; I9; I15; I17; I20) and waste management-related concerns (I1; I4; I6; I8; I12; I13). Those SSHC-NPOs with a recent environmental certification (EMAS: I9; I15, other eco-certificates: I1; I4; I7; I19) or who have voluntarily implemented a greenhouse gas accounting covering scope 1 and scope 2 (I3; I8; I9; I15; I17; I20) are better prepared in some areas. Two SSHC-NPOs (I14; I17) which use the balanced scorecard as a strategic management tool stressed that they intend to incorporate environmental matters into their balanced scorecards in the future. In two other organizations, a specific environmental cockpit is in place to monitor their environmental performance. The integration of environmental risks into risk management is still in its infancy (I2; I3; I4; I8; I11; I14). Due to the adverse effects of the war in Ukraine on energy delivery security and energy prices, three NPOs had established a risk management task force for energy safety risks at the time of the interviews (I8; I9; I11). Reward and Compensation: Financial incentives for employees were rejected by all interviewed CEOs due to a decremental crowding out effect as expressed in the following quote: "Incentives are generally difficult in NPOs because, as I said, people are very strongly intrinsically motivated. And then often such a reward or a monetary incentive system is even counterproductive." (I2). The interviewees stressed that their organizations offer a variety of voluntary social benefits for environmental sustainability, such as support for environmental tickets for local public transport (I3; I16; I17; I18), electric bicycles and to a much lesser extent electric cars (I2; I7; I10; I12; I16; I17; I19). Activities are also offered to promote health in the workplace (I7; I10; I12; I16; I17). Furthermore, two SSHC-NPOs are offering bonuses to individuals who do not use parking spaces (I15; I16). A fairly common measure is financial support for the purchase of electric bicycles. (Covered) bicycle parking spaces are being created to incentivize bicycle commuting (I2; I3; I12; I16; I18; I20) in addition to providing free bicycle repair services (I2; I3; I12; I16; I18; I19). Six interviewees (I12; I14; I15; I16; I18; I20) emphasized their well-established idea management system and their culture of appreciation. Rewards range from non-monetary benefits to a budget for implementing award-winning measures. The focus on environmental matters in idea management activities was generally vague, with a few exceptions related to food waste reduction or the use of green anesthesia. Administrative Controls: The CEO or top management board is ultimately responsible for organizational governance for sustainability matters (I2; I5; I6; I7; I9; I11; I12; I13; I14; I15; I16; I18; I19; I20). Table 4 displays organizational governance structures in hierarchical lower management levels. Two organizations are in the process of establishing a climate protection officer (I4; I7), as the following quote illustrates: “[a]nd I say as it looks at the moment, we will have someone as sustainability or climate protection manager also in the future, who will also have the responsibility for sustainability reporting." (I6). Regarding policies and procedures, a few SSHC-NPOs mentioned that they have anti-corruption, anti-fraud guidelines, information technology and health and safety policies, guidelines for ethical investments as well as regional and green procurement and green building. Environmental policies and procedures were the least addressed ones. Table 5 provides an overview of the key findings regarding the five MC types of Malmi and Brown [ 34 ]. Non-financial reporting As Fig. 2 illustrates, and as discussed in the section on cultural controls, employees play a critical role in communicating environmental issues ( Fig. 2 ). Followed by clients and patients, funders, suppliers, municipalities, banks, volunteers and ministries. For internal stakeholders, environmental matters are communicated through newsletters (I1; I2; I8; I9; I11; I16; I18), the intranet (I4; I6; I9), information mails (I7) and online platforms (I12). Environmental matters are also part of the information provided by the top management to the lower management levels regarding organizational strategic priorities (I4; I11; I18). External stakeholder communication on environmental matters include website appearances (I6; I8; I9; I16), annual reports (I6; I9), poster campaigns (I12), and social media channels (I7; I12; I16). The integration of environmental criteria into the requirements of public funders is still very limited, with the exception of public funding earmarked for green building initiatives, programs for supporting a green energy transformation, or energy saving programs. Gaining access to the EU-sustainable finance fund was not addressed by the interviewees. None of the SSHC-NPOs currently meet the advanced ESG-reporting requirements for environmental issues. The six SSHC-NPOs with existing greenhouse gas accounting (I3; I8; I9; I15; I17; I20) are in a slightly better starting position but even for these organizations, a fully-fledged ESG-reporting on environmental matters will be a challenge. The roadmaps presented in the interviews by those 16 SSHC-NPOs that have to publish an ESG-report (see Table 4 ) are very ambitious, considering the present status of environmental KPI. There is still a lot to do, as the quote illustrates. "[I]t's a relative monster project [...] you need an insane amount of numbers, data, facts [...]. Not only energy figures but really quite a lot of other things as well [...]." (I1) None of the interviewees made reporting on environmental matters their top priority, as the quote shows: "[ ...] I don't think it is the highest priority, even if I am honest, to produce the report." (I4) . Other interviewees stressed that the organisation will be a lazy follower in implementing the CSRD.

Discussion The awareness of the general relevance of the environmental dimension of sustainability is high within the focused SSHC-NPOs. Regarding our overall research question „How well are Austrian SSHC-NPOs prepared to comply with the CSRD requirements regarding environmental matters?“, the findings of our study show that all organizations still have a huge amount of work ahead regarding their environmental management (strategy, SMCP and ESG-reporting). Austrian SSHC-NPOs take a pragmatic approach towards environmental matters, incorporating them as much as economically feasible. If there are trade-offs between the environmental dimension and the social purpose, they are solved at the expense of the former. The way the interviewees referred to the trade-offs between the ecological dimension and the other two sustainability dimensions suggests a compartmentalized, non-integrated approach. Since NPM, SSHC-NPOs have already a tradition of choosing a compartmentalization approach. This strategic response is usually chosen to become more business-like, while not comprising their social mission identity [ 13 , 20 ]. Regarding sub-question one „How is the environmental dimension already implemented in their strategy and management controls and what steps are taken to improve the integration of the environmental dimension?“ , the interviewed SSHC-NPOs are in an early stage of systematic integration of environmental sustainability into their strategy and SMCP. The interviewees stressed the crucial role of cultural controls. The analyzed SSHC-NPOs are in an early instrumental stage regarding planning, environmental cybernetic controls and administrative controls. While steps are taken to improve the inclusion of selective environmental commitments into long-term planning, the integration of environmental matters into short-term planning and therefore steering the operational practices is nearly non-existent. Linking environmental matters to financial rewards was opposed because of a decremental crowding-out of intrinsic motivation. Recalling the idea of MC as a package [ 34 ], the current implementation status of SMCP is far too selective to do justice to the idea of interlinking the various SMCP for the implementation of environmental matters. While the international literature review by Lueg and Radlach [ 2 ] showed cybernetic controls to be the first choice followed by administrative SMCP, our sample of SSHC-NPOs prioritize cultural controls. This is in line with their self-image as social value-driven organizations and the significance of the social dimension of sustainable matters [ 13 ]. The deliberate and selective adoption of SMCP while neglecting others becomes evident, highlighting that SSHC-NPOs are in a very early stage of organizational change process for integrating environmental matters. Concerning sub-question two „How are Austrian SSHC-NPOs preparing themselves to improve their reporting on environmental matters?“ , the findings show that those 16 SSHC-NPOs, which must publish an ESG-report in 2025, are in the early stages of their journey towards ESG-reporting. That aligns with prior literature which portrays NPOs as having a slow and highly selective approach to implementing non-financial reporting [ 21 , 22 , 23 , 24 ]. In line with that none of the interviewees identified ESG-reporting and therefore improving the reporting on environmental matters as their top priority. The findings also show that employees are the top communication addressees for environmental matters. The relaxed approach to environmental reporting by the 16 SSHC-NPOs, which must comply with CSRD in 2025, corresponds with the pragmatic approach towards the environmental dimension. Based on an evaluation of the findings from the perspective of Shabana et al.'s [ 32 ] three-stage model, the coercive pressure to improve non-financial reporting on environmental matters was still very weak in the perception of our interviewees. This was the case regardless of whether they have to comply with the CSRD in 2025 or are voluntarily establishing non-financial reporting on environmental issues (e.g., in the form of greenhouse gas accounting, carbon footprints). The roadmap of what needs to be done to comply with the CSRD requirements is quite ambitious, given the short time frame before the publication of the first ESG-report in 2025. More in line with the second stage of Shabana and co-authors [ 32 ] is, that nine SSHC-NPOs have already embedded environmental aspects in their strategy and, with one exception, the remaining SSHC-NPOs plan to improve the integration of environmental matters into their strategy. Additionally, the interviewees stressed the great importance of cultural controls in advancing their SMCP. That behavior is directed toward making the environmental dimension an essential part of the organizational values. As the CSRD was only based in 2022, mimetic isomorphism is the leased addressed one. Recalling Gond et al. [ 3 ], SSHC-NPOs heavily rely on a cognitive integration. Therefore, they put not the technical integration of MC and sustainability management activities first, as suggested by Gond et al. [ 3 ]. The integration of environmental KPI is still in its early stages. Regarding the organisational integration of environmental matters, first steps are taken for adapting governance structures. Environmental matters are the least addresses once in the sustainable management policies and procedures. The importance attached to cognitive integration [ 3 ] and cultural controls [ 34 ] by the analyzed SSHC-NPOs allows two interpretations. First, the unfriendly one: Due to the rudimentary stage of integrating environmental matters into their strategy and the deplorable status of environmental KPI, one could conclude that most of the SSHC-NPOs have moved to the next talk level by emphasizing the need to first get the organizational strategy right and by acknowledging that green transformation requires a long organizational change process. Fitting with the next talk level is also the relaxed approach of the CSRD-affected SSHC-NPOs towards improving their non-financial reporting till 2025. In line with that is also the highly pragmatic approach towards advancing the environmental dimension only if that does not come at the costs of social and financial objectives. The more friendly interpretation is that the SSHC-NPOs do the right thing by first aiming to integrate environmental matters into their organizational values and culture due to the fact that they are mission-driven organizations. The managers of the SSHC-NPOs stressed many times that the commitment towards the social dimension of sustainability or their societal mandates are the cornerstone for their existence, belief systems and organizational identity. That is a positive sign in the debate about mission-drifts of social service providing NPOs due to NPM reforms [ 13 , 20 ].

Conclusions The paper contributes to the academic debate in the following ways: Empirically, by focusing on the status quo and the envisaged steps of SSHC-NPOs towards the CSRD, the paper addresses a timely and highly under-researched field. Austrian SSHC-NPOs have a wait-and-see attitude toward the coercive pressures of advancing their reporting on environmental matters. Furthermore, our findings indicate that there are no signs that environmental matters will be put on equal footing with the other two sustainability dimensions in the near future. Social and financial matters are prioritized at the cost of environmental matters. A highly pragmatic approach prevails regarding environmental matters. The study examines the integration level to which environmental aspects are considered at the instrumental level, particularly in non-financial reporting and SMCP. In particular, this study represents the first attempt to assess the preparation of SSHC-NPOs within a corporatist welfare state for the forthcoming EU CSRD with a focus on environmental concerns. From a theoretical point of view , the paper indicates, that “doing things differently” also challenges the three stages by Shabana et al. [ 32 ] and Gond et al. [ 3 ]. The analyzed SSHC-NPOs have a clear preference for integrating the environmental dimension into their organizational culture while being far less willing to invest in elaborate environmental KPI. That also impacts the financial reporting side on environmental matters. The in-betweenness had to be expected, as the phases are ideal types, but the particular blend offered by SSHC-NPOs tentatively indicates that NPOs values or social mission orientation need to be taken into account as a significant context factor. Concerning practical implications , the implementation of environmental MC and improving the non-financial reporting in SSHC-NPOs is crucial for several reasons. SSHC-NPOs heavily rely on public support and trust, necessitating the enhancement of their public image as responsible environmental stewards. By adopting environmental MC and communicating their environmental commitment, SSHC-NPOs could show their dedication to responsible practices which also plays a major role in employer branding. Implementing environmental MC would allow them to assess and monitor their activities, identify areas of concern and opportunities for cost savings, as well as implement strategies to reduce their negative impact on the environment. Recalling the Austrian government's commitment to Agenda 2030, the greening of their procurement practices will most likely come earlier than expected for SSHC-NPOs. Finance institutions will also increase pressure for sustainable finance in the near future. This study has several limitations, especially with regard to sample size, study period, and the regional focus on Austrian SSHC-NPOs. The focus on large SSHC-NPOs allows us to gain insights into the adoption of environmental matters in the most important fields of the Austrian NPOs with respect number of employees, the economic value creation and the increase of employees in the past decade [ 33 ]. The drawback is that the results are not transferable to other fields of the NPO sector, such as advocacy organizations, cultural NPOs, or NPOs where the main actors are volunteers. Another caveat is that Austria is a corporatist welfare state and therefore the findings might be different from liberal welfare states or welfare states with a social-democratic tradition. Although we had the impression that the interview richness had reached a reasonable saturation, the selection of the interview partners can be criticized because our paper is based on 21 NPOs. We have so far done deductive and inductive coding and have not yet integrated a documentary analysis of all the publications of the 21 SSHC-NPOs addressing the environmental dimension. A next step for further research could be to look at how another corporatist welfare states, as the SSHC-NPOs are under the same green transformation pressures.

The climate crisis requires the systematic integration of environmental matters into the management practices of companies across all sectors. The 2022 Corporate Sustainability Reporting Directive (CSRD) has created a need in many large social service and healthcare non-profits (SSHC–NPOs) within the European Union to extend the integration of environmental matters by 2025. While there is plenty of research on environment management practices of large for-profit enterprises, the research on environmental matters in SSHC-NPOs, which gain their legitimacy from social value creation, has been neglected. This study examines how large Austrian SSHC-NPOs are preparing for the environmental requirements set by the CSRD. The integration of environmental considerations into their core strategy, sustainable management control practices, and non-financial reporting poses a significant challenge. To evaluate the status of integration of environmental matters, the paper uses two sequential stage models based on institutional theory. The study is based on data from interviews with 21 Austrian SSHC-NPOs. The findings reveal that the integration of environmental matters is at an early stage, driven by a pragmatic approach with a strong emphasis on social and financial concerns. Cultural controls take precedence in management control practices, while administrative and cybernetic controls lag behind. Environmental reporting does not meet CSRD requirements, and the studied SSHC-NPOs aim for minimal compliance only, when CSRD comes into force in 2025. Additionally, it highlights that these organizations do not conform to the sequential stages proposed by two institutionalist stage models, emphasizing the role of the SSHC sector's context in shaping their behavior and practices. Keywords

Conceptual and theoretical background Conceptual background There are several approaches which highlight the different MC types within management control systems (MCS) [e.g., [34] , [35] , [36] ]. The most holistic one is the approach by Malmi and Brown [ 37 , 38 ], which places a special emphasis on the package idea of MC instruments and a wide range of informal (e.g., values, code of ethics) as well as formal control (e.g., key performance indicators (KPI), long and short term planning, bonus plans) [ 34 , Table 1 ]. The package idea consists of five types of controls: cultural, planning, cybernetic, reward and compensation, and administrative. These controls must be integrated holistically to be effective as a package [ 34 ]. This framework is the one which has been often adopted when discussing the interplay between sustainability reporting and MC and the specific design of sustainability management control systems (SMCS) [ [2] , [4] , [10] , [39] , Table 1 ]. With the specific focus on organizational values within the cultural controls, the Malmi and Brown [ 34 ] framework is particularly suitable for SSHC-NPOs, where the social value orientations are of greater importance than in for-profit enterprises. The main objective of for-profit organizations is to generate profits, and social responsibility activities are often part of a strategy to reduce reputational risk and therefore have a serving function for financial value creation. As outlined above, social purpose is paramount to the organizational legitimacy of SSHC-NPOs with their mandate to create social value for society as well as for their clients and patients. Regarding the academic debate on SMCP in NPOs, Daub et al. [ 29 ] concluded that there is hardly any research on the integration of corporate social responsibility or corporate sustainability into the management systems and processes of NPOs. Since 2014, only very few papers have been published which refer explicitly to SMCP and environmental sustainability in NPOs. If one takes a look at the existing literature on SMCP and environmental sustainability in NPOs and NGOs, it becomes apparent that the focus of the papers has been on selective SMCP in NPOs and NGOs [ [27] , [40] , [41] , 42 , 43 , 44 , 45 , 46 , 47 , 15 , 48 ]. The main topics of the research on sustainability management in general are: • Partnerships and collaborations [ 27 , 40 , 42 , 43 , 45 , 46 , 48 ]: These articles show, how NPOs and NGOs promote sustainability through strategic partnerships and collaborations. By collaborating with different entities, resources are pooled, impact is amplified and innovation is encouraged. These alliances strengthen advocacy, mitigate risk and contribute to long-term viability by diversifying funding sources and building organizational capacity. • Organizational structures and culture [ 46 , 47 ]: The findings of these studies indicate that the organizational structure of an NGO should be carefully designed in order to achieve effectiveness and sustainability [ 46 , 47 ]. Furthermore, organizational culture significantly impacts sustainability [ 47 ]. In addition, a high employee retention rate and a stable workforce promote sustainability [ 44 ]. • Strategic planning [ 27 , 46 , 15 ]: Strategic planning guides NPOs to efficiently allocate resources, foster long-term impact, and align actions with sustainable development goals [ 27 , 46 , 15 ]. Weerawardena et al. [ 15 ] emphasized the relationship between sustainability and strategy and claimed that the creation of a sustainable organisation is closely linked to the strategic direction of the NGO. Furthermore, there is hardly any academic research on environmental sustainability in NPOs and NGOs. Previous research has only dealt with the topics of: • Environmental orientation in general [ [49] , [50] , [51] , [52] , [53] ]: These articles deal with the topic of environmental sustainability in NGOs in general. • Environmental management systems [ 30 , 31 ]: RT White et al. [ 31 ] examined the implementation of the eco-management and audit scheme (EMAS) in an NPO. It is shown that by implementing EMAS, the organisation was able to identify operational improvements and make significant efforts to improve its environmental performance, reducing the carbon footprint per year. Chakraborty & Roy [ 30 ] investigated the implementation of environmental accounting (EMA), which provides combined monetary and physical information aimed at sustainable development, pollution prevention, and cleaner production methods. If one evaluates these results from the perspective of Malmi and Brown's [ 34 ] package concept, it quickly becomes clear that there is a lack of a holistic view of SMCP. Only individual forms of controls are addressed. Furthermore, to the best of our knowledge, this research is the first, to assess the readiness of SSHC-NPOs for the EU CSRD concerning environmental matters, or how SSHC-NPOs are preparing themselves for the integration of environmental matters into their strategy and non-financial reporting. Theoretical background Research gaps exist in terms of evaluating the readiness of SSHC-NPOs to incorporate environmental concerns into MC practices and non-financial reporting. To address this gap, the paper employs institutional theory, in particular the three-stage model of Shabana et al. [ 32 ], which identifies a dominant isomorphism for each stage that shapes the non-financial reporting of organizations. Shabana's framework serves as a valuable tool to assess the progress of SSHC-NPOs in their adoption of non-financial reporting [see e.g., [54] , [55] ] but is also suitable for integrating sustainability matters into MC, as the implementation status of SMCP can also be categorized according to the three levels. For extending the evaluation of the progress SSHC-NPOs made in integrating environmental matters into their SMCP, this paper additionally draws on the three dimensions of integration of sustainable matters by Gond et al. [ 3 ]. By focusing on the three levels of integration, Gond et al. [ 3 ] complement the three-level model of Shabana et al. [ 32 ] by focusing on the extent of socio-technical levels of integration of SMCP and MC practices [e.g., [56] , [57] ], or as Gond et al. [ 3 , p. 2012] put it: “the thick interface between both types of management controls”. Both classifications distinguish three levels of integration. Table 2 provides an overview of the linkage of the three-phase model of Shabana et al. [ 32 ] with Gond et al. [ 3 ]. In the first stage of Shabana et al. [ 32 ] coercive isomorphism, i.e., new regulatory requirements or societal expectations are the dominant pressure. Coercive isomorphism leads to defensive and selective non-financial reporting. Firms fail to meet stakeholder expectations and at most, narrow the stakeholder expectation gap [ 32 ]. Transferring that reasoning onto SMCP, one could argue that only very selective SMCP are implemented during this stage. According to Gond et al. [ 3 ] technical integration is characterized by the fact that only selective environmental MCS are implemented. Although there might be several environmental KPI in place, there is no systematic approach to steering and monitoring environmental matters. A broader set of MC practices is in place which does not focus on SMCP [ 3 ]. As a consequence, two parallel systems exist. Sustainable matters are primarily dealt with outside the MC function [ 3 ]. In the second phase, normative isomorphism drives non-financial reporting and SMCP resulting in highly ambitious approaches regarding sustainability matters. All three dimensions of sustainability are essential to the mission, strategy and managerial practices of SSHC-NPOs. Normative isomorphism leads to the broad application of SMCP [ 32 ]. At this stage, the development of shared understanding requires cognitive integration [ 3 ]. Opportunities for discussion are created between people who bring different ways of thinking, mentalities, and practical viewpoints regarding environmental matters. The outcome of these processes should be reflected in a shared understanding of sustainability objectives and strategy [ 3 ]. In the third phase, mimetic isomorphism dominates. Approaches of other organizations are imitated if they are perceived to be successful, and the benefits of sustainable management begin to outweigh the costs [ 32 ]. Organizations take a pragmatic stance toward the integration of environmental matters into their strategy and SMCP. According to Gond et al. [ 3 ], appropriate governance and organizational structures are fully established and organizational actors with varying levels of professional socialization and from different hierarchical level differences have developed joint SMCP and non-financial reporting. The integration of environmental matters into MC and strategy should not only be seen as something an organisation possesses, but also as something that people do [ 3 ]. Data availability statement The data has not been deposited into a public available repository. The authors do not have permission to share data. Ethics approval statement All participants/patients provided informed consent to participate in the study. All participants/patients provided informed consent for the publication of their anonymized case details and images. CRediT authorship contribution statement Philumena Bauer: Writing – review & editing, Writing – original draft, Visualization, Software, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Dorothea Greiling: Writing – review & editing, Writing – original draft, Resources, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

CC BY

no

2024-01-16 23:41:57

Heliyon. 2023 Dec 24; 10(1):e23767

oa_package/b0/a0/PMC10788281.tar.gz

PMC10788286

38226207

Introduction Processed excipients are essential components of pharmaceutical and medical products. They serve various functions, such as ensuring stability, improving drug delivery, and enhancing patient compliance. Excipients are inactive substances that are combined with active pharmaceutical ingredients (APIs) to create a final dosage form. Processed excipients are excipients that undergo specific treatments and processing to meet the requirements of a particular drug formulation. They can be natural or synthetic materials and are added to pharmaceutical formulations for their various functional properties. Advances in pharmaceutical technology have led to the development of novel processed excipients. These include nanoparticles, liposomes, and other specialized carriers designed to enhance drug delivery and therapeutic outcomes. The ocular system converts visual information into electrical signals, making it one of the most remarkable sensory systems in the human body [ 1 ]. Since the eye is a highly isolated organ with formidable protection, it is a challenging target for the delivery of drugs. The blood-retina barricade and the blood-aqueous humor barricade, which the ocular vascular barrier system severely limits systemic ophthalmic drug delivery. As a result, even though the cornea and conjunctiva constitute significant epithelial barriers, ocular drugs for treating the conjunctiva, cornea, and anterior chamber are given topically [ 1 ], further illustrating the basic anatomy of the eye in Fig. 1 . Ophthalmic drug therapies were limited to topical treatments, injections around or inside the eye, or systemic delivery. Although ophthalmic drops are the most accessible way to deliver ocular remedies, achieving an adequate level of therapeutic substance in the targeted region for an extended period is challenging [ 2 ]. Conventional medicines are rapidly removed from ocular surfaces by baseline and reflexive lachrymation, blinking, and drainage, even before the medication contacts the optic nerve or tissues in adequate concentrations. As a result, the primary limitations in ocular medication delivery via local application include low mucosal tissue permeability and limited mucosal residence time [ 3 ]. Due to the demand for advancements in drug conveyance strategies for ocular assertions, there is a mandate to develop novel and enhanced techniques. Among the different drug delivery techniques that improve ocular bioavailability are those that provide a more extended residence period on the cornea and conjunctiva [ 2 ]. One way of achieving this is by adding new additions to existing works of art or inventing novel medication delivery systems. Drugs have a better chance of operating locally on those membranes or penetrating deeper ocular tissues because of the extended drug residence time, which allows them to achieve their target effectively. The first effective attempts at broadening the use of ocular therapies were made in the 1980s. The studies resulted in the discovery of mucoadhesive polymers that can adhere to the eye's surface, increasing the viscosity of eye drops and slowing their drainage rate [ 3 ]. The conventional mucoadhesive polymers were restricted to forming weak reversible connections with the mucus layer covering ocular surfaces during its initial phase. A new class of mucoadhesive polymers emerged due to later developments, which can form covalent bonds with mucus glycoproteins, predominantly through disulfide linkages [ 4 ]. Nanomedicine acquired significant attention in the late 1990s, prompting the development of innovative nanocarrier formulations [ 5 ]. Their small dimensions facilitate effective passage through protective layers in the eye, thereby improving the amount of drug that can be absorbed and reducing the need for frequent dosing. Through the containment of medications in tiny carriers, they safeguard sensitive eye tissues from harm, decreasing harmful reactions and optimizing the effectiveness of treatment [ 6 ]. Furthermore, their ability to circumvent the barrier between the bloodstream and the eye enables the direct administration of treatments to the specific location in need, presenting a hopeful resolution for complex eye ailments [ 7 ]. Various empirical investigations have shown that high-dosage forms such as contact lenses and eye inserts can adhere to the eye for long periods, ranging from hours to days. This insight has led to the development of multiple drug delivery systems that aim to prolong medication retention in the precorneal region. These systems have effectively enhanced bioavailability by extending the drug contact time with the eye after topical administration. As a result, many commercially available medicines have become possible through these systems [ 4 ]. Ocular drug delivery presents unique challenges due to the complex anatomy and physiology of the eye. To improve drug delivery to the eye, various processed excipients and drug delivery systems have been developed. These excipients can enhance drug solubility, improve bioavailability, extend drug retention time, and provide controlled release. Processed excipients and drug delivery systems are designed to address specific challenges in ocular drug delivery, such as limited drug absorption, rapid clearance, and the need for prolonged drug release. The choice of excipient or delivery system depends on the drug's characteristics and the desired therapeutic outcome. Additionally, regulatory approval and safety considerations are crucial when developing ocular drug delivery systems. In the past few years, there has been extensive utilization of processed excipients in the advancement of sophisticated techniques for administering medications to the eyes, aimed at addressing a range of ocular ailments. A simple search on “PubMed” for the keywords “Excipients” and “Ocular” using “AND” as a Boolean operator resulted in 286 articles from 2010 to 2022. Furthermore, the keywords “Nanomedicine” and “Ocular” using “AND” as a Boolean operator resulted in 437 articles from 2010 to 2022. This indicates that there are ample opportunities for further advancements in creating efficient methods for delivering drugs to the eyes, employing a diverse array of processed excipients.

Isolated/enucleated organ/organotypic methods Optometric and spectroscopic methods are commonly employed in protocols for evaluating ocular effects to quantitatively assess changes in the isolated cornea caused by a test substance. These methods are usually followed by histological analysis to provide further insights. Corneal imperviousness is an important In vivo terminus for assessing the things on the cornea, although the data obtained is often based on observational and subjective assessments. Corneal opacity indicates various changes in the cornea, such as protein denaturation, bump, and vacuolization, then hurt to the epithelium, and corneal stroma [ 188 ].

Conclusion Ocular preparatory transmission is a complex and continuing experience that necessitates the development of effective and targeted strategies. Although several methods are available, such as topical drops, injections, and systemic delivery, sustaining effective medication concentration for a prolonged time creates challenges. Significant advances in materials engineering and pharmacological delivery strategies are being pursued to address these constraints. Innovative drug conveyance arrangements intended for ocular preparation conveyance have emerged as viable options, intending to increase bioavailability, improve treatment effects, and overcome obstacles associated with conventional therapy. Nanoparticles, hydrogels, microparticles, and implantable devices are examples of materials that provide prolonged release, controlled delivery, and targeted dispersion within ocular tissues. As a vital part of ocular formulations, excipients influence drug conveyance qualities. Researchers are looking for new excipients that are changing current ones to improve medication release, solubility, bioavailability, and residence time inside ocular tissues. To guarantee patient safety and limit unwanted effects, these excipients' safety profiles and possible toxicology must be thoroughly investigated. Despite tremendous advances, problems such as enhancing eye penetration, effective drug release kinetics, biocompatibility, and system stability persist. The combination of nanotechnology and targeted drug delivery systems shows promise, taking advantage of the distinctive characteristics of nanoparticles for precise and streamlined drug release. The future of ocular medication administration depends on the continuing research of innovative methods, excipient optimization, and nanotechnology breakthroughs. Challenges associated with ocular preparation conveyance can be addressed through cross-disciplinary collaboration and thorough research, increasing handling outcomes for affected roles with ocular diseases.

Ocular drug delivery presents a unique set of challenges owing to the complex anatomy and physiology of the eye. Processed excipients have emerged as crucial components in overcoming these challenges and improving the efficacy and safety of ocular drug delivery systems. This comprehensive overview examines the opportunities that processed excipients offer in enhancing drug delivery to the eye. By analyzing the current landscape, this review highlights the successful applications of processed excipients, such as micro- and nano-formulations, sustained-release systems, and targeted delivery strategies. Furthermore, this article delves into the bottlenecks that have impeded the widespread adoption of these excipients, including formulation stability, biocompatibility, regulatory constraints, and cost-effectiveness. Through a critical evaluation of existing research and industry practices, this review aims to provide insights into the potential avenues for innovation and development in ocular drug delivery, with a focus on addressing the existing challenges associated with processed excipients. This synthesis contributes to a deeper understanding of the promising role of processed excipients in improving ocular drug delivery systems and encourages further research and development in this rapidly evolving field. Keywords

Effects of physiological factors on ocular drug delivery approaches The designing and development of drug delivery systems targeting ocular tissues poses a significant challenge for scientists. The eye is alienated into frontal and later chambers, and each layer of ocular tissue presents structural variations that hinder drug absorption regardless of the administration route (i.e., topical, systemic, periocular) [ 8 ]. Due to various inherent barriers specific to the optic system anatomy and physiology, ophthalmic drug delivery offers a unique encumbrance to systemic drug delivery [ 9 ]. These barriers present significant hurdles for drug delivery scientists. Depending on the route of administration (e.g., local, enteral, parenteral, etc.), as shown in Table 1 , specific biological barriers exist, which serve to shelter the eye from conceivably detrimental rudiments [ 10 ]. Overcoming these barriers requires careful consideration and innovative approaches in sending remedies to the eye [ 10 ]. The treatments for anterior eye conditions usually involve the use of topical administration, which is primarily accomplished through ophthalmic drops. Several factors could improve the efficacy of locally applied medications. Precorneal factors, i.e., tear film dynamics, blinking, tear drainage, and induced lacrimation, can significantly impact the bio-readiness of topically directed formulations [ 11 ]. The tear crust acts as a protective barrier and quickly removes administered solutions due to its high turnover rate. Due to limited contact time with absorptive membranes, the amount of the applied dose that reaches intraocular tissues is relatively low [ 8 , 12 ]. The cornea, conjunctiva, and sclera, which form the ocular surface, present additional barriers to drug permeation. The cornea entails different coats, including the epithelium, stroma, and endothelium, separately posing encounters for drug penetration due to their specific properties and structures [ 12 ]. The attendance of lipoidal characteristics and skintight junctional complexes in the corneal surface layer poses a barricade to the pervasion of aquatic-solvable remedies. The stroma's well-hydrated structure presents a limitation for non-polar drugs. The endothelium controls the movement of fluids between the stroma and the aqueous humor. Furthermore, the blood vessels and tight junctions in the conjunctiva cause medicines to be drained into the bloodstream, which reduces their effectiveness in treating eye conditions [ 25 ]. The sclera, contiguous with the cornea, has permeability comparable to the corneal stroma. During ocular drug delivery through systemic administration (via parenteral routes), drugs face the challenge of crossing intraocular barriers [ 1 ]. The blood-eye barricade is assembled by the tight connections between the cells in the endothelium of the iris/ciliary venules and the non-pigmented ciliary epithelium. These connections act as a protective barrier that prevents solutes, including the aqueous humor, from entering the intraocular space [ 26 ]. The blood-retinal barricade, which entails retinal capillary endothelial cells and retinal pigment epithelium (RPE) cells, acts as a protective barrier that confines the admittance of drugs to the later ocular region [ 27 ]. Situated between the neurosensory retina and the choroid, the retinal pigment epithelium (RPE) executes a fundamental character in the sustenance of the visual system. It facilitates crucial functions within the visual pathway by preferentially transporting molecules between the photoreceptors and the choriocapillaris. The intercellular permeation is limited by the skintight junctions in the RPE that restrict the passage of molecules between adjacent cells [ 27 ]. Although drugs can enter the choroid quickly through oral or intravenous dosing, the superficial blood-retinal barricade confines further admittance into the retina. Overcoming the blood-retinal barricade has been explored through nanotechnology, with studies showing the successful passage of inorganic or metal oxide-based nanoparticles and gene delivery systems [ 28 ]. Functionalized nanoparticles, when delivered intravenously, have been shown to target choroidal neovascularization lesions specifically. While specific drugs have demonstrated distribution in ocular tissues following intravenous administration, the utilization of systemic administration for treating ocular disorders is restricted due to concerns related to toxicity and delivery. Researchers have explored oral administration as a non-disruptive and patient-preferred method for administering drugs to the eye [ 29 , 30 ]. This method can be used alone or concurrently with topical delivery methods. Although topical delivery alone may not attain therapeutic concentrations in the retinal region, oral administration presents potential benefits for handling enduring retinal ailments when equated to parenteral routes [ 31 ]. The operation of oral sending for ocular drug administration encounters challenges due to restricted access to targeted ocular tissues. Generally, the higher dosages are habitually obligatory to achieve the desired beneficial worth, potentially consequential in non-localized side effects [ 28 ]. Safety and toxicity considerations are paramount when seeking a response to therapy in the eye through oral administration [ 28 ]. Some drugs, such as oral carbonic anhydrase enzyme blockers used in glaucoma therapy, have been discontinued due to their circulatory venomousness [ 32 ]. The use of the insolence pathway for delivering medicines to the eyes is not widely practiced, and the research on drugs for this route is limited [ 33 ]. The periocular and intravitreal administrations are alternative routes to overcome the limitations of mucosal dosing and systemic dosing methods to achieve optimal drug concentrations in the retinal region. These routes offer targeted delivery options for effectively treating ocular conditions. These routes are particularly beneficial in avoiding systemic side effects and catering to geriatric patients [ 34 ]. The periocular route comprises techniques such as Sub-Tenon injections, Retrobulbar space injections, and Periocular injections, which are relatively less intrusive than intravitreal injections [ 34 ]. Subconjunctival injections bypass the rate-limiting conjunctival epithelial barrier and the cornea-conjunctiva barrier, allowing drug permeation over the sclera and the choroid to spread the neural retina and photoreceptor cells [ 16 ]. Intravitreal shots provide the lead of direct drug conveyance into the vitreous humor. Drug distribution within the vitreous can be uneven, with smaller molecules diffusing more rapidly than larger ones. The vitreous humor obliges as a barricade for the conveyance of retinal inheritable influence remedy, as it can impede drug mobility [ 1 , 35 ]. In general, the interactions between negatively charged glycosaminoglycans and cationic complexes can further hinder the movement of drugs. In addition to the vitreous, the inner restrictive crust performs as a barricade that confines the conveyance of medications from the vitreous to the retina. The movement of therapeutics from the vitreous to the distal retinal layers and choroid is a multifaceted process, mainly allied with the RPE [ 36 ]. Fig. 2 further summarizes the various routes of the ocular drug delivery system. It is crucial to determine the therapeutic effectiveness of a drug by understanding its elimination half-life in the vitreous. The elimination process can take different routes, including diffusion into the aqueous humor and passage through the blood-retinal barrier. The half-life of compounds in vitreous humor is extended through their hydrophilicity and higher molecular weight [ 37 ]. Various transporters play a critical role in facilitating the transport of nutrients across biological membranes, and they can be used to improve the delivery of ocular medications. Various efflux transporters, i.e., P-gp, MRP, and BCRP lower drug bioavailability by expelling drugs from the cell membrane. Influx transporters are essential in promoting foreign substances and vital nutrients entry into the ocular tissues. The Solute Carrier Family 1 (SLC1), Solute Carrier Family 6 (SLC6), and Solute Carrier Family 7 (SLC7) gene families' unique amino acid and peptide transporters have been identified in ocular tissues and potential targets for drug delivery strategies [ 38 ]. Table 2 further summarizes the existence of additional trailers used in transporting ocular drugs, i.e., organic cation/anion, monocarboxylate, and nucleoside trailers. These teasers depict pivotal roles in the absorption and subsequent transport of particular compounds across ocular tissues, which support the comprehensive method of ocular medicine delivery. Prodrugs developed specifically targeting these trailers can increase drug allure, enhance physical and chemical properties, get around efflux pumps, augment drug fascination, improve physicochemical properties, and bypass efflux pumps [ 38 ]. Melanin, found in ophthalmic tissues like the RPE and uvea, can interact with medications and change how they are metabolized. Reduced pharmacological efficacy could arise from a considerable reduction in the optimum medication at the effective site due to melanin binding brought about by electrostatic and dispersion forces from London. Preparations that are simple and lipotropic are more inclined to mix with melanin [ 38 ]. The binding process is crucial in administering ophthalmic drugs because it could alter therapeutic dosages in the anterior ocular fleshes and affect the projected therapeutic efficacy. Following peri scleral drug administration, choroidal melanin, and RPE impact how therapeutics enter the retina and vitreous. Contrary to the melanin-free sclera, the presence of melanin can lead to a prolonged permeation lag time for lipophilic molecules and reduced solute permeability across the choroid-Bruch's membrane [ 38 ]. Advancement in drug delivery strategies to the anterior segment of the eyes Ophthalmic drops are the preferred treatment option for ocular diseases due to their ease of use, accessibility, and non-invasiveness [ 48 ]. Various techniques have been employed to enhance the effectiveness of such eye drops. The use of cyclodextrins to improve the aqueous solubility of hydrophobic therapeutics/drugs illustrates advancements in ophthalmic therapy [ 8 , 48 ]. This strategy has resulted in better medication penetration into the eyes. To extend the time that eye drops stay on the ocular surface, viscosity additives, i.e., polyvinyl alcohol (PVA), and cellulose derivatives are generally utilized. These substances improve the efficiency of drug administration without seriously harming ocular cells by lowering drainage and encouraging attraction across the cornea [ [48] , [49] , [50] ]. Fig. 3 further describes the various advanced approaches for ocular drug delivery systems in recent years. Liposomes Liposomes are specialized colloidal carrier systems that can be utilized for the fore slice of the eye. These microvesicles entail an aquatic core encased in by a hydrophobic lipid layer, allowing for the ease of administration through eye drops. They also offer the advantage of sustained and controlled release mechanisms, reducing the frequency of drug application. When designing liposomes, several factors are taken into consideration, i.e., the negative charge of the corneal mucosa, the functional pH of the eye, drug solubility, molecular mass, lipid bilayer mobility, vesicle integrity, and drug loading efficiency [ 51 ]. In a recent year, Dong et al., examined the morphology of liposomes with silk fibroin coating (SLs) for ophthalmic therapy to assess their physical characteristics. The drug encapsulation efficiency of SLs was investigated to determine their ability to contain ibuprofen effectively. The release profile of ibuprofen from solid lipid nanoparticles (SLNs) was investigated through release kinetics analysis in assessment to a drug solution rather than conventional liposome. The learning also appraised the corneal penetration studies of SLs to assess their ability to penetrate the cornea [ 52 ]. In addition, Hosny et al., investigated optimizing the incorporation of gatifloxacin in a liposomal aqua gel matrix. Through their investigation, the researchers identified that a specific fraction of phosphatidylcholine, cholesterol, and stearyl amine resulted in the highest permeability. The inclusion of stearyl amine, a positive electrostatic modifier, improved the loading efficiency. Therefore, incorporating carbopol 940 as a hydrogel facilitated sustained and prolonged drug release, enhancing gatifloxacin transcorneal transport and increasing efficiency [ 53 ]. In another study, Li et al. investigated the utilization of anionic liposomes coated with modified low molecular weight chitosan (LCH)-loaded diclofenac sodium for corneal delivery. The application of low molecular weight chitosan (LCH)-coated liposomes demonstrated improved average residence time, corneal penetration, and concentration levels compared to uncoated liposomes or DS dissolved in water. The inclusion LCH in the formulation potentially enhanced permeability by interacting with the Cornea having a negative surface charge and potentially tight junction permeabilization. Furthermore, the LCH improved the stability of the liposomes by reducing hydrolysis and oxidation [ 54 ]. Microemulsion Microemulsions are stable mixtures of oil and water that use non-ionic surfactants and co-surfactants. Microemulsions offer several advantages, including transparency, thermodynamic stability, and controlled drug release. Various hydrophobic drugs can be loaded more effectively by formulating microemulsions in different ways, and their residence time in the eye can be extended [ 55 , 56 ]. A study on mucoadhesive chitosan-coated cationic dexamethasone microemulsion system demonstrated that CH-MEs (mucoadhesive chitosan-based microemulsions) give favorable physicochemical characteristics, mucoadhesive solid properties, and excellent stability over three months. The sustained drug-release properties of mucoadhesive chitosan microemulsions (CH-MEs) were also observed. Several In vivo experiments on rabbit eyes, utilizing a uveitis-tempted rabbit eye archetypal, displayed a notable upsurge in the anti-provocative worth of the eyes entertained with mucoadhesive CH-MEs Relative to suspension formulation currently in the market [ 57 ]. Microemulsion formulations have been found to increase the bioavailability of not just hydrophobic drugs but also non-hydrophobic drugs. For instance, in the case of drugs like pilocarpine, microemulsion formulations have been developed to reduce dosing frequency, resulting in improved controlled release and extended drug retention in the eye [ 58 ]. Nanosuspensions Ocular drug delivery systems utilize colloidal particles falling within the 10–1000 nm size threshold. These particles serve as carriers for entrapping, adsorbing, or encapsulating the drug along with a bio-adhesive polymer for the occurrence of polyacrylic acid (PAA) or its unoriginal [ 59 , 60 ]. Upon application, these nanoparticles agglomerate on the cornea, resulting in an extended retention period on the corneal superficial. This agglomeration augments the diffusion of the non-ionized drug into the cornea, improving drug delivery efficiency. When compared to a viscosity agent like polyvinyl alcohol (PVA), the use of PAA in topical delivery systems increases sustained drug absorption observed at later time points following administration. Sub-micron particles have demonstrated greater absorption compared to larger microparticles [ 20 , 21 , 61 ]. In research by Adibkia et al., the practice of piroxicam: Endragit® RS100 nanosuspension formulation in reducing inflammation was investigated in rabbits with endotoxin-tempted uveitis (ETU). The nanosuspension construction exhibited superior anti-provocative effects to the microemulsion construction and the unprocessed control set. The nanosuspension's effects lasted up to a 12-h interval, with the maximum therapeutic impact during a 24-h duration. The cationic copolymer, Endragit® RS100, potentially enhanced the residence period in ophthalmic tissues by interacting with negatively charged cells through ionic interactions. This drug-sending scheme has the skill to curtail the required dosage and dosing interval to once daily, which could enhance patient obedience [ 62 ]. In recent years, Kassem et al. conducted a study comparing the worth of frequently castoff glucocorticoid therapeutics in nanosuspensions plus micro suspensions in the eyes of rabbits. The nanosuspensions exhibited the highest drug absorption and maximum intraocular pressure (IOP) reduction, indicating superior efficacy. This research suggests that the claim of nanoparticle suspensions in innovative ocular drug-sending structures could enhance patient adherence by offering once-daily dosing options [ 63 ]. Ocular inserts and minitablets The specialized hard or semi-hard wafers or tablets created specifically for ophthalmic uses are called ocular inserts plus minitablets. The front of the eye segment and cornea can get continuous drug delivery thanks to the placement of these inserts in a cul-de-sac [ 64 ]. They provide numerous benefits, such as changed therapeutics pharmacokinetic profiles and clinical outcomes related to extended exchange length and ongoing drug announcements. They can lessen systemic toxicity and reduce the frequency of instructions, improving patient adherence. Ophthalmic inserts made of soluble hydroxypropyl cellulose have been tested and proven to reduce the symptoms of dry eyes significantly. However, these inserts can occasionally cause extraneous-build consciousness, discomfort, and blurred vision, possibly leading to the end of therapy [ 64 ]. To improve the ocular administration of medications, ocular bio-erodible minitablets, which are similar to ocular inserts, were developed and used. The insertion of these minitablets into the fornix of the eye has been shown to increase gentamicin delivery for treating infectious keratitis [ 65 ]. Ciprofloxacin minitablets without preservatives are considered bioequivalent to Ciloxan® eye drops used on a 30-min basis because they have been demonstrated to maintain bactericidal levels in the tear film for up to 8 h [ 66 ]. Another novel delivery method is Ophthacoil, which uses a metallic wire wound around an aqua gel coating containing a medication. When the hydrogel comes into touch with tears, it expands, enabling gradually controlled medication release. Ophthacoil displayed persistent pradofloxacin problems over the minimal inhibitory concentration (MIC) for 16 h in a trial on dogs [ 67 ]. Ocular implants for the anterior chamber SurodexTM is a bioerodible matrix implant developed by Allergan Inc., specifically designed to release 60 μg of dexamethasone over 7 days. The implant entails a rod-fashioned polymer matrix finished with PLGA (poly (lactic- co -glycolic acid)), HPMC (hydroxypropyl methylcellulose), and dexamethasone [ 68 , 69 ]. It is placed within the front eye chamber and serves as an anti-provocative negotiator for the affected role undergoing cataract surgery. SurodexTM is currently undergoing late-stage scientific hearings and has demonstrated good patient tolerability. However, its effectiveness appears comparable to topical steroids, as distinguished in the learning [ 70 ]. Punctal plugs, also denoted as punctum plugs or lacrimal plugs, also occluders, are small biocompatible inserts used for sustained ophthalmic therapy [ 71 ]. These plugs are typically the size of a rice grain and can be inserted into the lacrimal duct. They come in various designs and shapes, often made of silicone, collagen, hydrophobic acrylic polymer, aqua gel, and polydioxanone. Once inserted, intracanalicular plugs may adopt a semi-hard state or expand to block the cavity. An advantage of this ocular device is its ease of insertion in elderly patients due to age-related enlargement of the puncta [ 72 ]. Advancement in drug delivery strategies to the posterior segment of the eyes The treatment of ocular disorders that mainly affect the back part of the eye, like diabetic macular edema, senile macular degeneration, and posterior uveitis, poses significant difficulties when relying on traditional methods of topical or systemic drug delivery. The issue is mainly caused by ocular vascular barriers that prevent drug absorption into the eye tissue [ 1 ]. The aquatic humor fence and the retinal vascular barricade protect the subsequent slice in addition to the corneal barrier already stated. The blood and intraocular fluid, flanked by the anterior eye barrier and created by the ciliary body, respectively, play a vital part in maintaining homeostasis. Similarly, the neurovascular retinal barrier controls the vitreous humor chemicals' removal while allowing access to a crucial metabolic substrate. Because of these obstacles, researchers have been forced to consider and create other methods of treating diseases that affect the latter part of the eye [ 1 , 19 , 73 , 74 ]. Sustained release ocular drug implants Various sustained-release ocular drug implants have proven effective in treating chronic ocular ailments that distress the later eye sector, such as uveitis, cytomegalovirus (CMV) retinitis, and neovascular age-linked macular deterioration. These implants, whether non-biodegradable or biodegradable, offer significant advantages by bypassing the blood-ocular barrier, providing extended drug delivery, and minimizing toxicity [ 75 , 76 ]. Non-biodegradable implants, including Retisert®, Vitrasert®, and Iluvien®, have been developed explicitly for specific indications. Retisert® is designed for CMV retinitis, Vitrasert® for idiopathic uveitis, and Iluvien® for diabetic macular edema (DME) [ 77 , 78 ]. This surgical implantation into the later eye sector distributes the treatment over an extended period. However, they can also possible adverse reactions and complications like retinal detachment and endophthalmitis. Iluvien® has demonstrated positive results in clinical trials for DME, with significant visual improvements lasting over 30 months [ 79 ]. Nevertheless, specific safety issues highlighted by the FDA have led to the request for additional clinical trials [ 56 ]. Biodegradable ocular implants utilize a matrix composed of biocompatible polymers, for illustration, polylactic acid (PLA), polyglycolic acid (PGA), and otherwise polylactic co-glycolic acid (PLGA), along with the drug. These implants are converted into harmless metabolites within the body and offer various advantages. They do not entail deletion once the drug is depleted from the implant. Ozurdex® is an illustration of a green scion operating a dexamethasone intravitreal PLGA scion to handle macular edema and then non-infectious uveitis [ 57 ]. It employs the NOVADURTM platform to sustain drug release, initially showing a burst followed by slower diffusion and polymer degradation. Innovations in ocular drug delivery include non-polymer-based systems like VerisomeTM developed by Icon Bioscience Inc. The system is injected into the target site using a needle and degrades simultaneously with drug release without causing toxicity [ 76 ]. Gene therapy Gene therapy refers to the inherited quantifiable conveyance, for illustration, DNA or RNA, into target cells to manipulate their inheritable influence expression. Initially, viral vectors delivered gene-modifying nucleic acids, allowing for stable expression within target cells [ 80 ]. However, viral vectors have limitations, including immunogenicity and the need for invasive administration methods. Different growth factors have been targeted in ocular gene delivery, and two FDA-approved drugs, Intraocular implants® and Macugen®, have been developed. Vitravene® is used to treat human cytomegalovirus (HCMV) in AIDS patients, while Macugen® is used for wet age-interrelated macular deterioration. These drugs bind to specific targets, inhibiting viral replication or suppressing ocular neovascularization [ 80 ]. For minimally invasive alternatives, RXi Pharmaceuticals has partnered with EyeGate Pharma to explore using electro-assisted, non-invasive drug transport technology for administering sd-rxRNATM (self-delivering rxRNA) [ 81 ]. Spark Therapeutics is in early-phase clinical trials and Phase III trials for choroideremia and Leber congenital amaurosis, respectively [ 82 ]. Avalanche Biotechnologies is in Segment II hearings for neovascular age-interrelated macular deterioration [ 83 ]. Oxford BioMedica, in collaboration with Sanofi, is conducting Phase I trials for Stargardt ailment and Segment I/II hearings for Usher syndrome [ 84 ]. These advancements in gene therapy hold promise for addressing these debilitating ocular conditions. Drug-eluting punctal plugs Typically, most of the Food and Drug Administration (FDA) approved drug-eluting punctal plugs block tear drainage when inserted into eyelid lacrimal puncta. These plugs, typically made from polymers, have traditionally been used to treat dry eye disease. Companies such as Mati Therapeutics and Ocular Therapeutic are developing them as sustained drug delivery systems for various ocular conditions. Mati Therapeutics is currently engaged in the clinical development of a punctal plug called Evolute latanoprost. This plug releases latanoprost over three months to treat wide-angle glaucoma and elevated intraocular pressure (IOP) [ 85 , 86 ]. On the other hand, Ocular Therapeutic is evaluating their swellable, degradable hydrogel plugs that are impregnated with drugs like dexamethasone for post-operative inflammation and pain, as well as timolol for wide-angle glaucoma or elevated intraocular pressure (IOP) [ 87 ]. These drug-eluting punctal plugs address the challenge of patient compliance by providing sustained drug release. They can be customized to enhance therapeutic effectiveness while reducing side effects. The insertion of these plugs is a simple office procedure performed by eye care professionals. This approach can optimize drug delivery, resulting in improved treatment outcomes for patients with various ocular conditions [ 86 ]. Vascular targeted photodynamic therapy (PDT) Vascular Targeted Photodynamic Therapy (PDT) is a handling repetition operated in ocular drug therapy, with Visudyne® being the sole FDA-authorized ocular drug incorporating PDT. Visudyne® consists of the photosensitizer compound verteporfin formulated in liposomes. The treatment involves IV administration of the photosensitizer, followed by non-thermal laser irradiation of the retina [ 88 ]. Upon laser activation, the photosensitizer generates cytotoxic free radicals that selectively target neovascularization, causing the blockage of the targeted blood vessels. However, Visudyne® has been found to originate hurt to retinal pigment epithelium (RPE) cells and then photoreceptor lesions [ 88 , 89 ]. Steba Biotech S.A. is conducting Segment II medical hearings to investigate the well-being and treatment outcomes of Stakel®-based vascular-directed photodynamic therapy for the affected role with choroidal neovascularization allied with age-linked macular degeneration (AMD) [ 90 ]. This ongoing research aims to evaluate the latent of Stakel® to illustrate an alternative treatment option for this specific ocular condition and assess its effectiveness and safety compared to existing therapies [ 90 ]. Intraocular implants Commercialized ocular implants made with modified polymers and a coating on intraocular lenses (IOLs) can help overcome complications after cataract surgery, including posterior capsule opacification [ 91 ]. Blue light-blocking lens According to a patent from 2015, a combination of a monomer used in the production of plastic lenses and a compound consisting of thiophene and benzene can be advantageous for eye protection by dispersing blue light. In addition to the material dispersing blue light, the mixture may include an ultraviolet light-blocking substance, a pigment, and a polymerization initiator, which are further mixed into the monomer. This formulation aims to provide enhanced protection against harmful blue light and ultraviolet radiation while incorporating other beneficial properties for eye health [ 92 ]. MicroShunt minimally invasive glaucoma surgery The RESERFLO®/MicroShunt is a glaucoma treatment device with a polymer dubbed poly(styrene- block -isobutylene- block -styrene) (SIBS). This stratagem is designed for minimally invasive glaucoma surgery (MIGS) and offers some reduction in intraocular pressure (IOP) and relief from glaucoma medications [ 92 ]. However, one limitation of the MicroShunt is its relatively large plate size. To overcome this challenge, there is a requirement for a compact glaucoma drainage device that does not rely on plates [ 92 ]. The device should be constructed from a polymer material that maintains its integrity over time and does not contribute to the development of substantial scar tissue or encapsulation. Researchers are exploring alternative materials and designs to develop more compact and efficient glaucoma drainage devices with improved long-term outcomes [ 92 ]. INVITE C in diabetic retinopathy The curcumin is loaded in polymer inulin- d -α-tocopherol succinate bio couples (INVITE), so the combination is called INVITE C. INVITE C nano micelles have shown promise in intraocular therapy for diabetic eye diseases, including diabetic retinopathy [ 93 ]. These nano micelles possess antioxidant activity that aids in protecting human retinal pigment epithelium cells, particularly under high glucose conditions. This characteristic enhances their potential as a therapeutic option for diabetic eye diseases [ 93 ]. There is still a constant search for polymers that are biocompatible and biodegradable and have better results and preclinical and clinical trials. The usage of modified polymers assured more residency time for eye drop formulation. Despite the ongoing challenges in corneal penetration, there is still significant potential for further exploration in the field of ophthalmic therapies using polymers. Combining polymer science with interventional studies in preventive and therapeutic approaches in ophthalmology has opened up new possibilities for the future of this field [ 23 ]. Table 3 highlights the need for alternative patient-compliant delivery systems and optimizing the potency of drug therapy for ophthalmic medications before administration. Excipients castoff in ocular inventions for modified drug conveyance Viscosity increasing polymers Enhancing the viscosity of the vehicle is a technique used to extend the residence time of ophthalmic drugs [ 97 ]. Synthetic polymers such as polyethylene oxides, polyacrylates and polyvinyl alcohols, polyesters, polyolefins, collagen, gelatin, and dextran are used to accomplish this increase in viscosity. Viscoelastic eye drops, on the other hand, may promote reflex tearing, resulting in faster medication clearance. More significant viscosity eye drops irritate patients, sometimes do not allow for repeatable drug doses, and produce blurred vision after delivery. An ideal viscosity range of 15–55 P (P) is thus advised, allowing for a longer residence time and avoiding adverse effects, as previously stated [ 98 ]. Polyethylene oxides Polyethylene glycol (PEG) is a colorless, transparent hydrophilic polymer derived from ethylene oxide monomers, which can improve the solvability, biological compatibility, and efficacy of the therapeutic moiety. PEG has various configurations (e.g., multi-armed linear) and molecular weights. The FDA considers it to be generally safe (GRAS) and has approved it for various purposes, including ophthalmic usage [ 99 ]. Short-acting implants, like Dextenza®, a PEG-based tubular implant, make use of PEG's controlled release features by preventing erosion by hydrolysis to give 1 month extended release of dexamethasone for both pain and inflammation control following surgery [ 100 ]. PEG is also used in other Ocular Therapeutix medications, including intracanalicular ocular insert of dexamethasone (OTX-DED) and tyrosine kinase inhibitor. OTX-DED is a low-dose, short-stretch therapy that uses the identical PEG technologies as Dextenza to deliver dexamethasone [ 101 , 102 ]. Thermosensitive polymers have received the most significant attention in terms of administration. To be appropriate, a thermogelling polymer must possess a gel formation temperature in an array of 32–36 °C so that it is molten at room temp but performs an immediate conversion to gel over the ocular shallow [ 101 ]. These polymers can be natural and synthetic. Poloxamer 407 is the most often used and readily available synthetic polymer. The general formula is PEOx- PPOy-PEOx, and it is made up of polypropylene oxide (PPO) and polyethylene oxide (PEO) units [ 103 ]. To enhance the drug tenure of dwelling in the precorneal region, a blend of mucoadhesive polymers and situ gelling polymers with high cohesive and mucoadhesive qualities looks favorable [ 104 ] like chitosan/poloxamer [ 105 ], polycarbophil/poloxamer [ 106 ], carboxymethyl cellulose/poloxamer [ 107 ], polyacrylate/poloxamer [ 108 ], gellan gum/poloxamer/polyacrylate [ 108 ] or hydroxypropylmethycellulose/poloxamer [ 109 ]. The DuraSite® system appears acceptable as a specimen of commercially accessible eye drops incorporating this combination [ 110 ]. It shows a temp-sensitive sol-gel conversion due to poloxamers and excellent much-gummy features due to a polyacrylate. A second variant of this arrangement (DuraSite 2®) was created in another study. Adding a positive-charged polymer to the current generation of eye drops ensured an extended dwelling period of the therapeutics in the precorneal precinct. These podia were utilized in the production of eye drops of azithromycin (AzaSite®) [ 111 ], and moxifloxacin (Besivance®) [ 112 ]. In the study, Lakhani et al. demonstrated that incorporating PEG (polyethylene glycol) into nanostructured lipid carriers enabled the anti-mycotic drug amphotericin B solubilization. This formulation showed potential for ocular antifungal topical treatment. PEG is currently being investigated in clinical trials for its therapeutic PEGylation properties [ 113 ]. Polyvinyl alcohols (PVA) Polyvinyl alcohol (PVA) is a synthetic polymer that is biodegradable and water-soluble. It has various applications in polymer manufacturing, medicine, and nutrition industries. PVA (polyvinyl alcohol) is frequently employed to enhance the drug's solubilization in aqueous media because of its resistance and processability. It has a low acute oral toxicity, with an LD50 (median lethal dose) ranging from 15 to 20 g/kg [ 114 ]. Furthermore, PVA is poorly absorbed through the digestive tract. The unique structure of PVA allows for variable permeability, making it suitable for sustained formulations. The recent approval of Yutiq®, an implant composed of polyimide/PVA, further expands the possibilities for controlled drug sending, mainly for handling uveitis. PVA is also painstaking for superficial ocular medication healing using wafer formulations. For illustration, a study by Dipak et al., developed PVA nanofibers loaded with gatifloxacin, demonstrating prolonged precorneal contact duration and a two-step announcement rough draft with a rapid preliminary sector tailed by the continual announcement for up to 7 h. This extended-release profile reduces the prerequisite for recurrent direction in treating parched eyes [ 115 ]. Similarly, Javid et al., demonstrated that utilizing nanofibers placid of levofloxacin-conjugated chitosan incorporated into PVA showed improved sustained release characteristics. The findings highlight the effectiveness of drug conjugation to the polymer core in reducing burst release and achieving prolonged release of levofloxacin [ 116 ]. Polyesters The artificial polymers PLA (polylactic acid), PGA (polyglycolic acid), and PLGA (poly (lactic- co -glycolic acid)) are frequently used in ocular preparation conveyance arrangements. The FDA has given clearance to these hydrophobic polyesters for use in ocular applications. Lactic acid is the source of PLA, whereas glycolic acid is the source of PGA [ 117 , 118 ]. Both are then generated by ring-inaugural polymerization. PGA breaks down more quickly than PLA because it is more hydrolyzable [ 117 ]. The phase 2 clinical trials for slow-release brimonidine in treating geographic withering associated with age-linked macular degeneration were conducted on Brimo DDS®, an intravitreal implant comprising PLA [ 119 ]. Based on the monomer ratio and end groups, PLGA, a copolymer of PGA and PLA, offers configurable biodegradation properties. It enables better bioavailability, increased biocompatibility, and controlled release of medicines. The adjustment of crystallinity, degradation time, and hydrophobicity is possible by adjusting the PGA/PLA ratios. The FDA-approved intravitreal implant Ozurdex® uses PLGA to release dexamethasone gradually over 4–6 months before entirely disintegrating in the body. The FDA has approved Durysta®, another PLGA-based implant, for IOP-lowering therapies in wide-angle glaucoma patients [ 120 , 121 ]. Researchers have looked into several modifications and polymer combinations in the search for aesthetically pleasing ocular drug conveyance. For treating inflammation, PLGA-containing nanoparticle formulations, in combination with other excipients, demonstrated therapeutic advantages, tolerability, and better corneal penetration. Additionally, PLGA nanocore lipid polymer nanoparticles have been demonstrated to boost corneal penetration, prolong drug release, and enhance therapeutic advantages [ 121 ]. Polymethacrylates Poly (methyl methacrylate) (PMMA) then added methacrylate-grounded polymers are acrylic, green thermoplastics acknowledged for their water transmissibility, mechanical strength, and thermal stability [ 122 ]. These polymers have been used in ophthalmic applications since 1936 and have gained FDA approval for intraocular use [ 77 ]. Methacrylate-based polymeric biomaterials have been somewhat cast off in ocular healings, plus the development of nanoparticles, micelles, ocular implants, and hydrogels for equally topical conveyance than intravitreal shot. Methacrylate derivatives such as poly(2-(dimethylamino) ethyl methacrylate) (DMAEM) and poly (2-hydroxyethyl methacrylate) (HEMA) have been investigated in various studies. For instance, PMMA (polymethyl methacrylate) has been utilized as a non-degradable substance coating in intravitreal implants like I-vacation, allowing for extended release of triamcinolone acetonide to extravagance diabetic macular edema [ 123 ]. HEMA and other methacrylates have been explored to develop atorvastatin-eluting lenses to manage ocular disorders associated with diabetes [ 124 ]. Poly(amidoamine) (PAMAM), a dendrimer, has received approval from the FDA for specific applications. However, its inclusion in the Generally Recognized as Safe (GRAS) list is limited due to toxicity concerns. PAMAM has been studied with dexamethasone for extended drug release in the posterior eye for diabetic retinal dysfunction and macular impairment. PAMAM conjugation to fluocinolone acetonide has been investigated for managing retinal vasculitis in conditions like macular impairment and retinal pigmentary degeneration [ 125 ]. These studies demonstrated the effective conjugation of therapeutic agents to the dendrimer, which was given to rats via intraocular injection. Polyolefins Poly (acrylic acid) (PAA), acknowledged by its commercial name Carbopol®, is a non-natural polymer made up of acrylic acid monomers. It exhibits excellent aquatic miscibility and thickening capabilities [ 126 ]. While PAA is biodegradable, it is imperative to note that the metabolites of acrylic acid can potentially induce inflammation. PAA carries a charge in physiological environments, contributing to its favorable mucoadhesive properties. In experimental hydrogel applications for anterior ocular delivery, Poly (acrylic acid) (PAA) has been combined with other polymers, such as PNIPAAm, to create temperature-sensitive in situ hydrogels. These have been employed for sustained release of epinephrine, particularly in treating glaucoma [ 127 ]. PAA has received FDA approval for various epidermal applications, including ophthalmic delivery. It is marketed in different eye drop formulations. One is Restasis®, an ocular emulsion that utilizes a carbomer copolymer sort A-grounded arrangement to convey cyclosporine in managing parched eye settings [ 128 ]. In research by Moustafa et al., fluconazole liposomal hydrogels were developed and synthesized as innovative arrangements to enhance ocular remedy conveyance and prolong the dwelling time. The results demonstrated that these liposomes exhibited an extended dwelling stretch in the eye, lasting more than 1080 min [ 129 ]. Collagen A fibrous protein called collagen is found naturally in many connective tissues, such as the sclera, cornea, lens capsule, and vitreous humor. Collagen is readily available from mainly bovine and porcine origins as it occurs naturally, is enzymatically degradable, biocompatible, and relatively simple to make. Recombinant collagen is now widely available, which has decreased reliance on animal sources for collagen manufacturing. Plant and yeast cells can produce recombinant collagen, which has benefits such as increased uniformity and safety throughout production [ 130 ]. Following ocular injury or cataract surgery, collagen shields have been used to protect the eyes for a long time. In various methods, it has continued to be used as a drug delivery system in ocular research. It has been utilized for encapsulated cell treatment, in which collagen encases cells and delivers them intravitreally. Collagen-based gels have been used to achieve extended drug release. Similarly, collagen has been utilized to achieve the same effect. Recent research has implemented it as a platform for regenerating retinal tissue, demonstrating its promise for tissue engineering uses in the retina [ 126 ]. Photrexa® is a contemporary ocular medicine delivery system that uses collagen [ 131 ]. A riboflavin eye solution filled with collagen crosslinks the biopolymer when subjected to ultraviolet A light. Gelatin Gelatin is a proteic polymer fashioned by the irreversible process of collagen breakdown. It is biodegradable, non-toxic, viscosity boosting, gel-forming, readily available, and low in cost, yet it has the advantage of a lower gel formation temperature and better aqueous solubility [ 102 , 132 ]. It is GRAS-certified and produced from avian, mammalian, and ichthyoid collagen I sources, enabling a wide variety of accessible molecular weights. One can use recombinant gelatin with specific molecular masses and isoelectric points to minimize immunogenicity. Gelatin has been used in ocular drug administration in eye drops to act as demulcent, posteriorly and anteriorly administered hydrogels, nano subdivisions for continual announcement, ophthalmic flesh manipulation, and then siRNA carter for inherited healing [ 23 ]. Dextran Dextran is a polysaccharide biopolymer placid of d -glucose slices and is lactic acid bacteria-produced dextran. Dextran can create hydrogels and serve as a carrier for siRNA in inheritable influence healing claims [ 133 ]. Dextran is a biopolymer authorized by the FDA for use in ocular eye drops like Tears Natural II® and Tears Natural Forte® as treatment options for dry eye [ 134 ]. It is conveniently chemically crosslinked and is being tested for intravitreal and topical administration of ocular therapies with chitosan, PLGA/PLA, then PEG. An experimental study demonstrated the effective administration of lutein, an antioxidant, using crosslinked dextran-chitosan nano subdivisions for topical claim [ 135 ]. Dextran can also be used to conjugate drugs for ocular administration. With a low molecular weight, Dextran has been tested as a carrier of cationic DNA, enabling targeted gene therapies for managing X-linked juvenile retinoschisis. Following intravitreal injection of rats, researchers successfully transfected and expressed a complex of dextran-protamine-DNA deposited over the surface of solid-lipid nanoparticles [ 23 ]. Mucoadhesive polymers Mucins comprise no fewer than 20 anionic O-glycosyl proteins, making up the membrane gel layer that protects the eye surface. Mucins create a glycocalyx layer on the conjunctival and ocular surfaces [ 136 ]. Excipients that provide adherence to this mucous membrane have a long ocular residence duration [ 137 ]. To enhance mucoadhesive properties, polymers can possess specific structural elements. These include: (i) Strong Charges: Polymers with robust charges promote ionic interactions, contributing to mucoadhesion. (ii) Hydrogen Bonding Groups: Functional groups like carboxyls, hydroxyls, amino, and sulfate groups have the skill to fashion robust hydrogen ties. These assemblies facilitate mucoadhesion. (iii) Molecular Weight and Chain Flexibility: Polymers with high molecular weight and chain flexibility facilitate crosslinking with the mucus membrane. This enables the formation of chain entanglements, enhancing mucoadhesion. (iv) Favorable Surface Energy: Polymers with surface energy that supports the spreading of mucus enhance their contact with the mucosal surface, improving mucoadhesion. These structural elements collectively contribute to the mucoadhesive properties of polymers, enabling effective interaction with mucus membranes. This interaction is crucial for applications such as drug delivery systems [ 138 ]. Mucoadhesive polymers must cling to the ocular mucus membrane while simultaneously releasing the medicine coupled to the polymeric chains. Cellulose derivatives Cellulose derivatives can be categorized into cellulose ether and cellulose ester. Cellulose ether includes hydroxypropyl methyl cellulose (HPMC), methylcellulose (MC), carboxymethyl cellulose (CMC), ethyl cellulose (EC), hydroxyethyl cellulose (HEC), then hydroxypropyl cellulose (HPC). In contrast, cellulose ester comprises cellulose acetate phthalate (CAP) plus cellulose acetate (CA) [ 139 ]. Among these derivatives, carboxymethyl cellulose (CMC) is a commonly used polysaccharide as it exhibits improved water solubility due to the incorporation of carboxy groups into the biopolymer chains. CMC often appears in topically delivered eye solutions such as Optive® or Refresh® in managing dry eye owing to its hydrophilicity and biocompatibility. However, numerous additional brands and formulations are available [ 96 ]. In an investigation by Yuan et al., CMC-based Nano wafers were created and characterized for prolonged anterior drug administration axitinib. Compared to typical eye drop delivery, the topically administered transparent Nano wafers feature therapeutic nano reservoirs that provide extended drug release and enhanced bioavailability [ 140 ]. The initial cellulose polymer to be developed was methylcellulose more than 60 years ago. Methylcellulose has no taste or odor, is water soluble, and is non-toxic. As a result, cellulose ethers are frequently employed in eye drops as viscosity-increasing agents and for wetting properties, which can shorten connection times due to their crust-establishing properties [ 141 ]. Cellulose spin-offs have LCST (lower critical solution temperature) behavior, which is a temperature-dependent sol-gel phase conversion, and the gelation process involves hydrophobic interactions between molecules with methoxy substitution [ 142 ]. This phase transition feature provides the ability to apply thermoresponsive ophthalmic therapy. Electrolytes, sugars, surface-energetic agents, and natural gums can all affect gelation by lowering the gelation temperature and reducing polymer hydration. An increase in the concentration of methyl cellulose results in a linear reduction in gelation temperature [ 139 ]. Hydroxypropyl methyl cellulose (HPMC), also acknowledged as Hypromellose is a hydrophilic polymer with a white or pale white appearance. It finds various applications as a meticulous announcement mechanism in oral and oro-mucosal drug conveyance. HPMC has the skill to swell and then fashion a gel, and it exhibits stability within the pH range of 3–11. It is resistant to gastric enzymes and is commonly used as an excipient for extended-release formulations and thickening, emulsifying, and stabilizing purposes [ 143 ]. HPMC, a cellulose-based semi-synthetic dietary fiber, consists of anhydrous glucose units. When exposed to water or gastrointestinal fluids, it forms a viscous solution. HPMC is utilized to enhance the permeation of hydrophobic therapeutic molecules, thereby improving their bioavailability. Liu et al., conducted a study on the ocular administration of FK506-encumbered nano micelles composed of amino-finished polyethylene glycol block poly (D, l )-lactic acid, then HPMC. This study demonstrated the boosted pharmacokinetics and therapeutic worth of FK506 in treating ocular athwart-allograft rebuff [ 144 ]. Nanda et al., inspected the enhancement of corneal permeability for the anti-inflammatory drug amlodipine using HPMC. They examined the upshot of sulphobutyl ether-cyclodextrin on a rabbit archetypal and then found that sulphobutyl ether-cyclodextrin improves the permeability of amlodipine-HPMC film [ 145 ]. Hyaluronic acid (HA) In various tissues, Hyaluronic acid (HA) is a glycosaminoglycan with a dominant character in the extracellular matrix. It possesses the skill to bind aquatic jots over hydrogen connection, which augments the stabilization of the lacrimal film and reduces the blinking reflex. However, HA-based ophthalmic formulations typically have a brief duration of ocular residence despite the influence of molecular weight on mucoadhesiveness. It has been observed that higher molecular weight HA exhibits greater mucoadhesiveness [ 146 ]. In a study by Salzillo et al., six commercially available formulations, including Blugel®, Bluyal®, Artelac Splash MDSC®, Hyabak®, Octilia Natural®, and Hyalistil Bio®, were compared to newly formulated HA-based eye drops. Among them, Bluegel® demonstrated enhanced fluid drainage grounded on nix-shear viscosity optimization of 24.2 mPas. This was selected as the bull's eye worth for consequent optimization trials [ 109 ]. In addition, Liu et al., also demonstrated the transplantation of retinal precursor cells consuming an HA-grounded hydrogel for accurate placement within the sub-retinal region without impairing their functionality. After complete HA degradation, the cells unveiled mature photoreceptor pointers [ 147 ]. Additionally, HA has been operated as a self-regenerating internal needle glaze for intravitreal injections to curtail drug spillage beyond the eye [ 148 ]. HA is commonly used as a lubricant in parched eye drops, acting stimulated tear crust in harvests for illustration, Vismed Multi, Optive Fusion, Hyalistil®, DROPSTAR®, then Neop. SolarazeTM gel is another application where HA gel is used to create a diclofenac prolonged-release formulation, providing management of eye inflammation and soreness [ 23 ]. Researchers have developed a thermosensitive in-situ hydrogel using HA and poly(N-isopropyl acrylamide). This hydrogel, incorporating ketoconazole as an active ingredient, has demonstrated enhanced therapeutic efficacy while minimizing adverse effects in In vivo trials [ 149 ]. Chitosan Chitosan is a green biocompatible polymer obtained by deacetylating chitin. It shows promise as an additive in ophthalmic formulations because its reactive amino clusters can intermingle with the cornea and then conjunctiva. However, unmodified chitosan is insoluble at physiological pH, particularly pH values above 5, which limits its topical application in ocular formulations [ 4 ]. To address this challenge, various spinoffs of chitosan have been developed for illustration N-carboxymethyl chitosan, then quaternary ammonium-chitosan like N, O-[N, N-dimethylaminomethyl (diethyldimethylene ammonium)n]methyl chitosan or N-trimethyl chitosan (TMC) [ 150 ]. While chitosan has limited FDA clearance, it is not approved for ocular use. Pharmacokinetic studies suggest efficacy [ 4 ]. Chitosan liposomes and micelles are skilled in carrying a high preparation payload and providing extended medication announcements, making them suitable for intravitreal injection. Chitosan is extensively exploited as a polymer encapsulation for less biocompatible anionic polymers in the step-by-step fabrication of core-shell biomaterials, for transporting anionic medications, and then inherited things owing to its cationic properties [ 151 ]. Recent research has focused on chitosan-based hydrogels to enhance the bioavailability of topically applied antibiotics like levofloxacin. Temperature-sensitive hydrogels containing hexanoyl glycol chitosan demonstrated reduced ocular discomfort and 1.92-fold advanced worth in rabbit aquatic humor equated to conventional antibiotic suspensions [ 152 ]. Researchers have developed glucocorticoid-containing solutions, such as an aquatic-unsolvable methyl-cyclodextrin-ammonium chitosan combination, which exhibits mucotackiness and cytocamaraderie [ 153 ]. Guar gum Guar gum is a biopolymer consequential from seeds that consist of rectilinear spine chains of β- d -mannose parts with branching points of α- d -galactose parts. It is a biocompatible, water-soluble polymer with viscosity enhancement, non-ionic nature, mucoadhesion, and hydrolytic degradability [ 23 ]. Due to its gelling properties, guar gum is commonly used in moisturizing eye drops and has received FDA approval for ocular repetition. It is sparingly soluble in certain solvents and alcohol-containing formulations, and its stability in solution can be challenging. To address these limitations, derivatives, including hydroxymethyl-guar gum, o -carboxymethyl o -hydroxypropyl-guar gum, and hydroxypropyl-guar gum, have been fashioned to rally solvability and permanency [ 154 ]. Hydroxypropyl-guar gum is often incorporated into lubricating eye drops [ 155 ]. In experimental studies, guar gum is being explored to enhance the bioavailability of natamycin, an antifungal agent used to manage ocular fungal contagions. This is achieved by incorporating guar gum into polyethylene glycol (PEG) nano-lipid carriers, which enable controlled release through a gelling system integrated with carboxy vinyl polymer and borate [ 156 ]. Guar gum-based micelles composed of poly(ε-caprolactone) (PCL) are being studied for extended release of ofloxacin, an antibiotic. These micelles and biotinylated glutathione, retinol, and cell-selective targeting molecules have shown continual medicine announcements for the tiniest 8 h [ 157 ]. Pullulan Pullulan is a polysaccharide receipted from the yeast Aureobasidium pullulans, entailing maltotriose slices connected by α-1,6 glycosidic bonds. It possesses several beneficial properties, including being non-ionic, biocompatible, stable across a wide range of temperatures then pH assortment, aquatic miscible, non-solvable in most organic diluters, effortlessly manageable, oxygen barricade, viscidness-boosting, then environmentally degradable [ 158 ]. It has received Generally Recognized as Safe (GRAS) certification from the FDA and then finds applications in various biopolymer-based systems, including ocular drug delivery. Modification techniques like sulfation or amination are often employed to introduce charges and enhance their reactivity due to their non-ionic nature. Co-polymerization of pullulan with synthetic polymers or other biopolymers has publicized talent in outspreading the bio dilapidation frequency equated to using pullulan alone. For instance, the combination of pullulan and gellan gum has been utilized to create electrospun nanofibers for developing in-situ gels, enhancing the bioavailability of topically administered therapeutics. The gelation assets of pullulan in aquatic make it apt for applications such as hydrogel inserts and thin films. A study by Pai and Reddy evaluated a 10 % pullulan gel insert designed for conjunctival preparation direction for its In vitro characteristics. The insert exhibited a complete breakdown and preparation announcement within 3 h of the claim in the In vitro setup [ 159 ]. Cyclodextrins Cyclodextrins have become the focus of various research studies for decades as substances capable of increasing drug solubility. Because of the development of inclusion complexes, they solubilize sparingly soluble actives [ 160 ]. Cyclodextrins are cyclic oligosaccharides that have been relatively studied for their skill to augment drug bioavailability, solvability, therapeutic activity, permeability, and physicochemical stability while reducing toxicity and tissue irritation. The primary mechanism by which cyclodextrins improve the effectiveness of ocular systems is through their skill to upsurge the dissolution of lipotropic drugs in tears. Additionally, β-cyclodextrin has been found to increase corneal perviousness by confiscating cholesterol from corneal cells, thus enhancing drug conveyance [ 161 ]. Researchers have explored various approaches to utilize cyclodextrins in ocular preparation conveyance. For example, the invention of a cyclodextrin-thiolated-poly-aspartic-acid couple for the ocular conveyance of prednisolone established protracted drug announcement equated to the physical amalgamation of polymer then cyclodextrin [ 162 ]. Sulphobuthyether-cyclodextrin (SBE-CD) has also been used to increase ocular drug residency and enhance endo-ocular bioavailability, as shown by the inclusion complex of SBE-CD with chloramphenicol [ 160 ]. Thiolated cyclodextrins have shown promise in enhancing drug delivery. For instance, α-cyclodextrin coupled with cysteamine (α-CD-Cys) demonstrated significantly higher mucoadhesive and improved coverage of irritations compared to α-cyclodextrin alone, enabling efficient delivery of the parsimoniously solvable drug cetirizine [ 163 ]. Another strategy involves incorporating cyclodextrins into nanosystems, as demonstrated by the preparation of nanoparticles treated with an immunopositive therapeutic for the managing of parched eye ailment, where the fraction of α-CD-cyclosporine A to γ-cyclodextrin was carefully modulated to act as an aggregating agent [ 164 ]. Table 4 describes excipients used in modified-release ocular formulations. Fig. 4 describes various excipients used for ophthalmic formulation development. Limitations of various excipients The main obstacle in ocular drug delivery systems is attaining and maintaining an appropriate amount of drug in the bull's eye turf for a sufficient stretch period. So far, it is estimated that the topical route of administration could not achieve more than 5 % drug penetration in the eyeball. As a result, excipients play an essential role [ 169 ]. Excipients such as viscosity enhancers, chelating agents, and surfactants promote permeability; however, the dangers outweigh the advantages, resulting in eye discomfort, impaired vision, and altered stability. Additionally, additional preservatives impact safe medication consideration and medicine packaging compatibility [ 170 ]. Ocular excipients have various drawbacks, including low solubility, instability, or toxicity [ 171 ], which make them unsuitable for long-term safe use, universal adoption, or desired efficacy. Excipient constraints in ocular devices for drug delivery are addressed in Table 5 . When utilized for an extended period, preservatives such as Benzalkonium chloride and Thiomersal can induce eye irritation, dry eye symptoms, and potentially harmful effects on the tear crust and corneal surface. The current trend of substituting Boric acid and Borax for the items mentioned above is unsafe, as it is not specified in the USFDA monograph for ophthalmic medicines. Furthermore, most excipients have solubility issues, and formulation variation has been recorded if the polymer is modified to improve release. The research is focusing on biopolymers; however, the supply is uncertain. Fig. 5 describes the several limitations of excipients used in ocular drug delivery. Toxicity studies of excipients Toxicities primarily transpire in the fore slice of the eyes due to the high numbers of trial jots at the spot of topical direction. As a result, toxic effects are most frequently pragmatic in the conjunctiva, cornea, iris, surrounding fleshes and organs for illustration, eyelids, skin, and lacrimal glands. It is crucial to develop and deliver safe and well-tolerated dosage forms. Below are various tests to ensure the ocular excipients' safety [ 184 ]. MTT assay The short-term exposure test method involves the quantitative measurement of cell viability by assessing the catalytic renovation of the vital tint MTT (3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide), also acknowledged as Thiazolyl Blue Tetrazolium Bromide, which results in the production of a blue formazan salt by living cells. This measurement is conducted after extracting the cells. The resulting cell viability, obtained after a 5-min acquaintance, is equated to a solvent control (relative viability) and is cast-off to judge the latent eye peril of the trial chemical conferring to the OECD guidelines [ 185 ]. Draize eye trial Albino rabbits, particularly New Zealand whites, are often used for ocular testing. Before exposure, the animals must be acclimated to the testing environment. It is also essential to assess the animals beforehand to ensure they have normal eyes, and the cages should be designed to minimize the risk of accidental injury. Typically, 3 to 6 creatures are cast off for each trial prescription, and in some cases, a solo creature may oblige as a sentry beforehand for further acquaintances [ 185 , 186 ]. Supplementary trial creatures may be counted to guesstimate the upshot of the vehicle if it is not hitherto acknowledged. During the experiment, one eye is assigned as the experimental eye, while the opposite eye is a matched control and is generally untreated. The experimental substance is administered in a volume of 0.1 ml per eye, although smaller doses (e.g., 0.01 ml) or single eye drops (approximately 0.04 ml) may be additional suitable. Solid amalgams are typically pulverized into fine dust before application. The experimental substance is applied to the lower conjunctival fornix, allowing natural blinking. Sometimes, the eyelids can be gently closed for a few seconds after application [ 185 , 186 ]. If needed, hygienic saline or a sensible brackish solution can be operated to flush the ocular outward. Pre- and post-exposure assessments are conducted through external visual examination under suitable lighting conditions [ 185 , 186 ]. Additional information may be gathered through magnification using binocular loupes or slit-lamp biomicroscopy. Evaluations are typically carried out at specific time intervals, such as 1, 24, 48, and 72 h after exposure, and if necessary, at 7 and 21 days. Ocular changes are evaluated using a scoring system that considers any eyelids, conjunctiva, cornea, and iris modifications. Supplementary tests, such as measuring levels of the test substance in tissues and fluids, may also be performed. Histopathological examination of removed eyes is generally reserved for severe reactions [ [185] , [186] , [187] ]. HET CAM test To prepare the eggs for testing, fresh, fertile eggs undergo a cleaning process using 70 % alcohol. The eggs are nurtured at a controlled temp. of 37 ± 0.5 °C plus a relative humidity of 66.0 ± 5.0 % for a duration of 8 diurnal. During the nurture period, the eggs are physically rotated every 720 min around the equatorial axis to thwart the embryo from sticking to the shell. On the one-eighth diurnal, the eggs are taken out of the incubator and inspected using a flashlight to confirm the presence of embryo development. Careful incisions are made in the eggshells to expose the Chorioallantoic Membrane (CAM) surface to create windows. Positive control solutions such as 0.5 M NaOH are used to assess coagulation, acetone is used to evaluate hemorrhage, and propylene glycol (PG) is used to observe hyperemia. A negative control solution, phosphate buffer solution (PBS) per a pH of 7.4, is also employed. The testing solutions are applied to the clearest vessel of the CAM, and their effects are compared to the control solutions. After each sample application, the CAM morphology is evaluated for irritation reactions, including blood coagulation, bleeding, and hyperemia. These reactions are assessed at 30 s, 2 min, and 5 min of exposure, and the irritation potential of the sample is calculated based on these observations [ 189 ]. In silico methods In silico mockups exploit storehouses of prevailing toxicology data derived from Ex vivo and I n vivo tryouts to forecast the venomousness of samples. These models rely on quantitative structure-activity relationships (QSAR) to launch the linkage mid a trial's chemical assembly and biological possessions. QSAR is rooted in the principle that the bustle of jots can be anticipated based on their assembly, plus these projections can be quantified [ 190 ]. By incorporating biological responses into computational algorithms, predictive models can be generated. Developing a dependable QSAR model necessitates accumulating, assessing, and weighing a sufficient volume of statistics for a definite noxious logic terminus [ 190 ]. This process facilitates a comprehensive connection between the jots under scrutiny and their biological bustle. Excipients augment drug conveyance by taming solvability, governing pharmacokinetics, and prolonging announcement. However, in ocular delivery, safety is of utmost concern [ 190 , 191 ]. Furthermore, Table 6 describes the list of excipients along with their toxicity studies. Future directions Ocular drug delivery is a complex process, making ensuring patient compliance and user-friendliness difficult. However, there is significant potential for improvement through research and development. Intelligent drug delivery systems offer promising solutions, including intravitreal injections, in situ gels, nanoparticles, implants, eye devices, and coils [ 59 ]. Another area of emphasis is the creation of processing excipients that enhance medication stability. Ocular medications frequently suffer degradation issues in the ocular environment. Future research can focus on developing excipients that preserve sensitive medications against degradation, allowing them to maintain therapeutic efficacy over long periods [ 169 ]. This could entail the introduction of innovative polymers or additives to form a protective barrier around the medication molecule, sheltering it from environmental conditions and extending its shelf life. Another crucial element of ocular medication administration is increasing drug solubility [ 18 ]. Many medicines used in ophthalmic therapy are poorly soluble, limiting their efficient distribution to target tissues. Processed excipients can be adjusted to improve drug solubility, allowing for more efficient and effective ocular delivery. Advanced formulation approaches, i.e., the creation of self-emulsifying systems or lipid-based carriers, could be investigated to improve drug solubilization and absorption inside the ocular tissues [ 169 ]. Adding permeation enhancers or mucoadhesive excipients into ocular formulations may enhance drug capture and retention, allowing improved penetration into ocular tissues. Nanotechnology has enormous promise for ocular preparatory conveyance in processed excipients. Nanoscale excipients can be designed to provide besieged and continuous drug announcements, allowing for more significant positive results [ 18 ]. Furthermore, processed excipients with improved healing, such as in situ gels, implants, or contact lenses, can revolutionize ocular treatment delivery. These arrangements can be planned to bring together processed excipients, allowing for sustained medication release, extended residence time, and increased patient compliance. Continued research and development in this subject will result in the creation of personalized excipients that optimize drug distribution, improve therapy effects, and rally for the overall handling of ocular illnesses [ 18 , 169 , 200 , 201 ]. Funding Not Available. Data availability statement There is no research related data stored in publicly available repositories and the data will be made available upon request. CRediT authorship contribution statement Sumel Ashique: Writing – review & editing, Funding acquisition, Formal analysis, Data curation, Conceptualization. Neeraj Mishra: Validation, Methodology, Investigation, Funding acquisition. Sourav Mohanto: Writing – review & editing, Visualization. B.H. Jaswanth Gowda: Funding acquisition, Formal analysis, Data curation. Shubneesh Kumar: Visualization, Validation, Resources. Amisha S. Raikar: Project administration, Methodology, Formal analysis, Conceptualization. Priya Masand: Formal analysis, Data curation. Ashish Garg: Visualization, Validation, Resources, Project administration. Priyanka Goswami: Visualization, Validation, Formal analysis. Ivan Kahwa: Writing – review & editing, Writing – original draft, Supervision, Resources, Data curation, Conceptualization. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments Sumel Ashique would like to acknowledge the Bharat Institute of Technology, India for providing all the facilities to draft the manuscript. Sourav Mohanto would like to acknowledge Yenepoya (Deemed to be University), India for providing the facilities to carry out the research-oriented activities.

CC BY

no

2024-01-16 23:41:58

Heliyon. 2023 Dec 23; 10(1):e23810

oa_package/2f/8d/PMC10788286.tar.gz

PMC10788288

38226328

Introduction Depression is one of the most prevalent mental health problems among young adults (18–25 years of age). ( National Institute of Mental Health, 2022 ) While the prevalence of depression among children is low without any sex differences, notable differences among females and males begin during adolescence and throughout young adulthood. ( Yang et al., 2007 ) Additionally, depression continues to either persist or reappear during later life. ( Mills et al., 2017 ) In the United States, an estimated 17 % of young adults had depression in 2020, with more females reporting depression (10.5 %) than males (6.2 %). ( National Institute of Mental Health, 2022 ) Due to the poor mental and physical health outcomes (e.g., anxiety, illicit drug disorders, migraine, poor self-rated health, and increased work impairment) associated with depression, ( Paradis et al., 2006 ) identifying potential modifiable risk factors, such as iron deficiency, are needed. Iron deficiency, which is treatable, may contribute to depression, ( Mills et al., 2017 ) yet there is limited research examining this association among young adults, a developmental period in which the brain is rapidly developing and individuals are becoming more autonomous therefore subject to environmental and lifestyle changes, including dietary changes. ( Poobalan et al., 2014 ) Iron deficiency is a major cause of anemia, one of the most common nutritional disorders worldwide. ( Kassebaum et al., 2014 ) Iron deficiency anemia is the impaired hemoglobin synthesis due to lack of iron, which can induce shortness of breath, dizziness, and fatigue. ( Khedr et al., 2008 ) Iron also plays a critical role in brain development and the functioning of neurological, autoimmune, endocrine, and cardiovascular system, all of which can result in multiple health problems such as stroke, coronary heart disease, and endocrine disorders. ( Khedr et al., 2008 , Mirza et al., 2018 ) Additionally, many researchers have reported that iron deficiency is associated with poorer physical performance and lower work productivity of adults of all ages. ( Khedr et al., 2008 ) However, the literature on the prevalence of iron deficiency among young adults in the United States and throughout the world is limited. These few studies focused on a large age range; one study reported that female adults 18 to 49 years of age have a higher prevalence of iron deficiency (33 %) compared to males (1.5 %) in Korea, ( Kim et al., 2011 ) with similar findings reported in Belgium, Tehran and Japan. ( Asakura et al., 2009 , Shams et al., 2010 , Pynaert et al., 2007 ). Some growth periods require higher amounts of iron, especially those going through puberty or pregnancy, therefore individuals in these groups are more prone to iron deficiency. ( Eicher-Miller et al., 2009 ) Specifically, many studies identified that women of reproductive age, typically defined as 12 to 49 years of age, thus including young adult females, is a risk factor for iron deficiency. ( Sekhar et al., 2016 ) A study by Shariatpanaahi et al. (2007) analyzed ferritin levels (a measurement of iron levels) of 192 female medical students (24.5 years mean age) and reported that ferritin levels were significantly lower in students with depression compared to healthy students. ( Vahdat Shariatpanaahi et al., 2007 ) Additionally, since young adults are now more independent and unsupervised, their diets may differ and they may be making poorer dietary choices, leading to an unbalanced diet and possibly deficiencies in key nutrients, vitamins, and minerals. ( Winpenny et al., 2017 ). While no studies have focused on the associations between iron and depression among this young adult age group, several studies have reported associations in adults, especially among older age adults 65 and above. Stewart and Hirani (2012) reported associations between higher serum transferrin receptor levels but not ferritin levels, both measurements of iron deficiency, with more depressive symptoms in older adults. ( Stewart and Hirani, 2012 ) Similarly, Hosseini et al. (2018) reported that mean serum iron levels were lower in the depressive symptom group while there was no difference among ferritin levels between the groups. ( Hosseini et al., 2018 ) Consistent with these findings, Su et al. (2016) reported no associations between ferritin levels and depressive symptoms among a cohort of Chinese adults. ( Su et al., 2016 ) Studies used different measurements of iron deficiency among broader and older age groups, making findings difficult to compare, thus our goal is to determine whether a comprehensive set of measurements of iron status (ferritin, serum iron, and transferring) is associated with depression among this younger age group. Due to the limited research on iron deficiency among young adults, as well as the associations between iron deficiency and depression, with known differences among males and females, our study aimed to describe the prevalence of these two important health outcomes among young adults, and stratified among males and females, as well as report on their associations using the National Health and Nutrition Examination Survey (NHANES) 2017 to 2020 dataset. We hypothesize that young adult females have both higher rates of iron deficiency and depression and while iron deficiency is associated with depression for both males and females, females with iron deficiency are at higher risk for depression. Research into iron deficiency, a modifiable risk factor of depressive symptoms and depression, is important since iron deficiency can both be prevented and also treated with increased iron-rich foods or over-the-counter supplements, ( Semba, 2003 ) thereby reducing risk for depression.

Methods Participants Our study utilized the United States’ NHANES 2017 to 2020 dataset. The NHANES, conducted by the National Center for Health Statistics (NCHS) on 2-year cycles, assesses health and nutritional status of the United States population. The NHANES constitutes a nationally representative sample of non-institutionalized civilians selected by a stratified multistage probability sampling using phone interview and physical examination. The NHANES phone interview includes sociodemographic, dietary, and health related questions. At the end of the phone interview, a physical examination is scheduled at a NHANES Mobile Exam Center. The physical examination consists of medical, dental, and physiological measurements, as well as laboratory tests. We anticipated utilizing the NHANES 2017–2018 and 2019–2020 datasets. However, due to the COVID-19 pandemic, field operations were suspended in March 2020. Therefore, data from 2019 to March 2020 were used to represent the 2019–2020 dataset. A total of 15,560 survey of all ages (birth to 80 years or more) were completed, with an interview response rate of 51 % and physical examination response rate of 46.9 %. Of these, there were 1,281 young adults aged 18 to 25 years. Since ferritin, a protein that stores iron, is one of the acute phase reactants accompanied by an inflammatory process, ( Vahdat Shariatpanaahi et al., 2007 ) our study excluded 293 individuals who had inflammatory disorders, which were defined as having serum total folate less than 4 ng/mL (number; n = 4) or high-sensitive C-reactive protein more than 10 mg/L (n = 289). We also excluded pregnant individuals (n = 16). Fifty-five participants were excluded due to not having data on depression. Thus, 917 participants were included in our analysis. We received exemption for this study from the University Institutional Review Board since our study is a secondary analysis of the NHANES dataset. Study variables and measures Exposures Ferritin ( g/L), frozen serum iron ( g/L), and transferrin saturation (%) were measured as indicators of iron status. Ferritin is a protein that contains iron found in red blood cells and primarily used in evaluating iron metabolism and iron storage deficiency in the body. ( Wick et al., 1995 ) Fresh serum or plasma were collected for analysis and measured by the Roche Cobas® e601 assay with total duration time of 18 min. After processing, ferritin specimens were stored at frozen temperature and protected from light. ( National Health and Nutrition Examination Survey, n.d. a , National Health and Nutrition Examination Survey, n.d. b ) Serum iron is circulating iron that is bound to transferrin (90 %) and serum ferritin (10 %). A three step process of the Roche method was used to measure frozen serum iron and transferrin saturation. ( National Health and Nutrition Examination Survey, n.d. a , National Health and Nutrition Examination Survey, n.d. b ) Specimens of frozen serum iron and transferrin saturation were stored at frozen or refrigerated temperatures, respectively. ( National Health and Nutrition Examination Survey, n.d. a , National Health and Nutrition Examination Survey, n.d. b ) Both frozen serum iron and transferrin saturation are useful in the measurement of iron deficiency along with ferritin, ( National Health and Nutrition Examination Survey, n.d. a , National Health and Nutrition Examination Survey, n.d. b ) therefore all three measurements were included in our study. Young adults who had ferritin levels of less than 30 g/L ( Camaschella, 2015 ), frozen serum iron levels of less than 60 g/L, ( Wu et al., 2002 ) or transferrin saturation level of less than 16 % ( Camaschella, 2015 ) were defined as iron deficient. Outcomes Depressive symptoms over the past two weeks were assessed using the nine-item Patient Health Questionnaire-9 (PHQ-9). ( Spitzer et al., 1999 ) Depressive symptoms consist of: 1) having little interest in doing things, 2) feeling down, depressed, or hopeless; 3) trouble sleeping or sleeping too much; 4) feeling tired or having little energy; 5) poor appetite or overeating; 6) feeling bad about yourself; 7) trouble concentrating on things; 8) moving or speaking slowly or too fast; 9) thoughts you would be better off dead. The PHQ-9 scale has been validated, with good sensitivity (0.90) and specificity (0.99) in detecting depression in young adults. ( Adewuya et al., 2006 ) Each response ranged from 0 for ‘not at all’ to 3 for ‘nearly every day,’ and a total score ranged from 0 to 27, with higher values indicative of more depressive symptoms. Depression was defined as having a PHQ-9 score of 10 or more based on the Diagnostic Statistical Manual of Mental Disorders, fourth edition criteria. ( Adewuya et al., 2006 ). Confounders Confounders included sex (male or female), age, race/ethnicity (non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, Hispanic, and other), ratio of family income to poverty, mental health professional, body mass index (BMI) categories, physical activity (inadequate and adequate), dietary iron intake, and dietary supplements. The ratio of family income to poverty was an indicator of income established by the Department of Health and Human Services to represent the ratio of household income to the poverty guidelines, after controlling for family size and inflation. ( U.S. Department of Health and Human Services, 2021 ) It was calculated by dividing family income by the poverty guidelines for the survey years and the score ranged from 1 to 5 with a lower score indicating lower socioeconomic status. BMI categories were defined as underweight (<18.5 kg/m 2 ), healthy weight (18.5 to < 25 kg/m 2 ), overweight (25 to < 30 kg/m 2 ), and obese (>30 kg/m 2 ). Physical activity was measured by the Global Physical Activity Questionnaire (GPAQ). ( Armstrong and Bull, 2006 ) Moderate physical activity was defined as an activity that causes small increases in breathing or heart rate such as brisk walking or carrying light loads for at least 10 min continuously. Moderate physical activity also included going on a walk and using a bicycle. ( Schuna et al., 2013 ) Vigorous physical activity was defined as an activity that causes large increases in breathing or heart rate like carrying or lifting heavy loads, digging or construction work for at least 10 min continuously. Respondents were asked to indicate the frequency (“In a typical week, on how many days do vigorous-intensity activities?”) and the duration of vigorous activity (“How much time spend doing vigorous-intensity?”) during work (e.g., paid or unpaid work, household chores, and yard work) and during recreational activities (e.g., sports, fitness, and recreational activities). In our study, total time spent in physical activity was calculated by multiplying the duration and frequency of both moderate and vigorous activity, where two minutes of vigorous activity was the equivalent to one minute of moderate activity. Based on the current physical activity guidelines, respondents were grouped as ‘inadequate’ (<150 min/week of physical activity) and 'adequate’ (>=150 min/week of physical activity). ( U.S. Department of Health and Human Services, 2018 ) Mental health professional was derived from the question “Have you ever seen a mental health professional in the past 1 year?” with ‘yes’ and ‘no’ responses. Dietary information was obtained through two 24-hour dietary recall interviews, the first was in-person in the Mobile Examination Center and the second interview was by telephone 3 to 10 days later. Dietary iron (mg) intake was calculated from nutrients obtained from foods, beverages, and water. Our study calculated dietary iron intake based on the average intake across the two interviews; if the participant had only one record, then this record was used for analysis. Dietary supplements were derived from the question “Any dietary supplements taken in the past 24 h” with ‘yes’ and ‘no’ responses. Data analysis Descriptive statistics, such as frequencies, weighted percentages (%), weighted mean (m), and weighted standard deviations (SD), were used to describe the exposure, outcome, and demographic variables. We used weighted multivariable Poisson regression models to examine the associations between the three iron deficiency measures (ferritin, serum iron, and transferrin saturation) and depressive symptoms. We adjusted for characteristics such as age, sex, race/ethnicity, ratio of family income to poverty, mental health professional, BMI categories, physical activity, dietary iron intake, and dietary supplements. The results were reported as beta-coefficients ( β) with 95 % confidence intervals (CIs). Next, we used weighted multivariable logistic regression models to examine the associations between the three iron deficiency measures (ferritin, serum iron, and transferrin) and depression. We adjusted for characteristics such as age, sex, race/ethnicity, ratio of family income to poverty, mental health professional, BMI categories, physical activity, dietary iron intake, and dietary supplements. The results were reported as odds ratios (OR) with 95 % CIs. Data was not transformed to correct for the distribution of the dependent variable, depressive symptoms, since we have a large dataset and linear regression is fairly robust to the violation of normality assumption with a large sample size. ( Hoffman, 2003 ) Data analysis was performed using STATA version 16.0 (Stata Cooperation, College station, TX).

Results Of the 987 young adults between 18 and 25 years of age, 917 had complete data on iron deficiency and PHQ-9 depressive symptoms and depression. The study sample characteristics and weighted prevalence of PHQ-9 depression by personal characteristics are shown in Table 1 . Of the participants included in our study, 9.8 % reported depression, with a weighted prevalence of 9.5 % using the PHQ-9 survey. While there were more males (54 %) than females (46 %) in the sample, 13.5 % of females (12.5 % weighted) and 6.6 % of males (6.8 % weighted) had PHQ-9 depression. Using the three measures of iron deficiency, more depressed young adults, compared to non-depressed, had serum iron and transferrin deficiencies, although the differences were not statistically different. More obese participants were depressed (38 %) compared to not depressed (29 %). More depressed participants identified as having seen a mental health professional in the past year (31.4 %) compared to the non-depressed participants (12.1 %). Additionally, while there were no differences among dietary supplements among the depressed and non-depressed group, depressed participants had lower levels of iron intake through foods compared to the depressed group. Supplementary Table S1 shows the characteristics of young adults by sex. There were more females that met the criteria for the three measurements of iron deficiency: ferritin (168 females, 13 males), serum iron (154 females, 49 males), and transferrin saturation (120 females, 22 males). The mean depressive symptom score for the sample was 3.5 (SD = 4.3), with males having a mean score of 2.9 (SD = 3.9) and females having a mean score of 4.1 (SD = 4.7). While more males were overweight or obese, more females were underweight. Lastly, more females engaged in sufficient physical activity, had less dietary iron intake, and took dietary supplements. Table 2 shows that serum iron deficiency ( β = 0.22, p-value < 0.01) and transferrin deficiency ( β = 0.33, p-value < 0.001) were associated with more PHQ-9 depressive symptoms after adjusting for sex, age, race/ethnicity, ratio of family income to poverty, mental health professional, BMI, physical activity, dietary iron intake, and dietary supplements. In similarly adjusted models for males and females, only transferrin deficiency was associated with more depressive symptoms among males ( β = 0.88, p-value < 0.001). Table 3 shows the results of the associations of the three iron deficiency measurements with PHQ-9 depression while adjusting for sex, age, race/ethnicity, ratio of family income to poverty, mental health professional, BMI, physical activity, dietary iron intake, and dietary supplements. There were no associations in the overall models with all young adults (both males and females). In the adjusted model for males, serum iron deficiency (OR = 4.84, p-value < 0.01) and transferrin deficiency (OR = 13.79, p-value < 0.001), compared to no deficiency, were associated with a higher odds of PHQ-9 depression. In the adjusted model for females, compared to no ferritin deficiency was associated with a lower odds of PHQ-9 depression (OR = 0.34, p-value < 0.01).

Discussion To the best of our knowledge, this is the first study to include three measurements of iron deficiency to examine their associations with young adult depressive symptoms and depression as well as examine sex differences. Our study showed that about 9.5 % of young adults reported at least moderate depression (PHQ-9 score >=10) in this NHANES 2017–2020 cohort. Mean depressive symptoms scores were low, not meeting the threshold for mild depression (PHQ-9 score >= 5), ( Kroenke et al., 2001 ) although the range of depressive symptom scores varied from 0 to 27. Additionally, more females reported having PHQ-9 depression and having more depressive symptoms. Overall, among all three measurements of iron, more females had ferritin, serum iron, and transferrin saturation deficiencies. While both serum iron deficiency and transferrin deficiency were associated with a higher risk of PHQ-9 depressive symptoms for young adults, only males with transferrin deficiency had a higher PHQ-9 depression score compared to no deficiency, with a 0.88 point difference in scores. On the other hand, while there were no associations between the iron measures and PHQ-9 depression for young adults as a whole, males with ferritin, serum iron, and transferrin deficiencies, compared to no deficiency, were at 14.13, 4.84, and 13.79 times higher odds of having PHQ-9 depression, respectively, whereas females with ferritin deficiency had a lower odds (OR = 0.34) of PHQ-9 depression. Our findings may be an underestimation of the prevalence of depression (9.5 %) in this age group, which is less than the estimated 17 % of young adults who reported having a depressive episode in the past year according to the National Institutes of Mental Health (NIMH) in 2020. ( National Institute of Mental Health, 2022 ) While the PHQ-9 questionnaire is well-validated among several populations, including this young adult population, ( Adewuya et al., 2006 , Spitzer et al., 1999 ) reporting bias may still exist due to the sensitive nature of mental and emotional health questions. Additionally, the measurement of depression in our study was self-reported, whereas a clinician diagnosis is often preferred. Nonetheless, in large epidemiological studies, clinician diagnosis and report are not feasible. Further, information on mental health professionals indicated that non-depressed participants saw a mental health professional in the past year, indicating the possibility that depressive symptoms may be under good control, thus the PHQ-9 questionnaire did not fully capture these participants since the questionnaire asks about symptoms within the past two weeks. Nonetheless, depression screening with a validated tool, such as the PHQ-9, should be considered at primary care settings due to ease of administration and scoring as well as per the recommendation of the US Preventive Services Task Force. ( U.S. Preventive Services Task Force, 2023 ) Consistent with the literature regarding sex differences, ( Herreen et al., 2022 , Salk et al., 2017 ) the prevalence of depression among young adult females (12.5 %) was nearly double that of young adult males (6.8 %), highlighting the need to determine and clarify risk factors, whether, biological, psychological or social, associated with depression stratified by sex. Similarly, females compared to males in this sample had higher rates of iron deficiency, supporting that females during the reproductive age have lower iron levels likely due to menstruation, especially when females have heavy menstrual bleeding. ( Munro et al., 2023 ) Other conditions affecting iron levels include pregnancy and childbirth, but our study excluded pregnant women. Our findings that males with transferrin deficiency had higher depressive symptom scores and males with ferritin, serum iron, and transferrin deficiency had a higher risk of depression are similar to findings of Yi et al. (2011). ( Yi et al., 2011 ) Yi and colleagues reported no significant associations among women (mean age 41.6) but found that men (mean age 44.2) with lower serum ferritin concentrations had more depressive symptoms. ( Yi et al., 2011 ) On the other hand, other studies consisting of pre-menopausal women ( Hunt and Penland, 1999 ) and older adults 65–83 years of age ( Baune et al., 2006 ) reported no significant associations between iron deficiency and depression. Of the few studies examining depressive symptoms, the age of participants ranged from young adults to middle or older adults, making comparisons difficult. Nonetheless, similar to our findings, but with a focus on females, Noorazar et al.’s (2015) study examining depressive symptoms among females with a Major Depressive Disorder reported no associations between serum ferritin deficiency and depressive symptoms. ( Noorazar et al., 2015 ) Our study found that, contrary to the literature, ( Vahdat Shariatpanaahi et al., 2007 ) females with ferritin deficiency had a lower risk of depression. However, our findings with gender differences may have been due to chance since there were a small number of females with PHQ-9 depression as well as a small number of males meeting criteria for both iron deficiency and PHQ-9 depression. Our findings with the association between the three measures of iron deficiency and depression may differ due to the different functions of ferritin, serum iron, and transferrin. Per our findings, it is a possibility that ferritin had the most influence on depression when stratified by sex due to our significant associations among both males and females. While the underlying mechanisms are still unknown, iron may play a role in neurotransmission, where iron has been hypothesized to synthesize the neurotransmitters dopamine, serotonin, and norepinephrine, all of which have been implicated in depression. ( Berthou et al., 2022 ) Iron has also been implicated in cytokine-mediated neuroinflammation, which leads to the dysregulation of neurotransmitters. ( Berthou et al., 2022 ). While our study has many strengths, it nonetheless has limitations. First, our study was cross-sectional therefore does not allow for causal inference. Second, our study consisted of participants with few depressive symptoms and a lower prevalence of depression compared to the NIMH estimates; ( National Institute of Mental Health, 2022 ) therefore, our study findings cannot be generalized to participants with more severe depressive symptoms. However, the PHQ-9 questionnaire provides us with information on symptomology. While the participants did not meet criteria for moderate depression or mild depression, most participants had some symptoms, reiterating the need to screen and follow participants closely at the primary care level. Third, our study relied on self-reports which can lead to recall and desirability bias, which may influence our findings. Fourth, although our study excluded those with inflammation based on CRP levels and controlled for several confounders, residual confounding may still exist. We cannot preclude that all participants with inflammatory disorders, such as metabolic syndrome and diabetes mellitus, were excluded. Fifth, we neither excluded nor adjusted for participants who were on anti-depressant medications since only the 2017–2018 dataset, but not the 2019–2020 dataset, had this information. Nonetheless, only one participant reported anti-depressant use. While we adjusted for dietary supplements where 34 % of our adolescents reported taking supplements, we did not have any information on iron supplements. Including nuanced information about supplements and diet is necessary for the research, management, and treatment of iron deficiency as well as further research on depression. Sixth, our significant findings with wide ORs and 95 % CIs may be inflated due to the small sample of participants with depression and in each of the iron deficiency groups, thus additional research focused on those with depression and stratified by sex should be considered.

Conclusion Our study provides evidence that iron deficiency is associated with young adult depression when stratified by sex. Our study findings indicate that young adult females with ferritin deficiency have lower odds of depression while young adult males with ferritin, serum iron, and transferrin deficiencies have higher odds of depression. Additionally, we found that young adult females have a higher prevalence of both iron deficiency and depression. It is crucial that future studies examine depression stratified by sex and the differing risk factors, including biological, behavioral, and social determinants. Additionally, our study highlights the need for more resources allocated to young adults, especially females, to identify both iron deficiency and other modifiable nutritional risk factors, including supplement intake, related to depression, as iron deficiency can be prevented with adequate nutritional intake. Lastly, young adults regardless of low iron levels should be considered for depression screening.

Depression is one of the most prevalent mental health conditions throughout the lifespan. Notable differences in the prevalence of depression among females and males arise during adolescence and may peak during young adulthood. Since iron deficiency is a treatable condition that may contribute to depression, this topic among youth (18 to 25 years of age) needs to be further explored. Thus, our study examines the associations between three measures of iron (ferritin, serum iron, and transferrin saturation levels) with Patient Health Questionnaire-9 (PHQ-9) depressive symptoms and depression among young adult males and females using the National Health and Nutrition Examination Survey (NHANES) 2017–2020. Using multivariable Poisson and logistic regression models, adjusting for several demographic and clinical variables, we report 1) the prevalence of depression and 2) the associations between iron deficiency and depressive symptoms and depression among males and females. 917 participants were included in our study. More females (12.5 %) than males (6.8 %) had PHQ-9 depression. Males with ferritin (adjusted odds ratio [AOR] = 14.13, 95 % confidence interval [CI]: 1.51, 132.21), serum iron (AOR = 4.84, 95 % CI: 1.02, 22.92), and transferrin (AOR = 13.79, 95 % CI: 3.59, 53.06) deficiencies were at higher risk for depression, while females with ferritin deficiency (AOR = 0.34, 95 % CI: 0.11, 0.97) had a lower risk for depression. Our study highlights the need to focus on depression screening among young adults as well as risk factors for depression among this age group. Identifying risk factors and screening for iron deficiency, especially among females, should be considered as well. Keywords

Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Supplementary data The following are the Supplementary data to this article: Data availability I have shared the link to the NHANES dataset. Acknowledgements This work was supported by the 10.13039/100000002 National Institutes of Health 10.13039/100000056 National Institute of Nursing Research, United States [5K01NR017207].

CC BY

no

2024-01-16 23:41:58

Prev Med Rep. 2023 Dec 12; 37:102549

oa_package/16/bb/PMC10788288.tar.gz

PMC10788296

38104652

Introduction Human skeletal muscle differentiates into two major fiber types based on their myosin heavy chain content, each with highly specialized contractile and metabolic profiles. Slow-twitch type I fibers have a slow peak shortening velocity, a low recruitment threshold, and a reliance on oxidative metabolism due to their high mitochondrial content. In contrast, fast-twitch type II fibers have faster shortening velocity and a higher recruitment threshold. However, they fatigue more easily in part due to protein allocation favoring glycolytic enzymes over mitochondrial content to enable higher catalytic capacity [ 1 , 2 ]. Investigations on the mitochondria of the oxidative and glycolytic fibers in mostly animal models have revealed some differences in intrinsic fat oxidation, reactive oxygen species (ROS) production, and mitochondrial calcium handling [ 3 ]. Subsequent work by Mishra et al. [ 4 ] showed a more fused mitochondrial network in oxidative (IIA) versus glycolytic (IIX/IIB) fibers. This was followed by Bleck et al. [ 5 ] who used 3D renderings to show large differences in mitochondrial network connectivity, intermyofibrillar location, and linked muscle fiber types to inter-mitochondrial junction morphology, which was previously suggested to alter mitochondrial cristae density [ 6 ]. Most of the data on fiber type-specific mitochondrial form or function are generated in animal models comparing oxidative to glycolytic muscle fibers [ 4 , 5 , 7 ]. Type IIA fibers generally tend to fall under the oxidative category together with type I fibers in animal models of metabolism, whereas investigations in human muscle, by contrast, more often categorize muscle according to their twitch speed, thus separating type I from type IIA and IIX, i.e. slow-, versus fast-twitch fibers [ [8] , [9] , [10] ]. This makes a translation from animal models to human skeletal muscle difficult as humans lack the glycolytic type IIB fibers and the population of intermediate type IIX fibers is highly limited [ 11 , 12 ]. Adding further complexity to the translation, when compared, complex II levels are significantly greater in type IIA fibers compared to type I in rat hindlimbs [ 13 , 14 ], with respiration showing similar tendencies [ 15 ]. Thus, inferring large differences found between oxidative and glycolytic muscle fibers in animals onto human slow- and fast-twitch fibers may be inappropriate. Unfortunately, due to the laborious nature of single muscle fiber analysis, fiber type-specific findings in animal models are seldom further investigated in human muscle. Recent analysis of the 3D characteristics of the mitochondrial network within human skeletal muscle has indicated two distinct morphologies of intermyofibrillar mitochondria: a high-volume, elongated network mostly aligned parallel to the myofibrils, and a low-volume, fragmented transverse network [ 16 ]. We suspect these two mitochondrial populations are a consequence of the different metabolic demands of the two main fiber types. To investigate this, we performed fiber type-specific single muscle fiber analysis on intrinsic molecular markers regulating mitochondrial structure-, and ultrastructure as well as oxidative phosphorylation (OXPHOS) enzyme content. Furthermore, using the novel THRIFTY technique developed in our lab [ 17 , 18 ], we were able to rapidly identify fiber types in individually dissected muscle fibers and for the first time, assess mitochondrial respiration in fiber-type specific pools from human skeletal muscle. This approach coupled with single fiber analysis of molecular markers allowed us to provide novel insights into muscle fiber type-specific mitochondrial function in human skeletal muscle.

Methods Ethics statement This study was conducted in accordance with the declaration of Helsinki. The study was approved by the regional ethics committee in Stockholm (Dnr: 2017/2034-31/2 and Dnr: 2021-00364). All participants were volunteers who gave their written and oral consent prior to enrolment. Participant characteristics and included muscle fibers Seven healthy participants were recruited for the study, four women and three men (age 28 ± 3 years, height 171 ± 8 cm, BMI 24 ± 1.7 kg/m 2 , and VO 2 max 45 ± 4 ml/kg/min). A median of 48 (range 15–52) slow-twitch (type I) and 31 (21–54) fast-twitch (type II) fibers were included per participant in the respiratory measurement following fiber typing, with 285 slow-twitch fibers and 232 fast-twitch fibers included across all participants. Muscle biopsies All muscle biopsies were collected in a fasted state between 9- and 12 am. Biopsies were taken from the vastus lateralis muscle under local anesthesia (2 % Carbocain, AstraZeneca, Södertälje, Sweden) using a 5 mm Bergström needle with manually applied suction [ 19 ]. Each sample was immediately blotted free of excess blood and divided into two pieces, one quickly snap-frozen in liquid nitrogen, freeze-dried, and processed as described under the section Single muscle fiber analysis . The second half of the muscle biopsy was processed as described below under the sections Muscle fiber typing and Fiber type-specific respirometry . Muscle fiber typing Immediately following biopsy sampling, the muscle sample was placed in a petri dish filled with ice-cooled biopsy preservation media (BIOPS; 10 mM Ca-EGTA, 0.1 μM free Ca, 20 mM imidazole, 20 mM taurine, 50 mM K-MES, 0.5 mM Dithiothreitol, 6.56 MgCl 2 , 5.77 mM ATP, 15 mM phosphocreatine, pH 7.1). The petri dish was placed on a bed of ice under a stereomicroscope and 80 individual muscle fibers were dissected out of the sample and placed in individual droplets of BIOPS allocated in a gridded plastic container. The dissected muscle fibers were fiber typed according to our recently developed THRIFTY protocol [ 17 , 18 ]. Briefly, an end of each muscle fiber (≈0.5 mm) was cut off and placed on a microscope slide pre-printed with a coordinate grid system. Fiber ends were stained for myosin heavy chain (MHC) I & II (BA-F8, SC-71; specifications for solutions in Supplementary Table 1 ) and identification was performed in a fluorescent microscope. The remaining sections of the fibers were subsequently pooled according to their fiber type. The whole procedure, from biopsy sampling to typed and pooled fibers, took approximately 5–6 h, during which fibers were constantly kept in ice-cooled BIOPS, which is sufficiently fast to not significantly affect respiratory output of the fibers [ 20 ]. Fiber type-specific respirometry The slow- and fast-twitch muscle fiber pools were permeabilized for 15 min in saponin solution (50 μg/1 ml BIOPS) and subsequently washed in BIOPS before measurement. Mitochondrial respiration and H 2 O 2 emission were measured using a two-channel high-resolution respirometer with an attached fluorescent probe (Oxygraph-2k, Oroboros Instruments Corporation, Innsbruck, Austria). Pools of muscle fibers were placed into the 2 ml wells containing respiration medium MIR05 (EGTA 0.5 mM, MgCl 2 6H 2 O 3 mM, K-lactobionate 60 mM, Taurine 20 mM, KH 2 PO 4 10 mM, HEPES 20 mM, Sucrose 110 mM, and BSA 1g L −1 ). The type I and & II fiber pools were loaded onto different chambers in a randomized order and the same protocol was carried out simultaneously in both chambers; 0.2 mM octanoyl carnitine + 0.5 mM malate (fat leak), 2.5 mM ADP (fat respiration), 10 mM glutamate + 5 mM pyruvate (Complex I), 10 mM succinate (Complex I + II), 0.5 μM rotenone (Complex II), 10 μM cytochrome C (membrane integrity), and 0.05 μM FCCP titration (uncoupled respiration). All experiments were performed at 37 °C and prior O 2 calibration was completed per the manufacturers' instructions. Following respirometry, the entirety of the 2 ml chambers was collected to determine protein loading control and to measure specific proteins). The wells were rinsed in dH 2 O approximately ten times to make sure all muscle fibers were collected. The collected material was stored in falcon tubes at −80 °C until further analysis. All respiratory data collected are related to calculated wet weight based on Pan actin immunostaining or mitochondrial content calculated from VDAC1+2, TOMM20, CIII & CIV immunostaining, as described below. All measurements and analyses were performed in DatLab 5.2 software (Oroboros, Paar, Graz, Austria). The addition of cytochrome c increased respiration by 5- to 8 % in type I and type II fibers (p < 0.05), indicating only minor damage to the mitochondrial membrane. No difference in the increase in respiratory rates between the fiber types was observed following cytochrome c addition. Homogenization of fiber pools Fiber samples collected from the wells of the Oxygraph-2k were centrifuged at 4 °C for 10 min at 3 000 g. Most of the supernatant was removed from the sample, and approximately 1 ml was left in each of the falcon tubes. The 1 ml solution was transferred into a microtube and spun at 4 °C for 15 min at 16 000 g. The supernatant was carefully removed, and the pelleted fibers were rinsed in 500 μl of dH2O. The centrifugation and washing procedure were repeated once. ZrO 0.5 mm beads were then added to the sample, followed by 200 μl of ice-cooled homogenization buffer (2 mM HEPES pH 7.4, 1 mM EDTA, 5 mM EGTA, 10 mM MgCl 2 , 50 mM B-glycerophosphate, 1 % Triton X, 1 % Phosphatase Inhibitor Cocktail 3 (Sigma–Aldrich P0044) and 1 % Halt Protease Inhibitor Cocktail (Thermo Scientific, Rockford IL)). The samples were immediately processed in a BulletBlender (NextAdvance, Averill Park, NY) until completely homogenized. Samples were then left to shake for 30 min and subsequently rotated for an additional 30 min. Homogenization, shaking, and rotation was carried out in a cold room at 4 °C. Samples were then diluted in 4× Laemmli sample buffer and denatured at 95 °C for 5 min before storage at −30 °C until immunoblotting. Calibration curve for calculation of wet weight and protein quantification To calculate the wet weight of each fiber pool, a calibration curve was prepared from freeze-dried muscle samples from five participants. The samples were manually dissected free from blood and connective tissue under a stereo microscope, pooled and subsequently weighed on an ultra-micro balance with readability of 0.1 μg (Cubis® MCA2.7S – 2S00 – M, Sartorius Lab Instruments GmbH & Co, Göttingen, Germany). The pooled sample was homogenized in 1 μl buffer per 1 μg sample according to the protocol described below and serially diluted five times. The calibration curve was loaded onto each gel together with the pooled fiber samples recovered from the respirometry chambers. The calibration curve yielded near-perfect linearity for actin immunostaining (R 2 = 0.9999–0.9997, Supplemental Figure 1 ). Thus, the actin immunostaining from each sample was used to calculate the sample wet weight [ 21 ]. A dry-to-wet weight ratio of 1⁄4 was used for all samples. Immunoblotting of fiber pools Together with the calibration curve, samples were loaded onto 18-well Criterion TGX gradient gels (4–20% acrylamide or AnykD; BioRad). Electrophoresis was performed for 28 min at 300 V in electrophoresis buffer (25 mM Tris base, 192 mM glycine, and 3.5 mM SDS) on a bed of ice in a cold room at 4 °C. The gels were then equilibrated in transfer buffer (25 mM Tris base, 192 mM glycine, and 10% methanol) for 30 min at 4 °C. Transfer of proteins to polyvinylidene difluoride membranes (BioRad) was carried out at 300 mA for 3 h at 4 °C. Membranes were stained with MemCode Reversible Protein Stain Kit (Thermo Scientific) to confirm successful transfer and to aid in sectioning the membrane before primary antibody incubation. After membrane sectioning and destaining, blocking was carried out for 1 h using 5 % milk in Tris-buffered saline with Tween (TBST; 20 mM Tris base, 137 mM NaCl, and 0.1% tween). Membranes were subsequently washed 3 × 3 min in TBST prior to overnight incubations with primary antibodies. Membranes were stained for MHC I and MHC II to check fiber pool purity, Pan actin for calculation of wet weight and VDAC1+2, TOMM20, CIII and CIV for mitochondrial volume ( Figure 1 C). Antibody information is presented in Supplemental Table 1 . Following overnight incubation, membranes were rewashed 3 × 5 min in TBST, followed by 1 h of secondary antibody incubation at room temperature. Again, membranes were washed 3 × 5 min in TBST and then finally incubated in SuperSignal West Femto Maximum Sensitivity Substrate (Thermo Scientific) before visualization in a ChemiDoc MP with Quantity One software (Version 6.0.1; Biorad). Blots were quantified in Image Lab (Version 6.0.1; Biorad). Calculation of mitochondrial loading control The mito-control for respiratory measurements ( Figure 1 B,C) were calculated as an average of proteins VDAC1+2, TOMM20, CIII and CIV (antibodies specified in Supplemental Table 1 ) obtained during western blotting. These staining's were done on the same fibers that were used in the Oroboros O2k chamber. A ratio between slow- and fast twitch fiber expression of each protein, where slow-twitch fiber expression was set to 1 ( Figure 1 C) was initially created. Next, an average ratio of the four proteins for each volunteer was calculated and used as mitochondrial loading control for the respiratory rates. Likewise, for the single muscle fiber western blots on lyophilized muscle fibers, a mito-control was created for each individual fiber as an average of VDAC1+2, CIII and CIV. TOMM20 was not stained for on these fibers since both CI (22 kDa), FIS1 (17 kDa), and SOD2 (22 kDa) were prioritized on the PVDF-membrane of each fiber, making an additional TOMM20 staining (16 kDa) difficult. While we recognize that choosing protein markers to match mitochondrial volume across different cell types may be difficult, we had a couple of considerations in mind prior to choosing staining protocol for the western blot. First and foremost, we wanted membrane bound mitochondrial markers, as we were uncertain of how running respiratory measurements as well as homogenization of fibers would affect non-membrane bound protein leakage from the mitochondrial matrix. Since the fibers recovered from the respiratory chamber were incubated in MIR05 containing BSA, the fibers had to be carefully washed so that the high concentration of BSA would not interfere with the subsequent western blot procedures, thus risking some non-membrane proteins to be washed from the sample. To increase the reliability of the data, an average of four (fibers from respirometry chamber) or three (single fiber analysis) membrane bound proteins were chosen as loading control. Proteins located on both the inner and outer mitochondrial membrane were chosen to account for variations in mitochondrial outer membrane volume and cristae density. Proteins within CIII and CIV were chosen as they linearly couple the redox reaction from CoQ regardless of where electrons enter the ETC. Thus, CIII and CIV should be dimensioned evenly within mitochondria due to their co-dependance during respiration. Moreover, both CIII and CIV have been shown to be fair predictors of mitochondrial volume and cristae area (r ≈ 0.6, respectively) [ 22 ]. As for protein markers of mitochondrial volume on the outer mitochondrial membrane, a myriad of previously published papers use either TOMM20 or VDAC1+2 as markers for OMM quantity or mitochondrial volume. Both markers have also been shown to parallel total mitochondrial protein levels [ 23 ]. Thus, we choose these proteins to keep the mitochondrial loading control somewhat consistent with a large portion of the literature that utilizes the content of a specific protein as a mitochondrial marker. Importantly, all proteins showed similar patterns of expression which lends further confidence to an accurate estimation of mitochondrial content in the two fiber types. Single muscle fiber western blot Freeze-dried samples were manually dissected under a stereomicroscope. Approximately 50–100 fibers were dissected from each biopsy sample and subsequently typed according to the THRIFTY protocol [ 17 , 18 ]. Ten fibers of each type from every volunteer were homogenized individually in 20 μl of ice-cooled homogenization buffer with 1× Laemmli sample buffer. The entirety of each sample was then loaded onto each well of 26-well Criterion TGX gradient gels (AnykD; BioRad), separated and transferred to PVDF membranes as described above. After blocking, antibody incubation and visualization, membranes were stripped using Restore PLUS Western Blot Stripping Buffer (Thermo Scientific) for 45 min at room temperature for staining of additional protein targets. All antibody information is compiled in Supplemental Table 1 . One type I fiber was excluded from analysis due to a slight MHC II cross-contamination. The first fully analyzed volunteer was not stained for OPA1 and MIC60. One participant also had MFN2, MIC60, and FIS1 stained on separate fibers compared to the rest of the markers. For these fibers, control proteins (MHC I, MHC II, Pan actin and VDAC 1 + 2) were stained, but not CIII and CIV. Thus, based on the previous 20 fibers from this volunteer, a control ratio was used to estimate Mito-control based on VDAC1+2 staining alone. Statistics Values are presented as mean ± SD for respirometry data and as individual values for single fiber data. Normality was tested with Shapiro–Wilks. Respirometry data were analyzed using paired samples t-test and single fiber data were analyzed with a Mann–Whitney U test. Correlations were calculated with Pearson correlation coefficient.

Results Total- and intrinsic mitochondrial respiratory capacity in human skeletal muscle fiber types Previous investigations on mitochondrial respiratory properties in the different muscle fiber types have all used various animal models comparing muscle predominantly expressing oxidative or glycolytic fibers [ 3 , 24 ]. Here, we succeeded in measuring respiration in completely pure pools of permeabilized human slow- and fast-twitch muscle fibers in seven healthy volunteers. As expected, full ADP stimulated respiration, i.e. fatty acid oxidation (FAO) plus OXPHOS capacity (substrate; octanoyl carnitine, malate, glutamate, pyruvate, and succinate) was higher in the slow-twitch fibers at 59 pmol x s −1 x mg −1 compared to the observed respiratory rate of 47 pmol x s −1 x mg −1 in the fast-twitch fibers, Figure 1 A. Following the high resolution respirometry in the Oxygraph, the fibers analyzed in the respiratory chamber were recovered and analyzed for mitochondrial volume control proteins (henceforth referred to as mito-control; described in more detail in the methods section). The mito-control proteins include proteins on the outer- (voltage-dependent anion channel 1 & 2; VDAC1+2, translocase of the outer mitochondrial membrane complex subunit 20; TOMM20) and inner mitochondrial membrane (ubiquinol-cytochrome C oxidoreductase; CIII, cytochrome C oxidase; CIV). These proteins jointly indicated an approximately three-fold higher mitochondrial volume in the slow-twitch fibers ( Figure 1 B,C). This large difference in mitochondrial volume is consistent with a previous report using focused ion beam-scanning electron microscopy [ 16 ] in humans, however, others before that have indicated smaller fiber type differences with approximately 60% higher mitochondrial volume in slow-twitch fibers using conventional electron microscopy [ 25 ]. Three of the mito-control proteins (VDAC1+2, CIII, & CIV) were again stained for during subsequent immunoblotting of lyophilized single muscle fibers from the same volunteers ( Figure 1 E–G) to obtain a mitochondrial loading control from single fiber protein analysis (Data presented in Figure 2 , Figure 3 ). The lyophilized fibers revealed a similar pattern of approximately three-fold higher mito-control in the slow-twitch fibers ( Figure 1 E–G) confirming the difference in mitochondrial content in the fibers recovered from the respiratory chambers. Next, utilizing the mito-control presented in Figure 1 B, we investigated the FAO + OXPHOS capacity per mitochondrial volume in the two fiber types. This revealed a significantly greater maximal respiratory rate in the mitochondria of fast-twitch fibers compared to in slow-twitch fibers ( Figure 1 D). Electron entry points to the inner mitochondrial membrane We discovered a 50% higher capacity for intrinsic FAO and OXPHOS in the fast-twitch fiber mitochondria ( Figure 1 D). This finding contrasts prior work in animal models showing similar [ 24 ] or higher [ 7 ] mitochondrial capacity in smaller mammals' red muscle tissue. An explanation for the discrepancy between our finding to that of the literature in animals may be that oxidative ‘red’ muscle in animals often includes a mixture of type IIA and type I fibers, whereas type IIA and type I fibers in this paper, and much of the human literature, are on opposing ends of the comparisons. During our analysis, no difference in the respiratory rates was however observed during states of FAO (substrate; octanoyl carnitine and malate) and complex I capacity (FAO substrates plus glutamate and pyruvate) between the mitochondria of the slow- and fast-twitch fibers ( Figure 2 A,B). This suggests that the higher FAO + OXPHOS capacity in the fast-twitch fibers is primarily due to an elevated capability to metabolize succinate ( Figure 2 C). As for protein expression, the two dehydrogenases of the electron transport chain (ETC), NADH dehydrogenase; complex I (CI), and succinate dehydrogenase; complex II (CII), were differentially expressed in the two fiber types when related both to our mito-control ( Figure 2 D–F) as well as when related to the ETC's electron demanding CIII & CIV protein complexes (data not shown). In reference to the existing literature, the higher CII/CI protein ratio in the fast-twitch fibers is partly reflected in one [ 26 ], but not both [ 2 ] proteomics data sets currently available in fast- and slow-twitch fibers of young individuals. The levels of NADPH oxidase 4 (NOX4) were nearly ten-fold higher in fast-twitch as compared to slow-twitch fibers related to their respective mitochondrial content ( Figure 2 G) and approximately 40% higher in relation to their respective fiber size ( Supplemental Figure 1J ). This may suggest that the fast-twitch fibers have an additional route to alleviate NADH oxidation pressure onto CI. Partly located at the IMM and the sarcoplasmic reticulum in muscle, NOX4 oxidizes NADH and consequently produces O 2 - and H 2 O 2 [ 27 ]. It has recently been suggested that NADH oxidation by NOX4 is the main origin of exercise-induced reactive oxygen species (ROS) in skeletal muscle [ 28 ]. As the deletion of NOX4 has been linked to reductions in antioxidant defenses and reduced superoxide dismutase 2 (SOD2) content [ 28 ], we also investigated SOD2 levels in the two fiber types. However, despite a much lower NOX4 expression in the slow-twitch fibers, we found SOD2 to be equally dimensioned in the mitochondria of slow- and fast-twitch muscle ( Supplemental Figure 3 ), which could suggest the higher NOX4 in fast-twitch fibers may be reflecting cytosolic, rather than mitochondrial content. Mitochondrial membrane dynamics and intrinsic respiratory rate In striking contrast to respiration ( Figure 1 A), OXPHOS enzyme content, and other mitochondrial proteins on a per fiber basis ( Supplemental Figure 2A -L), we found that levels of inner mitochondrial membrane (IMM) fusion and mitochondrial cristae remodeling protein optic atrophy 1 (OPA1) [ 29 ] were significantly greater expressed in fast-twitch fibers ( Supplemental Figure 2F ) despite the three-fold higher mito-control in the slow-twitch fibers ( Figure 1 E). Related to mitochondrial volume, OPA1 expression in fast-twitch fibers was accentuated even more, with levels 10-fold over that found in mitochondria of slow-twitch fibers ( Figure 3 A). Likewise, MIC60, a key component in the mitochondrial inner membrane organizing system (MINOS) [ 30 , 31 ], was also expressed in a fiber type-dependent manner with nearly three-fold higher expression in fast-twitch fiber mitochondria ( Figure 3 B). Additionally, fast-twitch mitochondria expressed double the ATP-synthase (Complex V; CV) levels of mitochondria in slow-twitch fibers ( Figure 3 C). Though the central function of CV is phosphorylation of ADP, it has additional roles in regulating cristae formation. CV exists as a V-shaped dimeric complex. The V-shaped structure of the hydrophobic F 0 membrane-embedded body directly imposes a folding of the IMM, forming the cristae rims [ 32 ]. Taken together, we observed three proteins that have been reported to work separately [ [31] , [32] , [33] ] and synergistically [ [34] , [35] , [36] ] to uphold a tight and energetically efficient cristae structure, which are more abundant in fast-twitch mitochondria. Fast-twitch mitochondria express higher levels of pro-fusion protein mitofusin 2 (MFN2) but not fusion inhibitor protein (mitochondrial fission protein 1; FIS1) ( Figure 3 D,E) [ 37 , 38 ]. We suspect that the higher MFN2 levels in fast-twitch mitochondria to some extent reflect a larger proportion of mitochondria tethered to the sarcoplasmic reticulum located between the myofibrils [ 39 ] to aid in handling the greater cytosolic calcium release seen with contraction in fast-twitch fibers [ 40 ]. This notion is also in line with our finding that CV is more abundant in fast-twitch fiber mitochondria, as intermyofibrillar mitochondria have previously been reported to contain more CV and less CIV compared to subsarcolemmal mitochondria [ 41 ] ( Supplemental Figure 4 ).

Discussion Here we employed a novel technique to measure human muscle fiber type specific mitochondrial respiration, coupled with protein expression analysis of key proteins regulating mitochondrial form and function, to gain insight into mitochondrial specialization in human slow- and fast-twitch muscle fibers. We observed that the mitochondria of the fast-twitch muscle fibers are significantly less abundant, but compensate for their relatively low volume by upregulating crucial proteins regulating mitochondrial bioenergetics and dynamics which ultimately leads to a faster succinate metabolism in the fast-twitch mitochondria. We suspect that a mechanism of elevated CII content in the fast-twitch mitochondria may in part be that the fast-twitch muscle fibers utilize the REDOX regulated bidirectionality of CII [ 42 ]. During exhaustive exercise where local hypoxia occurs, especially in the center of the larger and less perfused fast-twitch myofiber, CII may short-circuit the ETC to oxidize the ubiquinol (CoQH 2 ) pool when oxygen diffusion to the centrally located mitochondria does not suffice. The fast-twitch fibers would thus need an additionally large pool of CII by the sarcolemma to metabolize both the succinate produced by CII reversal in intermyofibrillar mitochondria and for oxidizing succinate during regular substrate metabolism. The hypothesis that oxidative phosphorylation runs, at least in part, more anaerobically the further from the sarcolemma it is located, has support in both O 2 diffusion into the myofiber being disproportionately limited past the first few microns from the sarcolemma [ 43 ], as well as the mitochondrial CV/CIV ratio increasing with the distance from the sarcolemma [ 41 ]. Alternatively, a metabolic state may be occurring at the onset of anaerobic respiration, where the fast- and slow-twitch fibers exchange lactate for succinate to favor their respective ETC composition ( Figure 2 C–F). As lactate yields additional NADH from complete oxidation compared to pyruvate, this may add additional urgency on CI to reduce the ubiquinone (CoQ) pool to CoQH 2 prior to CII involvement. Depending on the severity of the CoQH 2 REDOX pressure within the IMM of the slow-twitch fibers, CII may either halt succinate oxidation or even reverse CII to produce succinate and alleviate some of the CoQH 2 pressure [ 42 ]. In either case, succinate levels will subsequently accumulate and get released into the interstitium of the muscle [ 44 ]. The succinate from slow-twitch fibers could then act as a substrate for fast-twitch fibers tailored towards succinate oxidation. Thus, fast- and slow-twitch fibers might exchange lactate and succinate during heavy exercise to symbiotically favor their respective predisposition towards CI or CII respiration. Regardless of whether succinate is utilized as an electron carrier between intermyofibrillar and subsarcolemmal mitochondria within fast-twitch fibers, or shuttled between fiber types, both would indicate a metabolic strategy that generates ATP faster, but at the cost of a reduced O 2 efficiency once fast-twitch fibers are recruited. Such a mechanism would fit the previously suggested metabolic state of an increased bypass of CI during heavy exercise to prioritize catalytic speed over substrate efficiency prior to any substantial lactate accumulation [ 45 ]. Moreover, our results of increased CII-respiration in the fast-twitch fibers may explain the increased O 2 cost per produced watt during heavy aerobic exercise, causing a non-linear relationship in O 2 consumption during the final minutes of gradient exercise tests to exhaustion [ 46 ]. Protein expression of OPA1, MIC60, and CV, all synergistically orchestrating cristae folding, were significantly more abundant in fast twitch mitochondria ( Figure 3 A–C). A tighter crista folding has been associated with an increased respirasome assembly [ 33 ] and, consequently, an elevated respiratory rate [ 33 , 47 ] which may in part explain the unexpected finding that the mitochondria of fast-twitch muscle fibers have an approximately 50% higher intrinsic respiratory rate ( Figure 1 D). Additionally, less respirasome formation due to a slower cristae structure regulation may also explain why the slow-twitch fibers could not translate the higher CI protein expression ( Figure. 2 D) into an elevated capacity for oxidation of NADH through CI ( Figure 2 B). Moreover, the OPA1 and MINOS-dependent constriction of the cristae junction is specifically suggested to uphold membrane potential over the IMM by preventing proton leakage into the intermembrane space, thus driving proton re-entry to the matrix faster through CV [ 29 , 36 ], possibly affecting the P/O ratio of the different mitochondria ( Figure 3 G). A greater cristae density may be an obvious explanation as to why many of the cristae-upholding proteins are more highly expressed in fast-twitch mitochondria. As space within the muscle cell is a scarce commodity, whereby fast-twitch fibers prioritize organelles related to contraction over mitochondria [ 48 ], cristae density, rather than mitochondrial volume, may be the preferred adaptation to increase muscle oxidative capacity in the fast-twitch fibers specifically. However, the limited available data on fiber type-specific cristae density in human muscle (using z-line width to identify muscle fiber type) does not support this notion [ 49 ]. Since we find both OMM fusion factors (MFN2; Figure 3 D) and IMM fusion- and cristae modulating factors (OPA1 & MIC60; Figure 3 A,B) to be more highly expressed in fast-twitch mitochondria, we suspect this is a compensatory mechanism by which aerobic ATP-production can be rapidly upregulated in the fast-twitch fibers despite their lower mitochondrial abundance compared to slow-twitch fibers. As mitochondria are more abundant in slow-twitch fibers, it should more easily uphold a larger fused grid-like structure [ 4 , 16 ]. A smaller pool of mitochondria, might more readily rely on MFN2 and other OMM fusion proteins to initiate rapid morphological changes upon fast-twitch fiber recruitment during exercise. This may imply either inducing transient mitochondrial membrane interactions without initiating full mitochondrial fusion [ 50 ], which highly influences mitochondrial cristae morphology [ 6 ], or, the formation of nanotunnels [ 51 ] to increase respiratory efficiency [ 52 ]. In conclusion, human fast- and slow-twitch muscle fibers host distinct mitochondrial populations. Slow-twitch mitochondria favor sheer mitochondrial volume and rely on CI for efficient oxidative coupling and NADH oxidation. By contrast, fast-twitch mitochondria are tailored for speed rather than efficiency through faster succinate metabolism and significantly elevated CII expression. The fast-twitch mitochondrial proteome may enable faster and more efficient mitochondrial morphology changes during high energy demands, likely as a compensatory mechanism for low total volume.

Filip J. Larsen and William Apró contributed equally to this work. Objective Human skeletal muscle consists of a mixture of slow- and fast-twitch fibers with distinct capacities for contraction mechanics, fermentation, and oxidative phosphorylation. While the divergence in mitochondrial volume favoring slow-twitch fibers is well established, data on the fiber type-specific intrinsic mitochondrial function and morphology are highly limited with existing data mainly being generated in animal models. This highlights the need for more human data on the topic. Methods Here, we utilized THRIFTY, a rapid fiber type identification protocol to detect, sort, and pool fast- and slow-twitch fibers within 6 h of muscle biopsy sampling. Respiration of permeabilized fast- and slow-twitch fiber pools was then analyzed with high-resolution respirometry. Using standardized western blot procedures, muscle fiber pools were subsequently analyzed for control proteins and key proteins related to respiratory capacity. Results Maximal complex I+II respiration was 25% higher in human slow-twitch fibers compared to fast-twitch fibers. However, per mitochondrial volume, the respiratory rate of mitochondria in fast-twitch fibers was approximately 50% higher for complex I+II, which was primarily mediated through elevated complex II respiration. Furthermore, the abundance of complex II protein and proteins regulating cristae structure were disproportionally elevated in mitochondria of the fast-twitch fibers. The difference in intrinsic respiratory rate was not reflected in fatty acid–or complex I respiration. Conclusion Mitochondria of human fast-twitch muscle fibers compensate for their lack of volume by substantially elevating intrinsic respiratory rate through increased reliance on complex II. Highlights • Human slow-twitch muscle fibers possess greater respiratory capacity than fast-twitch fibers • Fast-twitch muscle fibers have lower mitochondrial density than slow-twitch fibers • Mitochondria in fast-twitch fibers compensate for their low volume by higher intrinsic respiratory capacity • The higher intrinsic mitochondrial capacity in fast-twitch fibers is driven by faster succinate metabolism • Mitochondria of the fast-twitch muscle express greater levels of proteins controlling cristae structure Keywords

Limitations of the study A limitation of the current study is the use of western blot methodology to indirectly assess mitochondrial content. Despite extensive efforts to make this method quantitative, we fully acknowledge that the magnitude of difference in mitochondrial content, as assessed here, is unusually large. While similar differences in human [ 16 ] and mouse [ 5 ] muscle have been reported previously, a large part of the literature reports considerably smaller differences between the fiber types. It therefore remains to be determined if our assessment using multiple membrane-bound markers reflects actual mitochondrial volume. Given the risk of overestimating the differences in mitochondrial content between type I and type II fibers, the intrinsic difference in fiber type-specific respiration should be interpreted with caution. Thus, to fully confirm the findings presented here, additional studies are required in which intrinsic fiber type-specific respiration is normalized using additional methods of assessing mitochondrial content such as proteomics [ 53 ], transmission- [ 54 ], or focused ion beam scanning electron microscopy [ 5 , 16 ]. Funding This study was funded by project grants awarded to W.A. (P2021-0173, P2022-0022) and F.L. (P2022-0049), respectively, from the 10.13039/501100005350 Swedish Research Council for Sport Science . During this work W.A. was also supported by an Early Career Research Fellowship from the 10.13039/501100005350 Swedish Research Council for Sport Science (no. D2019-0050). In addition, SE has been awarded project grants from Elisabeth and Gunnar Liljedahls foundation. CRediT authorship contribution statement Sebastian Edman: Writing – review & editing, Writing – original draft, Visualization, Validation, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Mikael Flockhart: Writing – review & editing, Investigation. Filip J. Larsen: Writing – review & editing, Supervision, Investigation, Funding acquisition, Conceptualization. William Apró: Writing – review & editing, Supervision, Methodology, Funding acquisition, Conceptualization. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Sebastian Edman reports financial support was provided by Elisabeth och Gunnar Liljedahls foundation. William Apro reports financial support was provided by Centrum för idrottsforskning. Filip Larsen reports financial support was provided by Centrum för idrottsforskning.

Supplementary data The following are the Supplementary data to this article: Data availability Data will be made available on request. Acknowledgments No acknowledgements.

CC BY

no

2024-01-16 23:41:58

Mol Metab. 2023 Dec 15; 79:101854

oa_package/63/4b/PMC10788296.tar.gz

PMC10788298

38226340

This story reflects a personal narrative of a graduate student who experienced a disturbing incident during Mardi Gras in New Orleans, shedding light on the enduring issue of racism. The author's journey to graduate school and the challenges faced along the way provide context for the pivotal moment of racial discrimination. The narrative then shifts to the author's decision to prioritize education over anger, highlighting the sacrifices made to protect their future as a graduate student. The incident serves as a stark reminder that, despite personal achievements and aspirations, racial prejudice persists. In conclusion, the author calls for resilience and focus in the pursuit of personal goals while acknowledging the ongoing struggle against racism and other forms of discrimination in society. This personal story serves as a poignant reminder of the challenges faced by individuals and the need for continued efforts to combat systemic intolerance. Keywords

I was marched around the corner and asked for my identification in front of a small bus. The bus looked like one of the very short public transport vehicles. It was dark. Minimal illumination from the nearby streetlight reached the spot where we were standing. The officer looked at my identification, handed it back, and commanded me to get on the bus. Three other policemen surrounded me in silence. Graduate school, at least for me, was an education beyond my class. It was a time of making lifelong friends, self-discovery, and maturation. But much like testing gravity by parachuting, I have also learned things that should have never been part of my college or life curriculum. As far as I can remember, hard work was always part of my upbringing. Its value was instilled in me by the example of my parents, who took pride in earning what they had. From selling flower seeds door-to-door to having my own paper route and serving customers at local food establishments, my experiences taught me aspiration and resilience. After getting my undergraduate degree, I worked for 6 years in corporate America, building a stable career. Yet, I felt drawn to another path. I wanted to do more. To be more. Earning a PhD was the path I chose to follow. I decided on a school in Dallas, Texas. My 6-year outdated Graduate Record Exam allowed me to be accepted and receive a tuition reduction but did not help me get a teaching assistantship. I had to provide for my room and board, but I still saw this as an opportunity. So, I resigned from my job, packed my things, and moved to Dallas. The first semester fully tested my resilience. I managed to get a job at Radio Shack, which was flexible to my school schedule and paid me enough to get by, that is, if I didn’t own a car or visit any local watering establishments. Yet, it was taking time away from studying. For months, I lived on 2-3 hours of sleep: a class at 8 am , the laboratory from noon to 4 pm , then work until 9:30 pm . At home, I would eat and study past 2 am , only to wake up to repeat the same routine. I learned to focus and manage my time, and I passed all classes, earning my teaching assistantship for the next semester. From there, things began to look up. I joined a laboratory that studied drosophila genetics. As the student graduate representative, I became a hall director, which gave me housing and an additional stipend. I joined the local rugby club. I made friends, not just in biology but throughout the university community. Together, we worked hard, learned about Dallas culture, and, over the years, watched the city blossom into a busy metropolis. It was a great time to be in Dallas. The idea of going to New Orleans, Louisiana, for Mardi Gras came as an unexpected opportunity. A friend, a senior premedical student, and New Orleans native, had just purchased a new Saturn and was eager to show it off on a 7-hour drive to the city. There were 4 of us on this trip: in addition to me and the driver, there was a residence life coordinator and a colleague of the driver. We were all staying with our New Orleans friend’s parents – a perfect solution for a graduate student budget. We set off on a lovely Friday afternoon and arrived in time to enjoy the evening's festivities. After stopping at the house for a quick meal, we headed to Bourbon Street. It was dusk. I knew that Mardi Gras was one of those legendary gatherings akin to New Year’s Eve in New York City or the Full Moon Party in Koh Phangan. Oddly enough, even though I grew up in the suburbs of New York, I had never attended the New Year’s Eve celebration. So, nothing I had done to this point in my life prepared me for what I saw. The crowd was unreal, alive. Standing on a balcony, looking down, I watched the crowd pulsate like a heartbeat. It seemed to swell, then shrink and swell again—An organized chaos. Within the crowd, thousands of little lives were doing their own little things. Some were moving up the street while others were moving down. Groups of people were standing along the edges; some stood in the middle of the road, and the crowds walked around them. Crossing the street in a straight line was impossible. It would have been easier to swim across the Mississippi. And there we were, walking at the fringes of the crowd to our next destination. We walked a single line between those standing against the establishments and those meandering in the street. Our leader had a plan: we were going to visit friends, so our movement through the crowd was more directed than most. The crowd pulsated as we moved in and out. My toes were stepped on a thousand times, but I didn’t mind it at all since I too, accidentally stepped on more than one toe. In most instances, people were too busy to notice; other times, they replied with “No worries; it is Mardi Gras” when I stopped to apologize. I stepped on one toe, however, that put things in an entirely different perspective. I was following my friend's single line through the crowd when I felt my foot land on someone else’s. Smiling apologetically, I turned around to a look I would never forget. Anger was spreading across a man’s face and growing into rage. “I will teach you what it is to be sorry!” he barked, piercing me with his eyes. I stood motionless – this person was an officer of the law. My friends would not see me for the rest of that evening. He told me to get on the bus as the other 3 officers surrounded me in silence. I had no choice. I walked up the stairs. There were 2 people arrested for urinating in public. They sat quietly, deflated, and intoxicated. I could not wrap my mind around this. What was I doing there? Just as I asked the officers why I was there, I felt a dull thud across my back. A wave of pain was followed by another thud. And then one after the other. Instinctively, I covered my head for some protection, and then I heard the words: “If you think Rodney King was beaten badly...” It was then that I realized why I was there: it was not for what I did wrong; it was for why I was wrong. I was black. The last thing I remember was everything fading to darkness while I was trying to reach for my license, now on the floor. The thuds on my back were getting duller until I felt no more. I regained consciousness the next morning to the smell of urine and alcohol saturating a jail-holding cell. I was lying on my back, staring at a key lime green concrete ceiling. People were stepping over me. There were about 2 dozen of us; some sitting along the wall, some pacing back and forth. My head felt like I had 2 of them, one inside the other. I cannot recall the extent of the pain, but I remember the reflection in the acrylic bulletproof glass: a distorted face with one side badly swollen and an eye in a new shade of night. I was torn somewhere between anger and fear. How could this happen to me? What did I do to deserve this? I wanted retribution. I wanted those officers to pay; I wanted the city to pay. Despite that, I had to be back at school on Monday. I could not let anyone know about my arrest and face the possibility of losing my spot as a graduate student. I had worked too hard to follow my anger. I chose my education over my indignation and rage. This was a hard pill to swallow, a pill wrapped in a story of a rugby tournament brawl and getting hit in the eye. I was lucky. My friend from New Orleans was able to pull some strings, and I was let out with a $50 fine and a sizable donation to the judge’s campaign. What I saved by staying at a friend's house, I paid to be legally assaulted. In retrospect, I was lucky. George Floyd did not wake up at all. Just like for many of my peers for me, graduate school was a time of hard work and scientific exploration. I learned how to ask and answer questions in the context of scientific reason. Yet, what I learned that night was beyond reason: regardless of my work, education, or personal qualities, I was still a black male. Over the years, I have come to realize that, sadly, my experiences as a black male in America were not very different from those of other groups that fall outside of the default. Any social default in its core is a moral defect, leading to racism, sexism, bullying, intolerance, bias, and prejudice. Intergroup conflict – a constant companion of human existence – stems from a disproportionate amount of power given to dominant groups, marginalizing many people within a country or society. Girls and women, people of color, religious communities, citizens in poverty, LGBTQ people, youth, and children—all have endured the consequences of the default. That being said, there is still a very small percentage of males in science, technology, engineering and mathematics (STEM) fields, pointing to added social pressures that we face. Many black males do not have 2 parental models. Often, the black male role model is missing. I was fortunate to have 2 strong role models in my parents. I have always fought to be a role model in the lives of my children. The hardest battles to fight were those that were institutionalized. As much as I try to bury what happened to me in New Orleans, the reoccurrence of such stories as George Floyd, Tamir Rice, Eric Garner, and many others remains a constant reminder that I cannot give up that fight and these events are not so once upon a time. For the last 2 decades, our society has been moving toward eradicating racism. What happened to me a few years after the Rodney King incident was over a quarter of century ago. It should have been a thing of the past, but to some extent, it seems worse today than in years past. For as long as we as a society continue designating a particular alternative as the default, nothing will move this needle. What I can say is—be resilient, remember your personal goals, stay focused on them, and continue moving onward and upward toward your own light.

Acknowledgments I would like to thank Arina Bokas for allowing me to bounce these ideas off of her as I put them to paper.

CC BY

no

2024-01-16 23:41:58

Res Pract Thromb Haemost. 2023 Nov 30; 8(1):102288

oa_package/58/2f/PMC10788298.tar.gz

PMC10788300

38225954

Introduction The long-term impact of natural disasters on self-harm and suicidal behaviour in adolescents are unclear, with varied responses to a range of natural hazards around the globe. For example, studies examining the impacts of earthquakes in Japan have reported an increased risk of children and adolescents’ suicidal attempts, plans and ideation six- and ten-years after the earthquake ( Chen, Zhou, Shi, Ma, & Fan, 2020 ; Tanaka et al., 2016 ). A Canadian study also reported that suicidal thinking increased in 11 to 19-year olds in the three years following exposure to wildfires ( Brown et al., 2021 ). Another study of the 2004 South-East Asian tsunami on Swedish children aged 10–15 years on holidays compared to a matched comparison group reported that those who were exposed to the tsunami had significantly higher suicidal ideation but not suicide attempts at 18–23 years ( Adebäck, Schulman, & Nilsson, 2018 ). However, to date, the number and type of natural hazards examined has been limited. Most studies have lacked comparison groups and have had small unrepresentative samples of children and youth. Furthermore, there have been even fewer studies of the implications of multiple disasters for suicidal and self-harm behaviours of children and adolescents. The influence of multiple and co-occurring disasters on self-harm and suicidal thoughts and behaviour of young people is largely unknown. For example, the only study that examines adolescents and multiple disasters used data from telephone calls to a mental health crisis line during the 2020 wildfire season in the United States ( Sugg, Runkle, Hajnos, Green, & Michael, 2022 ). In this study wildfires were not associated with elevated reporting of suicidal thoughts and self-harm ( Sugg et al., 2022 ) but the authors did note that the COVID-19 pandemic could have overshadowed impacts of fires given the already elevated levels of calls to the crisis line. The only study of children examining the impacts of multiple disasters and suicidal ideation reported that the Great East Japan Earthquake and tsunami increased the risk of suicidal ideation for girls who experienced trauma related to an earthquake at preschool age, but not boys ( Fujiwara et al., 2017 ). Exposure to multiple disasters has also been linked to greater lifetime risk of suicide attempts in Australian adults by Reifels, Spittal, Dückers, Mills, and Pirkis (2018) . However, their representative cross-sectional survey did not report the link extended to suicidal ideation or plans. Having a better understanding of the impact of exposure to natural disasters, and particularly multiple disasters, on children and adolescents is becoming increasingly important given the projections that climate change will increase the frequency of natural disasters and that the co-occurrence of natural disasters will become the “new normal” ( National Academy of National Academies of Sciences, 2022 ). Potential mechanisms that may explain how youth mental health could be undermined by exposure to natural disasters include the direct trauma of exposure, a deterioration of parental mental health and reduced parenting capacity, the financial strain on households and on local communities ( Hunter, Gray, & Edwards, 2012 ; Taylor & Edwards, 2022 ; Conger, Ge, Elder, Lorenz, & Simons, 1994 ; IPCC, 2022 ). More broadly, climate change has also been associated with climate anxiety in youth which could have implications for depression, self-harm and suicide ( Crandon, Scott, Charlson, & Thomas, 2022 ; Sanson, Van Hoorn, & Burke, 2019 ; Scribberas & Fernando, 2022 ). This paper examines the impacts of natural disasters on suicidal behaviour and self-harm among Australian youth. It uses parent-reported exposures to natural disasters from the Longitudinal Study of Australian Children (LSAC), a nationally representative cohort study of adolescents followed from ages.14-15 to 18–19. The study has two aims. First, to document the association between exposure to a fire, flood or drought on suicide and self-harm from 14 to 19 years of age. Second, to test whether exposure to compound disasters (two or more disasters occurring in the previous 12 months), cascading disasters (drought followed by fires) and consecutive disasters (multiple disasters within the last two years or over an eight-year period) are associated with suicidal behaviour or self-harm over the course of adolescence. This study extends prior research by including many types of natural hazards, longitudinal data on suicidal behaviour and self-harm over a six-year period, a nationally representative comparison group not exposed to disasters, and different types of multiple disaster exposures.

Methods Study design and participants LSAC, also known as Growing Up in Australia , is a longitudinal cohort study of Australian children that began in 2004 ( Edwards, 2014 ). Written informed consent was provided by parents at the Wave 1 interview and this was provided by young people at 14–15 years as well. In this study we use the K-Cohort, born between March 1999 and February 2000, because there were three waves of self-harm and suicidal behaviour information collected via self-administered computerised surveys of adolescents when they were 14–15 to 18–19 years. The parents of these children were first interviewed between March and November 2004 ( Mohal et al., 2023 ). The LSAC sample is a clustered design, based on postcodes, with stratification across capital cities and balance of state for each state and territory from Medicare Australia's enrolment database, considered the most comprehensive database of Australia's population ( Edwards, 2014 ). Medicare is a universal healthcare system that provides government subsidies to all population for primary and secondary care (98% of parents register their children with Medicare by 12 months of age; Soloff, Lawrence, & Johnstone, 2005 ). LSAC data is publicly available to the scientific community through the Australian Data Archives. Procedures The first wave included data on 4983 children in the K-Cohort and the eighth wave of LSAC retains 60.9% of the wave 1 K-Cohort. Our key outcomes were collected in waves 6 to 8, when the K-Cohort was aged 14–15 to 18–19 years from Computer Assisted Self Interview. For this study we focus on exposure to natural disasters reported by biological mothers or a parent who knew the child best. The data used is from when the K-Cohort was aged 10–11 to 18–19 years. Thus, disaster exposure includes the period when outcomes were collected, but also collects exposures up to 4 years previous to when suicide and self-harm were first measured. This is an important feature of the design, as it enables cumulative effects of disaster exposure to be estimated. The LSAC Data User Guide provides a detailed description of the study design and procedures ( Mohal et al., 2023 ). The Australian Institute of Family Studies Ethics Committee provided ethics approval for the LSAC, and all participants provided written informed consent. Dependent variables Self-harm: measures of self-harm were derived from the Avon Longitudinal Study of Parents and Children (ALSPAC): Life of a 16+ Teenager questionnaire ( ALSPAC, 2007 ). Young people were asked to respond to Yes or No to the following questions “During the last 12 months have you thought about hurting yourself on purpose in any way (i.e. by taking an overdose of pills, or by cutting or burning yourself)?” and “During the past 12 months have you hurt yourself on purpose in any way (i.e. by taking an overdose of pills, or by cutting or burning yourself)?” This was measured from 14 to 15 years. Suicide-related behaviour: Measures of suicide-related behaviours were derived from the National Survey of Mental Health & Wellbeing and measured from 14 to 15 years ( Australian Bureau of Statistics, 2007 ). We focussed on suicidal ideation (‘During the past 12 months did you ever consider attempting suicide?“) and attempts (“During the past 12 months, how many times did you actually attempt suicide?“). Disaster exposures Exposure to fires, floods and droughts: In this study we used primary caregivers report of exposure to bushfires/floods and droughts in the previous year. Parents were asked whether their home or local area was affected in the last 12 months. Although the suicide and self-harm data are only available from waves 6 to 8, we use disaster exposure data from wave 4 onwards because of our interest in understanding multiple disaster exposures on these outcomes. In this study geospatial data was only available for large geographic areas and therefore caregiver self-reports of disaster were a better measure of disaster exposure given that they were more localised. Moreover, there is some evidence that self-report disaster is more strongly correlated with disaster impacts on household finances ( Hunter, Gray, & Edwards, 2012 ; Edwards, Gray & Borja, 2021a ) and child mental health and family functioning was more strongly correlated with caregiver reports of disaster exposure than geospatial measures of disasters ( Edwards, Gray & Borja, 2021a ). However, geospatial data may well be more appropriate for other studies such as those examining the impacts of smoke from wildfire ( Molitor, Mullins, & White, 2023 ). Multiple natural disaster exposures Following Leppold, Gibbs, Block, Reifels, and Quinn (2022) , we operationalise compound, cascading and consecutive disasters in this longitudinal cohort ( Lawrence et al., 2015 ; National Academy of Science, 2022 ). Compound disasters: Two or more natural disasters that occur simultaneously, operationalised as two or more disasters that occurred in the previous 12-months. Cascading disasters: Disasters that increase in progression over time and generate unexpected secondary events. In this study we focus on instances where fire follows a drought in the previous wave. Consecutive disasters: Consecutive disasters are two or more disasters that occur in successive waves of the LSAC survey in the same area. We generate two types of variables. Firstly, the number of disasters in each survey wave for each natural hazard - this includes separate variables for fire/flood and for drought. Secondly, we generate a separate cumulative consecutive measure of disasters for all natural hazards measured - this measure aggregates all disasters up until that survey wave. Covariates Statistical models also included child age (in years), child sex at birth, whether the child was born overseas, and whether the child was from an Aboriginal or Torres Strait Islander background. Other covariates included whether the family had moved house since the last wave, whether they lived in a regional or metropolitan area, and their state of residence. Neighbourhood Socio-Economic Advantage/Disadvantage is measured by the Socio-Economic Indexes for Areas which is a standardized score for socioeconomic position by geographic area (postcode of family domicile) compiled from 2011 Australian Census data. We used the Index of Relative Socioeconomic Advantage/Disadvantage to capture neighbourhood advantage and disadvantage, which numerically summarizes the social and economic conditions of Australian neighborhoods (national mean of 1000 and a standard deviation (SD) of 100, where higher values represent greater advantage and less disadvantage). Statistical analyses We estimated a random effects model to account for clustering of waves within adolescents. We used the xtlogit command in STATA 15. The following covariates were included: child age, child sex, child Aboriginal or Torres Strait Islander background, whether the family had moved since the last wave, neighbourhood advantage/disadvantage, regional area and state of residence. For each outcome we estimated five random effects models for self-reported disasters. The first model tests the impact of disasters on outcomes at the same wave. The next four models test the impact of compound, cascading, consecutive and multiple consecutive disasters, respectively. Compound disasters are two disasters that occur in the last 12-months. For cascading disasters, we are testing the impact of a fire/flood occurring at the same wave as the outcome variable that was preceded by a drought in wave t −1. For consecutive disasters we are testing the cumulative influence of a single disaster from wave 4 through to wave 8 (so there is lagged influence as well as a contemporaneous influence of a disaster). Multiple consecutive disasters are the cumulative risk of exposure to multiple disaster types from wave 4 through to 8. Population attributable fractions are used in public health to understand how important a particular factor is in contributing to ill health. While not assuming a causal effect, the technique provides a ‘thought experiment’ on the degree to which an illness or a natural disaster contributes to rate of ill health in the population (in this case rates of self-harm and suicide ideation and behaviour in Australian youth). To estimate a population attributable fraction for the impact of natural hazards on suicide and self-harm we used the Stata program ‘punafcc’ developed by Roger Newsom using methods recommended by Greenland and Drescher (1993) . Funding source This research was supported by the Australian Medical Research Future Fund (APP1201335). The funders had no influence on the research in this paper.

Results Given that consecutive disaster measures incorporate disaster exposures from the previous waves, we report on disaster exposure in waves 4 to 8. Table 1 shows that on average, 5% of children were directly affected by fire or floods and a further 5% were exposed to drought from wave 4 to 8. However, Fig. 1 shows that there was variation in exposure to fire/floods from wave 4 to 8, with higher levels of exposure in wave 4 and 5 (7–8%) and then in wave 7 (7%). For drought there were fewer peaks over wave 4 to 8 with a peak in wave 4 (9%) and in wave 8 (7%). The suicide and self-harm measures were collected in waves 6 to 8 from 2903 to 2908 children and include 8708 to 8714 person-waves of data. However, cascading disasters and consecutive disaster measures included natural disaster data from wave 4 as well. Findings from random-effects logistic regression models suggest that fire/floods in the last year were associated with statistically significantly higher rates of self-harm and self-harm ideation and suicidal thoughts ( Table 2 ). Droughts were not associated with self-harm or suicide. To understand how much variation in outcomes could be explained by fire or floods, population attributable fractions were calculated while holding all other covariates equal. Population attributable fractions suggested that eradicating fires or floods would reduce the rates of suicide ideation by 2.0 percentage points (95% CI: 0.54 to 3.84) from a base of 10.3%, self-harm ideation by 1.7 percentage points (95% CI: 0.55 to 2.76) from a base of 18.35 and self-harm by 2.0 percentage points (95% CI: 0.28 to 3.62) from a base of 9.5%. Multiple exposures to natural disasters based on parent self-reports was significantly associated with suicidal ideation for compound disasters only. In this case, the population attributable fraction suggested that for every compound disaster prevented, there was a reduction of suicidal ideation by 1 percentage point (95%CI: 0.2–1.9%) from a base of 10.3% in the population of Australian youth. Cascading impacts of drought to fire/flood was associated with an increased risk of self-harm ( p < 0.10). For consecutive disasters that were recurrent over several waves, children exposed to recurrent droughts were at significantly lower risk of suicidal ideation. There was no evidence of consecutive multiple disasters being associated with any self-harm or suicide variables.

Discussion In this study we find that sudden onset disasters were associated with increased self-harm ideation, self-harm and suicidal ideation of adolescents. Fire or floods in the last year increased the risk of self-harm, self-harm ideation and suicidal thoughts of adolescents. Exposure to some types of multiple disasters were associated with suicide ideation and self-harm. Compound disasters of fire/flood and drought were also associated with increased risk of adolescents having suicidal thoughts. Cascading disasters of drought followed by fire/flood were associated with increased risks of adolescents’ self-harm but recurrent consecutive droughts were associated with significantly lower risks of suicidal ideation of youth. To understand the potential impacts of natural disasters for the mental health of the population of young people in Australia we also used population attributable fractions. While our estimates are not causal, they provide an upper bound estimate of the capacity to address mental health challenges by eliminating the impacts of disasters entirely. Our findings suggest that eradication of sudden onset disasters could have some substantial benefits for population health. For instance, this would lead to a reduction in suicidal ideation by 2 percentage points from a base rate of 10%. Findings in the current study are consistent with previous research that have also found disasters were associated with higher levels of suicidal ideation ( Brown et al., 2021 ; Reifels et al., 2018 ). The only other longitudinal study of youth also showed small increased risks of suicidal ideation in response to fire ( Brown et al., 2021 ). A retrospective cross-sectional study reported much greater lifetime exposure to multiple natural disasters in Australia than in our study ( Reifels et al., 2018 ). In comparison our study reported modest impacts on suicidal ideation and suicide attempts. Differences in findings may suggest that further life time exposures accumulated to create greater risks, or simply reflect a delayed impact on self-harm and suicidal ideation and behaviour of natural disasters. Future research is important to help clarify the relationship between exposure and suicidal ideation and attempts across the lifetime, thereby helping to inform strategies to protect against the mental health consequences associated with natural disasters. This study provides the first longitudinal analysis of multiple disaster exposure on mental health in adolescents and explicitly testing and operationalising compound, cascading and consecutive disasters for longitudinal studies. The increased risks of cascading disasters for self-harm has not been documented previously, although one other study examining cascading disasters and suicidal behaviour reported greater risks of suicidal ideation for girls in the Great East Japan Earthquake and tsunami ( Fujiwara et al., 2017 ). The varied influence of different types of multiple disasters (compound, cascading, consecutive) on self-harm and suicidal behaviour could be a function of the timing in disaster exposures combined with variable impacts of sudden onset (fire/flood) and slow onset disasters. For example, in this study sudden onset multiple disasters led to increased risks when disasters were relatively recent (compound disasters or cascading disasters). However, there were no increased risks for recurrent sudden onset disasters and lower risks of suicidal ideation with recurrent drought (a slow onset disaster). Perhaps exposure to a slow onset chronic stressor such as drought that can span several years (the Millennium drought in Australia was from 2001 to 2009) may lead to habituation to the stressor or the mustering of community and family resilience to improve wellbeing in the long term ( Masten, 2021 ). An important element of adaption and resilience to future multiple disasters will be to identify factors that protect against poor mental health and build resilience. Prior research suggests that strengthening financial supports to disaster will offset the negative consequences for parental mental health and parenting ( Conger et al., 1994 ), but that family, schools and communities all may play a role in supporting young people's mental health ( Masten, 2021 ). For public policy and government responses, local community engagement has been considered to be an important element in any successful response to multiple disasters ( Leppold et al., 2022 ). Future research focused on identifying protective factors are warranted, as well as further replication in other contexts. Despite its strengths, there were some limitations of the current study. The random effects regressions likely reflect associations rather than robust causal estimates. However, testing multiple exposures to disasters over time meant that fixed effects models could not be estimated ( McNeish & Kelley, 2019 ). Given that the disaster exposure could be considered to be a random occurrence, there is some support for the causality of our estimates. There were also some limitations in the measurement of outcomes and disaster exposures. Like many studies, the items measuring suicidal thoughts and behaviour do not attempt to strictly define a suicide attempt and there is a risk that self-harm is conflated with suicidal thoughts and behaviours ( Ammerman, Burke, Jacobucci, & McClure, 2021 ). This risk is limited in LSAC as self-harm items were asked first, and then suicide questions second. Comparative national studies, like the Australian Child and Adolescent Survey of Mental Health and Wellbeing ( Lawrence et al., 2015 ) generated similar rates, with rates of suicidal ideation, at 8%, a little lower than in LSAC, and the rates self-harm consistent between the two studies. Another potential limitation in our measures is the risk of under-reporting disaster exposure. Caregivers were asked to report on disaster exposure in the last year but waves of data collection occur every two years thereby increasing the risk of under-reporting. Despite this limitation, the current study is an advance on retrospective reporting of multiple natural disasters which suffer from retrospective recall bias ( Galea, Maxwell, & Norris, 2008 ).

Conclusions The 2019–20 wildfires attracted world attention (author et al., 2020) and highlighted that Australia has a high risk of disaster exposure. However, the 2023 wildfires in North America underscore that our findings are increasingly applicable to other countries. Beyond wildfires, the IPCC predicts that there are increased risks of drought in Australia but also in South America, the Mediterranean, China, North America and Eurasia ( IPCC, 2022 ). Despite our study findings, it is important to note there was considerable resilience to disaster exposure suggesting that youth, families and communities may well develop protective strategies to support mental health. Further research to identify these protective factors is urgently needed to inform better disaster mitigation policy and service responses (NASEM, 2022).

Few studies have examined the relationship between exposure to natural hazards and suicide and self-harm in youth. We extend prior research by investigating the association between multiple disasters and the risks of self-harm and suicide longitudinally in a nationally representative longitudinal cohort of adolescents 14 to 15 years to 18-19 years of age. Natural disasters were identified through parental self-reports for the local area. Different types of multiple disaster exposures were investigated including compound disasters (two or more disasters occurring in the last 12 months), cascading disasters (a disaster that leads to another disaster in the subsequent wave) and consecutive disasters (multiple disasters within the last two years or over an eight-year period). Using 8,714 person-waves of data from 2,908 adolescents, findings from random effect models suggest that parental reports of fire or floods increase the risk of self-harm ideation, self-harm, and suicidal ideation. Compound disasters of fire/flood and drought were also associated with increased risk of suicidal thoughts. Cascading disasters of drought followed by fire/flood increased the risks of self-harm but recurrent consecutive droughts were associated with lower risks of suicidal ideation. Australian adolescents are exposed to high rates of natural disasters that increase the risk of self-harm and thoughts of self-harm and suicide. Climate change will increase risk of natural disaster exposure for all countries. Despite these increased risks, there was resilience to disaster exposure particularly in the case of recurrent drought suggesting that youth, families and communities may well develop protective strategies to support mental health. Highlights • First national cohort study to examine disasters and adolescent suicide. • Fire or floods increased the risk self-harm, self-harm and suicidal ideation. • Compound disasters increased the risk of suicidal ideation. • Cascading disasters increased the risks of self-harm. • Repeated droughts lowered risks of suicidal ideation. Keywords

Funding source This research was supported by the Australian Medical Research Future Fund (MRF1201355). The funders had no influence on the research in this paper. Competing interests The authors declare that they have no competing interest. Ethical statement The Australian Institute of Family Studies Ethics Committee provided ethics approval for the LSAC, and all participants provided written informed consent. Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. The authors declare the following financial interests/personal relationships which may be considered as potential competing interests.

Supplementary data The following is the Supplementary data to this article. Data availability LSAC data is available from the Australian Data Archives

CC BY

no

2024-01-16 23:41:58

SSM Popul Health. 2023 Dec 9; 25:101576

oa_package/ba/d5/PMC10788300.tar.gz

PMC10788301

38150745

Introduction There are at least 5 million people currently living with age-related dementia in the United States, posing a major public health problem. Alzheimer’s disease (AD) is the most common form of dementia, accounting for about 70 % of all dementia cases in the elderly (Alzheimer's Association, 2016 ). AD care is expensive for the medical and social services sector, as well as the entire US economy. In contrast to the decreasing death rate of other major diseases over recent years, such as cardiovascular disease, the proportion of death attributed to AD has substantially increased, and it is now the third leading cause of death in the United States (Alzheimer's Association, 2016 ). Previous research has highlighted that AD has a long preclinical phase lasting 10–20 years ( Vermunt et al., 2019 ). This preclinical phase is characterized by an absence of overt clinical symptoms despite an accumulation of pathological changes within the brain. These changes typically begin with the accumulation of β-amyloid (Aβ), followed by the accumulation of pathological tau, structural disturbances in grey- and white-matter, and finally, a decline in cognition ( Jack et al., 2010 , Jack et al., 2013 ). Some cognitive decline may occur during this preclinical phase, but not an extent that disrupts an individual’s daily routine ( Price and Morris, 1999 ). It has been proposed that the acceleration of cognitive decline may begin months, or even years, before individuals are normally diagnosed with AD-related dementia ( Karr et al., 2018 ). Given that the pathological changes associated with AD begin years prior to cognitive symptoms, there is an increasing need to develop predictors of AD for use in the asymptomatic and early clinical phases. In this study, we utilized the Washington University Knight Alzheimer's Disease Research Center (ADRC) dataset, which is dedicated to gathering data on preclinical participants, the majority of whom exhibit normal cognitive function at the baseline (Clinical Dementia Rating [CDR]® = 0). Our primary objective was to integrate imaging biomarker data with cognitive testing data to detect the earliest signs of cognitive decline in individuals with preclinical AD. There is a pressing need amongst AD researchers and clinicians to develop an optimal approach for the identification of individuals who are at the highest risk of progressing from a state of cognitive normalcy to mild cognitive impairment (MCI), and subsequently to AD-related dementia. Notably, the recent approval by the US FDA of two disease-modifying monoclonal antibody (mAb) treatments, which have proven effective in the very early stages of the disease, underscores the urgency of this area of neuroscience research. There is a compelling need to develop tools that aid in the identification of AD during its preclinical phase, ensuring that patients can access treatment at the earliest possible juncture. Focusing on the earliest measurable pathological changes in preclinical AD, Aβ plaque deposition, previous studies have utilized PET for detecting and quantifying the presence of Aβ. For example, researchers have shown that Aβ-specific PET tracers, such as 11 C-Pittsburgh compound B (PiB), can be used to detect Aβ plaques and measure the rate of Aβ accumulation over time throughout the disease process, including in the preclinical stage ( Vlassenko et al., 2012 , Villemagne et al., 2013 , Villemagne et al., 2017 ). Recent studies have hypothesized that Aβ-PET scans could be useful for understanding the relationship between Aβ and cognition ( Wang et al., 2018 , Fodero-Tavoletti et al., 2009 , Farrell et al., 2017 ). However, researchers must carefully consider the limitations of PET when considering whether baseline Aβ measurements are more sensitive to cognitive deterioration to avoid unnecessary repeated scans. Importantly, PET imaging requires patients to be exposed to ionizing radiation and the performance of multiple PET scans does pose risks to individuals, although these risks are very low ( Brix et al., 2009 , Nievelstein et al., 2012 ). Therefore, we were interested in evaluating whether a single measure of Aβ was better predictors of eventual cognitive decline, as compared to longitudinal measures of Aβ. We used a random coefficient model to evaluate the relationship between PIB-PET and cognition measures statistically in a cohort of individuals participating in research studies at the Charles F. and Joanne Knight Alzheimer Disease Research Center (Knight ADRC). This statistical method allows us to accommodate heterogeneous numbers of visits and intervals between visits, for participants in this study. Given that sporadic AD does not have a predetermined “estimated age of onset” and the rate of cognitive decline varies between subjects, it is difficult to determine when biomarker changes are occurring relative to the onset of dementia. These factors justify the use of a random coefficient approach, as this allows us to model separate time until dementia onset, and rates of change, for each individual. By including model terms that accommodate variability in these characteristics of dementia onset, we are able to increase the validity of conclusions drawn from these analyses.

Methods Participants Participants were enrolled in longitudinal studies of memory and aging at Knight ADRC at Washington University in St. Louis. Data was obtained through the Open Access Series of Imaging Studies (OASIS; https://www.oasis-brains.org ) ( Marcus et al., 2010 ). Individuals were excluded from analyses if they presented with dementia that was not primarily caused by AD, had active neurologic or psychiatric illness, had a history of serious head injury or clinically meaningful stroke, or used psychoactive drugs. Participants were required to have at least two longitudinal clinical and cognitive assessments and longitudinal Aβ-PET scans, resulting in a sample of 153 participants. Clinical and cognitive assessment In addition to PET imaging, each individual underwent clinical and cognitive assessments. The presence and severity of dementia was determined by experienced clinicians using the clinical dementia rating® (CDR®; Morris, 1993 ). A CDR of 0 indicates absence of dementia, and ratings of 0.5, 1, 2, and 3 indicate very mild, mild, moderate, and severe dementia, respectively. Each cohort at the Knight ADRC receives slightly different cognitive batteries, so only tests across all cohorts were considered in this study. A cognitive composite score was calculated from the average z-scores of 10 neuropsychological tests (Logical Memory, Logical Memory Delayed tests ( Wechsler, 1997 ), Digit Span Forward, Digit Span Backward Tasks ( Wechsler and Stone, 1973 ), Boston Naming Test ( Kaplan et al., 1976 ), two subsets of the Trail Making Test: Part A and Part B ( Armitage, 1945 ), two category fluency tasks: The Vegetable Naming and Animal Naming tests ( Goodglass and Kaplan, 1983 ), and subscales of the revised Wechsler Adult Intelligence Scale (WAIS): Information and Block tests ( Wechsler, 1997 ). PET acquisition and processing For the OASIS dataset, dynamic PIB PET scans were acquired on a Siemens Biograph 40 PET/CT or a Siemens/CTI EXACT HR + scanner for 60 min after tracer administration and reconstructed using standard iterative methods, with attenuation and scatter correction. All PET scans were processed with the PET Unified Pipeline (PUP, https://github.com/ysu001/PUP ) using FreeSurfer derived ROIs ( Su et al., 2018 , Su et al., 2013 ) to calculate the standardized uptake value ratio (SUVR), using the cerebellar cortex as a reference region. Quantitative PET analysis used peak time windows of 30 to 60 min’ post-injection for PiB ( Fischl, 2012 , Su et al., 2013 ). Partial volume correction was performed using a geometric transfer matrix (Rousset et al., 1998, Su et al., 2015 ). To measure amyloid burden, we calculated the mean cortical SUVR (mcSUVR) by averaging the partial volume corrected SUVRs from the FreeSurfer ROIs in the lateral orbitofrontal, medial orbitofrontal, rostral middle frontal, superior frontal, superior temporal, middle temporal, and precuneus regions, as previously defined ( Su et al., 2013 ). It is necessary to standardize methods for data collection and analysis to better aid cross-center, multi-tracer utility. The Centiloid Project was initiated to derive a standardized quantitative amyloid imaging measurement scale based upon normalization of data from the 18F-tracers to that of PiB. In this linear scale, young controls (≤45 years) have a mean of zero centiloid units (CL) and typical mild to moderate AD patients score on average 100CL ( Klunk et al., 2015 ). In this study, we used the centiloid derived from mcSUVR. Genotyping Genomic DNA was isolated from peripheral blood samples using standard procedures. APOE genotyping was performed as described in previous work ( Talbot et al., 1994 ). Statistical analysis We used a random coefficient model to statistically evaluate the relationship between PIB-PET and cognition measures. This statistical method allows us to accommodate heterogeneous numbers of visits and intervals between visits for participants in this study. This is an important feature because it increases the ecological validity of our conclusions, given that real-world patients are unlikely to have the same number of PET scans collected or the same time interval between scans. First, we were interested in quantifying how well baseline Aβ predicts cognitive performance. In order to accomplish this, a one-step random coefficient model was used (equation (1) . In this model, time is set as a continuous variable (measured in years), representing the interval between the baseline cognitive assessment and each subsequent visit. Within this model, time is treated as both a fixed and random effect. In order to examine if longitudinal Aβ predict cognitive decline, a separate two-step random coefficient model was conducted. The first step of this model involved calculating the rate of change in Aβ (slope), treating time as both a random and a fixed effect (equation (2) . The second step of this model quantified whether the rate of change in Aβ levels (slope) was predictive of longitudinal cognitive decline (equation (3) ). By adding a three-way interaction term to equation 3, mixed models were also used to determine the effect of APOE ε4 carrier status (Aβ slope*time* APOE ε4 carrier status) and sex (Aβ slope*time*sex) on modifying the effect of longitudinal Aβ on cognitive decline. Last, we examine cognitive changes over time for three different Centiloid levels: low (Centiloid = 0), medium (Centiloid = 50), and high (Centiloid = 100). Models were fitted for cognition as a function of time and baseline Aβ, using mixed model output to obtain three different equations (Centiloid = 0, 50, 100). All statistical analyses were performed using PROC MIXED in SAS Version 9.4 (SAS Institute Inc., Cary, NC), and p < 0.05 was regarded as statistically significant.

Results Demographic characteristics of the participants at baseline are presented in Table 1 . There were 153 participants in the study cohort, with an average age of 69.9 ± 5.9 years and average years of education of 15.5 ± 2.5. Of the 153 participants, 87 (56.8 %) identified as female and 52 (34 %) were APOE ε4 carriers. Among the APOE ε4 carriers, 62 % are female. Participants were followed for an average of 6.01 ± 2.0 years’ assessments. The mean cognitive composite score was 0.37 ± 0.42. The baseline scores and main results for the individual cognitive tests are presented in Supplementary Tables S1 and S2 , respectively. Aβ derived from a single PET image predicts cognitive decline Our first analysis aimed to characterize whether Aβ derived from a single baseline PET scan could accurately predict cognitive decline. Our results showed that baseline Aβ significantly predicts the rate of change in cognitive composite score ( p = 0.029, t = -2.2, Table 2 , see Table S2 for detail for the individual tests. ). For all participants, older individuals ( p < 0.0001), males ( p < 0.0001), individuals with less education ( p < 0.0001) had significantly lower baseline performance on the cognitive composite. We generated a scatter plot to investigate the relationship between baseline Aβ and cognitive decline. This analysis revealed a significant association between the rate of change in cognitive composite score and baseline Aβ ( p = 0.019, estimate β = -0.00038, Fig. 1 A ). Rate of changes in Aβ derived from longitudinal PET scans does not predict cognitive decline Building from the prior analyses, we were also interested in characterizing how well the rate of change in Aβ, derived from longitudinal PET scans, predicts cognitive decline. Longitudinal Aβ was not significantly associated with the rate of change in cognitive composite score ( p = 0.917, t = 0.08, Table 3 ). We also generated a scatter plot to characterize the relationship between rate of change in Aβ and cognitive decline. Fig. 1 B shows that rate of change in Aβ is not significantly associated with the rate of change in the cognitive composite score ( p = 0.984, estimate β = 0.00005, Fig. 1 B ). Impact of covariates on the relationship between longitudinal Aβ and cognitive change Finally, we examined the impact of important covariates on the relationship between longitudinal Aβ and cognitive change. Table 4 shows that the presence of the APOE ε4 allele increases AD risk by accelerating cognitive decline ( p = 0.028, t = 2.2). Sex was found to have no significant effects on the association between amyloid with cognitive decline ( p = 0.088, t = -1.7). Comparing the baseline and rate of change in Aβ In order to understand why the baseline centiloid is a better predictor of cognitive decline than longitudinal centiloid, we plotted the relationship between these measures using a scatterplot. Fig. 2 shows that the rate of centiloid was low when baseline centiloid was also low, while accumulation rates increased to a maximum when baseline centiloidwas around 50. Accumulation rates declined above this maximum, approaching zero when baseline centiloid was greater than 100. Given this inverted U finding, we were motivated to predict cognition changes over time for three different baseline centiloid levels. Fig. 3 shows that cognition decreases with time for the three different baseline centiloid levels examined. More specifically, scores on the cognitive composite score decrease with time for different levels of baseline centiloid, when baseline centiloid increases, cognition decreases more rapidly.

Discussion We aimed to investigate the relationship between Aβ and cognition in the preclinical and early impairment stages of AD. More specifically, we compared the utility of a single baseline measure of Aβ versus longitudinal measures of Aβ for predicting cognitive decline, as defined by a cognitive composite score. This investigation revealed that baseline Aβ is a better predictor of cognitive decline than longitudinal Aβ. This is potentially due to the non-linear, inverted U-shaped relationship we observed between baseline and longitudinal PET measures (see Fig. 2 ). This finding is consistent with previous studies that also reported a U-shaped function between these two measures (Clifford et al., 2013; Mishra et al., 2018 , Schindler et al., 2021 ). This inverted U-shaped pattern may indicate that the rate of change in Aβ may be less informative at all points in the disease course than overall levels of pathology. Importantly, the values of baseline Aβ are much greater than the absolute values for longitudinal change in Aβ, and thus, longitudinal Aβ changes can be easily affected by signal noise. Together, these results suggest that baseline Aβ is more accurate than longitudinal Aβ for predicting cognitive decline. This holds significant practical implications, as undergoing multiple PET scans necessitates patients being repeatedly subjected to the potential risks of exposure to ionizing radiation ( Brix et al., 2009 , Nievelstein et al., 2012 ). Thus, it is imperative to explore whether conducting multiple amyloid PET scans on individuals with initial amyloid positivity is clinically necessary or beneficial. Our findings support the concept that a single amyloid PET scan may be adequate for evaluating dementia risk in the context of AD. This has substantial ramifications for public health, particularly in the current era of AD treatment with monoclonal antibodies (mAbs), as it has the potential alleviate the burden on patients and healthcare systems through earlier diagnosis and treatment of dementia. Several previous studies have also investigated the relationship between AD biomarkers and subsequent cognitive changes in individuals initially classified as cognitively normal. For instance, a recent study by Hanseeuw and colleagues revealed a link between amyloid accumulation and tau accumulation, with subsequent associations with the rate of cognitive decline also observed ( Hanseeuw et al., 2019 ). However, this study did not establish a direct relationship between longitudinal Aβ levels and cognition. Another study suggested that Aβ load is predictive of future cognitive decline, but this study solely investigated how baseline amyloid predicts cognitive decline with no longitudinal analyses performed ( Dumurgier et al., 2017 ). The novelty of our study lies in our comparative evaluation of baseline and longitudinal amyloid to determine which holds greater value for the prediction of cognitive outcomes. Given the cost and radiation exposure risks associated with PET scans, it is essential to ascertain whether a single scan is sufficient for the early detection of the disease. In our current study, we also illustrated our model fit on cognitive change over time for three different amyloid levels, defined by low, medium, and high centiloid units, using equation (1) and the parameters obtained from the model. Our results show that cognition decreased over time and that when baseline Aβ is higher, cognition decreases more rapidly. This finding is consistent with a recent paper ( Farrell et al., 2017 ), which also reported that higher baseline Aβ burden predicted a steeper decline in episodic memory, processing speed, vocabulary, and Mini-Mental State Examination performance. Furthermore, another study by Lim et al ( Lim et al., 2013 ) focusing on Aβ-positive adults combined cognitively normal participants with those with mild cognitive impairment and dichotomizing them into low- and high-Aβ groups. They reported a greater memory decline for the group with higher Aβ levels compared to the group with lower Aβ levels. These findings suggest that the magnitude of Aβ burden may be useful in predicting the rate of future cognitive decline, with those with the greatest burdens showing the most decline. Our study also reported that important covariates, such as the presence of the APOE ε4 allele, can significantly increases AD risk by accelerating amyloid deposition, which is consistent with some previous studies ( Liu et al., 2019 ). One recent study suggested that longitudinal Aβ measures in a group of APOE ε4 negative individuals were significantly predictive of cognitive decline ( Hsiung et al., 2004 , Paranjpe et al., 2019 ). Furthermore, these authors also found that in individuals with at least one copy of the APOE ε4 allele, the rate of change in Aβ is not a significant predictor of cognitive decline. This may be because APOE ε4 status will play a primary role in predicting cognitive decline in individuals who carry an APOE ε4 allele, but when this allele is absent, Aβ becomes the primary predictor of cognitive decline. This could also be because Aβ levels increase earlier in individuals who are APOE e4 carriers, whereas Aβ does not increase until later in APOE e4 negative individuals ( Schindler et al., 2021 ). The use of a random coefficient model (RCM) is a major statistical strength of this study, as this approach enables the use of different number of visits and visit intervals for each individual across variables of interest in our analyses. Most previous work in this field focused on using general linear mixed models (LMM), which assume a homogenous number of visits and inter-visit intervals for all study participants ( Lawrence et al., 2018; Shokouhi et al., 2013). The patient cohort for this study had highly variable numbers of visits and inter-visit intervals between participants, with some individuals attending only two visits, while others had up to seven, precluding the use of LMM for our analyses. Thus, we adopted an RCM approach in this study to examine the individual-level variation in the relationship between predictor variables (e.g., rate of change in amyloid) and outcomes (e.g., cognition). LMM has notable restrictions that also render it unsuitable for this study, such as its requirement for linear disease progression during the follow-up period, which is unlikely in an AD patient cohort. In addition, our two-step model is very intuitive and simple, compared to complex existing models used in previous longitudinal studies. Patients outside of research settings are very unlikely to have homogenous numbers of clinical visits or consistent intervals between visits, thus, a random coefficient model is better able to statistically control for this expected variability. By using this approach, we are confident that our conclusions are as rigorous, reproducible, and clinically relevant as possible, facilitating their translation towards clinical use. Despite the major advantages of our study, there are some limitations to note. For example, the Aβ level used in this study is an average of measures across several critical brain regions. This measure was chosen as it is standard in this field, however, future iterations of this work should investigate Aβ levels across a wider number of regions in order to fully characterize the relationship of interest. Next, there are some limitations to our two-step model as estimated slopes, which are used for predictive calculations, are subject to errors. But our two-step model is very intuitive and simple, comparing to existing complex models used in previous longitudinal studies. Finally, this project only included PiB-PET as a potential biomarker of cognitive decline, as a way to better mimic potential clinical data, which is often limited to single biomarker. In order to fully understand what biological changes are best predictive of the eventual cognitive decline associated with AD, it may be helpful in future analyses to include a wider selection of relevant biomarkers, such as volumetric Magnetic Resonance Imaging (MRI), tau PET, or fluid biomarkers. In summary, the current study provides evidence that baseline Aβ is more accurate than longitudinal Aβ for predicting cognition decline. Importantly, this study also confirms that covariates, such as the presence of the APOE ε4 allele, have deleterious group level effects on the ability of longitudinal Aβ measures to predict cognitive decline. From a clinical point of view, these results are encouraging as they provide evidence that only one amyloid positive PET scan is necessary in order to make predictions about future cognitive decline in individuals at risk for developing AD. This is especially important, as PET imaging is a limited resource that will be difficult to scale to large clinical populations, and requires the use of radioactive radiotracers, so minimizing the exposure to these for patients is an important step forward for understanding and treating AD. Finally, given that PET is able to image one of the earliest pathological changes known to occur in the preclinical phase of AD, Aβ deposition, our study provides strong support for the use of this technology for early identification of those who are most at risk of developing AD-related dementia.

The use of biomarkers for the early detection of Alzheimer’s disease (AD) is crucial for developing potential therapeutic treatments. Positron Emission Tomography (PET) is a well-established tool used to detect β-amyloid (Aβ) plaques in the brain. Previous studies have shown that cross-sectional biomarkers can predict cognitive decline (Schindler et al.,2021). However, it is still unclear whether longitudinal Aβ-PET may have additional value for predicting time to cognitive impairment in AD. The current study aims to evaluate the ability of baseline- versus longitudinal rate of change in- 11 C-Pittsburgh compound B (PiB) Aβ-PET to predict cognitive decline. A cohort of 153 participants who previously underwent PiB-PET scans and comprehensive clinical assessments were used in this study. Our analyses revealed that baseline Aβ is significantly associated with the rate of change in cognitive composite scores, with cognition declining more rapidly when baseline PiB Aβ levels were higher. In contrast, no signification association was identified between the rate of change in PiB-PET Aβ and cognitive decline. Additionally, the ability of the rate of change in the PiB-PET measures to predict cognitive decline was significantly influenced by APOE ε4 carrier status. These results suggest that a single PiB-PET scan is sufficient to predict cognitive decline and that longitudinal measures of Aβ accumulation do not improve the prediction of cognitive decline once someone is amyloid positive. Keywords Abbreviations β-amyloid Alzheimer’s disease Clinical Dementia Rating Pittsburgh compound B standardized uptake value ratio Mild cognitive impairment

CRediT authorship contribution statement Gengsheng Chen: . Nicole S. McKay: Writing – review & editing. Brian A. Gordon: Conceptualization, Methodology, Writing – review & editing. Jingxia Liu: Methodology, Writing – review & editing. Nelly Joseph-Mathurin: Writing – review & editing. Suzanne E. Schindler: Writing – review & editing. Jason Hassenstab: Writing – review & editing. Andrew J. Aschenbrenner: Writing – review & editing. Qing Wang: Writing – review & editing. Stephanie A. Schultz: Writing – review & editing. Yi Su: Writing – review & editing. Pamela J LaMontagne: Data curation. Sarah J. Keefe: Data curation. Parinaz Massoumzadeh: Writing – review & editing. Carlos Cruchaga: Writing – review & editing. Chengjie Xiong: Methodology. John C. Morris: Data curation, Funding acquisition, Project administration. Tammie L.S. Benzinger: Conceptualization, Funding acquisition, Investigation, Project administration, Resources, Supervision, Writing – review & editing. Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Dr. Benzinger has held investigator-initiated research funding from the NIH, the Alzheimer’s Association, the Barnes-Jewish Hospital Foundation, Siemens Healthiness and Avid Radiopharmaceuticals (a wholly owned subsidiary of Eli Lilly). Dr. Benzinger participates as a site investigator in clinical trials sponsored by Avid Radiopharmaceuticals, Eli Lilly, Biogen, Eisai, Jansen, and Roche. Dr. Benzinger performs paid and unpaid consulting for Biogen, Eli Lilly, Eisai, Roche and Siemens. SES is analyzing biomarker data provided by C2N Diagnostics to Washington University.

Further reading Supplementary data The following are the Supplementary data to this article: Data availability Data will be made available on request. Acknowledgments This study was supported by NIH P50AG00561, P30NS09857781, P01AG026276, UL1TR002345, P01AG003991, R01AG043434, UL1TR000448 and R01EB009352. Freesurfer computations were performed using the Washington University Center for High Performance Computing, which is partially supported through grant 10.13039/100000097 NCRR 1S10RR022984-01A1.

CC BY

no

2024-01-16 23:41:58

Neuroimage Clin. 2023 Dec 15; 41:103551

oa_package/3c/a4/PMC10788301.tar.gz

PMC10788305

38150746

Introduction Brain structure and function undergo drastic transformations over the first two decades of life ( Stiles & Jernigan, 2010 ). Over the same period, there are also profound changes in multiple facets of sleep, potentially reflecting developmental programs of brain reorganization ( Knoop et al., 2021 ). Given abundant evidence for the critical impact that sleep has on memory consolidation and working memory, deviations from normal sleep patterns during childhood and adolescence – the most learning-intensive periods in life – might have long-lasting consequences ( Kopasz et al., 2010 ). For example, active processes taking place during sleep support synaptic pruning and connectivity restructuring ( Buchmann et al., 2011 , Tononi and Cirelli, 2014 ). A more precise characterization of normative childhood trajectories of sleep may therefore aid our understanding of brain developmental processes and shed light on how subsequent cognitive and behavioral dysfunction can emerge. Children with neurodevelopmental disorders (NDD) and related diagnoses including epilepsy are exceptionally vulnerable to sleep problems ( Kamara and Beauchaine, 2020 , Robinson-Shelton and Malow, 2015 ). Sleep dysregulation has been documented in individuals with autism spectrum disorder (ASD) ( Souders et al., 2017 ), attention deficit / hyperactivity disorder (ADHD) ( Becker, 2020 ), cerebral palsy ( Simard-Tremblay et al., 2011 ), Down syndrome ( Stores & Stores, 2013 ), and epilepsy ( Larson et al., 2012 ). Comorbid conditions that commonly accompany clinical manifestation of NDDs, including intellectual disability, have also been associated with sleep abnormalities ( Surtees et al., 2018 ). However, the majority of these reports focused on a single disorder despite evidence of trans-diagnostically shared risk factors and pathogenic mechanisms, especially for psychiatric disorders ( Gandal et al., 2016 , Thapar et al., 2017 ). A second limitation is that the most-studied aspects of sleep have been macro-level metrics (e.g. time in bed) approximated by parental report. Such measures are inherently limited with respect to characterizing brain activity. In contrast, electrophysiological characterizations of sleep micro-architecture reflect underlying processes during different sleep stages more directly. One recent review described the presence of sleep oscillation (e.g. spindles, slow oscillations) abnormalities among NDDs but also highlighted the scarcity of relevant studies and limited sample sizes which, combined with varying age ranges and inconsistent methodologies, precluded strong conclusions ( Gorgoni et al., 2020 ). Across the lifespan, when investigating age-related change numerous studies have adopted the paradigm of predicting chronological age based on neuroimaging or other data sources, on the assumption that its deviation from observed age – the brain age gap – is a putative marker of overall brain health. Compared to healthy controls, altered “brain age gaps” have been observed in various adult-onset disorders and conditions including Alzheimer disease, mild cognitive impairment, schizophrenia and multiple sclerosis ( Baecker et al., 2021 ). Accelerated aging – larger positive brain age gaps – has also been shown to follow traumatic brain injury ( Cole et al., 2015 ). Conversely, opposite patterns in children born very prematurely have been interpreted as reflections of delayed brain development ( Franke et al., 2012 ). Although the majority of studies on brain age used structural MRI data, with predictions usually achieving high correlations ( r > 0.9) with chronological age ( Franke & Gaser, 2019 ), recent reports in adults demonstrated that the sleep EEG can also give comparable results ( Nygate et al., 2021 , Sun et al., 2019 ). In aggregate, these findings suggest that sleep EEG features track strongly with age and may be valuable for mapping typical and atypical childhood development. In the present study, we used polysomnography (PSG) data from a large clinical cohort to comprehensively chart the developmental trajectories of metrics derived from sleep electroencephalograms (EEG) across childhood and adolescence. Based on clinical records, we initially removed individuals with any of six major clinical groups including NDDs and used data from remaining individuals ( N = 1,828, 2.5 – 17.5 years) to characterize neurodevelopment in non-NDD sample. We further validated findings in an independent sample of generally healthy children with snoring. This latter group of children participated in a clinical trial and were screened to be free of severe neurodevelopmental delay (estimated by the Differential Ability Scales II), although they had a more restricted age range ( N = 1,213, 4.5 – 10 years). Based on the profile of associations we detected, we then developed a multivariate, joint model to predict chronological age as a function of sleep macro- and micro-architecture. Specifically, we tested 1) whether such a model was transferable between different studies and populations, and 2) whether deviations between predicted and chronological age distinguished children with NDD from non-NDD children. For the last two goals we employed two additional datasets: the Pediatric Adenotonsillectomy Trial for Snoring, (PATS, N = 627 children with snoring) and the Cleveland Family Study (CFS, N = 730), a family-based study covering a wide age range from 6 to 88 years.

Methods Participants We primarily used PSG data from two pediatric samples – the Nationwide Children’s Hospital Sleep DataBank (NCH) and the Child Adenotonsillectomy Trial (CHAT) – both available via the National Sleep Research Resource ( http://sleepdata.org ). The NCH sample was created to facilitate pediatric sleep research. It was composed of patients (from infants to some adults) who underwent clinical PSG from 2017 to 2019 at the Nationwide Children’s Hospital ( Lee et al., 2022 ), and contained diagnostic (ICD 9/10 codes) and medication data. All the data were de-identified prior to NSRR deposition, and received NCH Institutional Review Board exemption with HIPAA waiver. The CHAT sample was derived from data collected from six US pediatric clinical centers as part of sleep screening procedures for a clinical trial of children aged 5 to 9 with snoring who were candidates for adenotonsillectomy. All participants were without severe chronic medical conditions or ADHD requiring medications (in total 12 participants had ADHD diagnosis), presented with snoring and were potential candidates for adenotonsillectomy ( Marcus et al., 2013 , Weinstock et al., 2014 ). All children (n = 1,244) were screened with a PSG (for detailed description of inclusion criteria for the data acquisition see Marcus et. al., 2013) and here we used the sample of all screened children excluding the follow-up recordings that were available for participants with mild to moderate obstructive apnea. Data collection for CHAT was approved by local Institutional Review Boards and written informed consent was obtained from each individual or their legal guardians. Primary exclusion criteria for the NCH sample were i) age younger than 2.5 years (due to potential differences in infant and toddler EEGs), ii) age above 17.5 years (due to sparsity of data, and iii) a narcolepsy diagnosis (n = 42). In CHAT, we removed individuals for whom age information was missing (N = 19). In both samples, if the same individual had multiple recordings available, we only used a single (the first) recording. For the brain age analyses only, to expand the age range of the validation set, we additionally included individuals ( N = 1008) from the Pediatric Adenotonsillectomy Trial for Snoring (PATS) dataset comprised of children between 3 and 13 years with snoring but with an AHI < 3 ( Wang et al., 2020 ). This allowed us to test the transferability of the model. Cleveland Family Study sample ( N = 730, including individuals between 6 and 88 years) was used for analysis testing the possibility of accelerated aging in the DS subgroup. Clinical information for NCH sample The DIAGNOSIS.csv file (available via NSRR) was used to delineate clinical sub-groups in the NCH sample. Following recommendations from the original description of the dataset ( Lee et al., 2021 ), we only used final diagnosis codes (DX_ENC_TYPE & DX_SOURCE_TYPE columns equal to “Final Dx”). Since diagnostic codes provided for the sample were either according ICD9 or ICD10, we searched for specific diagnoses using the string search based on the diagnosis description (DX_NAME). For example, searching a string “[Aa]utis” and visually checking all unique matching diagnoses as well as the ICD codes. The information for all matching diagnoses for each condition is provided in the Supplementary table 1. To control for medication use, we used records available in the MEDICATION.csv file. Specifically, we identified participants whose PSG was performed between the prescribed medication start and end date. We identified four therapeutic classes of medication out of 42 that could potentially affect sleep: 1) antihistamines, 2) psychotherapeutic drugs, 3) CNS drugs, including anticonvulsants, 4) hormones, and 5) sedatives/hypnotics. We summarized them by therapeutic class and subclass (THERA_CLASS and THERA_SUBCLASS), pharmaceutical class (PHARMA_CLASS). EEG preprocessing All steps of sleep EEG data processing were performed using Luna ( http://zzz.bwh.harvard.edu/luna/ ), an open-source package developed by us (S.M.P). All NCH, CHAT and PATS PSGs contained six EEG channels (F3, F4, C3, C4, O1, O2). In CHAT, two temporal channels (T3, T4) were also available. The CFS cohort contained C3 and C4 EEG channels only. We first selected 30-seccond epochs of a particular stage (N2, N3, REM) according to manual, AASM-based staging (in the CFS it was performed using only central channels) in all datasets. Since the original sampling rates varied between and within the datasets (256–512 Hz in NCH and 200–512 Hz in CHAT, 200 Hz in PATS and 128 Hz in CFS), all EEG signals exceeding 200 Hz were datasets were down-sampled to 200 Hz and for CFS sample rate of 128 Hz was kept unchanged. In all datasets, EEG signals were referenced to contralateral mastoids (M1 or M2), converted to uV units and bandpass filtered between 0.5 and 35 Hz (raw signals were exported without any additional filters being set). Due to excessive line noise interference observed in many NCH samples, we applied an approach to remove it based on spectrum interpolation ( Leske & Dalal, 2019 ), as implemented in Luna. Next, within each stage, we identified all epochs with maximum amplitudes above 200 uV, or with flat or clipped signals for more than 10 % of the epoch; further, epochs were marked as outliers if they were i) more than 3 SDs from the mean (for that individual) of all channels for any of the three Hjorth parameters, activity, mobility and complexity ( Hjorth, 1970 ), ii) 4 SDs from the mean of other epochs of the same channel or iii) 4 SDs from the mean of all epochs across all channels. Hjorth-based epoch outlier removal was performed twice for each individual. Channels and/or epochs were removed if more than 50 % of epochs were outliers. Such thresholds were selected empirically to remove gross artifacts from the signals but also to avoid removing too many epochs. The quality of the signals was inspected visually in several randomly selected EEG recordings and visualizing spectral power across frequency for all signals across all studies. The final averaged number of epochs per participant for the NCH dataset were –M(SD, range) – N2: 370 (95, from 35 to 869), N3: 230 (74, from 24 to 654), R: 151 (58, from 10 to 379); and for CHAT N2: 374 (89, from 96 to 686), N3: 282 (79, from 53 to 652), R: 165 (47, from 10 to 349). Spectral power estimation Spectral power was estimated using Welch’s method separately for N2, N3 and REM, summarized by classical frequency bands – slow (0.5–1 Hz), delta (1–4 Hz), theta (4–8 Hz), alpha (8–12 Hz), sigma (12–15 Hz), beta (15–30 Hz), and total power (0.5 to 35 Hz). Specifically, for each 30-seccond epoch, we applied the Fast Fourier Transform with 4 s segments (0.25 Hz spectral resolution) windowing with a Tukey (50 %) taper, with consecutive segments overlapping by 50 % (2 s). We then averaged power across all segments per epoch. Subsequently, epoch-wise power was averaged across all epochs for a particular channel and stage. Relative power was computed with respect to the total absolute power. Absolute power values were then log-transformed prior to analysis. Spindle detection Motivated by recent findings that two classes of spindles – slow frontal and fast central – emerge as early as 18 months after birth ( Kwon et al., 2022 ), we detected both separately. Slow and fast spindles were detected using 7-cycle wavelets with center frequencies of 11 Hz and 15 Hz (with approximately +/-2 Hz around each central frequency) respectively as previously described ( Purcell et al., 2017 , Warby et al., 2014 ). Specifically, putative spindles were identified based on temporally smoothed (window duration = 0.1 s) wavelet coefficients (from a complex Morlet wavelet transform) using following criteria. Intervals exceeding 1) 4.5 times the mean for at least 300 ms and also 2) 2 times the mean for at least 500 ms were selected as putative spindles. Intervals over 3 s were rejected; consecutive intervals within 500 ms were merged (unless the resulting spindle was greater than 3 s). Subsequently, additional quality check (QC) procedure was applied. Putative spindles were discarded if the relative increase in non-spindle bands activity (delta, theta, and beta) was greater than the relative increase in spindle frequency activity (i.e. relative to all N2 sleep), thereby ensuring putative spindles preferentially reflect sigma band activity, and not general increases in signal amplitude, which is often due to artifact or other non-spindle activities. Based on the set of spindles that passed QC, we computed spindle density (count per minute), amplitude, duration, observed frequency, and chirp (intra-spindle frequency change computed as a difference in frequency between the first and the last half of a spindle with zero values meaning no change and negative values meaning intra-spindle frequency deceleration). SO detection Zero-crossings were identified based on the EEG signals band-pass filtered between 0.5 and 4 Hz. To define putative SO the following temporal criteria were satisfied: 1) a consecutive zero-crossing leading to negative peak was between 0.3 and 1.5 s; 2) a zero-crossing leading to positive peak were not longer than 1 s. With respect to amplitude criteria, two separate approaches were used, similar to Djonlagic et al (2020) . First, an adaptive/relative threshold (our default) such that negative peak and peak-to-peak amplitudes were required to be greater than twice the mean (for that individual/channel). Second, an absolute threshold requiring a negative peak amplitude larger −40 uV, and peak-to-peak amplitude larger then 75 uV. SO density (count per minute) as well as the mean amplitude of the negative peak, peak-to-peak amplitude, duration and the upward slope of negative peak were estimated for each channel. Coupling between SO and spindles For each channel we identified spindles that overlapped with detected SO and characterized their coupling using the following three metrics. First, we computed the proportion of spindles that overlapped with a SO (“gross overlap”). Using the filter-Hilbert method, we also estimated SO phase at the spindle peak, which was averaged (circular mean) across SOs for each channel (coupling angle at spindle peak). In addition, the inter-trial phase clustering assessed the consistency of non-uniform phase coupling between SO and spindles (coupling magnitude). Overlap and magnitude metrics were z-transformed using a null distribution of same metrics generated during 10,000 random permutations where time indices of the time series were shuffled in a manner that preserved the overall number of SOs, spindles and the gross overlap between SO and spindles (the latter is true only for the coupling magnitude). Exclusion criteria based on sleep data For all used datasets (NCH, CHAT, PATS and CFS), an additional exclusion criterion applied for the macro-architecture analysis was TST < 180 mins. For analysis of spectral power and spindles, additional exclusion criteria were applied: i) N of available epochs for each stage (N2, N3, REM) after outlier removal less than 10, ii) persistent line noise interference (SPK measure > 5 SD in least one channel at any stage), iii) outlier spectral power at 1 Hz (<4 SD or > 4 SD in least one channel at any stage) to target movement, ocular artifacts, general low signal to noise ratio, or at 25 Hz (>4 SD in least one channel at any stage) to target muscle activity artifacts. Signal polarity flips were observed in a portion of recordings in all samples (for details on polarity in several NSRR samples: https://zzz.bwh.harvard.edu/luna/vignettes/nsrr-polarity/ ) and additional exclusion criteria were necessary for analyses dependent on signal polarity – those involving SO and coupling between SO and spindles. Recordings with ambiguous polarity were removed for these analyses (-1 < T_DIFF < 1 at C3 or C4 during N2 stage from Luna's POL command) and polarity of all recordings with T_DIFF > 1 at C3 or C4 during N2 stage was flipped. Final sample size and demographic characteristics for the primary samples used (for NCH and CHAT) are in Table 1 . Characteristics of the final analytic sample of PATS that was included for brain age prediction were: total N = 627, with 307 females, 384 White, 198 Black or African American individuals and 45 individuals of other ancestry, with a mean age of 6.3 years (range 3 – 13 years); and for CFS: total N = 635, with 361 females, 271 Whites, 343 Blacks and 21 individuals of other ancestry, with a mean age of 39.4 years (range 6.7 – 88.5 years). Statistical analysis We used linear regression models of each sleep metric regressed on age, also controlling for sex and race/ethnicity. Outliers (using a 3 SD criterion) were removed for each sleep metric (repeated twice). In a control analysis we added Apnea-hypopnea index (AHI) computed as an average number of apnea and/or hypopnea events per hour of sleep as an additional covariate to linear regression models, which yielded almost identical results (Pearson correlation between signed -log 10p-value controlling and not controlling for AHI was > 0.99 in both NCH and CHAT samples). We also performed similar analysis using arousal index (AI) for a subset of individuals from the CHAT sample (N = 374, age range 5–10 years) for whom precomputed AI was available on NSRR. To describe effect sizes of age-related changes in sleep metrics, we also calculated Pearson correlations between sleep metrics and age. For metrics with suspected non-linear trajectories, we used Akaike’s and Bayesian information criteria to formally test if the quadratic model was a better fit to describe age-related change. In analyses of NCH clinical subgroups, we used linear regression models controlling for race, sex, AHI, co-occurring sleep diagnoses, other NDD diagnoses and medication use. For Sup. Fig. 5 , prior to running a linear regression analysis, 3-SD outliers were replaced with NAs in two rounds and sleep metrics’ estimates were z-transformed to obtained standardized linear regression coefficients. P-values were adjusted for multiple comparisons (all sleep estimates n = 321) separately for each subgroup and tested effect (subgroup, subgroup by age interaction) using FDR method ( Benjamini & Yekutieli, 2001 ). To predict individuals' ages, we trained a multiple linear regression model using the sleep metrics studied here. After excluding subjects from NDD subgroups, we randomly split the non-NDD NCH sample into training (70 % of subjects) and held-out (30 %) sets. CHAT, PATS and NDD-NCH subgroups sample were retained as additional, independent testing sets. To reduce the number of features for prediction model, we removed highly correlated variables (abs r > 0.9 in non-NDD NCH training set). Remaining features (list is provided in Supp. Table 7) for all datasets were z-transformed using the mean and standard deviation of the training set. Sex and race were included as covariates in all models. Initially, the model performance in the non-NDD NCH training sample was estimated using 10-fold cross validations using mean explained variance (R2) and mean absolute error and their SD across folds. Further, the model was tested in four held-out validation sets (non-NDD NCH held-out sample, CHAT, PATS and NDD NCH set) and pearson’s correlation, mean absolute error and mean error between predicted and true chronological age were reported. We additionally applied an alternative to conventional linear regression: specifically, gradient descent boosting machines as implemented in the LightGBM machine learning library as implemented in Luna. Performance was near identical to the linear regression model, in this particular case, and so we retained the simpler model, as performance in terms of age prediction was already high. In the NDD NCH subgroups analysis of brain age gap we estimated the variability in the prediction of results by repeating the step of fitting the model 100 times with different, non-overlapping individuals from non-NDD NCH randomly assigned to held-out testing and training set each round. The NDD NCH subgroups were kept the same. Estimates of model performance – mean absolute error and mean error between predicted and true age – were then averaged across 100 rounds, as well as their min and max values. We used a two-sample t -test to test if there was significant difference in MAE and ME (mean error) between non-NDD NCH held-out testing set and each clinical subgroup in each round and reported the median p-value in Fig. 5 D. We applied the following steps to estimate similarity in all sleep variables (macroarchitecture, absolute and relative band power, spindles, SO estimates and coupling metrics) across different age bins of combined sample of PATS and CFS cohorts and DS group ( Fig. 6 D). First, we z-scored all sleep variables across individuals of the combined sample and NCH sample using mean and standard deviation of the combined sample of each sleep metric Then, we defined age bins in the combined sample (two-year non-overlapping windows centered at 3, 5, 7, 9, 11, 13 and four-year non-overlapping windows centered at 16, 20, ..., 80 years). Larger age windows for older ages were chosen due to expectation of lower rate of change. Another reason was to ensure that there were sufficient number of participants for each bin (at least 10). Then means were computed for each sleep metric across individuals belonging to a particular age bin in the combined sample, as well as the DS subgroup and non-NDD group of the NCH sample. Finally, the average absolute difference between DS (and non-NDD) means and each age bin means were computed and plotted as a function of age. CRediT authorship contribution statement N. Kozhemiako: Conceptualization, Methodology, Formal analysis, Writing – original draft. A.W. Buckley: Conceptualization, Writing – review & editing. R.D. Chervin: Conceptualization, Writing – review & editing. S. Redline: Conceptualization, Resources, Writing – review & editing. S.M. Purcell: Conceptualization, Methodology, Software, Writing – original draft, Funding acquisition.

Results The primary discovery dataset comprised 2,800 individuals with whole-night PSGs from the Nationwide Children’s Hospital (NCH) Sleep Databank (accessed via the National Sleep Research Resource, NSRR ( Zhang et al., 2018 )). Based on ICD codes, we defined six subsets within the NCH sample based on the presence of a diagnosis of the following neurodevelopmental and childhood onset disorders: ASD, ADHD, intellectual disabilities, Down syndrome (DS), cerebral palsy (CP) and epilepsy (for brevity, below we use “NDD group” to refer to these sub-cohorts although we acknowledge that epilepsy is usually not classified as an NDD). We selected these six subgroups based on each having N > 100 subjects in the full dataset and previous reports of alterations in sleep ( Becker, 2020 , Larson et al., 2012 , Simard-Tremblay et al., 2011 , Souders et al., 2017 , Stores and Stores, 2013 , Surtees et al., 2018 ) (see Table 2 for demographic details, and Sup. Table 1 for diagnostic details). There was a considerable overlap between NDD diagnoses (Sup. Fig. 1a). As expected in this clinically-referred and ascertained sample, most NCH individuals had a sleep-related clinical diagnosis, precluding a straightforward definition of a “healthy control” comparison group. Although we refer to the “non-NDD sample” of NCH, we note that the individuals therein collectively had more than 9,000 unique diagnostic codes in their medical records, for both acute and chronic disorders (although not necessarily contemporaneous with the PSG). For example, there were diagnoses of cough ( N = 1176), obesity (N = 793), skin rash ( N = 633), unspecified disturbance of conduct ( N = 215) and anxiety disorder ( N = 210). Excluding the abovementioned six NDD groups, 93 % of participants had one or more sleep-related diagnostic codes including sleep apnea, insomnia, hypersomnia and others, although sleep disorders were more prevalent still among the NDD groups (Sup. Fig. 1b). To address possible medication effects on sleep architecture, we further identified individuals across the NCH sample who were prescribed medications likely to affect sleep at a time overlapping the PSG recording. Only 14 % (262 out of 1,829 non-NDD NCH individuals) of the NCH sample had such medication prescribed at the night of PSG, with a majority being antihistamines (9 %) (Sup. Fig. 1c, Sup. Table 2). The proportion of individuals prescribed sleep-impacting medication during PSG was substantially higher among the NDD subgroups (42 %). Medication use was therefore added as a covariate in our primary analyses (see Methods for more details). For our initial analyses of normative age-related changes in sleep we excluded the NDD groups, resulting in a final sample of N = 1,828 (detailed demographic information in Table 1 ). To provide an independent replication cohort, we obtained PSGs from N = 1,213 individuals from the Child Adenotonsillectomy Trial (CHAT) study, comprising children with reported snoring who were screened to be without neurodevelopmental delay. Sleep macro-architecture in children without NDD Macro-architecture metrics (i.e. those derived from the hypnogram based on manual staging) showed substantial age-related changes. Congruently in both samples, total sleep time (TST) exhibited a marked linear reduction with age (controlling for race/ethnicity and sex) in both NCH ( r = -0.26, p < 10 -15 ) and CHAT ( r = -0.13, p = 9 × 10 -6 ), as did sleep maintenance efficiency (SME) ( r = -0.17, p = 2 × 10 -12 in NCH, r = -0.07, p = 0.021 in CHAT) ( Fig. 1 A). Age-related effect sizes in CHAT are expected to be attenuated and/or more variable than in NCH, due to the narrower age range. Sleep also grew more fragmented with age in both samples, based on increases in the sleep fragmentation index (SFI) and duration of wake after sleep onset (WASO). The macro-architectural feature showing the strongest age-related change in both datasets was the number of NREM sleep cycles ( r = -0.38, p < 10 -15 in NCH and r = -0.13, p = 4 × 10 -6 in CHAT), which remained significant even after covarying for TST (p < 10 -15 in NCH and p = 0.002 in CHAT). At the same time, the average duration of sleep cycles increased with age in both samples, although the number of cycles was still the greatest determinant of TST (e.g. in NCH, r = 0.5 for cycle number compared to r = 0.12 for cycle duration). As others have reported ( Baker et al., 2016 , Feinberg et al., 2012 ), sleep stage composition changed profoundly across this developmental period ( Fig. 1 B). We observed similar effects in both cohorts, with one exception for N1 ( Fig. 1 B, top row). N1 duration also showed the largest absolute difference between the datasets, possibly reflecting variations in manual staging protocols and the intrinsically low construct validity of N1 as a distinct and atomic physiological state. Stage N2 duration increased with age from 177 to 201 min between 4 and 16 years of age in NCH sample ( p < 10 -13 in NCH, p = 0.003 in CHAT). In contrast, N3 sleep reduced with age from 126 min in 4-year-olds to 81 min in 16-year-olds (p < 10 -15 in NCH and p = 10 -5 in CHAT) suggesting an age-related reduction in mean NREM depth. In general, the proportion of time spent in all NREM (stages N1, N2 and N3) increased with age ( r = 0.29, p < 10 -15 in NCH and r = 0.19, p = 3 × 10 -10 in CHAT) while stage R displayed age-associated reduction in both duration and proportion ( p < 10 -15 in NCH and p < 10 -13 in CHAT) from 92 to 64 min between 4 and 16 years. Further, REM latency computed with regard to sleep onset increased with age ( r = 0.2, p < 10 -15 in NCH, r = 0.13, p = 7 × 10 -6 in CHAT). Although there was evident difference in R latency between the two cohort, the estimates were comparable to the previous reports summarized in ( Scholle et al., 2011 ). There were also fewer transitions between NREM and R periods with age in both datasets. Sleep EEG spectral characteristics Within sleep stage, spectral power across classical frequency bands displayed large age-dependent changes ( Fig. 2 ), most notably a reduction in absolute spectral power in slower frequency bands (e.g. the largest effect size is illustrated in Supp. Fig. 2 A – delta band absolute power during R stage, r = -0.82 in NCH and r = -0.3 in CHAT). These effects likely reflect gross changes in the amplitude of the sleep EEG (most evident for slower bands that have higher power due to the 1/ f nature of the power spectrum); that they were observed uniformly across sleep stages and channels suggests that these effects may not be specific to sleep neurophysiology (versus gross anatomical changes, for example). In contrast, absolute sigma power increased with age most strongly during N2 ( p < 10 -15 in NCH and p < 0.001 in CHAT across all channels with strongest effects at O1, r = 0.39 and r = 0.16 for NCH and CHAT respectively). A smaller but still significant increase was observed during N3, but not REM. Modelling with higher-order age terms suggested a nonlinear trajectory (based on Akaike’s and Bayesian information criteria), reflecting a slight decline in power occurring in adolescence ( Fig. 2 ). Relative sigma power (normalized by total power 0.5 – 35 Hz to account for age-related changes in total power) similarly showed the strongest age-related changes (e.g. during N2 at O1, r = 0.76 in NCH and r = 0.4 in CHAT, Supp. Fig. 2 B, with similar effects in N3, all p < 10 -15 in NCH and p < 10 -5 in CHAT). Other frequency bands and stages showed marked developmental changes in relative power with qualitatively distinct stage- and topographically specific developmental trajectories ( Fig. 3 ). Considering NCH only, delta power decreased with age during R across all channels (all p < 10 -15 , max effect size at F4 r = -0.67), but increased with age in frontal channels, during N2/N3, with a peak around the age of puberty. In occipital channels, delta power decreased with age across all stages (all p < 10 -15 , with r ranging from r = -0.4 to 0.66). In contrast, theta power increased with age during R (all p < 10 -10 , max effect size at C4 r = -0.23) but decreased with age during N2/N3 (all p < 10 -15 , max effect size at F3 r = -0.63 during N2) in frontal and central channels. For the comparable age range, we observed broadly consistent patterns of NREM age-related change in CHAT. Finally, during REM, relative alpha, sigma and beta power increased with age (all p < 10 -15 , effect sizes ranging from r = 0.6 to 0.76 in NCH and all p < 10 -10 , effect sizes ranging from r = 0.21 to 0.35 in CHAT). Spindles, slow oscillations and their coupling Across childhood and adolescence, we observed multiple changes in NREM sleep spindles ( Fig. 4 ). Although the density (count per minute) of both slow and fast spindles (SS and FS respectively, targeting 11 Hz and 15 Hz activity, with approximately +/-2 Hz around each central frequency) increased with age in frontal channels (all p < 10 -15 , from r = 0.21 to 0.45 in NCH and p < 10 -14 , from r = 0.22 to 0.26 in CHAT), their developmental trajectories were distinct ( Fig. 4 ). Whereas FS density linearly increased across all channels (all p < 10 -15 in NCH and p < 10 -10 in CHAT) from 0.9 spindles per minute at age of 4 to 1.9 at age of 16 at C3, SS density displayed an inverted-U profile, most pronounced in frontal channels and peaking around 10 years with 2.4 spindles per minute at F3 (compared to 1.5 and 1.9 spindles at 4 and 16 years of age, respectively). Spindle morphology, not rate of occurrence, showed the most marked age-related changes, however. In particular, intra-spindle deceleration (sometimes called “chirp”, a typical characteristic of both fast and slow spindles) grew more pronounced with age, especially for SS ( Fig. 4 , p across all channels < 10 -15 in NCH and p < 10 -7 in CHAT, with strongest effects at F3: r = -0.66 and r = -0.38 in NCH and CHAT respectively (Sup. Fig. 2 C) as opposed to largest effect in FS density r = 0.53 and r = 0.29 in NCH and CHAT). Average spindle frequency varied markedly with increasing age, but differently for slow and fast spindles ( Fig. 4 ): whereas SS frequency became faster with age (across all channels p < 10 -15 , max effect size at F3 r = 0.5 in NCH and p < 10 -10 , max effect size at F4 r = 0.3 in CHAT), FS frequency slowed down (all p < 10 -15 , max effect size at F4 r = -0.5 in NCH and p < 10 -8 , max effect size at F4 r = -0.29 in CHAT). Opposing directions of change in frequency of SS and FS was also observed in older adults ( Djonlagic et al., 2020 ). Notably, both SS and FS developmental trajectories were highly non-linear with SS maximum and FS minimum frequency observed around 13 years of age. Note, that the total number of detected spindles (especially FS in occipital channels) was relatively low at younger ages somewhat limiting estimation robustness of their morphology characteristics. Nevertheless, they still displayed very consistent age-related trends and channel-specific patterns in both datasets. We detected slow oscillations (SOs) during N2 sleep by identifying zero-crossings in 0.5–4 Hz bandpass-filtered signals and applying fixed-duration but relative-amplitude thresholds (see Methods for details). As for spindles, SO density increased with age ( Fig. 4 ) across all channels in both cohorts (all p < 10 -15 in NCH and p < 10 -15 in CHAT). The steepest increase observed was at F4 with 10 SOs per minute at age of 4 and 15 at 16 years of age ( r = 0.7 in NCH and r = 0.38 in CHAT, Sup. Fig. 2D). With respect to SO morphology, average negative peak amplitude, peak-to-peak amplitude, and the slope between negative and positive peaks decreased linearly across all channels (p < 10 -15 in NCH and p < 10 -5 in CHAT, with the largest effect for peak-to-peak amplitude at O1, r = -0.74 in NCH and r = -0.23 in CHAT, Sup. Fig. 2F). SO duration was the only parameter to express a marked non-linear trajectory, with a steep increase from 4 to 10 years and a steady decline after age 10–12 years ( Fig. 4 ). With respect to inter-cohort differences, SO duration estimates (as well as relative power in slow frequency band across all stages) expressed some of the largest absolute differences between the NCH and CHAT cohort. That could potentially be due to distinct technical characteristics of PSG recording devises (e.g. inherent filtering setting). However, it is noteworthy that the age-related trends remained consistent across both cohorts. SO properties are necessarily dependent upon the criteria used to detect them. In our primary analyses, based on optimizing the strength of observed spindle/SO coupling, we elected to use a relative amplitude threshold (see Methods for details). For example, 63 % of individuals had significant FS/SO overlap at C3 in NCH (66 % in CHAT) versus 58 % and 56 % when an absolute SO amplitude threshold was used. The choice of SO detection criteria can impact patterns of age-related change, however. Indeed, using an absolute threshold, age-related trends for SO rate, slope and amplitude were reversed – the former decreased with age while the latter two increased (Sup. Fig. 3). This apparent contradiction reflects a general decrease in SO activity with age, which is congruent with the observed age-related decrease in slow and delta band power. Reduced SO activity consequently lowers any relative threshold, which can in turn lead to relatively more events being detected. The optimal choice of threshold is an empirical question that will depend on the subsequent analyses, which underscores the importance of always explicitly reporting the type of detection thresholds used. Returning to the original relative-threshold set of SO, we assessed spindle/SO coupling in three ways: 1) the tendency for spindles to preferentially occur non-uniformly with respect to SO phase (coupling magnitude), 2) the preferred SO phase at spindle peaks (coupling angle), and 3) the extent of any above-chance overlap between spindle and SO events, ignoring SO phase (coupling overlap). In general, SS showed more marked age-related changes, compared to FS ( Fig. 4 ). SS coupling increased in frontal and central channels in both cohorts (all p < 10 -15 in NCH and p < 10 -1 in CHAT: e.g. at F3 r = 0.56 in NCH and r = 0.44 in CHAT, Supp. Fig. 2 E). With respect to FS coupling, an age-related increase was observed only in the NCH sample (across all channels p < 10 -15 ), primarily driven by a steep increase in adolescence ( Fig. 4 ). The lack of CHAT replication here likely reflects the restricted age range: indeed, among NCH individuals 10 or under associations were greatly attenuated, compared to the same tests in the older NCH subsample (data not shown). Next, we found that individual's preferential SO phase angle at spindle peak (circular mean) shifted across development. In both cohorts, FS tended to occur before the SO positive peak, whereas SS tended to occur after it. However, with increasing age, both SS and FS shifted closer to the SO positive peak. Again, the most rapid change in preferred FS SO phase was during adolescence. Finally, rates of above-chance gross overlap between SS and SOs increased with age at frontal channels (F3/F4 p < 10 -15 , max r = 0.3 in NCH and p < 0.05, max r = 0.1 in CHAT) whereas FS overlap showed a modest (albeit significant) decrease at central channels (C3/C4 p < 0.01, max r = 0.17 in NCH and p < 0.05, max r = 0.13 in CHAT). Unlike other SO metrics, age-related changes in coupling were generally similar despite different approaches for SO detection (Sup. Fig. 3). Given that the majority of individuals in both cohorts had sleep apnea and/or snoring, we additionally retested all sleep variables for association with age after including apnea-hypopnea index (AHI) as a covariate. Results remained effectively identical, with a Pearson’s correlation r > 0.99 between signed log-transformed p -values in original and AHI-controlled analyses, also with comparable levels of significance. Same results were obtained using arousal index instead of AHI. Brain age prediction using sleep macro- and micro-architecture measures Above, we demonstrated 1) that age was strongly associated with multiple sleep macro- and micro-architecture metrics, and 2) that findings were congruent for two samples from distinct (clinical versus research) sources. We next aimed to condense these multivariate developmental patterns into a single model to estimate chronological age from the sleep EEG, and then to test whether its deviation from observed age - i.e. brain age gap - could identify pathological neurodevelopment. As our starting point, we fit a simple linear regression of age on sleep macro- and micro-architecture metrics adjusting for sex and race, using 70 % of the non-NDD NCH sample (i.e. after first excluding all clinical subgroups). We estimated performance using 10-fold cross-validation ( Fig. 4 A). An initial model including 258 sleep variables (see Supp. table 7 for the detailed list) and two covariates (sex and race) achieved R 2 = 0.89 (0.03 SD) and mean absolute error (MAE) of 1.08 years (0.11 SD). Whereas models trained on micro-architectural features only (151 spectral power, or 95 sleep spindle, SO and coupling metrics) performed almost as well as the full model, a model including only macro-architecture metrics displayed much lower performance ( Fig. 4 B). To validate and test model transferability, we applied the full NCH-derived model to three held-out validation sets, none of which included any NDD individuals ( Fig. 4 A): i.) the remaining 30 % of the non-NDD NCH sample, ii.) the CHAT sample, which was used as the replication dataset in the previous sections and iii.) a new pediatric dataset PATS (Pediatric Adenotonsillectomy Trial for Snoring, PATS: N = 627 [307 females]. The PATS dataset, similarly to CHAT, comprised children with snoring from diverse ethnic backgrounds participating in a clinical trial. Compared to CHAT, PATS had a wider age range that was closer to NCH’s (mean age 6.3 years, range 3 to 13 years). In all three test samples, the model predicted age with relatively high accuracy, indicating a high degree of transferability ( Fig. 5 C). The highest performance was observed in the NCH testing set, where predicted and observed age correlated r = 0.93. Although the correlations were lower in CHAT and PATS ( r = 0.56 and r = 0.85 respectively), this likely was in part due to the narrower age ranges; the MAEs were comparable to NCH (1.11–––1.31 years). Prediction accuracy was similar for boys and girls in all three testing samples ( p > 0.05), whether or not participant sex was not included in the model. Nonetheless there were still significant (albeit relatively subtle) sex differences and sex-by-age interactions in multiple measures of sleep macro and microarchitecture consistent with the previous literature ( Baker et al., 2020 , Campbell et al., 2012 ) (Sup. Fig. 4). Finally, the brain age gap (either MAE or ME) was also not significantly associated ( p > 0.05) with AHI in either NCH or CHAT. Despite significant linear associations across the majority of the sleep features, many of them displayed non-linear developmental trajectories. This prompted us to test whether our age prediction would improve using a non-linear approach. The results, however, were nearly identical when LightGBM – a tree-based learning algorithm – was employed using the same set of sleep EEG features (e.g. linear regression model vs LightGBM in the held-out PATS sample – r = 0.85 vs 0.85, MAE = 1.18 vs 1.1 years, ME = 0.48 vs 0.6 years). Brain age prediction and NDD Brain age gap was significantly negative (i.e. younger than expected) in individuals with DS (ME = -2.1 years, p = 9 × 10 -9 ) or intellectual disability (ME = -0.77 years, p = 0.02) subgroups ( Fig. 5 D). Interestingly, for DS the brain age gap effect was exacerbated with increasing age, suggesting greater developmental delays in older patients ( Fig. 6 E). Given prior reports of accelerated ageing (i.e. positive age gaps) in DS adults based on structural MRI and epigenetic markers ( Cole et al., 2017 , Horvath et al., 2015 ), we sought to contextualize our finding of delayed development in DS, especially given that the non-NDD NCH training sample was obligatorily limited to children and adolescents, combined with the fact that some sleep EEG metrics follow nonlinear, inverted-U trajectories across the lifespan (e.g. fast spindle density increases across childhood, peaks around 20 years of age and slowly declines thereafter, Fig. 6 B). We therefore estimated the developmental trajectories of sleep EEG metrics in a new dataset spanning both childhood and adulthood: PATS augmented by the Cleveland Family Study ( N = 730, 401 females, mean age of 41 spanning 6.8 to 89 years) ( Fig. 6 A). Given the nonlinear, inverted-U lifecourse trajectories of spindle density, one can identify groups of older adults with numerically equivalent mean spindle densities compared to young children, as well as compared to the DS group (e.g. average FS density at 3 and 76 years was 0.85 and 0.78n/min respectively compared to 0.63 in DS group, Fig. 6 B), which could be interpreted as (extreme) accelerated ageing in DS. However, looking across other sleep metrics, the childhood DS group did not resemble individuals of older age, but was instead close to profiles seen in younger children. This held both for individual metrics such as absolute theta power ( Fig. 6 C) as well as for similarity across a composite of sleep metrics ( Fig. 6 D). That is, considering the broad profile of all age-dependent sleep metrics, the DS group (mean age of 8 years) was more similar to the average for children aged 6 years (i.e. consistent with a delay) rather than any older adolescent or adult group. In absolute terms, the brain age gap was significantly larger across all NDD subgroups compared to the non-NDD NCH sample: DS (MAE = 2.48 years, p = 4 × 10 -7 ), intellectual disability (MAE = 2.06 years, p = 3 × 10 -5 ), CP (MAE = 1.6 years, p = 0.011), epilepsy (MAE = 1.57 years, p = 0.003), ASD (MAE = 1.46 years, p = 0.02) and ADHD (MAE = 1.3 years, p = 0.02) groups, consistent with greater levels of heterogeneity among NDD groups. To gain some insight into which sleep alterations were driving these brain age discrepancies in NDD, we performed a series of exploratory analyses within NCH, comparing each NDD subgroup with the non-NDD sample, statistically controlling for i) the other, potentially overlapping NDD subgroups, ii) diagnoses of major sleep disorders, iii) contemporaneous use of medications likely to impact sleep, iv) age, v) sex, vi) race and vii) AHI (see Methods for details). The results for all NDD groups are provided in Supp. Tables 3–6. Individuals with DS – the subgroup with the largest mean discrepancy between predicted and chronological age – showed the highest number of significant group differences and age-by-group interactions (Sup. Fig. 5 ). The largest effect sizes were for spindle characteristics – lower SS spindle density (1.1n/min in DS vs 1.9 in the non-NDD sample at F4, p < 10 -15 ), duration (at F3 b st = -1.1, p < 10 -15 ), less pronounced chirp (at C3 b st = 0.86, p < 10 -15 ) and absolute beta power (at F3 during N2 b st = 0.94, p < 10 -15 , Supp. Tables 3–5). In terms of age-related trajectories, coupling between SO and spindle metrics were among those showing stronger age-by-group interactions. For example, coupling overlap between SO and SS at F3 significantly decreased with age in DS ( Fig. 6 F, r = -0.28, p = 0.02) but increased in the non-NDD sample ( r = 0.17, p < 10 -10 ). Despite significant deviations in both MAE and ME, the number of altered sleep estimates were much lower in the intellectual disability group compared to the DS group; these were largely related to spindle frequency characteristics (Supp Fig. 5). SS density was associated with the degree of intellectual disability (the largest effect at C3 r = -0.39, p = 0.0009, Fig. 6 G) and was in general decreased compared to individuals without NDD in frontal channels (the largest effect at F3 b st = -0.29, p = 0.003, Supp. Tables 5, Fig. 5 H). Finally, multiple nominally significant group and age-by-group alterations were detected for ADHD, CP and Epilepsy ( Supp. Fig. 5 , Sup. Table 3–6 contain full reports of uncorrected results). Except for ASD, all NDD-subgroups had common significant ( p < 0.05) deficit of stage R sleep expressed as reduced duration (in DS), proportion (in intellectual disabilities) or both (in CP and epilepsy) and stage R latency (in ADHD) even after adjusting for sleep disorders and relevant medication.

Discussion Using both clinical and research datasets, we identified consistent patterns of age-related change during childhood and adolescence, for multiple facets of sleep macro- and micro-architecture. We further showed that, when combined, the different sleep EEG metrics we examined could reliably predict an individual's age in pediatric populations, and that the resulting models were broadly transferable across different cohorts. Finally, we showed that some NDD subgroups (primarily DS) exhibited systematic differences in their predictions of age, reflecting multiple disruptions of sleep architecture. Macro-architecture changes with age Confirming numerous previous reports of decreasing total sleep time during the first two decades of life ( Galland et al., 2012 , Iglowstein et al., 2003 , Ohayon et al., 2004 ), we additionally showed significant changes in other sleep macro-architecture metrics. Congruently with a large cross-sectional study of children without sleep complaints ( Scholle et al., 2011 ), we found an increase in sleep cycle duration, stage R latency and a decrease in a number of sleep cycles with age. With respect to sleep efficiency, for which both age-related increases ( Baker et al., 2012 , Scholle et al., 2011 ) as well as decreases ( Baker et al., 2016 ) have been reported, our findings pointed to a significant decline in both cohorts. Major developmental changes were also observed with respect to sleep stage composition. Prior reports of developmental trajectories of NREM sleep showed consistent findings of an increase in N2 and a reduction in N3 with age ( Baker et al., 2016 , Jenni and Carskadon, 2004 , Ohayon et al., 2004 , Scholle et al., 2011 , Tarokh et al., 2011 , Tarokh and Carskadon, 2010 ), as found in our analyses. In contrast, the reports diverge for stage R sleep showing either an increase ( Baker et al., 2016 , Feinberg et al., 2012 , Ohayon et al., 2004 ), no change ( Tarokh et al., 2011 , Tarokh and Carskadon, 2010 ), or a decrease ( Scholle et al., 2011 ) in stage R percentage and/or duration. Our analysis – which represents the largest study to date – supports a decrease in R sleep across ages 2.5 to 17.5 years. Despite statistically consistent associations with age, the age-adjusted absolute values of many macro-architectural metrics varied greatly between NCH and CHAT, a pattern observed in other contexts (e.g. comparing stage duration statistics between two elderly PSG cohorts, Djonlagic et al., 2021). Although such cohort effects may primarily reflect different recording contexts or other technical factors, rather than physiological differences between these populations, inasmuch as macro-architectural measures are susceptible to cohort-specific measurement biases, these issues may present challenges for the transferability of predictive models based on macro-architectural metrics. EEG spectral composition changes with age Multiple studies have reported age-related changes in absolute delta power during NREM sleep, suggesting an inverted-U trajectory: an increase from childhood to adolescence ( Feinberg et al., 2012 ) and reduction from adolescence to adulthood ( Baker et al., 2012 , Baker et al., 2016 , Feinberg and Campbell, 2013 , Jenni and Carskadon, 2004 , Tarokh and Carskadon, 2010 ). During R sleep, a linear decrease in delta power has also been documented ( Feinberg & Campbell, 2013 ). Theta power has also been reported to decline with age across all sleep stages ( Feinberg and Campbell, 2013 , Gaudreau et al., 2001 ). Despite limited number of channels, we confirmed topographical patterns in anterior-posterior direction of relative delta power change across a large and diverse sample. In a sample of 55 individuals (2.4–19.4 years), the scalp location with maximal SWA (1–4.5 Hz) power shifted with age in a posterior-to-anterior direction ( Kurth et al., 2010a , Kurth et al., 2010b ), mirroring the pattern of cortical thinning during childhood and adolescence( Shaw et al., 2008 ). Several studies utilizing both MRI and sleep EEG reported a link between slow-wave activity during NREM sleep and cortical thickness and/or gray matter volume ( Buchmann et al., 2011 , Goldstone et al., 2018 ). In general, the developmental decline in total power, as well as in delta and theta frequency bands, is not specific to sleep but also widely reported in wake ( Barriga-Paulino et al., 2011 , Boord et al., 2007 , Gasser et al., 1988 , Whitford et al., 2006 ), and has been similarly linked to reduction in cortical gray matter volume, and cerebral metabolic rate ( Boord et al., 2007 , Whitford et al., 2006 ). Thus, similar developmental changes in EEG spectral composition during both wake and sleep might reflect global structural changes in the brain, such as synaptic pruning that underlies normal cortical maturation and are not state-specific ( Segalowitz et al., 2010 ). In contrast to the general decline in total power, absolute sigma power increased with age. A similar increase was previously reported from 10 to 12 years in a longitudinal sample ( Tarokh & Carskadon, 2010 ) but there are also reports of higher absolute sigma in children compared to adults ( Gaudreau et al., 2001 ). Another longitudinal sample reported a complex trajectory of sigma power where it increased linearly from 6 to 12 years of age and then decreased from 12 to 16 ( Feinberg and Campbell, 2013 , Kurth et al., 2010a , Kurth et al., 2010b ), matching our observations of inverted-U trajectories in spindle activity. When normalized by total power, sigma power displayed one of the strongest age-related effects. During NREM sleep, this increase can be attributed to spindle maturation (discussed below). During R sleep, relative sigma power age-related increases were also accompanied by increases in relative alpha and beta power. While sleep studies reporting age-related changes in relative power are scarce, similar findings of an overall increase in higher frequencies are well reported during wake ( Dustman et al., 1999 , Gasser et al., 1988 , Soroko et al., 2014 ), once again pointing to brain maturation processes evident across both sleep and wake. Although, to our knowledge, there are no prior studies explicitly reporting age-related changes in relative power across classical frequency bands in sleep, our results indicate that relative power is a good metric to highlight topographical-, stage- and frequency-dependent aspects of developmental changes in the sleep EEG. For example, we observed distinct and sometimes contrasting trajectories between occipital and anterior channels as well as between NREM and R sleep stages (e.g. in delta and theta frequency bands), that can be practically leveraged when developing multi-channel tools for automatic stage classification in children and adolescents. NREM microarchitecture Rapid increases in spindle density during childhood have been linked to maturation of thalamocortical circuits ( Fernandez & Lüthi, 2020 ). Our results tend to confirm previous reports of slow spindle density increasing during childhood and reaching its peak around puberty ( Purcell et al., 2017 , Scholle et al., 2007 ). SS duration also was shown to exhibit a similar developmental pattern, peaking slightly earlier than density, confirming previous reports ( Scholle et al., 2007 ). In contrast, FS density increased through adolescence, congruent with recent studies ( Goldstone et al., 2019 , Purcell et al., 2017 ). One of the most frequently reported findings is that the spindle frequency peak estimated from examination of the power spectrum increases with age ( Campbell and Feinberg, 2016 , Hahn et al., 2020 , Tarokh et al., 2011 ). Although one early study showed that this is true for both frontal and centro-parietal spindles ( Shinomiya et al., 1999 ), tracking the spindle frequency based on the sigma peak for both slow and fast spindle can be problematic given that the sigma peak in children falls within a range of slow spindles ( Hoedlmoser et al., 2014 ). Alternatively, sigma peak increase could potentially reflect a shift in the ratio of slow versus fast spindle density with fast spindles becoming more prevalent after puberty. Our findings were based on spindle detection as discrete events, an approach that allowed us to estimate change in frequency for both slow and fast spindles in a more direct way. We showed that while SS frequency increases with age congruently with previous reports based on the sigma power peak, the FS frequency decreases. Interestingly, such a pattern appears to be opposite of the aging effect in aging adults (50 s to 80 s) where the SS and FS frequencies tend to diverge with SS becoming slower and FS faster ( Djonlagic et al., 2020 ). A recent study that investigated spindles in three groups of adolescents found that both slow and fast spindles frequency was increased with age ( Bocskai et al., 2022 ). However, in that study, slow and fast spindles were distinguished only using topographical differences (all spindles detected in frontal channels were declared slow and all spindles in centro-parietal channels fast). We also report a previously undocumented finding for intra-spindle frequency change (or spindle “chirp”). This metric is usually negative, reflecting the characteristic deceleration of both SS and FS ( Andrillon et al., 2011 ). Previous literature has suggested that chirp might be linked to spindle termination mechanisms and cortical modulation ( Carvalho et al., 2014 ). We found that intra-spindle deceleration intensifies over childhood and adolescence, showing some of the most pronounced age-related changes in our study. SOs also displayed some of the strongest changes within our cohort. SO rate per minute increased dramatically from childhood to adolescence while SO amplitude and slope decreased. Age-related decreases in SO amplitude and slope were reported previously in a small ( N = 14) sample (Kurth, Jenni, et al., 2010). We also reported, however, that the interpretation of age-related trends is obligatorily highly dependent on the choice of the SO amplitude detection threshold. Prior work showed that coupling between SOs and spindles also increased from childhood to adolescence in longitudinal sample of 33 individuals ( Hahn et al., 2020 ); this result was confirmed in our sample and extended to show that the strongest changes are for SS coupling magnitude. Estimating brain age from the sleep EEG We showed that individual differences in sleep macro- and micro-architecture can be summarized using simple methods to generate a highly accurate predictor of chronological age that is transferable across multiple independent samples. As has been widely employed by many groups using different brain imaging modalities (as well as epigenetics and other biomarkers), the difference between predicted “biological” age and observed chronological age can be interpreted as a measure of development and health ( Franke & Gaser, 2019 ). In pediatric populations, studies using structural MRI ( Franke et al., 2012 , Hong et al., 2020 ) have been able to predict age with high accuracy ( r > 0.9). Age has also been predicted using functional MRI ( Li et al., 2018 , Lund et al., 2022 , Qin et al., 2015 ), albeit with lower accuracy ( r = 0.54–––0.73). While a few studies using sleep EEG in adult cohorts achieve good results – r = 0.82–0.93 ( Nygate et al., 2021 , Sun et al., 2019 ), to our knowledge this is the first study to demonstrate this in a pediatric cohort. One recent report used resting state wake EEG spectral power to estimate brain age in a cohort of 5–18 year-olds, with an average MAE of 1.2 years ( Vandenbosch et al., 2019 ). This MAE is broadly similar to our results (MAE 1.1–1.3 years), although we note that are our results are based on performance in two independent samples, not only by means of cross-validation within the same sample, as in ( Vandenbosch et al., 2019 ). Given that many spectral age-related changes were evident in wake as well as sleep, we might expect similar performance for age prediction using either wake or sleep EEG. However, it remains an open and empirical question as to how highly correlated brain age estimates are when based on different modalities (MRI versus EEG) or different physiological states (wake versus sleep). Note that better prediction of chronological age is not, in itself, necessarily the most relevant factor: as a trivial conceptual example, a model that achieved perfect prediction (r = 1, MAE = 0) would be useless. Furthermore, even if two approaches have identical performance with respect to prediction of chronological age per se , they may still yield very different results with respect to how the model residuals (i.e. the so-called brain age gap) relate to brain development and health. To address the question of the biologically or clinically relevant properties of the brain age gap, we estimated it in NDD subgroups as well as non-NDD children. The largest deviations in both absolute and signed values (MAE and ME) were seen in individuals with DS and intellectual disability, with both groups showing negative gaps, consistent with delayed brain development. While we are not aware of other studies reporting on brain age in children with DS, a similar analysis was conducted for 46 adults (age range 28–––65 years) with DS using structural MRI. In contrast to our results, they reported a positive brain age gap interpreted as accelerated aging ( Cole et al., 2017 ), finding that it was related to increased beta amyloid deposition and cognitive decline. With the use of DNA methylation levels to calculate an ‘epigenetic clock,’ Horvath and colleagues also pointed to accelerating biological aging in brain and blood tissue ( Horvath et al., 2015 ), with evidence that such advanced ageing of blood samples begins prenatally ( Xu et al., 2022 ). Likewise, as well as shorter life expectancies generally, adults with DS display older biological age based on multiple physiological measures (e.g. BMI, blood pressure, etc) ( Nakamura & Tanaka, 1998 ). When cognitive and behavioral levels were assessed, however, individuals with DS tend to have lower developmental age ( Gameren-Oosterom et al., 2011 ), similar to our findings of children with DS being the most similar to younger age children with respect to their sleep macro and microarchitecture. Such results support the notion that accelerated/decelerated aging patterns are not universal and can be tissue and system-specific ( Horvath et al., 2015 ), as well as that brain age based on the sleep EEG may be reflective of cognitive and behavioral development. Additional analyses controlling for chronological age revealed alterations in multiple sleep macro and micro-architecture metrics in the DS subgroup, many of which were the opposite of typical age-related changes, suggesting altered developmental patterns in DS. For example, we saw a global increase in absolute spectral power in DS versus an age-related decrease in the control groups. Likewise, individuals with DS had reduced SS, FS and SO density across ages 4 to 16, counter to the marked age-related increases in these metrics in this age range. With the exception of one report of increased higher total spectral power ( Sibarani et al., 2022 ), the results of which we confirm here, a fuller assessment of sleep microarchitecture in DS has not been conducted and so our findings provide an important developmental perspective on abnormalities associated with DS. We also observed a consistently younger functional pattern in the sleeping brain in individuals with intellectual disabilities. In terms of sleep microarchitecture, spindle frequency metrics expressed the most marked alterations. While it is hard to compare our findings to the existing literature due to scarcity of reports available, two reports concluded that children with intellectual disabilities – especially those with more severe impairments – had decreased spindle density based on visual detection ( Shibagaki et al., 1980 , Shibagaki and Kiyono, 1983 ). Our findings also pointed to reduction in SS density in ID, that was more profound with the higher degree of intellectual disability. While other NDD subgroups – ASD, ADHD, CP and epilepsy – did not express consistent shifts towards either younger or older brain (ME), they all expressed larger brain age gaps in absolute terms (MAE). This may indicate considerable heterogeneity within these disorders: indeed, this has been previously reported in other contexts for ASD and ADHD ( Dajani et al., 2016 , Jeste et al., 2015 ) as well as CP ( Rosello et al., 2021 ) and epilepsy ( Pack, 2019 ). Alternatively, these results could reflect group-level differences in the sleep EEG leading to increased noise in the age prediction model, as NDD groups were excluded from the primary NCH model fitting. In general, future studies will be needed to fully evaluate the relative merits of different brain age metrics, and to determine whether ones based on the sleep EEG offer additional, unique information or not, as well as how brain age alterations may vary over the course of a disease. One important data-point to guide the development of possible clinical applications would be to determine how state-dependent (versus trait-like) these measures are: for example, considering children before versus after the onset of behavioral and cognitive symptoms, or in response to medication, or as a function of duration of illness. Except ASD, all subgroups displayed alterations in stage R sleep, either in terms of absolute duration, relative duration or stage R latency. This is consistent with previous reports of stage R deficits in DS ( Spanò et al., 2018 ), CP ( Hayashi et al., 1990 ), epilepsy ( Sadak et al., 2022 ), and intellectual disability ( Esposito & Carotenuto, 2014 ), supporting the notion that R deficits are common characteristics across NDDs. Previous studies reported that lower R duration was associated with worse cognitive performance and mortality in older individuals ( Djonlagic et al., 2020 , Leary et al., 2020 ). As such, R sleep metrics may not be good candidates for condition-specific biomarkers, but rather reflect pathophysiological alterations shared between distinct disorders. Caveats & conclusions In summary, the present study provides a comprehensive assessment of age-related changes in sleep macro and microarchitecture, based on a large sample from multiple cohorts spanning the first two decades of life. Nonetheless, certain constraints should be mentioned. One obvious limitation is that the NCH data are from a database of clinical encounters. Based on comparison to CHAT, all primary age-related changes appeared qualitatively (and often quantitatively) conserved across studies, suggesting that these robust effects reflect fundamental developmental processes that may transcend diagnostic status. The subjects in CHAT either qualified for a diagnosis of obstructive sleep apnea, or snored and on that basis had some form of sleep-disordered breathing, meaning that the subjects though not necessarily referred to a sleep disorders center still most likely did not have normal sleep. Therefore, our comparisons may be limited due to lack of truly normative controls. Nonetheless, the presence of sleep-disordered breathing or other sleep disorders in many of the subjects who contributed data for the present analyses does not invalidate the high likelihood that sleep in these individuals still reflected many aspects of normal sleep development across childhood and adolescence. For example, we did not find evidence of significant effects of AHI on the age-related changes of sleep metrics or brain age prediction. Another limitation is the absence of detailed cognitive and behavioral data in the NDD cohort, precluding more direct investigations of how age-related changes in sleep track with development and any subclinical traits. Similarly, puberty status was not available, limiting our ability to interpret developmental differences not captured by age. Due to the absence of resting state data, we also could not account for developmental shifts in the alpha peak that could have effect on power in neighboring frequency bands. We acknowledge that while we report robust age-related changes in NREM micro-architecture across multiple datasets as proof-of-principle, the performance of this algorithm could likely be further improved by optimizing details of the analytic approach. Our findings from retrospective clinical and clinical research data, meanwhile, appear to suggest very strong age-related changes across numerous sleep metrics in children ages 4 to 17 years, which could be robustly identified across independent samples despite the demographic, clinical and procedural/technical differences. As well as confirming previous reports based on smaller samples, we describe new metrics not previously studied from a developmental perspective, including stage- and channel-specific developmental trajectories of relative spectral power, intra-spindle frequency modulation, and temporal overlap between spindles and SOs. A model of multiple sleep metrics across different domains was able to predict chronological age with high precision in typically developing individuals, whereas this correspondence was lessened in individuals with NDDs, suggesting that these sleep metrics are sensitive to various functional abnormalities present in the (sleeping) brain. Taken together, our results indicated that sleep macro and microarchitecture offer important information about brain maturation which may facilitate a better understanding of the atypical neurodevelopment.

Caveats & conclusions In summary, the present study provides a comprehensive assessment of age-related changes in sleep macro and microarchitecture, based on a large sample from multiple cohorts spanning the first two decades of life. Nonetheless, certain constraints should be mentioned. One obvious limitation is that the NCH data are from a database of clinical encounters. Based on comparison to CHAT, all primary age-related changes appeared qualitatively (and often quantitatively) conserved across studies, suggesting that these robust effects reflect fundamental developmental processes that may transcend diagnostic status. The subjects in CHAT either qualified for a diagnosis of obstructive sleep apnea, or snored and on that basis had some form of sleep-disordered breathing, meaning that the subjects though not necessarily referred to a sleep disorders center still most likely did not have normal sleep. Therefore, our comparisons may be limited due to lack of truly normative controls. Nonetheless, the presence of sleep-disordered breathing or other sleep disorders in many of the subjects who contributed data for the present analyses does not invalidate the high likelihood that sleep in these individuals still reflected many aspects of normal sleep development across childhood and adolescence. For example, we did not find evidence of significant effects of AHI on the age-related changes of sleep metrics or brain age prediction. Another limitation is the absence of detailed cognitive and behavioral data in the NDD cohort, precluding more direct investigations of how age-related changes in sleep track with development and any subclinical traits. Similarly, puberty status was not available, limiting our ability to interpret developmental differences not captured by age. Due to the absence of resting state data, we also could not account for developmental shifts in the alpha peak that could have effect on power in neighboring frequency bands. We acknowledge that while we report robust age-related changes in NREM micro-architecture across multiple datasets as proof-of-principle, the performance of this algorithm could likely be further improved by optimizing details of the analytic approach. Our findings from retrospective clinical and clinical research data, meanwhile, appear to suggest very strong age-related changes across numerous sleep metrics in children ages 4 to 17 years, which could be robustly identified across independent samples despite the demographic, clinical and procedural/technical differences. As well as confirming previous reports based on smaller samples, we describe new metrics not previously studied from a developmental perspective, including stage- and channel-specific developmental trajectories of relative spectral power, intra-spindle frequency modulation, and temporal overlap between spindles and SOs. A model of multiple sleep metrics across different domains was able to predict chronological age with high precision in typically developing individuals, whereas this correspondence was lessened in individuals with NDDs, suggesting that these sleep metrics are sensitive to various functional abnormalities present in the (sleeping) brain. Taken together, our results indicated that sleep macro and microarchitecture offer important information about brain maturation which may facilitate a better understanding of the atypical neurodevelopment.

Highlights • This study demonstrates the potential of sleep-based features to provide practical markers for tracking neurodevelopment in children across multiple clinical populations. • Using whole-night polysomnography data, we comprehensively describe robust age-related changes in multiple sleep metrics derived from electroencephalogram and demonstrate their utility in predicting an individual's chronological age with high accuracy. • The differences between the predicted age and the chronological age can be indicative of neurodevelopmental alterations. • This study suggests that sleep patterns can provide a sensitive way to understand the process of brain maturation and could lead to the creation of objective sleep-based biomarkers that can be used on a larger scale to measure neurodevelopment. Profiles of sleep duration and timing and corresponding electroencephalographic activity reflect brain changes that support cognitive and behavioral maturation and may provide practical markers for tracking typical and atypical neurodevelopment. To build and evaluate a sleep-based, quantitative metric of brain maturation, we used whole-night polysomnography data, initially from two large National Sleep Research Resource samples, spanning childhood and adolescence (total N = 4,013, aged 2.5 to 17.5 years): the Childhood Adenotonsillectomy Trial (CHAT), a research study of children with snoring without neurodevelopmental delay, and Nationwide Children’s Hospital (NCH) Sleep Databank, a pediatric sleep clinic cohort. Among children without neurodevelopmental disorders (NDD), sleep metrics derived from the electroencephalogram (EEG) displayed robust age-related changes consistently across datasets. During non-rapid eye movement (NREM) sleep, spindles and slow oscillations further exhibited characteristic developmental patterns, with respect to their rate of occurrence, temporal coupling and morphology. Based on these metrics in NCH, we constructed a model to predict an individual's chronological age. The model performed with high accuracy ( r = 0.93 in the held-out NCH sample and r = 0.85 in a second independent replication sample – the Pediatric Adenotonsillectomy Trial for Snoring (PATS)). EEG-based age predictions reflected clinically meaningful neurodevelopmental differences; for example, children with NDD showed greater variability in predicted age, and children with Down syndrome or intellectual disability had significantly younger brain age predictions (respectively, 2.1 and 0.8 years less than their chronological age) compared to age-matched non-NDD children. Overall, our results indicate that sleep architecture offers a sensitive window for characterizing brain maturation, suggesting the potential for scalable, objective sleep-based biomarkers to measure neurodevelopment.

Declaration of competing interest The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Dr. Chervin reports the following financial and non-financial potentially competing interests: consultant (through contract with Michigan Medicine) for Eli Lilly & Company; editor and author, UpToDate; officer and board member for International Pediatric Sleep Association; member, advisory board for a non-profit organization, The Pajama Program. Other authors report no conflict of interests.

Supplementary data The following are the Supplementary data to this article: Data availability NCH-SDB & CHAT are available via NSRR (http://sleepdata.org). PATS will be posted on NSRR at a future date. Luna software is open and available via http://zzz.bwh.harvard.edu/luna Acknowledgements This work was supported by the National Institutes of Health (NIH)/National Institute of Mental Health (NIMH) Grant R03 MH108908 (to S.M.P.), NIH/National Heart, Lung, and Blood Institute (NHLBI) Grants R01 HL146339 and R21 HL145492 (to S.M.P.), and the NIH/National Institute on Minority Health and Health Disparities (NIMHD) Grant R21 MD012738 (to S.M.P.). This research was supported (in part) by the Intramural Research Program of the National Institute of Mental Health ZIC MH00296206. The National Sleep Research Resource was supported by the U.S. National Institutes of Health, National Heart Lung and Blood Institute (R24 HL114473, 75N92019R002). NCH Sleep DataBank was supported by the National Institute of Biomedical Imaging and Bioengineering of the National Institutes of Health under Award Number R01EB025018. The Childhood Adenotonsillectomy Trial (CHAT) was supported by the National Institutes of Health (HL083075, HL083129, UL1-RR-024134, UL1 RR024989). The Cleveland Family Study (CFS) was supported by grants from the National Institutes of Health (HL46380, M01 RR00080-39, T32-HL07567, RO1-46380.

CC BY

no

2024-01-16 23:41:58

Neuroimage Clin. 2023 Dec 19; 41:103552

oa_package/30/1c/PMC10788305.tar.gz

PMC10788309

38226291

Introduction Congenital anomalies of the inferior vena cava (IVC) are often recognised as incidental findings at abdominal imaging. Most congenital variations are asymptomatic, but knowledge of these abnormalities can be crucial for preoperative planning of surgeries and endovascular procedures, as well as to avoid misdiagnosis. 1 The embryonic development of the IVC is complex involving numerous sequences in formation, regression, and persistence of the embryonic veins, which eventually contribute to different segments. Any abnormality of this process can lead to an abnormal IVC and anomalous veins. Occasionally, congenital anomalies of the IVC can be associated with congenital cardiac abnormalities, the recognition of which may be crucial for patient management. 2 This article emphasises the understanding of congenital IVC variants and their clinical significance.

Conclusion Congenital anomalies of the IVC are often asymptomatic but may have clinical implications. Knowledge of these variations is crucial in preprocedural planning of IVC filter placement or preoperative planning for cardiovascular surgeries as well as nephrectomy. These congenital anomalies may predispose to certain conditions such as urinary obstruction and UTI. Moreover, some of these anomalies have been associated with other congenital anomalies. Thus, it seems prudent to be cognizant of these IVC variations and ensuring that treating physicians are aware of these findings so that they can plan their treatment appropriately.

The embryology of the inferior vena cava (IVC) is complex, involving the sequential appearance and regression of multiple segments that ultimately form the IVC. Any alteration in this process during embryogenesis can result in congenital anomalies of the IVC. This study aimed to recognise common as well as rare anomalies of the IVC and associated veins, and their clinical implications. The anomalies tend to have diverse appearances based on the timing and segments involved. The development of the IVC is intertwined with the development of other veins like the renal vein, azygos vein and portal vein, and these veins may also be anomalous. Additionally, IVC anomalies are associated with various other congenital anomalies including cardiac anomalies, the recognition of which may be important for patient care. The IVC tends to have multiple normal variants and anomalies because of a complex process involving multiple segments contributing to the adult IVC. Knowledge of these variants is crucial for preoperative planning of procedures. Contribution This study would help in understanding the embryogenesis of the IVC and correlation with the imaging appearances and the clinical implications of each of these common as well as rare types of congenital anomalies.

Anatomy and embryology The IVC is the main route of venous return from the lower extremities and abdomen, formed by the confluence of the two common iliac veins at L5. It has a retroperitoneal course in the abdomen, located to the right of aorta. It courses close to the liver where it receives the hepatic veins and then passes through the diaphragmatic hiatus at the T8 level to become intrathoracic, terminating in the right atrium. 3 The four segments of the IVC are the hepatic, suprarenal, renal, and infrarenal segments ( Table 1 ). 4 The development of the different segments of the IVC is complex and involves the sequential appearance, anastomoses and regression of three pairs of embryonic veins, namely the posterior cardinal, subcardinal, and supracardinal veins. During the fourth week of foetal life, the posterior cardinal veins develop and become dominant by the sixth week, responsible for the return of all the blood from the body wall caudal to the heart. 5 The vitelline vein, which drains the embryonal yolk sac, is responsible for blood return from the viscera and eventually forms the hepatic segment of the IVC. The subcardinal veins become the dominant venous system by the seventh week of foetal life. The posterior cardinal veins are dorsolateral and the aorta is ventromedial to the subcardinal veins. The suprarenal segment of the IVC develops from the right subcardinal vein. By 8 weeks, the supracardinal veins predominate, positioned dorsomedial to the posterior cardinal veins and dorsolateral to the aorta. The intrathoracic course of the supracardinal veins form the azygous and hemiazygos veins. The left supracardinal vein disappears, while the right supracardinal vein forms the infrarenal segment of the IVC. The renal segment of the IVC is formed by the anastomoses between the subcardinal and supracardinal veins. 1 , 5 , 6 The iliac veins derived from the persistent posterior cardinal veins form an anastomosis with the infrarenal IVC derived from the right supracardinal vein. 7 Although the posterior cardinal veins do not contribute to the adult IVC, abnormality in development of the posterior cardinal veins can lead to IVC anomalies. 5 Anomalies in this complex process can lead to various congenital anomalies of the IVC, which has been reported in nearly 4% of the general population ( Table 2 ). 2 Associated congenital abnormalities can occur particularly in cases with azygos continuation of the IVC, which are related to heterotaxy syndrome with left isomerism. 2 , 8 The ureter, during development, takes it course posterior to the posterior cardinal veins. The supracardinal vein is posteromedial to the embryonic ureter. The inter-supracardinal anastomosis posteriorly, inter-subcardinal anastomosis and post-subcardinal anastomoses anteriorly, and the supra-subcardinal anastomosis laterally together forms the renal collar. The paired ventral and dorsal limbs drain the kidneys during early development. Both dorsal limbs usually regress, while the ventral limb on right side is incorporated into the lateral wall of the renal segment of the IVC. The ventral limb on left side and the anterior limb of the renal collar form the normal adult left renal vein. 1 Imaging Ultrasonography (US) is a readily available, cheaper and safer imaging modality. Although often used for the initial evaluation of the IVC, particularly in paediatric patients, it has limitations of operator dependence and difficult visualisation because of obscuration by bowel gas. 2 , 7 , 9 Contrast enhanced CT abdomen is the preferred and most commonly performed imaging modality for evaluation of the IVC. The IVC is best evaluated in the venous phase with a delay time of 70 s – 90 s after intravenous contrast administration to allow for uniform enhancement of the infrarenal IVC. However, imaging is commonly performed in the portal venous phase during routine abdominal CT at 60 s – 70 s. Multiplanar reformation can be obtained with multidetector CT, which is often useful in delineating the course of the IVC and anastomosis. 2 , 7 , 9 , 10 Although MRI has the advantage of not utilising ionising radiation, it is an uncommonly used modality. Its limited role is because of high cost, need for anaesthesia in the paediatric population, and limited availability. Evaluation of the IVC can be performed using post-contrast three-dimensional breath-hold T1-weighted MRI. Balanced steady-state free precession is another useful MRI sequence for imaging of the IVC. 7 , 9 Magnetic resonance venography (MRV) can be performed using the time of flight (TOF) or phase contrast technique, which does not require administration of gadolinium contrast. 11 , 12 Congenital anomalies Most congenital anomalies are asymptomatic, but their identification is necessary to avoid mistaking them for pathology and for planning of vascular procedures ( Table 3 ). 13 Most of the congenital IVC abnormalities, particularly those that result in an abnormal vein or abnormal dilation of normal veins, can mimic lymphadenopathy. 1 Left-sided inferior vena cava A left-sided IVC is formed as a result of disappearance of the right supracardinal vein and persistence of the left supracardinal vein. The left-sided IVC typically terminates at the left renal vein. It then courses ventral to the aorta to the right side to form a normal right-sided suprarenal IVC ( Figure 1 ). The reported incidence is 0.2% – 0.5%. 1 , 14 , 15 It is a mirror image variant without an increased risk of any abnormality. However, it can lead to confusion between venous and arterial access. Also, transjugular IVC filter placement and pulmonary thrombolysis is more challenging with a left-sided IVC. In addition, it can mimic left paraaortic adenopathy. 1 , 9 Rare spontaneous rupture of an abdominal aortic aneurysm into a left IVC has also been reported. 16 , 17 Double inferior vena cava A double IVC has a left-sided IVC in addition to the normal right IVC. It results from persistence of both supracardinal veins. The left IVC crosses the midline anterior to the aorta to join the right IVC ( Figure 2 ). There may be variations in calibre of the left and right IVC. 1 , 9 The incidence of double IVC has been reported between 1% and 3%. 15 Recurrent pulmonary embolism despite IVC filter placement should raise suspicion of a double IVC. Review of prior cross-sectional images before IVC filter placement is useful to diagnose congenital IVC abnormalities such as double IVC. If no prior cross-sectional images are available for review, cavography through the left iliac vein should be performed prior to filter placement. A filter can be placed into each cava if a double IVC is present. The aberrant vessel may also mimic a lymph node. 1 , 7 , 9 Circumaortic left renal vein Two left renal veins are present on the left side with one coursing anterior to the aorta and other posterior to the aorta. It results from persistence of the posteriorly located inter-supracardinal vein anastomoses as well as the anteriorly located inter-subcardinal vein anastomoses that form a venous ring around the aorta ( Figure 3 ). The reported prevalence ranges from 2.4% to 8.7%. 1 The posterior renal vein is 1 cm – 2 cm inferior to the normal anterior vein. The left adrenal vein drains into the superior renal vein. The left gonadal vein drains into the inferior renal vein. Knowledge of this variation is significant in preoperative planning prior to nephrectomy and catheterisation for renal venous sampling. It can also mimic retroperitoneal adenopathy. 7 , 9 It is also important to recognise this variation prior to positioning of the IVC filter. 18 These patients may uncommonly present with hypertension, haematuria and varicoceles. 19 Retroaortic left renal vein A single left renal vein with a retroaortic course as the dorsal arch of the renal collar persists, while the ventral arch (inter-subcardinal anastomosis) regresses. Reported prevalence is 1.7% – 3.4%. 1 , 7 Although mostly asymptomatic, occasionally patients may present with symptoms of haematuria, flank pain and varicocele. 19 , 20 Retrocaval or circumcaval ureter The genitourinary system develops separately from the IVC. However, embryogenesis of the IVC determines the spatial relationship between the ureter and the IVC. In this anatomical variant, the infrarenal segment develops from the right posterior cardinal vein, which is anterolateral to the ureter, instead of the right supracardinal vein, which is posteromedial to the ureter. The anomaly almost always occurs on the right side with a handful of cases reported on left. 9 , 21 The reported prevalence of a retrocaval ureter is 0.13%. 22 In intravenous urography, the proximal part of retrocaval ureter makes a characteristic course. It projects over or medial to the lumbar pedicles. This gives the characteristic fish hook or reverse J appearance ( Figure 4 ). 7 , 9 , 19 Although most cases are asymptomatic, right flank pain is the most common symptom. 21 Compression of the ureter between the IVC and vertebra can result in hydronephrosis. Ureteral obstruction or recurrent urinary tract infection (UTI) may necessitate surgical relocation of the ureter anterior to the cava. 9 , 19 , 21 Coexistence of other congenital genitourinary, cardiovascular and spine abnormalities have been reported. 21 Interruption of the inferior vena cava with azygos or hemiazygos continuation Absence of the hepatic segment of the IVC is thought to be because of failure of the right subcardinal-hepatic anastomosis with resultant right subcardinal vein atrophy and shunting of blood from the supra-subcardinal anastomosis to the azygos vein. The azygos vein joins the superior vena cava at the right paratracheal space in the expected normal location. The hepatic segment is not entirely absent. It drains directly into the right atrium. The right gonadal vein drains to the ipsilateral renal vein as the post-subcardinal anastomosis does not contribute to the formation of the IVC. It has a reported prevalence of 0.6%. 1 , 7 , 9 Azygos continuation is more common than hemiazygos continuation. Azygos continuation of the IVC can be seen in asymptomatic patients or in association with other congenital abnormalities such as severe congenital heart disease and asplenia or polysplenia syndromes ( Figure 5 ). 1 , 8 , 23 An enlarged azygos vein can mimic a right paratracheal or retrocrural lymph node. Knowledge of this variation is useful for preoperative planning for cardiopulmonary bypass and vascular procedures. 1 , 9 Double inferior vena cava with retroaortic right renal vein and hemiazygos continuation of the inferior vena cava This rare anomaly results from anomalies in multiple steps during the development of the IVC and renal vein. Similar to azygos continuation of the IVC, absence of the right subcardinal-hepatic anastomosis results in failure of development of the hepatic segment. Left lumbar and thoracic supracardinal veins persist giving rise to a left IVC and blood is shunted into the hemiazygos vein via the left supra-subcardinal anastomosis. Persistence of the dorsal limb and regression of the ventral limb of the renal collar leads to a right renal vein that meets the right IVC and courses posterior to the aorta to join the left IVC. The azygos vein is rudimentary. The hemiazygos vein drains via alternate collateral pathways. The hemiazygos vein can drain into the azygos vein, the coronary vein of the heart in patients with a persistent left superior vena cava or via the accessory hemiazygos vein into the left brachiocephalic vein. These aberrant vessels may simulate a left mediastinal mass or aortic dissection. Like azygos continuation of the IVC, the hepatic segment of the IVC drains into the right atrium independently. 1 , 24 Double inferior vena cava with a retroaortic left renal vein and azygos continuation of the inferior vena cava This congenital variation occurs when the supracardinal vein and the dorsal limb of the renal collar persist while the ventral limb regresses and the subcardinal-hepatic anastomosis does not form. 1 , 25 These are extremely rare and detected incidentally. Very few cases of right-sided varicocele have been reported in patients with a double IVC and azygos or hemiazygos continuation. 25 , 26 The imaging findings are similar to azygos continuation of the IVC except there is a double IVC instead of the normal right-sided IVC and the left renal vein has a retroaortic course 25 ( Figure 6 ). Clinical implications of this variation are to avoid misdiagnosis as lymphadenopathy and preoperative surgical planning. 25 The absence of the inferior vena cava The entire IVC or only the infrarenal portion of the IVC can be absent. Absence of the entire posthepatic IVC results from failure of development of all three paired venous systems. These are extremely rare. The absence of the infrarenal IVC implies failure of development of the posterior cardinal and supracardinal veins. 9 As a single embryonic event cannot explain this abnormality, it is thought that it is not a true embryonic anomaly but the result of perinatal thrombosis and atrophy. 1 , 7 , 9 Imaging shows the absence of the IVC or a portion of the IVC with prominent venous channels providing collateral flow ( Figure 7 ). These veins are prone to idiopathic deep vein thrombosis and lower extremity venous insufficiency. Prominent lumbar collateral vessels may develop, which can mimic paraspinal masses. 9 The common iliac veins can be absent as well. The lower extremity blood drains to the azygos and hemiazygos veins via anterior paravertebral collateral veins. Anterior paravertebral collateral veins receive blood via enlarged ascending lumbar veins, which in turn are formed by the external and internal iliac veins. 1 Extrahepatic portocaval shunt or Abernethy malformation The portal vein is formed cranially from a segment of the prehepatic right vitelline vein, the intervitelline anastomosis, and caudally from the left vitelline vein around the fifth week of life. 19 The right vitelline vein thus serves as a common conduit during the development of both the IVC and the portal vein. The process of formation of hepatic tissue results in disruption of intervitelline vein anastomosis disconnecting the cranial and caudal segments of the portal vein. Eventually, the cranial segment of the portal vein disappears. Hence, there is no direct communication between the portal vein and the IVC. 19 These extrahepatic portocaval shunts may result either from excessive involution of the vitelline vein or failure of the vitelline vein to establish an anastomosis with the hepatic sinusoids or hepatic veins. 9 , 27 Abernethy malformations are of two types. Type 1 is more common in females and is characterised by complete shunting of portal blood into the IVC and an absent portal vein ( Figure 8 ). It is associated with congenital cardiac, gastrointestinal (e.g., biliary atresia) and genitourinary anomalies. 9 , 19 Type 2 is more often seen in males as an isolated abnormality with partial end-to side anastomosis between an intact portal vein and the IVC resulting in shunting of blood. The presence or absence of the portal vein is an important imaging finding because it helps to distinguish between the two types. The Abernethy malformation is associated with focal nodular hyperplasia and hepatocellular carcinoma (HCC). 7 , 9 Type I is managed with a liver transplant. The Type II shunt may require occlusion of the shunt if the patient develops hepatic encephalopathy or bleeding varices. It can be managed surgically or percutaneously using balloons or coils. 19 Inferior vena cava membranes and webs These are uncommon IVC findings attributed to congenital vascular anomalies or the sequela of thrombus formation. 9 North American and northern European populations are rarely affected. There is usually an underlying hypercoagulable state and they present acutely. Outcomes are usually fatal. Asian and South African populations are more commonly affected in whom it eventually leads to congestive cirrhosis. 28 The onset is insidious in these patients. 7 , 9 A complete or fenestrated membrane or a segment of fibrotic occlusion in the intrahepatic IVC with prominent intrahepatic and extrahepatic collaterals is seen on imaging ( Figure 9 ). Inferior venacavography can be helpful in confirming the diagnosis. Endovascular treatment to relieve portal pressure can be performed depending on the severity of the associated liver disease. 7 , 9 , 29 The association between membranous obstruction of the intrahepatic IVC, Budd-Chiari syndrome and HCC is well established. 7

Acknowledgements Competing interests The authors declare that they have no financial or personal relationships that may have inappropriately influenced them in writing this article. Authors’ contributions R.K.C. performed the manuscript preparation, literature review, and editing. P.N. conceptualised the idea, and performed the literature review, manuscript preparation, and editing. S.K., E.G. and N.S. were involved in literature review, manuscript preparation, and editing. A.N. conceptualised the idea, and performed the literature review, manuscript reviewing, and editing. V.O. conceptualised the idea, and was involved in literature review, manuscript reviewing and editing, and submission. Ethical considerations This article followed all ethical standards for research. Images from radiological investigations performed during patient care have been used in the publication but all patient-identifiable information was concealed to maintain patient anonymity. Funding information This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Data availability The authors confirm that the data supporting the findings of this study are available within the article. Disclaimer The views and opinions expressed in this article are those of the authors and do not necessarily reflect the official policy or position of any affiliated agency of the authors.

CC BY

no

2024-01-16 23:41:58

SA J Radiol. 2023 Nov 24; 27(1):2687

oa_package/62/b9/PMC10788309.tar.gz

PMC10788311

38225986

Introduction Glucagon‐like peptide (GLP)‐1 agonist therapy is a non‐insulin‐based approach that is combined with diet and exercise to treat type 2 diabetes mellitus and obesity. It has gained recognition as an effective weight loss treatment with only mild side effects. The most notable are nausea, vomiting and diarrhoea [ 1 ]. Here, we report two cases in which the use of GLP‐1 agonist therapy for weight loss may have resulted in delayed gastric emptying, with subsequent regurgitation and pulmonary aspiration of gastric contents.

Discussion Semaglutide is a GLP‐1 receptor agonist that was first marketed over 15 years ago, initially designed for glycaemic control in diabetic patients. By mimicking the GLP‐1 hormone released by the gut in response to food, semaglutide increases endogenous insulin secretion to lower blood glucose. The 2021 STEP 2 trial demonstrated semaglutide to be an effective weight loss drug, reporting 10% weight loss among subjects [ 2 ]. It was approved by the US Food and Drug Administration to be used as a weight loss agent among some overweight and obese patients. Since 2021, semaglutide has gained recognition as a ‘celebrity weight‐loss jab’. There has been great demand for it that has led to worldwide shortage. It is currently available in three different forms (subcutaneous injection for children, subcutaneous injection for adults and tablet form). We are not the first group to suggest a connection between semaglutide and peri‐operative aspiration [ 3 ]. However, we hope to add to the existing literature and encourage future research that might lead to improved peri‐operative practice for patients taking semaglutide. Semaglutide appears to delay gastric emptying by impeding antral and duodenal motility while at the same time stimulating pyloric tightening [ 4 ]. Inhibition of postprandial acid secretion also appears to contribute to this ‘ileal brake’ mechanism through direct effects on gastric smooth muscle [ 5 ]. The effect on gastric smooth muscle appears to be mediated largely by vagal afferent nerves [ 6 ]. Delayed gastric emptying is a known risk factor for pulmonary aspiration during anaesthesia [ 7 ]. Gudin et al. report a 3.5‐fold increase in the rate of peri‐operative pulmonary aspiration in patients using drugs such as GLP‐1 agonists [ 8 ]. It should be noted that both of our patients continued to use semaglutide in the week prior to anaesthesia. We think that GLP‐1 agonist therapy may have resulted in delayed gastric emptying and led to pulmonary aspiration of gastric contents. Although we are unable to prove causation, we agree with Klein and Hobai [ 3 ] that there is ‘significant cause for concern’, given the known mechanism of these drugs. Non‐compliance with the pre‐operative fasting regimen could have been the cause of peri‐operative aspiration. This seems unlikely in our cases because our patients were noted to be very engaged in safe preparation for their anaesthetic. The American Society of Anesthesiologists (ASA) released pilot guidelines in June 2023 [ 9 ], addressing GLP‐1 receptor agonist therapy and their relationship to peri‐operative aspiration events. The recommendation for patients taking semaglutide is to withhold the medication for a week prior to anaesthesia, based on its 7‐day half‐life. If on the day of the procedure gastrointestinal symptoms such as severe nausea, vomiting, bloating or abdominal pain are present despite having discontinued the drug, consideration should be given to delaying the procedure. If the patient has no gastrointestinal symptoms but did not stop the medication, it is advised to either use ‘full stomach’ precautions or to use ultrasound guidance to evaluate stomach contents. Pulmonary aspiration is one of the highest risk complications of airway management. According to the 2021 ASA Closed Claims analysis, 57% of aspiration incidents result in death, and another 15% result in permanent severe injury [ 10 ]. Bedside ultrasound appears to be emerging as an effective modality to aid in assessing gastric content, and future studies using gastric ultrasound may help to elucidate the effects of semaglutide and whether omitting for 7 days is sufficient. We hope our two cases help to raise awareness of this potential clinical problem with an increasingly common drug.

Summary Semaglutide is a new weight loss treatment that has received substantial media attention in recent years. Anaesthetists must be aware of a potentially dangerous side effect of the drug: decreased gastric emptying. This is caused by effects on gastric smooth muscle, mediated by the vagal afferent nerves. This is especially relevant in the peri‐operative setting where pulmonary aspiration of gastric contents is a recognised complication. Here, we report two cases of peri‐operative regurgitation of gastric contents in patients taking semaglutide. A patient taking semaglutide may have a full stomach despite compliance with routine pre‐operative fasting guidelines. We consider how to manage patients receiving glucagon‐like peptide‐1 agonist therapy in the peri‐operative period, including identifying those at high risk of regurgitation. Precautions such as rapid sequence induction and tracheal intubation can be used, but gastric ultrasound may also be useful in the pre‐operative environment to help identify patients at high risk of aspiration.

Report Case 1 A 70‐year‐old man with a body mass index (BMI) of 35 kg.m −2 and a medical history of ischaemic heart disease, type 2 diabetes mellitus, hypertension and chronic obstructive pulmonary disease presented for elective endoscopic retrograde cholangiopancreatography due to cholidocholelithiasis. His medications included spironolactone, aspirin, tamsulosin, metformin, empagliflozin, ramipril and semaglutide, taken as a 1 mg weekly subcutaneous injection. His last dose of semaglutide was 6 days prior to the procedure. Pre‐operative physical examination was unremarkable; he was orientated and alert, and had no nausea or vomiting. Blood glucose was 8.8 mmol.l −1 (HbA1C 7%). He had undergone surgical procedures in the past uneventfully. He gave informed consent to undergo surgery under general anaesthetic (GA). The anaesthetist decided to perform rapid sequence induction and intubation, a decision made because of the patient's BMI and history of diabetes mellitus. Prior to induction, the patient had been nil by mouth for food and fluids for over 12 h. He had stable respiratory and haemodynamic parameters, with oxygen saturation of 95% on room air, blood pressure 132/72 mmHg and a heart rate of 105 beats.min −1 . The patient received pre‐oxygenation with 100% oxygen for 5 min, after which etomidate 20 mg and rocuronium 100 mg were given intravenously. During laryngoscopy, a large volume of regurgitated particulate gastric content including undigested food was observed. Laryngoscopy and tracheal intubation were performed using a standard laryngoscope with a size 4 Macintosh blade and an 8‐mm tracheal tube. Nasogastric suction was performed. Oxygen/air was delivered, with an F I O 2 of 0.95. Bag ventilation was performed, initially requiring peak ventilatory pressures of up to 35 cm H 2 O. Despite this, the oxygen saturation deteriorated to 83%. After these measures were continued for 5 min, the situation improved. At this stage, anaesthesia was maintained with sevoflurane (1.1 minimum alveolar concentration, MAC), with a peak ventilatory pressure of 28 cm H 2 O; the oxygen saturation was 95%. A chest radiograph was ordered, showing bilateral infiltrates with fluid in the dependent segment of the right lower lobe. Because observations were stable, it was agreed that the procedure should go ahead, which was carried out successfully. The patient was then transferred to the intensive care unit and tracheal extubation was performed the next day. The patient was discharged after a week. Case 2 A 25‐year‐old woman with a BMI of 32 presented for incision and drainage of a breast abscess under general anaesthesia. The only medication she was taking was semaglutide 1 mg weekly subcutaneous injection, for weight loss. Her last dose of semaglutide was 4 days prior to surgery. Physical examination was unremarkable. She had received anaesthesia in the past without problems. She gave informed consent to undergo surgery under GA. The patient had fasted from food and fluids for over 8 h. General anaesthesia was induced using 3 mg midazolam, 200 mcg fentanyl and 200 mg propofol. A size 4 laryngeal mask airway was used, and anaesthesia was maintained with sevoflurane (1.0 MAC). At the end of uneventful surgery, it was confirmed that the patient was breathing spontaneously with normal observations, and the supraglottic airway was removed. Following supraglottic airway removal, the patient began to cough and showed signs of respiratory distress. Regurgitation of solid and liquid stomach content was observed, and her oxygen saturation fell to 85%; her respiratory rate was observed to be 14. The patient's head was tilted to the right and the oral cavity was suctioned with a nasogastric tube. Because of the desaturation, we decided to re‐induce general anaesthesia with propofol 200 mg and succinylcholine 100 mg, and perform tracheal intubation using a standard laryngoscope with a size 3 Macintosh blade and a 7.5 mm tracheal tube. General anaesthesia was maintained with sevoflurane (1.0 MAC) for 20 min once respiratory parameters were stabilised. Sevoflurane was stopped, and tracheal extubation was performed once the patient was fully awake. Her heart rate was 90 beats.min −1 , blood pressure was 135/88 mm Hg and oxygen saturation was 97%. In the post‐anaesthesia care unit, observations were normal, and the chest radiograph was unremarkable.

Acknowledgements This case report was published with the written consent of the two patients involved. There are no competing interests to declare.

CC BY

no

2024-01-16 23:41:58

Anaesth Rep. 2024 Jan 14; 12(1):e12278

oa_package/cd/c0/PMC10788311.tar.gz

PMC10788312

38226344

INTRODUCTION Animals sometimes possess ornamentation that seem to have negative effects on survivorship, such as long tails and colorful plumage in birds (Andersson, 1994 ). Many empirical studies have shown sexual selection for ornaments, explaining the evolution of seemingly costly ornaments (e.g., see Hill & McGraw, 2006 for a review on birds). However, the mere presence of sexual selection does not necessarily mean that sexual selection has been the major selection force causing ornament elaboration, because focal traits can have mainly evolved via viability selection and sexual selection may have caused ornament elaboration only slightly beyond viability optimum (i.e., possible exaptation; Bergstrom & Dugatkin, 2016 ). Then, it is not surprising to see that researchers have alternatively focused on the cost of ornamentation (e.g., Evans, 1998 ; see below), because the ornaments' cost function would clarify the relative importance of sexual and viability selection on the focal traits. A famous example is those testing aerodynamic performance of long outermost tail feathers in the barn swallows Hirundo rustica (Figure 1 ; e.g., Bro‐Jørgensen et al., 2007 ; Buchanan & Evans, 2000 ; Evans, 1998 ). Although long outermost tail feathers have been shown to be sexually selected (e.g., Møller, 1988 ; reviewed in Møller, 1994 ; Romano et al., 2017 ; Turner, 2006 ), it remains unclear how and why long tails evolved due to the possible aerodynamic function of long tails (Norberg, 1994 ). Evans and colleagues have experimentally shortened outermost tail feather length and measured aerodynamic performances of swallows (e.g., Buchanan & Evans, 2000 ; Evans, 1998 ; Rowe et al., 2001 ; also see Evans & Thomas, 1997 ; Thomas & Rowe, 1997 for the detailed descriptions of predictions). They predicted that, if long outermost tail feathers evolved purely through sexual selection, shortening the length should produce a decrease in aerodynamic cost (Figure 1 upper right). Or, if it evolved purely through viability selection due to the aerodynamic function, shortening the length should increase the aerodynamic cost (Figure 1 upper left). Lastly, if viability and sexual selection together favor the evolution of long tails, shortening the tail would decrease the aerodynamic cost first and then increase the cost beyond the aerodynamic optimum (Figure 1 upper middle). Their results were consistent with the last prediction with estimated peak values located around 10 mm shorter from the current tail length. Therefore, they concluded that tail feathers mainly evolved through viability selection and sexual selection would have elongated tails only around 10 mm (9%–20%) beyond the aerodynamic optimum (and thus many ornithological books follow their conclusions; e.g., Fjeldsa et al., 2022 ). However, this argument is problematic, because they do not consider compensatory traits (also known as cost‐reducing traits), which have been reported to affect aerodynamic cost of long tails in many bird species including barn swallows (e.g., Balmford et al., 1994 ; Barbosa & Møller, 1999 ; Møller, 1996 ; Møller et al., 1995 ). In fact, several compensatory traits have been reported in barn swallows, including greater wingspans, wing area, aspect ratios, and reduced wing loading (e.g., Møller et al., 1995 ; reviewed in Husak & Swallow, 2011 ; also see Moreno & Møller, 1996 ; Saino et al., 1997 ; Tubaro, 2003 for other kinds of compensation). Animal performance could be affected by compensatory traits, which is thought to be the case in the peacock Pavo cristatus , another model species of sexual selection: Peacocks with fully expressed trains, that is, a costly trait, have a lower metabolic cost of locomotion than those with rudimentary trains possibly due to the presence of compensatory traits (Thavarajah et al., 2016 ). Similar but less striking results, such as no detectable performance cost of gorgeous ornaments, are repeatedly reported in ornamented animals (e.g., Baumgartner et al., 2011 ; Kojima & Lin, 2019 ; Trappett et al., 2013 ). Assuming costly nature of ornamental traits, these findings are puzzling (although the possibility that all these seeming costly ornaments are virtually cost‐free cannot be excluded). The importance of compensatory traits has repeatedly been advocated (e.g., Husak et al., 2015 ; Møller, 1996 ; Oufiero & Garland, 2007 ). Using a hypothetical data set, Oufiero and Garland ( 2007 ) demonstrated that ignoring a compensatory trait led to an incorrect (and opposite) conclusion, at least in a correlational study, because the focal costly trait appeared to have a positive effect on performance if a compensatory trait was not taken into account. Likewise, Husak and Swallow ( 2011 ) reviewed compensatory traits and proposed that a simple test of relationships between ornamentation and performance can lead to misleading conclusions (also see Husak et al., 2015 for an updated review). However, although these studies stress that correlational and even manipulation experiments should consider compensatory traits carefully (e.g., Husak & Swallow, 2011 ), it is still unclear whether and how compensatory traits affect cost function concerning ornament manipulation. Particularly, whether the low performance of individuals with reduced ornaments (and hence concave cost function; see preceding paragraph) can be explained by compensatory traits is not known. As Oufiero and Garland ( 2007 ) used a hypothetical data set, a simple model experiment rather than complicated empirical data would be helpful to demonstrate the potential influence of compensatory traits. Here, using a simple “toy” model, I examined whether the concave cost function (see Figure 1 ) could be produced when the ornament has evolved purely through sexual selection for exaggerated ornamentation with the coevolution of compensatory traits. Although I verbally explained potential confounding effects of compensatory traits above, our simple model experiment would directly demonstrate how compensatory traits affects the performance of (virtual) animals with ornamentation. For this purpose, I used “chuonchuon” (Figure 2 upper panel), a traditional balancing toy in Vietnam, as a model system. Chuonchuon is a suitable model, because its balance is determined by the whole phenotype, as in the locomotor performance of animals (e.g., see Husak & Swallow, 2011 ). Different from purely verbal models, simulations, or mathematical models, this kind of simple, physical toy model would be intuitive (and thus suitable) for empirical researchers of animal performance to understand the importance of compensatory traits. Clearly, this toy model is NOT to test the function of particular traits (e.g., swallows' tails), but to test whether compensatory traits can cause reduced whole‐body performance under an experimental reduction of ornamental traits that has evolved purely through sexual selection. Rather than testing the performance of models with arbitrary values of ornamental and compensatory traits, I first determined an evolutionary stable form of chuonchuon based on game theory under the setting that long tails have been sexually selected (i.e., providing reproductive advantages) and its negative effects on balance (i.e., survival of chuonchuon) could be reduced by a compensatory trait, extended wings. Then, I tested the cost function of ornamentation by experimentally shortening tail length, as in the previous manipulation experiments in swallows listed above. To test the generality of the finding of chuonchuon, I also provided a simple (or even rudimentary) mathematical model. Even such a simple model would be useful to determine whether and, if any, in which condition the apparent concave cost function of ornament can be found. I discussed the evolutionary implications of the observed pattern.

MATERIALS AND METHODS Experimental setup I used four commercially available “chuonchuon” for my experiment (ca. 7 cm length: Tirakita, Japan). I used unmanipulated chuonchuon as an ancestral state, in which neither sexually selected ornamentation nor compensatory traits have yet been evolved. Then, I used a plastic straw (0.4 g, 4 mm width and 160 mm length; Strix design, Japan) as an elongated tail. As a compensatory trait, I made extended wings using clay (0.8 g per wing, i.e., twice as heavy as a plastic straw to account for the lever principle; Figure 2 middle panel). I here simply set the situation with a set of dichotomous events, that is, only balanced chuonchuon can “survive,” and only longer tailed chuonchuon in the population can “reproduce.” Then, the ancestral form can survive anytime, but the tail‐elongated form cannot survive when elongated tails make them unbalanced (I used a cut‐off point as 30° deviation from the horizontal plane, here). Tail elongation made models unbalanced in the range of the current study (i.e., maximum length to −80 mm shortened tails; see Figure 1 ), and thus, they cannot survive without compensatory traits, that is, extended wings. Even when it survived, the ancestral form cannot have descendants when any derived forms with elongated tails (i.e., those with longer tails than ancestral form) exist. Because my interest here is not in the actual evolutionary trajectory but the cost function of the derived form, I obtained the evolutionary stable form by comparing the “fitness” of each form. For simplicity, I assumed continuous variation in tail length but assumed two discrete states of the extended wings (i.e., with and without extended wings). In this case, the evolutionary stable form is a chuonchuon with maximum elongated tails with extended wings. This is because chuonchuon with maximum elongated tails and extended wings can survive (i.e., balanced) and thus reproduce under any circumstances (i.e., they have the longest tails compared to all other forms). I did not subdivide compensatory traits (e.g., small/medium/large), because the purpose of the current study is not to quantify the exact peak performance or identify the best size of compensatory traits but to examine the effect of compensatory traits. Measuring performance As in manipulation experiments in the barn swallow (Buchanan & Evans, 2000 ; Rowe et al., 2001 ), I shortened the extended tails of the evolutionary stable form (i.e., a chuonchuon with maximum elongated tails as ornamentation with extended wings as a compensatory trait; see above) of chuonchuon models by 0, 20, 40, 60, and 80 mm, and took two pictures each time. From each of the two pictures per individual chuonchuon, I measured the absolute angle of chuonchuon deviated from the horizontal plane as stability (or balance ability: Figure 2 middle panel) using ImageJ software. Then, I averaged the two measurements (repeatability = 0.99, F 19,20 = 270.24, p < .0001; Lessells & Boag, 1987 ) to have a representative estimate of stability for each treatment (i.e., tail shortened by 0/20/40/60/80 mm) in each chuonchuon. Statistics To account for individual variation in performance, I used a Bayesian linear mixed‐effects model with a normal error distribution to examine the stability of chuonchuon given tail shortening. Here, stability, measured as the absolute angle of chuonchuon (see above), was used as a response variable. Individual identity was used as a random effect. By calculating the Brooks–Gelman–Rubin statistic (Rhat), which must be <1.2 for all parameters (Kass et al., 1998 ), the reproducibility of the MCMC simulation was confirmed. Data analyses were conducted using the R statistical package (ver. 4.1.0; R Core Team, 2021 ), using the function “MCMCglmm” (with its default setting) in the package “MCMCglmm” (Hadfield, 2010 ). Mathematical model To test whether a quadratic cost function of purely sexual traits (see above) can be found in the presence of compensatory traits, I used a simple mathematical model. I used here an equation for a bivariate quadratic selection surface, characterized by slope and curvilinear terms, as commonly used in evolutionary biology (cf. Arnold, 2003 ; Arnold et al., 2001 ; Lande & Arnold, 1983 ; Stinchcombe et al., 2008 ; see Section 3 ). This selection surface can be used to approximate complex empirical performance surfaces regardless of the shape of the actual performance surface (Arnold, 2003 ). I used the same symbols for each term as those used in the literature on selection surface (e.g., 0.5* γ ii , instead of q ii , though the latter is often used in multiple regression models: Stinchcombe et al., 2008 ).

RESULTS Toy model I found a significant quadratic relationship between tail length manipulation and stability of the chuonchuon (Table 1 ; Figure 2 lower panel): The stability of the chuonchuon, measured as an absolute angle deviated from the horizontal plane, increased with decreasing tail length until the peak value where chuonchuon had the maximum estimated stability and further tail shortening beyond the peak value decreased stability. Mathematical model Suppose that animal performance can be approximated using the following equation (cf. Arnold, 2003 ; Arnold et al., 2001 ; Lande & Arnold, 1983 ; Stinchcombe et al., 2008 ): where x 1 denotes the expression of the ornament, x 2 denotes the expression of compensatory trait(s), and γ 11 , γ 22 , γ 12 , β 1 , and β 2 are coefficients for each variable with α as a constant. For simplicity, imagine an elliptical performance surface where the optimal performance can be set at ( x 1 , x 2 ) = (0, 0), so that any additions of the ornament and compensatory traits decrease performance, which is accomplished by the condition, β 1 = β 2 = 0, with γ 11 < 0, γ 22 < 0, and γ 12 2 < γ 11 * γ 22 (note that, when γ 12 2 > γ 11 * γ 22 , there will be a saddle in the performance surface: Stinchcombe et al., 2008 ). Assuming that the performance surface has a hill with an upward tilting axis (i.e., for each ornament expression, the decline of performance can be minimized by the presence of certain amounts of compensatory traits, which is accomplished by γ 12 > 0; see Figure 3 for an example), the above equation can be reduced as follows: with the condition, γ 11 < 0, γ 22 < 0, γ 12 2 < γ 11 * γ 22 , and γ 12 > 0. In this bivariate performance surface, the experimental manipulation of the ornament is equivalent to changing the x 1 value while x 2 value remains unchanged from the current value of compensatory traits (say, k ; note that k > 0 under the presence of compensatory traits). Then, by substituting y = k in the above equation, the quadratic function of x 1 is obtained: This can be rewritten as follows: Now, it is clear that we can have a convex performance function (and hence, concave cost function) of ornament, because k * γ 12 / γ 11 < 0 (note that k > 0, γ 12 > 0, and γ 11 < 0; see above). Given that sexual selection favors the exaggeration of the ornament (i.e., the bivariate sexual selection surface can be an increasing function of x 1 ), the current value of the ornament should be located on the right side of the ridge line (i.e., beyond the upward tilting axis of the hill; see above; also see Figure 3 ) where sexually selected advantages offset low performance of ornamented animals. When experimentally reducing the ornament expression, we find that performance increases until the ornament is reduced to be k * γ 12 / γ 11 and then decreases (Figure 3 ). Retrospectively, this equation indicates that an incremental cost function of sexual traits, which is supposed to be common under manipulative experiments of the ornament (see Figure 1c ), can be predicted only when k * γ 12 / γ 11 ≥ 0. This can be accomplished when k = 0 (i.e., when compensatory traits are absent for some reason, such as lack of genetic variation) or when γ 12 ≤ 0 (i.e., when seemingly compensatory traits are not actually compensatory traits but just costly traits such as ornament with their joint costs being additive or multiplicative). Although this condition is confined to an elliptical performance surface as assumed above (i.e., when γ 12 2 > γ 11 * γ 22 ), this is a reasonable assumption. When there is a saddle in the performance surface (e.g., see Figure S1 left panel), there is no optimal performance different from elliptical performance surface. Furthermore, it accompanies the changing function of the ornament, in which ornament is not a purely sexual trait but can be a functional trait in some ranges of variables, which is beyond the scope of the current study (because here we focused on purely sexual traits; see Section 1 ). Under the condition γ 12 2 = γ 11 * γ 22 , the cost of the ornament is perfectly compensated by compensatory traits, and thus, animals can have any ornament value together with the corresponding compensatory traits without performance costs (Figure S1 right panel). In other words, the cost of ornament can be virtually absent under this condition. As in elliptical performance surfaces (see above), performance increases until the ornament is reduced to k * γ 12 / γ 11 and then decreases in this case.

DISCUSSION The main finding from the chuonchuon experiment is that the apparent concave cost function can evolve purely through sexual selection in the presence of compensatory traits (i.e., with no viability advantage of ornamentation), which is further reinforced by a mathematical model. The importance of compensatory traits has repeatedly been advocated (reviewed in Husak et al., 2015 ; Husak & Swallow, 2011 ; Møller, 1996 ), but previous studies assume that compensatory traits still function as “cost‐reducing” traits regardless of the situation (e.g., see Husak et al., 2015 , p. 18). Here, I showed that “overcompensation” could decrease performance, producing a concave cost function, which is not surprising as individual performance is determined not by single traits but by the whole phenotype. In the current toy model, extended wings as a compensatory trait worsened the whole‐body performance measured as stability, once the part of the elongated tail was removed. As an old Chinese proverb say, “too much is as bad as too little” at least in some cases. The assumption that the cost function can reveal the relative importance of sexual and viability selection (e.g., Buchanan & Evans, 2000 ) is therefore unreliable. In other words, the prediction of sexual selection (i.e., Figure 1c ) is no longer valid and can be replaced by a concave cost function (as in Figure 1b ) at least under the presence of compensatory traits, indicating that the presence/absence of viability selection is undetectable (see Figure S2 for the corrected prediction). Of course, the current experiment using chuonchuon models does not mean that sexual selection always produces a concave cost function (and does not mean that the same physical property, i.e., stability, determined the evolution of ornamentation, such as swallows' tail and compensatory traits). The actual compensatory traits would include multiple traits (see Section 1 ), and hence, the current balancing toy model should be regarded as a “basic” model possessing single compensatory trait (or, more accurately, a set of compensatory traits, since extended wings in fact alter multiple dimensions of the wing morphology). Still, even without mathematical model (see Section 3 ), it seems likely that overcompensation will often reduce locomotor performance, particularly when locomotion requires co‐opted, integrated phenotype (e.g., aerial locomotion). In fact, concerning the function of long tails in barn swallows, Norberg ( 1994 ) mentioned that “any experimental shortening and lengthening of the outer tail feathers is likely to upset an original co‐adapted character set...” (p. 231). When compensatory traits co‐evolved with the sexually selected tail length to minimize the aerodynamic cost of tail length (Møller, 1996 ), an experimental reduction in tail length alone cannot clarify the aerodynamic cost of tail length, because the manipulation did not follow the actual evolutionary pathway (see Figure 3 ; also see Figure S3 for an alternative scenario in which ornament has some aerodynamic function). Also, all macroevolutionary studies of these aerial insectivores so far have supported sexual (rather than viability) selection explanation (e.g., Hasegawa et al., 2016 ; Hasegawa & Arai, 2017 , 2018 , 2020a , 2020b , 2021 , 2022 ), demanding further experimental studies, possibly using a smaller amount of manipulation (so that overcompensation can be avoided) in many species with various ornamentation (i.e., phylogenetic comparative “experiment”; Figure S4 ). An alternative approach, the manipulation of compensatory traits (sensu Møller, 1996 ; see also Figure 3 ) would in theory be valuable to show the effect of overcompensation, but is impractical due to the multi‐dimensionality of compensatory traits (e.g., manipulations of wing size and shape are often difficult to conduct). In addition, the current experiment provides useful insights. First, the cost of ornamentation can be transformed from one type to another. Extended wings enhanced the stability of chuonchuon with elongated tails, though they required additional investment (i.e., a production cost) instead (also see Møller, 1996 for a similar argument). Furthermore, the production cost of compensatory traits can be much higher than the production cost of the ornament itself (see Section 2 ). It is straightforward to conclude that a single measure of cost (e.g., aerodynamic cost) is inappropriate to approximate the total cost (and thus viability disadvantage) of ornamentation, even when the production cost of ornamentation itself seems negligible at first glance, as is the case for the outermost tail feathers in barn swallows (Figure 1 ). Although we discussed here about a simple compensatory trait (i.e., extended wings) with a production cost, alternative, but not mutually exclusive, kinds of costs can be involved, for example, by changing the position of pivot point via the change of the body shape in the toy model (i.e., possible developmental cost). Such morphological changes will accompany several physiological and behavioral changes (and thus may also require some form of maintenance cost) in living animals. We therefore should keep in mind that any, or at least many, kinds of costs can be involved to compensate costly ornaments and that the considerations of all these costs might be impractical. In summary, the current model experiment demonstrated that a concave cost function, which is often thought to result from viability advantage of focal ornaments, can be observed when sexual selection has favored the evolution of the costly ornaments, due to the presence of compensatory traits. Given that compensatory traits co‐evolved with sexual traits (Møller, 1996 ), the manipulation of sexual traits alone would not clarify the evolutionary force on sexual traits (see Figure 3 , Figure S3 ). Future studies should consider that overcompensation can be detrimental and that more sophisticated experimental design is needed when inferring selection pressure on (and the evolutionary history of) ornamentation.

Abstract The cost of ornamentation is often measured experimentally to study the relative importance of sexual and viability selection for ornamentation, but these experiments can lead to a misleading conclusion when compensatory trait is ignored. For example, a classic experiment on the outermost tail feathers in the barn swallow Hirundo rustica explains that the concave (or U‐shaped) aerodynamic performance cost of the outermost tail feathers would be the evolutionary outcome through viability selection for optimal tail length, but this conclusion depends on the assumption that compensatory traits do not cause reduced performance. Using a simple “toy model” experiment, I demonstrated that ornamentation evolved purely though sexual selection can produce a concave cost function under the presence of compensatory traits, which was further reinforced by a simple mathematical model. Therefore, concave cost function (and the low performance of individuals with reduced ornaments) cannot be used to infer the evolutionary force favoring ornamentation, due to a previously overlooked concept, “overcompensation,” which can worsen the whole body performance. A simple balancing toy model demonstrated that ornamentation evolved purely though sexual selection can produce a U‐shaped cost function, which is often thought to result from viability advantage of focal ornaments, under the presence of compensatory traits. U‐shaped cost function cannot be used to infer the evolutionary force favoring ornamentation. Hasegawa , M. ( 2024 ). Compensatory traits can explain the concave cost function of purely sexual traits . Ecology and Evolution , 14 , e10850 . 10.1002/ece3.10850

AUTHOR CONTRIBUTIONS Masaru Hasegawa: Conceptualization (lead); formal analysis (lead); funding acquisition (lead); methodology (lead); writing – original draft (lead). CONFLICT OF INTEREST STATEMENT The author declares no conflict of interest. Supporting information

ACKNOWLEDGMENTS I thank Dr. Emi Arai, Dr. Shumpei Kitamura and his lab members at Ishikawa Prefectural University for their kindest advices and supports. I also thank Prof. Michael D. Jennions, reviewers, and English‐proofing Company (Enago) for their helpful comments. MH was supported by the KAKENHI grant of the Japan Society for the Promotion of Science (JSPS, 19K06850). I am grateful to valuable comments by Dr Alex McQueen and anonymous reviewers, which greatly improve the quality of the manuscript. DATA AVAILABILITY STATEMENT Data attached as Table S1 will be deposited into osf.io once accepted.

CC BY

no

2024-01-16 23:41:58

Ecol Evol. 2024 Jan 14; 14(1):e10850

oa_package/9c/3b/PMC10788312.tar.gz

PMC10788314

38226113

Introduction A preventable blindness condition is a refractive error (RE). According to the World Health Organization (WHO), in 2006, approximately 153 million people over the age of five worldwide suffer from uncorrected refractive abnormalities; eight million of these people are blind [ 1 ]. Furthermore, 12.8 million persons worldwide, or 0.96% of the population between the ages of five and 15, suffer from vision impairment as a result of uncorrected or badly corrected refractive abnormalities; Southeast Asia and China show the greatest occurrence rates in metropolitan and highly developed areas [ 2 ]. Many research has been carried out in Saudi Arabia to evaluate the RE and other eye problems of the country's population. According to a study done in Jazan, south Saudi Arabia, astigmatism (31%), hyperopia (32.2%), and myopia (17.2%) were the most common REs. Mixed astigmatism (3.5%) and hyperopic astigmatism (16.1%) came next [ 3 - 5 ]. In the same region, another study found that the prevalence of hyperopia was 3.6% in healthy people and 27.6% in those with RE. Amblyopic eyes made up 30% of hyperopic eyes. Seventy percent of students had RE, while 9.3% of students had myopia. Amblyopic eyes made up just 9% of myopic eyes [ 4 ]. According to a study, myopia accounted for 24.4% of cases of RE in Arar, northern Saudi Arabia, with hyperopia accounting for 11.9% and astigmatism for 9.5% of cases. However, in a different study conducted in the nation, it was discovered that gender and age had a substantial impact on the prevalence of the various patterns of RE [ 6 ]. According to a cross-sectional study done in Al Hassa, the population under study had a prevalence of myopia that explained 9.0% of the hypermetropia that was found in 27 students, 1.4% of myopic 34, and hyperopic astigmatism 33 [ 7 ]. Previous research found a substantial correlation between the development of myopia and older age, as well as higher academic standing in the previous semester [ 3 ]. However, school-age youngsters in Bisha province, southwest Saudi Arabia, do not know anything about RE. Early identification of these health issues may help limit the subsequent impact they have on school-age children. This study aimed to determine the prevalence and types of RE among school-age children in Bisha, Saudi Arabia. In addition, the study sought to investigate the common risk factors associated risk associated with RE.

Materials and methods A descriptive questionnaire-based cross-sectional study was carried out during the period from December 2022 to November 2023 in Bisha province, southwest Saudi Arabia. All the governmental schools in the province were involved in a computer list that the Ministry of Education accessed; then, we used a computer to choose four schools randomly. Schoolchildren from both genders were selected from four primary public schools to participate in the study. The selected schools (two males and two females) are considered the main governmental public schools in Bisha. These are Angal Bisha Primary School, King Abdullah School, Bright International School Bisha, and 12th Elementary Female Primary School. Schoolchildren who were involved in our study were selected through a simple random sampling technique. Sample size Using the Richard Geiger equation (n0=Z2pq/d2), with a margin of error determined as 5%, a confidence level of 95%, and a 50% response distribution, to get the final total sample size, which equals to 288 children. We added 25% to avoid bias in the number of participants, so 360 children were involved in our study. Inclusion and exclusion criteria All schoolchildren aged from seven to 14 years old were included in the study. Children who were physically disabled, previously diagnosed with a vision problem, and psychologically ill were excluded. Data collection and procedure A validated questionnaire form was used to collect information about sociodemographic characteristics and clinical data on the history of the ocular problem, findings of the visual acuity test, cover-uncover test, and the result of the refractor machine. Procedure The results of the refractor machine were broken down into many stages: A certified optometrist conducted a 10-minute vision test as a part of the medical evaluation. Teens with an abnormal ocular movement, an eye problem (strabismus, nystagmus, or ptosis), or a visual acuity of 6/9 (20/28) or worse in one or both eyes were referred for a fuller 45-minute ophthalmic examination within a month. This examination included the following tests: an auto chart projector (ACP-8 Series, Topcon Corporation, Tokyo, Japan) and the Snellen "Tumbling E" eye chart were utilized to measure each student's uncorrected visual acuity fully. Six meters will separate them from the brightly lit Snellen chart. Every eye was tested independently for visual acuity. The pupil was able to read more than half of the letters on the line with the lowest font; therefore, it was noted. Cover-Uncover Test A cover-uncover test was used to measure eye alignment at near (40 cm) and far (3 m) distances. During the exam, the screener covered the student's left eye with a paddle and instructed the student to focus on a certain, standardized fixation target. We held the paddle in front of the eye for around three seconds. In order to ascertain whether refixation took place, the screener examined the unhindered right eye. At least three iterations of the cover-uncover test were conducted. A tabletop video/photo refractor, the Power Refractor II (version 3.11.01.24.00), evaluated eye alignment and the RE in eight meridians binocularly. Statistical methods Data were collected, arranged, and entered in an Excel sheet (Microsoft, USA) and then transferred to IBM SPSS Statistics for Windows, version 25 (released 2017; IBM Corp., Armonk, New York, United States). Descriptive statistics were carried out and presented as mean, standard deviation, proportions, and frequency tables. Every two variables were compared using the chi-square test. P-values were considered statistically significant if they had a value less than 0.05. Ethical approval Before completing the surveys, each parent was shown the purpose and significance of the study and asked for a written agreement. The participants' confidentiality was kept as the questionnaires contained no personal data referring to or implying the participants’ identity. All of the participants included in the study were anonymous. The University of Bisha College of Medicine's (UBCOM) ethical approval granted clearance (ref. no. UB-RELOC H -06-BH-087/ (0905.23)).

Results A total of 360 schoolchildren from primary schools were enrolled in the study. The age of the participants ranged from seven to 14 years, with a mean of 10.1 years (standard deviation (SD)=2.05). Most of the participants were boys (59.4%, n=214). About half of the participants are in age equal to or more than 10 years. Most participants were from urban areas (86.1%, n=310). Of the 360 pupils, the majority were from year five (22.2%, n=80) and year three (20.6%, m=74), followed by year six (16.7%, n=60) and year two (15.6%, n=56) (Table 1 ). Among the 360 pupils, the prevalence of hyperopia was 21% in the right eye and 23% in the left eye. In addition, the prevalence of myopia was 20% in the right eye and 22.5% in the left eye (Figure 1 ). Figure 2 illustrates the percentage of visual acuity of the participants. In our study, we define myopia and hyperopia in dioptres (D) as follows: mild myopia (-0.5 to less than -3 D), moderate myopia (-3 to less than -6 D), and severe myopia (greater than -6 D); mild hyperopia (+0.5 to less than +3D), moderate hyperopia (+3 to less than + 6D), and severe hyperopia (greater than +6 D). Most of the pupils had a normal visual acuity in the right eye of 20/20 (n=122, 33.9%); the lowest result was 20/100 and 20/200 (n=4, 1.1%). On the other hand, the visual acuity of the left eye was normal at 20/20 in 130 participants (36.1%); the least visual acuity was 20/100 in eight participants (2.2%). Table 2 presents the association between the visual acuity and baseline characteristics of the students. There is a significant association between visual acuity of the right eye and age group (p=0.036) and residency (p=0.028) of students. Pupils 10 years or above and residing in local areas were more likely to have squints in the left eye. In addition, those who had previous health problems were more likely to have abnormal acuity in the left eye (p=0.028). Table 3 presents the association between the type of RE and the general characteristics of the students. Hyperopia and myopia did not show insignificant associations related to the gender, age group, and educational levels of the schoolchildren. However, there was a significant association between myopia and residence (p=0.026), where the disorder was higher among students from urban areas. In addition, students with previous ocular problem were more likely to have hyperopia than those without ocular problems (p=0.004). Table 4 summarizes the association between the type of REs and visual acuity. There was a significant association between visual acuity and myopia (p=0.001). Meanwhile, there was no significant correlation between hyperopia and visual acuity (p=0.412).

Discussion In this study, the prevalence of REs among schoolchildren in Bisha, Saudi Arabia, between the ages of seven and 14 years, was compared to reports from other parts of the world. Understanding that RE is a complicated and multidimensional condition with a wide range of genetic, demographic (age, race, ethnicity, and geography), ocular, and extrinsic factors (pressure to pursue higher education, changes in lifestyle, and prolonged indoor and near activities) variations in its prevalence. Numerous studies have evaluated the frequency of different types of refractive defects in students [ 4 - 8 ]. To enable meaningful comparison, we restricted the comparison to a study that was published in 2000 and after [ 9 - 14 ]. The overall prevalence in our study was very high; with 55.8% of all participants, it was significantly associated with age, residence, history of previous ocular problems, and visual acuity test. A high prevalence of REs in schoolchildren can impose a significant economic burden on families and healthcare systems. The cost of eye examinations, eyeglasses, contact lenses, and other vision correction interventions can be substantial, particularly for families with limited financial resources. However, a history of squint was not common among the participants. Few of the students were using spectacles. Uncorrected RE can damage a person's vision in both children and adults, with both short- and long-term consequences, such as missed chances for school and employment; slower economic growth for individuals, families, and countries; and a lower standard of living [ 8 ]. REs go uncorrected for a variety of reasons, including lack of awareness and recognition of the issue at the individual, family, community, and public health levels; lack of accessibility to and/or affordability of refractive testing services; inadequate availability of reasonably priced corrective lenses; and cultural barriers to compliance [ 9 ]. Although refractive defects might be easily repaired, the estimated amount of visual impairment brought on by uncorrected REs is of public health concern [ 6 ]. RE is more common in some regions than others, ranging from 10% in Australia [ 3 ] to 50.3% in India [ 9 ]. In our study, there was a high prevalence of REs, which is similar to the study done in Taif region of Saudi Arabia, with a 50.9% prevalence rate. This is in contrast to what was reported in King Abdul-Aziz Medical City, Riyadh (9.8%) [ 5 , 14 ], Al-Hassa, eastern Saudi Arabia (13.7%) [ 7 ], and Qassim Province, Saudi Arabia (16.3%) [ 11 , 14 ]. Even among studies carried out in the same geographic area, there is a significant variation in the prevalence of RE overall. This wide variation may be due to variations in the operational definition, cut-off values used to identify various types of REs, and measurement techniques. It has been discovered that the prevalence of hyperopia varies significantly among different people [ 1 , 8 ]. The current study stated that the prevalence of hyperopia was higher (32.2%) than that previously reported from Saudi Arabia (0.7-17.63%) [ 11 , 14 ]. In addition, our study showed an increased tendency for hyperopia prevalence among those aged less than 10 years (62%) compared to those aged more than 10 years (54%), which is inconsistent with findings in Riyadh's study, where only 4.7% of them had hyperopia [ 15 ]. Similar surveys conducted in schools across the globe reveal that myopia prevalence varies from 0.8% to 65.7%. In the current study, myopia was detected in 26.9% of the individuals, which is similar to earlier findings from Al-Hassa, Saudi Arabia (23.4%) [ 7 ], Taif, Saudi Arabia (33.2%) [ 2 ], China (36.9%) [ 1 ], and Iran (29.3%) [ 10 , 16 ]. However, it is higher than those previously reported in Al-Qassim (5.8%) and Riyadh (5.7%). Given the global trend of myopia increasing, prevention programs, including limiting near activities and encouraging children to participate in more outdoor activities, may be necessary. Study limitations In order to highlight the necessity of establishing a school-based child eye care system in Saudi Arabia, the primary goals of this study were to estimate the prevalence of RE and compare the findings with those of similar studies conducted in Saudi Arabia and other nations. This research, however, necessitates additional studies of a similar nature that concentrate on the causes, risk factors, and associations between various REs, as these aspects will aid in the rationale and explanation of the findings and development of necessary preventive measures that will give future research greater weight. In addition to the high rate of student absenteeism and gender-related concerns, this study was conducted around the conclusion of the academic year, which is a very limited period to perform such a significant study with a large sample. Children with attention-deficit hyperactivity disorder (ADHD), autism, or other conditions that could interfere with their ability to cooperate and comprehend the tests being administered were not allowed. The lack of transportation in more rural locations also prevented schools from being included.

Conclusions The prevalence of REs among school-age children in Bisha, Saudi Arabia, was presented in the current study. The study results showed that half of the sample population in this area had at least some REs. According to the study's findings, about half of the participants in this research had some degree of REs. These results highlight the need for prompt and careful screening programs to detect and treat refractive problems in this age range.

Background: A widespread and serious eye condition is a refractive error (RE). Globally, uncorrected refractive defects affect numerous individuals, with some who are blind. Numerous studies in Saudi Arabia have been conducted to assess reflective error. However, there is a lack of knowledge regarding RE among school-age children in Bisha province, southwest Saudi Arabia. This study aimed to determine the prevalence and types of RE among school-age children in Bisha, Saudi Arabia. Methods: A cross-sectional study involved 360 schoolchildren from primary schools was carried out between December 2022 and November 2023 in Bisha. A validated questionnaire form was used to collect sociodemographic information and clinical data (history of the ocular problem, visual acuity test findings, and the refractor machine's result). Result: A total of 360 schoolchildren aged from seven to 14 years, with a mean of 10.1 years (standard deviation (SD)=2.05). The prevalence of hyperopia was 21% in the right eye and 23% in the left eye. In addition, the prevalence of myopia was 20% in the right eye and 22.5% in the left eye. A significant association between visual acuity and myopia (p=0.001). By contrast, there was no significant correlation between hyperopia and visual acuity (p=0.412). Conclusion: The current study summarized the prevalence of REs among school-age children in Bisha, Saudi Arabia. The study population included nearly half of those with at least some degree of RE. These results highlight the need for prompt and careful screening programs to detect and treat refractive disorders across this age range.

CC BY

no

2024-01-16 23:41:58

Cureus.; 15(12):e50530

oa_package/bd/62/PMC10788314.tar.gz

PMC10788315

38226093

Introduction The infratemporal fossa, situated immediately medial to the mandibular ramus, serves as a central hub for the distribution of the mandibular division of the trigeminal nerve. The four main branches of the mandibular nerve are lingual, inferior alveolar, long buccal, and auriculotemporal nerves, with the latter carrying postsynaptic parasympathetic fibers from the otic ganglion. The auriculotemporal nerve exhibits notable variability, including the presence of single, double, triple, or even quadruple roots [ 1 , 2 ]. The nerve may display a connecting branch between its upper and lower roots, may have between five to nine terminal branches, and may fenestrate the superficial temporal vein [ 1 , 3 ]. In some cases, it can also establish connections with adjacent nerves, such as the lesser occipital nerve temporofacial division or temporal branch of the facial nerve [ 1 , 4 ]. The nerve occasionally communicates with other branches of the trigeminal nerve, notably the zygomaticotemporal branch of the zygomatic branch of the maxillary nerve [ 5 ]. The precise functional significance of auriculotemporal nerve communications remains incompletely understood [ 4 ]. One study has observed that the fibers within the communicating auriculotemporal nerves in facial nerve branches consistently innervated certain upper muscles of facial expression [ 6 ]. In this report, we examine an instance of neural communication between the auriculotemporal and inferior alveolar nerves through both gross and histological studies.

Discussion The prevalence of neural communications between the auriculotemporal and inferior alveolar nerves varies in the literature, likely owing to the incidental nature of this observation and the limited number of specimens examined. Anecdotally, we have rarely observed this anastomosis during infratemporal dissections. Anil and colleagues noted the bilateral occurrence of communication between the auriculotemporal and inferior alveolar nerves in one out of 10 specimens they dissected [ 7 ]. The second part of the maxillary artery passes through the loop formed by the communicating branch of the auriculotemporal nerve and the mandibular nerve proximal to the origin of the inferior alveolar nerve. Gülekon and colleagues also observed a connecting branch between the auriculotemporal and inferior alveolar nerves in four out of 32 infratemporal fossae [ 2 ]. Buch and Agnihotri reported a recurrent branch of the inferior alveolar nerve, which existed in approximately 45% of cadavers [ 8 ]; in one of the cases they described, the recurrent branch bifurcated, the two divisions enclosed the maxillary artery and then joined the mandibular nerve. There was a communication between a division of the recurrent nerve and the auriculotemporal nerve. The present case shows some differences from those reported before. First, the maxillary artery did not pass through the loop formed by the communicating branch between the auriculotemporal and inferior alveolar nerves. Second, on histological examination, the auriculotemporal nerve, before and after the communicating branch, had 12 and 10 nerve fascicles, respectively. Given that both the auriculotemporal and inferior alveolar nerves are primarily sensory in function, this finding suggests the possibility of sensory fibers traveling from the inferior alveolar nerve to the auriculotemporal nerve. Last, we noted that the auriculotemporal nerve contained several ganglion cells. Whether these ganglions represent sensory or secretomotor neurons is not clear and should be investigated in future studies. Ganglion cells have been reported in the trunk of several other cranial nerves (e.g., glossopharyngeal, spinal accessory, and hypoglossal nerves) and their branches [ 9 - 11 ]. The neural communication between the auriculotemporal and inferior alveolar nerves is clinically significant as, for example, pain can be referred from the anatomical territory of one nerve to the other.

Conclusions Occasional neural communications exist between the branches of the mandibular nerve within the infratemporal fossa. The functional significance of these interconnections is yet to be fully understood. In this report, we illustrate neural communication between the auriculotemporal and inferior alveolar nerves. On histological examination, the auriculotemporal nerve proximal to the communicating branch exhibited a greater number of nerve fascicles compared to the distal portion beyond the communication. Given that both nerves primarily serve sensory functions, this observation indicates that the sensory fibers potentially pass from the inferior alveolar nerve to the auriculotemporal nerve. Consequently, this observed neural connection could be implicated in referred pain from the anatomical territory of the inferior alveolar nerve (mandibular teeth) to regions innervated by the auriculotemporal nerve, including the temporomandibular joint, auricle, external auditory meatus, tympanic membrane, and the temporal skin. Further research is crucial to elucidate the prevalence and functional significance of neural communications within the infratemporal fossa.

Communications between cranial nerves or their branches have been described previously. The exact functional significance of some of these neural communications remains to be fully understood. This paper reports a unique communication between the auriculotemporal and inferior alveolar nerves within the infratemporal fossa. The histological examination indicates an antegrade connection from the inferior alveolar nerve to the auriculotemporal nerve, which could potentially be implicated in referred pain from the anatomical territory of one nerve to the other.

Case presentation During dissection of the infratemporal fossa in a formalin-fixed, latex-injected, adult head specimen, a neural communication was found between the auriculotemporal and inferior alveolar nerves. Briefly, after the removal of the masseter muscle and outer cortex of the mandible, the inferior alveolar nerve and artery were isolated and traced to their origin. The condyle and coronoid process of the mandible and lateral pterygoid muscles were removed. The lingual nerve was located and traced to its origin from the otic ganglion. Next, the middle meningeal artery and its surrounding auriculotemporal nerve fibers were isolated. It was noted that a branch from the auriculotemporal nerve traveled anteroinferiorly over the medial pterygoid muscle to join the inferior alveolar nerve in the infratemporal fossa (Figure 1 ). Specimens from this communicating branch and auriculotemporal nerve before and after this communication were fixed in 10% buffered formalin solution and submitted for histological examination using Luxol fast blue with hematoxylin-eosin staining. On histology, the communicating branch contained 1 nerve fascicle without ganglion cells. Before communicating, the auriculotemporal nerve had approximately 12 nerve fascicles with a cluster of ganglion cells. The portion of the auriculotemporal nerve distal to the communication had 10 nerve fascicles with no ganglion cells.

The dissection was carried out by MMS and RST. TT, MMS, and RST conducted the literature review and drafted the manuscript. All authors reviewed and approved the final manuscript. The authors are grateful to Dr. Thomas Müller for providing them with a copy of his elegant article. Artificial intelligence (AI) (GPT 3.5) has been employed for the language editing of this manuscript. No part or scientific content within this manuscript has been generated by AI.

CC BY

no

2024-01-16 23:41:58

Cureus.; 15(12):e50526

oa_package/8d/81/PMC10788315.tar.gz

PMC10788316

38226096

Introduction Surgical training in the United Kingdom typically comprises two distinct programs: two years of core surgical training (CST), followed by six years of specialist training (ST) as a speciality registrar [ 1 ]. During CST, junior surgical trainees () undergo placement in several different, but usually complementary, specialities and are encouraged to develop both their operative and nonoperative skills in a supported environment [ 2 ]. They are also obligated to undertake surgical skills courses such as advanced trauma life support (ATLS) [ 2 ]. Completion of the Membership of the Royal College of Surgeons (MRCS) examination is an exit requirement for satisfactory CST training completion [ 2 , 3 ]. The advantage of this structure is junior trainees (CSTs) are provided with an extremely well-supported training environment surrounded and trained by their more senior colleagues (STs and consultants). CSTs mostly operate with the direct supervision of a trainer (scrubbed or unscrubbed), which is meant to support the development of good surgical techniques: this is defined by the Intercollegiate Surgical Curriculum Program (ISCP) as Level II supervision (able and trusted to act with direct supervision) [ 2 ]. As a result, it is assumed that once a trainee reaches the speciality training (ST3) level, they will be competent to act and operate with an element of indirect Level III supervision (defined by ISCP as able and trusted to act with indirect supervision) [ 2 ]. This is assessed from their final multi-consultant report (MCR) at the end of CST, where their level of competence and independence is formally signed off [ 2 ]. This structure heavily relies on proactive trainers in the theatre environment. Time pressures, individual attitudes, and the number of trainees competing for operating opportunities among other factors including maintaining compliance with the European Work Time Directive can contribute to the quality of training received [ 4 ]. The expectation of CSTs in theatre in terms of their operative ability and needs is often different from that of STs [ 2 , 5 ]. Given current NHS pressures, there is often the request for the CSTs (and less so of STs) to leave theatre and address nonoperative surgical issues in the hospital, which can come at the expense of their theatre exposure [ 6 , 7 ]. In busy periods such as winter or surrounding ongoing strike action, this may result in inadequate access to operative training opportunities [ 8 ]. One way in which core trainees (CT) throughout the NHS have traditionally tried to bridge the gap between core and ST is the concept of “acting up” [ 9 ]. The CST ‘steps up’ to the role of ST, typically in the latter part of their core training, when the opportunity arises. This gives them exposure to ST clinical work in preparation for their official ST3 year, which often follows [ 9 , 10 ]. Whilst this can be highly beneficial for training, it requires the support of all senior colleagues in understanding that more support will be necessary for the CST-registrar (CST-R). If this is not available, then acting up risks the CST-R feeling overwhelmed and burnt out, struggling to handle the increased responsibilities [ 11 ]. This can lead to errors with significant medicolegal consequences and negatively impact their confidence both inside and outside of the operating theatre [ 12 ]. The progression from practising under direct to indirect supervision is a large and potentially daunting one. There is a greater emphasis on decision-making, which is particularly evident in acute situations. Given the hierarchical nature of the surgery barriers for a CST to escalate to their ST are much lower than those for an ST to escalate to their consultant [ 13 ]. Thus, escalating from a CST-R to a consultant poses the greatest challenge, and a supportive and unintimidating consultant body is essential. It is relatively difficult to quantitively compare the nonoperative experience of trainees before and after acting up. Operative experience is much easier to quantitively compare as all UK core surgical trainees are required to keep a logbook of cases participated in [ 2 ]. This includes coding all procedures according to the degree of involvement in the Surgical eLogbook. The options are assisting (A), supervised trainer scrubbed (STS), supervised trainer unscrubbed but in theatre (STU), performed (P), and training a trainee (T) [ 10 ]. In “A,” the trainee performs less than 50% of the operation; in “STS,” they perform over 50%; and in “STU,” “P,” and “T,” they perform the whole procedure with or without the trainer present [ 10 ]. We compare the operative experience of a vascular surgery-themed trainee during six months as a core surgical trainee and six months of acting up as a CST-R.

Materials and methods A retrospective analysis of the Intercollegiate Surgical Curriculum Program (ISCP) eLogbook was performed. The inclusion criteria were all operations performed by one vascular surgery-themed trainee within a regional vascular hub between August 3, 2022, and January 31, 2023, and all operations performed between February 1, 2023, and August 1, 2023. This corresponds to the full CST year one. The first six-month block corresponded to time spent as a true core surgical trainee (year one) and the second six-month block acting as a registrar (CST-R). Noncoded operations or operations with any role other than (A, STS, STU, P, or T) were excluded. Volume and variation in the complexity of caseload populating operating lists throughout the year were taken to have remained consistent. Data were collected, and graphs were generated using Microsoft Excel. For each case, data were collected on the role played (A, STS, STU, P, T), the six-month period during which the case was performed, and whether the procedure corresponded to a vascular index procedure (amputation/debridement, vascular access procedure, varicose vein surgery) as defined by the ISCP Vascular Surgery Curriculum [ 5 ]. The proportion of cases logged at each level of involvement was compared between the two six-month blocks. Statistical analyses were conducted in Stata (version 18.0; StataCorp LLC, College Station, TX). Pearson’s chi-squared test was used to test the null hypothesis that there was no difference between the proportion of cases performed at each level during each six-month block to the 0.05 significance level. The number of low-difficulty cases where the trainee was the sole operator (cases logged as STU, P, or T) in each block was used as a baseline to ensure the progression seen was because of increasingly complex ST operating rather than the increased volume in CST level operating expected throughout CST. An abscess incision and drainage were used as the reference low-difficulty case. This was selected as within the Trust, CSTs routinely perform abscess incisions and drainages as the sole operator. The number of days the trainee was within the hospital and thus available to operate within each block was also compared to account for any reduced operating from days off because of industrial action.

Results The trainee conducted a total of 229 cases overall. One hundred twenty cases were performed in the first six-month block as a CST year one, and 109 cases in the second six-month block as a CST-R. Within the first six-month block, 89.5 days were spent in the hospital with the potential to operate, and, in the second, 77.5. This reduction was largely because of NHS industrial action. The 9.2% reduction in total cases performed corresponds to a 13.4% reduction in days spent in the hospital. As a CST-R, the trainee played a more prominent operative role in a greater number of cases. The number of cases the trainee performed as the sole operating surgeon (STU, P, or T) demonstrated a relative increase of 100% from 16 (13%) to 28 (26%) cases (p=0.018). The number where they were simply assisting demonstrated a relative decrease of 19% from 52 (43%) to 38 (35%) cases (p=0.19). The number of cases where the trainer remained scrubbed also demonstrated a relative decrease of 9% from 52 (43%) to 43 (39%) cases (p=0.55). The number of low-difficulty cases performed independently remained unchanged at 11 for each six-month block. A total of 91 index procedures were conducted. Forty-nine of these as a CST year one and 43 as a CST-R. The number of procedures recorded as A demonstrated a relative decrease of 33% from 20 (41%) to 12 (28%) cases (p=0.195). Those recorded as STS demonstrated a relative decrease of 12% from 28 (57%) to 22 (51%) cases (p=0.56). The number of cases performed as sole operating surgeons (STU, P, or T) demonstrated a relative increase of 950% from one (2%) to nine (21%) cases (p=0.004). The total number of cases along with the the breakdown of roles (A, STS, STU, P, and T) for the first- and second six-month blocks are shown in Table 1 . The number of index procedures and the breakdown of roles for the first and second six months are shown in Table 2 . The percentage breakdown of roles played for all cases for the first six-month block is shown in Figure 1 . The percentage breakdown of roles played for all cases for the second six-month block is shown in Figure 2 . The percentage breakdown of roles for index procedures for the first and second six-month blocks is demonstrated in Figure 3 .

Discussion There is a clear trend evident in the results. As a CST-R, the trainee had played a more significant role in a greater number of cases. The number of cases where they were operating independently (STU, P, or T) went from 13% to 26% of the total cases performed within the six-month block, a statistically significant relative increase of 100%. The number of independent low-difficulty cases remained unchanged between the two blocks. This indicates the CST-R continuing to perform typical CST level cases whilst adding more complex ST level operating to their repertoire. That the number of index procedures conducted (STU, P, or T) demonstrated a significant increase from 2% to 21% supports this within the caveat of the small sample size. Although not of statistical significance, the decrease in the number of index procedures recorded as A may indicate that CST-R plays a more significant role in vascular index procedures both operating alone and being trained. The relatively small decrease in overall cases recorded as A or STS demonstrates the increased independent operating is unlikely to have greatly detracted from the CST-R's one-to-one operating with a scrubbed trainer or their exposure to the most complex cases as an assistant: both important parts of surgical training. Notably, the P value for the change in both A and STS cases overall and for index procedures was >0.05. Therefore, it is only possible to attribute statistical significance to the changes in operation observed in this study where the CST-R was the sole operator. The longer a CST remains in a department, the more their abilities and skill level will be clear to their trainers. This tends to result in increasing levels of trust and independence with time. However, it is unusual for CSTs to be left to operate independently without a trainer in the operating theatre as this corresponds to a level III on the ISCP Multi Consultant Report, and it is an exit requirement of the program, not expected until the last placement of CST [ 3 ]. Operating without a trainer directly available to guide and instruct forces the development of operative decision-making and reasoning under pressure [ 14 ]. This is a very important skill that is expected of STs, and it is a source of significant stress amongst junior registrars when they begin their higher training [ 11 ]. This work shows that acting up provides opportunities to develop these skills in a more supportive and controlled environment, which could help in reducing the anxiety and stress experienced by trainees on beginning higher training. Expectations regarding the complexity of operations performed also change between CST and ST [ 2 , 5 ]. The CST curriculum requires competence to be achieved (defined as level III trusted to act with indirect supervision) in six operative procedures: scrubbing, gloving, and gowning, preparing an aseptic field, injecting local anesthetic, and incising and closing a wound [ 2 ]. As they progress in our Trust, CSTs may also be expected to perform “simple” procedures such as incision and drainage of an abscess unsupervised. STs are expected to perform more complex procedures and the ISCP vascular curriculum requires "index procedures" to be signed off as level IV (defined as able and trusted to act at the level expected of a day-one consultant) as trainees progress [ 5 ]. The step up as a CST-R offers exposure to operating at a higher level provided that the CST-R is competent to perform the more complex procedures. The experience of ST-level operating is extremely good preparation for higher training. For the trainee in this study, all cases performed “STU,” “P,” or “T” were discussed preoperatively in a step-by-step manner and then again postoperatively with a consultant. This allowed for reflection on difficulties encountered and how the trainee addressed them. Examples of cases performed in this way include above and below-knee amputations, forefoot and toe amputations, major wound debridements, peritoneal dialysis catheter insertions and removals, arteriovenous fistula formations, and varicose vein ablations. These are all vascular surgery "index procedures" [ 5 ]. No complications arose from any of the procedures undertaken by the trainee “STU,” “P,”, or “T” as a CST-R. There is also a benefit to other STs in the department in having a CST-R. In our department, STs cover the senior workload in three main areas: the acute Take, the ward, and the operating theatre. The acute take and the ward are covered by a “Reg of the Week” (RoW). The RoW has no scheduled operations during this period to ensure availability for their nonoperative duties, other than emergency vascular operations. The promotion of the CST-R ensured that the other STs did a reduced frequency of RoW and thus a corresponding increase in scheduled operating lists. It also improved opportunities for other STs to take study leave and attend courses because of the reduced frequency of scheduled commitments. Despite the numerous positives from the CST-R program, there are nonetheless some potential drawbacks. The relative lack of experience of the CST-R means that senior colleagues must provide more support than they otherwise would for an ST. The promotion of the CST results in one fewer member of junior staff on the ward for STs to delegate to. This may increase the overall workload of the RoW on the ward. The supervising consultant will also need to provide more support for the CST-R during their periods as RoW to ensure all senior ward and acute take duties are carried out safely. Finally, because of a lack of experience, the CST-R will inevitably operate slower than their more experienced colleagues, and this needs to be accounted for when planning lists. A buddy system was used for the trainees in this study. Whenever they were RoW and covering the acute take, a senior registrar was scheduled to be available as the first port of call. This enabled minor clinical issues and logistics to be addressed without the need to involve the consultant on call. A similar concept is often used in hyperacute surgical specialties such as neurosurgery or trauma surgery where first-year STs (ST3) are placed on a two-tier rota: the junior registrar is the tier one, and a senior registrar is their tier two; there is then a supervising consultant on call from home [ 15 ]. This structure ensures that approachable support is always available for the junior trainee whilst protecting the consultant body from being harassed unnecessarily when on call. This study has several limitations. These include a reliance on accurate theatre data being entered into the eLogbook. It was assumed that the trainee uploaded 100% of theatre cases accurately. If this study were to be repeated, there is a risk that minor procedures are omitted as the CST-R focuses on more complex procedures as they improve their surgical abilities. The second six-month period analyzed for this study corresponded with industrial action on a scale never before seen in the NHS [ 16 ]. There was an approximate 12% reduction in elective vascular operating lists at our Trust. Within our department, elective cases are frequently the cases where a CST-R would play a greater role because of the complexity of cases and time allocation. Many routine elective operating lists were canceled, and priority was given to emergencies [ 17 ]. Many of the routine elective operations canceled would have been cases where trainees would be operating. Another major limitation is that this study was only addressed from the perspective of the CST-R. The opinions of both the other STs and the consultant body are not reported. Similarly, the impact of losing an approachable CST on the ward for the foundation year one (F1) doctors was not reported. The final caveat relates to the structure of CST rotations in the UK. Most two-year programs are made up of four six-month rotations through different specialities [ 1 ]. For a trainee to achieve the experience and level of trust to act up within a speciality they need to be in post for a significant length of time. This ideally requires two six-month rotations within the same speciality and the same department. It is interesting to note that these changes were observed within the CT1 year rather than the traditional CT2 year and the trainee in question had undertaken a year of surgical training in a different speciality before starting CST.

Conclusions Acting up can be an invaluable way for core surgical trainees to develop their operative and nonoperative skills. It provides greater independence and endorses the development of clinical reasoning and decision-making. It enables the CST to practice their leadership skills and more senior-level communication skills. It also offers significantly more operative opportunities to better prepare a trainee for their higher specialty training. It nonetheless requires a supportive department and consultant body, and, if this is not ensured, there is potential for junior trainees to be rapidly overwhelmed with ensuing personal and medicolegal ramifications. The structure of CST in the UK with its frequent rotations between specialties and departments is not always conducive to creating this environment. If a trainee can remain in post for two six-month blocks then there is much to be gained from a formalized acting-up program and consideration should be given to designing core surgical rotations with this in mind. In our Trust, this opportunity is also available in urology and is thought to be successful.

Introduction United Kingdom surgical training consists of a two-year core surgical training (CST) followed by a six-year higher speciality training (ST). There is a significant step up in responsibility and operative skills when transitioning from core to higher training. One-way trainees can bridge this gap is to “act up” to registrar level “CST-R.” The CST “steps up” to the role of ST typically in the latter part of their core training and gains exposure at being the "reg of the week," primary assistant in theatre, managing MDTs, and taking speciality referrals. This can be an excellent training opportunity. This study aims to demonstrate a quantitative improvement in trainee operation as a result of stepping up. Methods This study compares the operative experience of one vascular surgery-themed trainee during six months as a CST and six months acting up as a CST-R. The trainee’s eLogbook was searched for all operations between August 3, 2022, and January 31, 2023, and between February 1, 2023, and August 1, 2023. The number of cases performed and the role played in each were analyzed. The number of low complexity cases conducted in each block was used as a baseline to ensure the progression seen was because of increasingly complex ST operating rather than the increase in CST level operating expected throughout CST. An abscess incision and drainage were used as the reference low-complexity case. Results The number of cases the trainee performed independently increased from 13% to 25%, while the number where they were simply assisting decreased from 43% to 35%. The number of cases where the trainer remained scrubbed decreased nonsignificantly from 43% to 39%. The number of low-complexity cases performed remained unchanged for each six-month block. Conclusion As a CST-R, the trainee played a more prominent operative role in a greater number of cases. The CST-R does require a supportive department and consultant body. It also enables other STs to gain more surgical exposure because of their reduced frequency of being the "reg of the week." If a trainee can remain in a post for two six-month blocks, then there is much to be gained from a formalised acting-up program, and consideration should be given to formally incorporating this into core surgical programs.

CC BY

no

2024-01-16 23:41:58

Cureus.; 15(12):e50517

oa_package/34/da/PMC10788316.tar.gz

PMC10788317

38226087

Introduction Osteochondromas are the most common type of benign bone tumor, accounting for 35% to 40% of all benign bone tumors [ 1 ]. However, their occurrence in nonstandard locations and association with developmental deformities present unique challenges in diagnosis and management [ 1 ]. This report details an uncommon case of ulnar exostosis with a concurrent developmental deformity of the left forearm in a 15-year-old female patient, highlighting the significance of recognizing atypical presentations of osteochondromas, particularly in pediatric orthopedics. The etiology of osteochondromas is multifactorial, often involving genetic predisposition and skeletal trauma, especially in growth plates and metaphyseal regions [ 2 ]. While these tumors primarily affect the metaphyseal regions, their manifestation in the ulna with an associated developmental deformity is rare. A comprehensive understanding of osteochondromas' pathogenesis and clinical presentations is essential for accurate diagnosis and effective management [ 3 ]. Radiographic and computed tomography (CT) imaging plays a pivotal role in characterizing the extent and nature of osteochondromas. Imaging findings offer valuable insights into these lesions' location, size, and potential complications [ 4 ]. The correlation between clinical presentation and imaging findings guides the selection of appropriate surgical interventions. Surgical excision is often necessary in symptomatic cases, with extraperiosteal en bloc resection considered the preferred approach to mitigate the risk of recurrence [ 5 ]. The surgical outcome is influenced by various factors, including the lesion's location, involvement of adjacent structures, and the patient's overall health [ 6 ]. This case report aims to contribute to understanding osteochondromas by presenting a rare case involving a unique combination of ulnar exostosis and developmental forearm deformity. By exploring the patient's clinical history, radiographic findings, and surgical intervention, we aim to enhance our understanding of the diagnostic and therapeutic challenges associated with osteochondromas in uncommon anatomical locations.

Discussion The presented case of ulnar exostosis with a developmental forearm deformity in a 15-year-old female patient raises several important considerations in the field of pediatric orthopedics. The occurrence of exostosis in the ulna, combined with a history of trauma and a subsequent supracondylar humerus fracture, introduces complexity to the understanding of musculoskeletal anomalies in this population. The patient's history of trauma and fracture aligns with previous studies that associate the development of osteochondromas with skeletal trauma, particularly in the context of growth plates and metaphyseal regions [ 7 ]. Although osteochondromas, as the most common benign bone tumors, typically originate in the metaphyseal regions, their manifestation in the ulna with a concurrent developmental deformity is relatively uncommon [ 8 ]. The specific correlation between trauma, fracture, and the subsequent development of an exostosis in this case warrants further exploration of the potential causal relationship. Radiographic and CT imaging played a pivotal role in the diagnostic process. The anteroposterior and lateral views of the left forearm revealed a solitary external bony protuberance over the ulna shaft, providing crucial visual evidence of the pathology. CT scans further elucidated the nature and extent of the lesion, demonstrating a large anteromedial bony projection involving the middle 1/3rd of the ulna with medullary cavity communication with the parent bone. These findings are consistent with the diagnostic utility of imaging modalities in characterizing osteochondromas [ 9 ]. The surgical approach involved extraperiosteal en bloc resection of the lesion under supraclavicular nerve block anesthesia. The success of this procedure is supported by existing literature advocating for surgical excision in symptomatic cases, emphasizing the significance of addressing the lesion extraperiosteally to prevent recurrence [ 5 ]. Histopathological examination confirmed the diagnosis of osteochondroma. This aligns with previous studies highlighting these tumors' benign nature, characterized by mature hyaline cartilage capped by a bony cortex [ 10 ]. The patient's favorable postoperative outcome, evidenced by the absence of pain at the one-month follow-up, underscores the efficacy of the surgical intervention in alleviating symptoms associated with such musculoskeletal anomalies.

Conclusions This case highlights the significance of considering unusual etiologies, such as exostoses, in the differential diagnosis of forearm swellings, particularly in the context of a relevant medical history. The radiographic and CT findings were crucial in establishing a precise diagnosis and guiding the appropriate surgical approach. The absence of complications postoperatively and the restoration of the patient's quality of life further underscore the importance of early recognition and intervention in managing such rare musculoskeletal conditions. As our understanding of atypical presentations of common pathologies continues to evolve, this case contributes to the expanding knowledge base in orthopedics. It emphasizes the need for a multidisciplinary approach, including clinical, radiological, and histopathological assessments, to ensure accurate diagnosis and effective management. Continued documentation and sharing of such cases contribute to the collective knowledge that informs clinical practice and enhances patient care in orthopedic medicine.

This case report presents a rare occurrence of exostosis of the ulna associated with a developmental deformity of the left forearm in a 15-year-old female. The patient reported a history of trauma resulting in a supracondylar humerus fracture managed conservatively eight years prior. The patient presented with a two-year history of pain and swelling over the left forearm. Clinical examination revealed a firm, non-tender, immobile swelling closely associated with the ulna, accompanied by a 20-degree cubitus varus deformity and forearm shortening. Radiographs and computed tomography scans confirmed the presence of a solitary external bony protuberance over the ulna shaft, communicating with the medullary cavity. A preliminary diagnosis of osteochondroma was established based on clinical and imaging findings. The patient underwent extraperiosteal en bloc resection of the lesion under supraclavicular nerve block anesthesia. A histopathological examination confirmed the diagnosis. Postoperative physiotherapy was initiated, and at the one-month follow-up, the patient reported being pain-free. This case highlights the rarity of exostosis of the ulna with associated developmental deformity, emphasizing the importance of a comprehensive diagnostic approach. Early surgical intervention resulted in a successful outcome, underscoring the significance of timely management in improving patient outcomes and quality of life.

Case presentation A 15-year-old female sought evaluation at the orthopedic outpatient clinic, reporting a two-year duration of pain and swelling in her left forearm. The patient recounted a traumatic incident from eight years prior, resulting in a supracondylar humerus fracture. The fracture had been treated with an above-elbow cast for three weeks. Upon examination, a firm, non-fluctuant, non-tender, and immobile swelling measuring 3 cm x 2 cm x 2 cm was identified (Figure 1 ). The swelling demonstrated a close association with the underlying bone and presented with normal overlying skin, showing no signs of neurovascular compromise. The left elbow displayed a 20-degree cubitus varus deformity, accompanied by noticeable shortening of both the radius and ulna in the left forearm. Radiographs, encompassing anterior-posterior and lateral views, unveiled a distinct external bony protuberance over the shaft of the ulna (Figure 2 ). Notably, the radial styloid was absent, and the radius shaft was bowing. CT imaging provided further clarity, revealing an isolated bony protuberance on the proximal ulna. This protuberance communicated with the medullary cavity without involving the surrounding soft tissue (Figure 3 ). A preliminary diagnosis of osteochondroma was established following clinical observations and imaging studies. The laboratory value of the bone is described in Table 1 . Hematological investigations yielded results within normal limits, and informed consent was obtained from the patient. Subsequently, the patient underwent extraperiosteal en bloc resection of the lesion, performed under supraclavicular nerve block anesthesia. Histopathological examination of the excised specimen conclusively confirmed the diagnosis as osteochondroma. Postoperatively, radiological X-rays were obtained (Figures 4 , 5 ). Initiation of postoperative physiotherapy exercises occurred on the first day, and suture removal was carried out on the 12th day. During the one-month follow-up, the patient reported being free of pain. This case underscores the rarity of ulnar exostosis with a concurrent developmental deformity of the forearm and highlights the importance of timely diagnosis and appropriate management for achieving optimal outcomes. Forearm deformities occur in 30% of patients with hereditary multiple exostoses, leading to radial head dislocation and loss of movement [ 1 ]. The early surgical intervention depicted in this case not only alleviated symptoms but also improved the overall quality of life for the patient.

CC BY

no

2024-01-16 23:41:58

Cureus.; 15(12):e50528

oa_package/5c/56/PMC10788317.tar.gz

PMC10788318

38221544

Introduction Technological advances in the orthodontic practice led clear aligner treatment (CAT) meet the demand for a minimally visible and customized treatment modality; however, it is still one of the main topics of scientific research of how to maximize treatment efficiency [ 1 , 2 ]. Orthodontic treatment without the use of bands, brackets, and wires was first described in 1945 by Kesling [ 3 ] who used a flexible tooth positioning appliance to move the teeth. Following his philosophy, the Invisalign® system (Align Technology Inc, Santa Clara, CA, USA) took it further by using computer-aided design and computer-aided manufacturing (CAD-CAM) technology to produce a series of transparent and removable appliances [ 4 , 5 ]. This new treatment approach offered the advantage of improved aesthetics, increased patient comfort, and better oral hygiene when compared to the conventional fixed mechanics [ 1 , 6 , 7 ]. In this system, digital scans are converted into virtual models via stereolithographic (STL) technology and processed with the ClinCheckTM software (Align Technology Inc, Santa Clara, CA, USA) to simulate virtual tooth movements and to plan where, when and how much interproximal reduction (IPR) is needed [ 4 , 8 , 9 ]. During the process, IPR is planned according to the clinical case requirements such as the amount of crowding, Bolton excess, molar and canine relationships, and overjet to achieve well-aligned teeth with optimal interproximal contacts [ 1 , 4 ]. The success of CAT relies on various factors. These are patient-related factors such as compliance to the appliances, bone turn-over rate, and crown and root morphology of the teeth, operator-related factors such as accurate execution of the planned IPR, and an appropriate and realistic treatment plan, and mechanical factors such as the shape and position of the attachments, and material and thickness of the clear aligners [ 10 – 14 ]. Furthermore, the malocclusion, tooth movements required to address it, and the type of tooth to be moved are also influential on the final result. IPR is used in the orthodontic practice to reduce the mesiodistal width of a tooth in order to resolve crowding and eliminate Bolton discrepancy, to treat black triangles by reshaping and approximating neighboring tooth, and, in the case of CAT, to provide teeth the space to carry out planned tooth movements [ 15 – 20 ]. Clinically, the most preferred IPR techniques include thin diamond burs or diamond-coated discs used with a handpiece, and hand- or motor-operated abrasive metal strips [ 15 , 17 ]. Diamond-coated metal strips adapt easily to the proximal contours of the teeth and bend without deformation owing to its flexible nature while providing optimum tactile control and protection for the lips and cheeks [ 21 ]. Although metal strip systems are claimed to offer more precise IPR, they require longer chair-time when compared to the motor-operated systems. On the other hand, motor-operated oscillating segmental discs are designed to be one-sixth the size (60°) of a standard disc and remove the enamel by making oscillating movements with a pivot angle of 30°. Like metal strips, oscillating segmental discs are also shown to protect soft-tissues which eliminates the need for lip and cheek protectors [ 22 ]. Precise execution of IPR is considered to be crucial for proper fitting of the aligners and complete realization of digitally planned tooth movements in CAT. However, there is a lack of reliable evidence on the accuracy of different IPR methods. Therefore, the aims of this study were to compare the consistency between planned IPR and executed IPR during CAT using 3 different IPR methods; hand-operated diamond strips, motor-driven 3/4 oscillating segmental discs, and motor-driven abrasive strips, and to evaluate patient perception of discomfort and anxiety with these methods. The null hypothesis was that there is no statistically significant difference between the planned and executed amounts of IPR, regardless of the method.

Materials and methods This prospective clinical study was approved by ...Başkent University Institutional Review Board and Ethics Committee (Project no: D-KA 21/13) and supported by ...Başkent University Research Fund. Inclusion criteria were (1) patients receiving Lite, Moderate or Comprehensive Invisalign® treatment packages (Align Technologies Inc, San Jose, CA, USA) with IPR prescription, (2) patients presenting mild to moderate crowding, and (3) full permanent dentition without impacted, missing or supernumerary teeth. Exclusion criteria were patients who (1) were candidates for extraction treatment, (2) received any dental procedure during the treatment which altered the mesiodistal width of the teeth other than IPR, (3) presented active periodontal disease, and (4) had undergone orthodontic treatment previously. Sample size calculation performed with 80% power to detect an outcome of 0.5 mm of difference between the planned and the executed IPR per arch, with a significance level of 0.05 and 10% potential drop-out suggested that 42 patients should be included in the study [ 1 ]. Patients were randomly assigned to one of the 3 IPR groups in the order of starting treatment. Patients in group 1 ( n = 14, 150 teeth) received IPR with hand-operated abrasive strips (ContacEZ, Ortho Classic®, Vancouver, WA, USA), patients in group 2 ( n = 14, 134 teeth) received IPR with motor-driven 3/4 oscillating segmental discs (KOMET, Sterisafe® A6, Rock Hill, SC, USA), and patients in group 3 ( n = 14, 133 teeth) received IPR with motor-driven abrasive strips (SWISS, Orthofile, Swiss Dentacare®, Switzerland) (Fig. 1 ). IPRs were planned to be performed when contacts were easily accessible using the ClinCheckTM software, and all IPRs were performed by the same experienced orthodontist (...A.A.Ö.) in a single center. The amount of crowding at the beginning of treatment and linear displacement between the contact points of the anterior teeth (Little’s Irregularity Index (LII)) at the session of the first IPR were calculated (MeshLab ver. 2022.02, ISTI-CNR, Rome, Italy) [ 23 ]. A metal interproximal gauge (KOMET, Rock Hill, SC, USA) was used to quantify the amount of stripping by entering the IPR site parallel to the long axes of the neighboring teeth without exerting pressure and pushing the teeth away from each other. Scans were taken at the beginning of treatment (T0) and at the end of first set of aligners (T1) using an intraoral scanner (iTero Element 5D, Align Technologies Inc, San Jose, CA, USA). Prior to the scans, patients were asked to brush their teeth if dental plaque was evident and teeth were dried thoroughly to remove any kind of remnants which may lead to false readings. Mesiodistal width of the teeth was noted using the arch length-tooth size discrepancy table (Bolton function) of the ClinCheckTM software at T0 and T1. It was assumed that prescribed IPR was carried out equally on either neighboring tooth. The difference between T1 and T0 gave the exact amount of executed IPR. Accuracy of IPR was assessed by calculating the absolute mean difference between planned and executed IPR, where absolute values prevented losing data. The reliability of the arch length-tooth size discrepancy table of the ClinCheckTM software was validated by using T0 and T1 widths of the teeth that were not subjected to IPR ( n = 315). As these teeth were kept intact throughout the treatment, both measurements were expected to be the same and ICC value equal to 1. Patients were asked to complete an anonymous questionnaire, on the session of the first IPR, to evaluate discomfort and anxiety levels induced by the IPR methods. The questionnaire statements were taken from validated questionnaire studies written in the native language of the patients [ 24 , 25 ]. The finalized questionnaire consisted of 17 questions, 2 for demographic data, and 15 for the evaluation of discomfort and anxiety ( Supplementary material ). Of these 15 questions, 3 were yes-or-no questions, and 12 were rating scale questions on a 100-mm visual analog scale (VAS). Cutoff points for rating scale questions were determined as follows; none (0–4), mild (5–44), moderate (45–74), and severe (75–100) [ 26 ]. Questionnaire statements were validated on a representative sample of 20 before commencement of the study. Statistical analysis Statistical analyses were performed using SPSS software package (version 22; IBM, Armonk, NY, USA). Shapiro–Wilk test showed that the data was non-normally distributed. Mann–Whitney U , Kruskal–Wallis H , Wilcoxon signed rank, and Chi-squared tests were used for comparisons of the data derived from the clinical step of the study, and Kruskal–Wallis H and Spearman’s signed rank tests were used for the results of the questionnaire.

Results A total of 42 patients and 417 teeth were included in the data analysis. The ICC value calculated for the reliability of the arch length-tooth size discrepancy table of the ClinCheckTM software was found to be 0.996 with good repeatability. The mean difference between T0 and T1 readings was − 0.09 mm (median, − 0.07 mm). The Cronbach’s alpha was 0.83 for the questionnaire statements. Demographic characteristics, distribution of the malocclusion type, the amount of crowding at the beginning of treatment, and LII values at the session of the first IPR were found similar between the groups (Table 1 ). Table 2 shows the comparison between planned and executed IPR within the groups. The overall average value of executed IPR was significantly less than the planned amount in group 1 ( p = 0.003), yet similar in groups 2 ( p = 0.511) and 3 ( p = 0.659). On quadrant basis, executed IPR in the upper left quadrant in group 1 was significantly less than the planned amount ( p = 0.036); however, it was significantly more in the upper right quadrant in group 2 ( p = 0.021). Table 3 presents the comparison of accuracy of the IPR methods by means of the average difference between planned and executed IPR. According to this, the accuracy of the IPR method in group 1 was significantly low for teeth 11, 21, 32, 33, and 43 ( p = 0.021). Table 4 presents the accuracy of IPR between the arches (upper/lower), sides (left/right), and groups of teeth (incisors/canines/premolars), where no significant difference was evident. The results of the questionnaire showed that patients’ discomfort and anxiety levels were similar between the groups regardless of the IPR method; however, scores in group 1 were slightly higher than the other 2 groups in general. Patients who were priorly informed about the procedure were less likely to make research (Table 5 ). Moreover, correlation tests revealed that a negative correlation existed between age, and the severities of pain ( r = − 0.548, p = 0.042) and tooth sensitivity ( r = − 0.540, p = 0.046).

Discussion IPR is an essential component of CAT which creates room for teeth to perform digitally planned movements and has to be executed precisely in order to achieve the desired outcome. Therefore, the primary aim of this study was to investigate the consistency between planned and executed IPR for 3 IPR methods including hand-operated abrasive diamond strip system, motor-driven oscillating segmental discs, and motor-driven abrasive strips, in patients receiving CAT with the Invisalign® system. The secondary aim was to assess discomfort and anxiety levels induced by these methods. This study stands out among the others in the literature with regard to its in vivo nature and prospective design which made standardization of the interventions possible. Furthermore, all IPRs were executed by an experienced orthodontist to minimize the risk of uncontrolled IPR, especially with the motor-driven methods that would lead to over-stripping and distortion of the contact surface morphology. This is also the first study to assess patients’ perceptions and experiences with different IPR methods during CAT. The null hypothesis is rejected. The overall executed IPR was less than the planned amount with the hand-operated abrasive strip system (ContacEZ), yet similar with the motor-driven systems (KOMET and SWISS), even though the amount of contact displacement was similar between the groups prior to the first IPR. These findings are in line with De Felice et al.’s [ 1 ] who showed that manual strip system fell short to fully execute the planned IPR. Although this system is claimed to provide a more reliable and precise stripping process due to the flexibility of the strips, they have to be used incrementally from thinner to thicker strips, and the clinician needs to enter the same region iteratively with a significant pressure [ 8 , 16 , 17 ]. This may lead to displacement of the tooth in its socket and lead to a false reading on the interproximal gauge. Furthermore, it is clinically more tiring and time consuming, especially when a marked amount of IPR is planned, which may give the clinician a false impression that the targeted amount is reached. On the contrary, motor-driven IPR methods are shown to provide more effective stripping than hand-operated methods by means of speed and the need for less amount of pressure entering the contact points [ 27 – 29 ]. Executed IPR was significantly deficient on the mandibular canines with the hand-operated abrasive strip system, which is in line with the findings of Kalemaj and Levrini [ 16 ], because canines are in tight contact with the adjacent teeth and are frequently pushed out of the dental arch in the case of crowding, which makes it challenging to precisely implement the planned IPR [ 30 ]. Furthermore, mandibular teeth are usually in a more crowded state than the maxillary teeth, which is supported by the findings of the present study showing that IPR was dominantly carried out on the mandibular arch (61.6%) when compared to the maxillary arch (38.3%) that was also documented by Hariharan et al. [ 31 ]. Executed IPR exceeded the planned amount on the upper right quadrant with motor-driven oscillating segmental discs and could not reach the planned amount on the upper left quadrant with the hand-operated abrasive strip system; however, the overall amount of executed IPR was similar between upper and lower arches, as well as left and right quadrants when the sample was evaluated regardless of the IPR system. Johner et al. [ 14 ] evaluated the accuracy of 3 different IPR methods (hand-operated abrasive strips, motor-driven oscillating discs, and motor-driven abrasive strips) on extracted premolar teeth with an in vitro study design. They showed that the amount of IPR was generally less than expected in all 3 groups. Although their results are in line with the abrasive strip group in the present study, their motor-driven IPR groups present conflicting results with ours. These differences seem to arise from the study designs, one being conducted in an actual biological setting with teeth surrounded by its periodontal ligament, differing in crown morphology and in a crowded state, and the other in an artificial setting with only premolar teeth mounted in silicone to mimic the periodontal ligament. Based on the findings of this study, motor-driven methods have proven to be more effective when compared to the hand-operated ones by means of precision, speed, and patient comfort. However, if the clinician favors a hand-operated method, it may be advised to perform slightly more IPR especially on mandibular canines and maxillary central incisors. Also, a brief preliminary alignment phase grants easy access to the contact surfaces which is believed to increase the accuracy of IPR with less damage to the tooth shape and contact surface morphology. Overall, a precisely executed IPR in clear aligner treatment is the way to complete realization of the planned tooth movements and ultimately to a successful treatment. According to the results of the questionnaire, patients were mildly anxious about IPR before the procedure, and their anxiety levels decreased after. A negative correlation existed between age and anxiety and discomfort levels. On the other hand, patients reported similar levels of pain and discomfort with both hand-operated and motor-driven IPR methods, which was unexpected because motor-driven IPR is thought to be more perturbative for the patients. Considering that motor-driven methods provide more accurate IPR, and that patients’ anxiety and discomfort levels are similar with the hand-operated method, motor-driven methods can be preferred over hand-operated methods. Today, websites and social media platforms are frequently referred by the patients for healthcare research. However, the lack of regulations and the fact that any individual can upload a content pose the risk of misinformation [ 32 ]. Results of the questionnaire showed that patients who were priorly informed about the IPR procedure tended to make less research. This means that patients can be kept away from the information pollution on the internet and the treatment can be led by the orthodontist in a more reliable environment.

Conclusions The overall results of this in vivo study showed a discrepancy between planned and executed IPR with the hand-operated abrasive strip system, tending to produce less enamel reduction than planned. The hand-operated abrasive strips fell short to fully realize the planned amount of IPR, especially on the maxillary central incisors and mandibular canines. The consistency of IPR with motor-driven oscillating segmental discs and motor-driven abrasive strips was high. Although the difference between the hand-operated and motor-driven methods was statistically significant, it may not be clinically relevant. According to the results of the questionnaire, both motor-driven and hand-operated methods caused similar levels of discomfort.

Objectives To comparatively assess 3 interproximal reduction (IPR) methods used in clear aligner treatment with regard to accuracy, and patient perception of discomfort and anxiety. Materials and methods A total of 42 patients, treated with the Invisalign® system, were included in this prospective trial and received one of the following IPR methods: hand-operated abrasive strips (group 1; 14 patients, 150 teeth), motor-driven 3/4 oscillating segmental discs (group 2; 14 patients, 134 teeth), or motor-driven abrasive strips (group 3; 14 patients, 133 teeth). Accuracy was evaluated using the difference between planned and executed IPR. Anxiety and discomfort levels experienced by the patients were evaluated using a questionnaire of 17 questions. Results The accuracy of IPR was high in groups 2 and 3; however, it was low in group 1 with the executed IPR significantly less than the planned amount. On quadrant-level, executed IPR was significantly less in the upper left quadrant in group 1, and significantly more in the upper right quadrant in group 2. The difference between planned IPR and executed IPR was significant for teeth 11, 21, 32, 33, and 43 in group 1, indicating deficiency. The average difference between planned IPR and executed IPR was 0.08 mm for group 1, 0.09 mm for group 2, and 0.1 mm for group 3. Anxiety and discomfort levels did not differ between the methods, but a negative correlation was observed between age and discomfort and anxiety levels. Conclusions The overall accuracy of the 2 motor-driven IPR methods was found to be better than the hand-operated system. Maxillary central incisors and mandibular canines were more prone to IPR deficiency when hand-operated abrasive strips were utilized. Patients were similarly comfortable with all 3 methods, and discomfort and anxiety levels decreased with age. Clinical relevance Motor-driven methods have proven to be more effective when compared to the hand-operated ones by means of precision, speed, and patient comfort. If the clinician favors a hand-operated method, it may be advised to perform slightly more IPR especially on mandibular canines and maxillary central incisors. Supplementary Information The online version contains supplementary material available at 10.1007/s00784-024-05499-4. Keywords Open access funding provided by the Scientific and Technological Research Council of Türkiye (TÜBİTAK).

Limitations Although the arch length-tooth size discrepancy table of the ClinCheckTM software is claimed to be prone to measurement errors, its repeatability was found to be very good which shows that the quantitative uncertainty was very low. There is some margin of error in all measurement methods (manual or digital), but the system can be accepted as reliable as long as this does not significantly alter the expected effect [ 16 , 33 , 34 ]. Blindness, on the other hand, is another favorable feature of the software. The demonstrated differences between the IPR methods are statistically significant; however, they may not be indicative of clinical significance. Supplementary Information Below is the link to the electronic supplementary material.

Author contribution P.G.E. contributed to the concept of this study, determined the methodology, explored the sources, performed the data curation and wrote the original text. A.A.Ö., A.A.K., and N.İ.T. contributed to the concept design of this study, determined the methodology, reviewed and edited the original text, and supervised. Funding Open access funding provided by the Scientific and Technological Research Council of Türkiye (TÜBİTAK). This work was supported by the University Research Fund (Project no: D-KA 21/13). Data availability Data availability is upon request to the corresponding author. Declarations Ethical approval This prospective clinical study was approved by ...Başkent University Institutional Review Board and Ethics Committee (Project no: D-KA 21/13). Consent to participate Written informed consent was obtained from all individual participants included in the study. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:58

Clin Oral Investig. 2024 Jan 15; 28(1):95

oa_package/94/b9/PMC10788318.tar.gz

PMC10788319

0

Introduction Over the past decades, scholars have dissected the manifold challenges of Artificial Intelligence (AI) in the health sector [ 1 , 2 ]. The attention has been often drawn to possible misuses or overuses of technology that entail problems of privacy and security, fairness and equity, data quality and data aggregation, bias and transparency [ 3 , 4 ]. These challenges clearly affect a pillar of the so-called international Bill of Rights, namely, the moral and legal right to health, as enshrined in Art. 25 of the Universal Declaration (UDHR) from 1948, and Art. 12 of the International Covenant on Economic, Social, and Cultural Rights (ICESCR). Scholars have accordingly stressed, “the need to adapt current evidence-based standards, to issues of privacy, oversight, accountability and public trust as well as national and international data governance and management” [ 5 ]. Whereas private companies, research institutions and national and international organizations with their public agencies have increasingly issued principles and guidelines for the use of AI in the health sector, lawmakers have been active too [ 6 – 8 ]. In April 2021, for example, the European Commission issued a proposal for a new Artificial Intelligence Act in EU law. Most of the medical devices, AI systems, and robotic applications under scrutiny in this paper should indeed be understood as ‘high-risk’ AI systems pursuant to the EU legislators [ 9 ]. In addition to threats and menaces of AI in the health sector that depend on misuses or overuses of the technology, the attention should be drawn to possible underuses of AI [ 10 ]. This is the subject of this paper. The claim is that the whole set of benefits and promises of AI can be missed or exploited far below its full potential in the health sector. For example, according to a press release of the European Parliament in September 2020, “underuse could derive from public and business’ mistrust in AI, poor infrastructure, lack of initiative, low investments, or, since AI’s machine learning is dependent on data, from fragmented digital markets [ 11 ]. Such drivers of technological underuse do not only regard easy replacements in current medical usages, but also new ones. Work in the field of Technology Use Theory provides the conceptual framework to dissect the “major threat” of AI underuse denounced by the European Parliament in its press release. Against this framework, we shall determine how much it costs to our societies not to use AI systems for health due to the wrong reasons. The remainder is structured as follows: First, Section 2 elaborates the theory of technological uses, followed by a discussion on the price for not using AI in health and a section detailing on current AI applications in health care. Section 5 addresses legal perspectives and other regulatory systems. Success stories are provided in Section 6 with insights into trends, followed by policy shortcomings and further to new recommendations for legal actors. The conclusion summarizes the limits but also lists recommendations for the future.

Conclusions This paper has mostly explored terra incognita, examining the underuse of AI for health, what it means and how much it may cost. Attention was drawn to the relevance of an issue that the European Parliament has presented as a ‘major threat’ in 2020. Scholars have increasingly debated the crucial role that AI innovation plays in the health sector, so that, drawing on this previous research, the focus has been on the risk that research and work in trustworthy AI systems for medicine and digital health may pile up in lab centers with no further impact. The analysis has dwelt on why scholars should be attentive to this phenomenon, inspecting the ways in which governments and public agencies have intended to tackle it. Since such governments and public agencies often admit the limits of their initiatives, this paper has examined the causes of this impasse, proposing some ways to ameliorate such initiatives. The exponential rate of AI advancements and the fact that prediction, prevention, and personalization are in the core of AI make risks of AI underuse increasingly threatening in this field [ 105 – 109 ]. Limits and drawbacks of today’s fight against the underuse of AI - also but not only for health - do not simply depend on the legal means employed by national governments and their public agencies, i.e., the soft tools of the law and methods of cooperation, rules of coordination, etc. Rather, the risk that an increasing set of AI systems for medicine and digital health may pile up in (the warehouse of) start-ups and research centers has to do with the lack of such coordination mechanisms and cooperation in several countries and jurisdictions. Even at their possible light, we should admit however that current policies and initiatives against the underuse of AI require time. Consider the modernization of such organizations, such as the FDA, or the WHO, or matters of public and business trust, or distrust, that are time demanding par excellence. Moreover, the development of new legal and technical standards for AI in medicine and health should not be expected to occur overnight. The assessment of misuses or underuses of AI systems, technical standards for connectivity and data processing, further standards of EU law for the class of ‘high-risk’ AI systems employed in medicine and for human healthcare, etc., should rather be understood as a part of the work that has to be done in the future. This medium-term perspective does not preclude short-term recommendations on the ways in which current mechanisms of cooperation, soft law, engagement, and cooperation for the fight against the underuse of AI can be improved. These recommendations regard models of legal governance, standards, evaluation of opportunity costs, and the overall relevance of the problem, i.e., the ‘major threat’ of AI underuse for health. The first recommendation has to do with a lesson learnt from medium- and low-income countries and their success stories. The first step is to clarify the specific threats of AI underuse that should be prioritized in a certain country. The initiatives should be scalable through the modularization of the projects. The model can be extended to high-income countries with no experience of co-regulatory approaches. The modularity and scalability of the approach should help the law tackling the mid-term issue of the modernization of public agencies and organizations. The second recommendation regards the development of new legal standards. They shall include the finetuning of different degrees of legal compliance. The more public agencies and governments (properly) insist on the soft law tools of cooperation and engagement, the less a traditional binary approach is fruitful. We need future work on how to evaluate the success, or unsuccess of their policies, through different degrees of legal compliance, between 0 (compliance) and 1 (non-compliance). The third recommendation brings us back to the evaluation of the opportunity costs that follow the underuse of AI systems in the health sector. We need future work on how much it costs not to use them, according to different classes and services of AI systems, in different jurisdictions, e.g., common law and civil law, and in different cultures and traditions, e.g., the civil law of France or Italy. Empirical research is critical to shed light on these differences and set up corresponding policies. The fourth recommendation regards the urgency of the problem. It depends on the exponential growth and advancements of AI. By considering the flourishment of AI systems for health, the alternative to the use of trustworthy AI systems is not the simple protection of the status quo. Due to advancements of AI techniques and the speed of AI innovation, the alternative is abating the level of protection of today’s rights, exponentially. We should take the underuse of AI and its opportunity costs seriously. The underuse of AI can be grasped as a waste of time, money, resources, and quality of life. This is a human tragedy. No country can afford it.

Purpose This contribution explores the underuse of artificial intelligence (AI) in the health sector, what this means for practice, and how much the underuse can cost. Attention is drawn to the relevance of an issue that the European Parliament has outlined as a "major threat" in 2020. At its heart is the risk that research and development on trusted AI systems for medicine and digital health will pile up in lab centers without generating further practical relevance. Our analysis highlights why researchers, practitioners and especially policymakers, should pay attention to this phenomenon. Methods The paper examines the ways in which governments and public agencies are addressing the underuse of AI. As governments and international organizations often acknowledge the limitations of their own initiatives, the contribution explores the causes of the current issues and suggests ways to improve initiatives for digital health. Results Recommendations address the development of standards, models of regulatory governance, assessment of the opportunity costs of underuse of technology, and the urgency of the problem. Conclusions The exponential pace of AI advances and innovations makes the risks of underuse of AI increasingly threatening. Graphical Abstract Keywords Open access funding provided by Medical University of Graz.

The theory of technological uses In its common sense, underuse corresponds to using less than what might be expected of a resource or technology. Non-use is considered a quasi-synonym of the term. In the field of technology use theory, the underuse of AI is directly linked to the notions of adoption, use and appropriation [ 12 ]. The notion of adoption consists in the acquisition of a technology, that of use refers to the concrete use of the technical object, while that of appropriation implies the technical and cognitive mastery of the tool [ 13 ]. The process of appropriation of a technique therefore corresponds to the transformation of a technology as it is envisaged by its designer into technology, as it is currently used [ 14 ]. Conversely, refusal and resistance to a technology are analyzed as non-use. The non-use of AI systems, factors of under-use, can be explained by external constraints by the specificities (missing trust, scepticism, fear, administrative burden with new innovations, and many more) linked to the clinical non-user (clinical staff, doctor or hospital). In the case of external constraints, the reasons may be economic, budgetary or strategic. The cost of investment in hardware, software, updates, maintenance and the need to have staff trained in robots and AI systems can be seen as an economic and budgetary disadvantage justifying the refusal to acquire innovative yet effective tools. The health care facility may still hesitate in the face of internal financial and budgetary requirements and difficulties in funding innovation within the structure itself. For example, in France (and in other countries where EU purchasing rules apply), the complexity of the rules of the Public Procurement Code applicable to public hospitals is possibly to discouraging the acquisition of CAPEX (capital expenditure) intensive robots and AI systems. It is still possible that AI has not been identified as a strategic priority in the hospital. A 2019 survey of French university hospital centers shows that 24% of them believe that it is not really a priority for their own establishment[ 15 ]. This likely will have changed though in the recent months with the introduction of generative language models like ChatGPT. From a normative viewpoint, the underuse of AI is critical because advancements of technology can be slowed down, or even opposed for the wrong reasons. Professional reluctance, greed of both public and private data keepers, lack of standards and infrastructures, down to public disbelief in times of conspiracy theories are among the main drivers of technological underuse in the health sector [ 16 ]. Non-use of AI systems linked to the specificities of the non-user may find its source in the concerns of doctors who use AI systems as to the legal consequences in the event of damage caused to a patient [ 17 ]. Keep in mind that underuses of technology may also depend on legal regulations. They can either hinder the advancement of technology through strict liability rules for accident control, or through provisions that require over-frequent revision to tackle such progress [ 18 ]. In the EU, the proposal for a new Artificial Intelligence Act and the proposal for an AI Liability Directive of September 2022 create a comprehensive legal framework in which the European Commission considers appropriate to contribute to innovation and development in the field of AI, in particular by enabling professionals to anticipate risks. The non-use related to specificities can also be largely explained by the lack of social acceptability of AI systems by doctors. For example, this is due to the complexity of using the AI system, the fear of the "black box" effect of Deep Learning, difficulties in evaluating and explaining the results provided by the AI system, concerns related to cybersecurity, fear of being replaced by the machine, resistance to change, fear of the dehumanization of the care relationship, and inadequacy of the AI system to the practices, etc. The 2019 survey carried out among French university hospital centers shows the ambiguity of the attitude of hospitals towards AI. While they are only a small majority to generally perceive the arrival of AI as very positive (57%), they overwhelmingly consider that AI is a very important subject for hospitals (81%). In all of these hypotheses related to technological acceptability, non-use then manifests itself either by an ab initio refusal to use an AI tool, or by abandoning or reducing use after having used the AI tool. However, even a dramatic drop can be reversible. For example, an adoption rate of an AI clinical decision support tool may drop dramatically due to an excessive system. By creating a tracking mechanism for monitoring trigger and adoption rates of a clinical decision support tool, one team found that adaptive modifications to the tool based on user feedback reduced trigger rates, thereby decreasing alert fatigue and increasing provider adoption of the tool [ 19 ]. Significantly, the acceptability of AI tools can improve when doctors see the tangible benefits. A study of the factors driving the adoption of a machine learning-based early warning system to detect sepsis proves that clinicians do not see the AI system as a substitute for their clinical judgment, but see themselves as a technology partner, even though they do not necessarily understand how this new tool works [ 20 ]. The underuse of AI in the health sector The ‘underuse of technology’ triggers that which economists call ‘opportunity costs’ [ 21 ], which in this context is the difference in productivity and quality gains of an AI supported health system and one that does not use or not use AI enough minus the cost for purchasing and maintaining the technology. The underuse entails lower standards in products and services, the redundancy or inefficiency of such products and services, down to the ‘shadow prices’ of the economy [ 22 ]. Work on national health services and their cost analysis estimated that the opportunity costs of ambulatory medical care in the U.S.A. are around 15%: “For every dollar spent in visit reimbursement, an additional 15 cents were spent in opportunity costs” [ 23 ]. In the U.K., the opportunity costs of the National Health Service may amount around 10 million pounds each year, whereas such figures could even underestimate the phenomenon [ 24 ]. Opportunity costs can be evaluated through thresholds for cost-effectiveness analysis [ 25 ], development of value frameworks for funding decisions [ 26 , 27 ], etc. However, work on the opportunity costs that follow the underuse of AI systems for medicine and healthcare is in its infancy [ 9 , 16 ]. The novelty of the issue, the difficulty of the task, the invisibility of the phenomenon, or the fact that scholars simply overlook the ‘major threat’ of AI underuse with its opportunity costs may explain the current state-of-the-art. Accordingly, to appreciate how much it costs the underuse of AI in the health sector, the focus must be on the whole set of AI systems for diagnostics and prevention, precision medicine and medical research, clinical decision-making and mobile health, healthcare management and service delivery, down to AI applications for personalized health care. We would not discuss any underuse of AI for health if a panoply of such AI systems were not available out there. How to specifically address the laggards and sceptics? It is very important to tackle AI implementations for example in a hospital as a group decision with stakeholders from nursing, administration, technical support, computer science, clinical staff, and also patient support. The first step in this direction concerns the digitization of data, starting from medical records up to the last possible complaint. Such digitization is not only indispensable for the use of AI systems in medicine and the health sector, but can effectively support functions and reductions of administrative burden as well as the reduction of costs that depend on traditional healthcare systems mainly hinging on paper records [ 28 ]. Several AI applications may help national health systems to dramatically decrease waiting times for specialist exams, unnecessary travels between home and hospital facilities, or people going to the emergency room of hospitals for unnecessary reasons. To specifically address the laggards and sceptics, it is worth mentioning that many of such AI systems regard administrative applications in healthcare. There are Robotic Process Automation (RPA) systems for medical records and revenue cycle management, clinical documentation, or claims processing [ 29 ]. Likewise, Natural Language Processing (NLP)-based systems can simplify such transactions as making appointments, or refilling prescriptions [ 30 ]. Other AI systems, such as Decision Support System (DSS) platforms embedding medical and AI algorithms can help decongest hospitals, by providing telemedicine services for homecare assistance [ 31 ]. The same holds true in the field of medical diagnostic investigation, in which AI systems can offer high quality services at low cost for preventive treatment services [ 32 ]. All in all, what all these examples show is that relatively easy installations can provide sizable and noticeable results, without posing any particular ethical dilemma or legal challenge. In addition to administrative applications, are there any further ‘low-hanging fruits’ for the use of AI systems in medicine and the health sector? A World of affordances There is almost no field of medicine and health that is not affected by advancements of AI technologies. In the fields of diagnostics and prevention, for example, deep learning techniques have been employed to identify breast cancer [ 33 ], or sight-threatening retinal diseases [ 34 ], to predict severe sepsis [ 35 ], patients with COVID-19 [ 36 ], complications in intensive care units [ 37 ], or which populations are at risk for particular diseases [ 38 ]; in echocardiography, AI is used to diagnose coronary artery problems through people’s heartbeat [ 39 ], or to predict the incidence of heart failure in asymptomatic people [ 40 ]; in neurology, to predict and prevent cases of psychosis, either analyzing the patient’s linguistic and expressive behavior, or controlling her symptoms [ 41 ]. Further big data-driven applications of AI aim to warn specialists about high-risk conditions, such as cardiac arrest or infections [ 42 ], or populations that tend to require readmission to hospitals [ 43 ]. A whole set of its own regards AI systems for precision medicine and medical research. Machine learning techniques have been proven to be effective in predicting which treatment protocols are more likely to succeed based on the context in which the treatment must take place and the characteristics of the patients [ 44 ]. Likewise, AI systems have been employed for the detection of tumors and potential health risks detectable from medical record data and DNA analysis [ 38 ]. In China, a feasibility study tested the potential of AI systems in cancer research to define the most effective treatment for each patient [ 45 ]. In Africa, the SophIA project has used AI systems for the analysis of genomic data, to identify genetic mutations that can cause diseases and, hence, establish the best therapy [ 46 ]. In oncology, AI techniques for massive data mining have made it possible to explore the biomedical scientific literature in order to identify among the millions of studies’ implicit links that cannot be detected by a human [ 47 ]. Of course, algorithms do not only help in the diagnosis of the case but can (advise to) make decisions based on the identification of the problem. The collection of data from patient records and the corresponding clinical evaluation of the AI system allow some AI systems to make predictions in real time, providing all links and sources of information for appropriate recommendations. For example, some researchers at the University of Barcelona, Spain, have devised an AI predictor: on the basis of data taken directly from medical records in electronic format, the AI systems identify which hematological patients with neutropenic fever could have infections deriving from resistant bacilli to antibiotic treatments, the so-called MDR-GNB infections [ 48 ]. In France, a team of researchers has created a scoring algorithm capable of analyzing a CT scan of the lungs and five biological parameters to precisely calculate a severity score allowing the patient to be classified according to the probable evolution of their health, their risk of transfer in intensive care and the need for respiratory assistance [ 49 ]. The use of AI systems in medical diagnostic investigations and clinical decision-making often makes it possible to offer high quality services at low costs or in any case proportionate to the accuracy of the treatment services for prevention [ 32 ]. Further AI applications have been developed for healthcare management and service delivery [ 30 ], up to AI applications for personalized health care and mobile health. There are devices that, processing the information on the patient’s symptoms, can provide diagnoses [ 50 ]; suggest healthier lifestyles [ 51 ]; or control the use of medicines prescribed to subjects [ 52 ], as in cases of tuberculosis [ 53 ], mental illness [ 54 ], etc. These systems can thus provide personalized assessment about the health status of the user, consumer or patient, potentially reducing the demand for assistance from medical or healthcare personnel, relieving the pressure or tasks that often fall on the families, by offering forms of assistance that are effective, but at low cost, even in countries with few resources. Admittedly, the collection and processing of sensitive data by means of AI applications for mobile health trigger some of the issues mentioned above in the introduction with the protection of privacy and data protection, confidentiality and transparency of the data processing, consent and non-discrimination. In addition, the use of AI systems in the healthcare ecosystem often raises the difficulty of integrating such systems into the organization and workflows of the sector with its doctors, nurses, patients, etc. Remarkably, there is a hot debate on possible losses of jobs due to the replacement of personnel with various AI devices and systems [ 55 ]. Researchers may overcome hurdles of controlled, possibly sensitive health data for AI models by the use of synthetic data generation pipelines [ 56 ]. The digitalization of the health sector as well as the prediction, prevention, and personalization of medicine are in any event in the core of AI, e.g., machine learning techniques. Scholars have increasingly stressed the manifold ways in which AI systems can contribute to this paradigm shift in digital health. To previous samples of AI systems for diagnostics and prevention, precision medicine and medical research, clinical decision making and mobile health, healthcare management and service delivery, we may add work on the applicability of predictive diagnostics for the development of 3P-Medicine for clinical application [ 57 ]; AI supported patient self-care systems in chronic heart failure [ 58 ]; non-invasive diagnostic tools in coronary artery disease [ 59 ]; mass spectrometry-based technologies [ 60 ]; predictive diagnostics of dementia [ 61 ]; different stages of cellulite [ 62 ], and so forth. Against this framework, we should question about how international organizations, national governments, and their public agencies aim to exploit the whole set of affordances and opportunities brought forth by AI for health, while protecting human rights and interests of all parties. In addition to declarations of principles and ethical guidelines, the focus of scholars has mostly been on acts and statutes against misuses and overuses of technology. For example, in EU law, the forthcoming Artificial Intelligence Act, the general data protection regulation, or GDPR, the machinery safety and cybersecurity regulations, the medical device regulation, etc. These normative acts represent a necessary, but insufficient element of the analysis. Such investigation should be complemented with the initiatives that have been taken by public agencies, governments, and organizations to tackle possible underuses of AI in the health sector with the soft tools of the law. The problem of how to implement the number of AI systems under scrutiny in this section can hardly be addressed on the basis of the top-down commands of hard law enforced through the threat of physical and/or pecuniary sanctions. The drivers of technological underuse, e.g., public distrust of the technology, cannot be simply tackled by legal order, or decree. What is then the current state of the legal art? Examples of uses of AI that have not been scaled up Artificial intelligence (AI) has made significant strides in revolutionizing various facets of the healthcare industry. Nonetheless, despite the considerable advancements and widespread implementation of AI technologies in medicine, there remain several areas where its deployment is constrained or underutilized. An area that has yet to fully leverage AI capabilities is the development and integration of intelligent clinical decision support systems (CDSS). CDSS employs AI algorithms and machine learning techniques to analyze patient data, encompassing medical records, laboratory results, and imaging data. Although specific CDSS have been developed for certain medical conditions, their integration into routine clinical practice on a large scale is still to be accomplished. The scalability of these systems is impeded by challenges such as data interoperability, limited integration with electronic health records (EHRs), and concerns pertaining to liability and accountability [ 63 ]. The domain of drug discovery and development also represents an area where AI’s full potential has not yet been realized. Traditional drug discovery processes are costly and time-consuming, often entailing years of research and clinical trials. AI algorithms have demonstrated promise in expediting and optimizing various stages of this process. However, the broad application of AI in drug discovery and development encounters obstacles related to data availability, regulatory approval processes, and the imperative for interdisciplinary collaboration. By harnessing AI technologies, researchers can identify potential drug candidates, forecast their efficacy and safety profiles, and refine drug formulations [ 64 ]. Precision medicine aims to customize medical interventions based on individual patients’ distinctive genetic, environmental, and lifestyle factors. AI-based predictive analytics holds immense potential in actualizing the vision of precision medicine by discerning patterns and correlations within extensive datasets. However, the integration of AI into routine clinical practice for precision medicine remains limited. Challenges such as concerns regarding data privacy, the absence of standardized guidelines, and restricted access to advanced computational resources impede the scalability of AI-driven predictive analytics. Nevertheless, as increasingly comprehensive genomic and clinical datasets become available, coupled with advancements in AI algorithms, the feasibility of precision medicine applications, including disease risk prediction, treatment optimization, and identification of novel therapeutic targets, becomes increasingly viable [ 65 ]. The COVID-19 pandemic has underscored the significance of remote monitoring and telehealth technologies in delivering healthcare services. AI can play a pivotal role in remote monitoring by analyzing patient-generated data, such as wearable sensor data, physiological signals, and patient-reported outcomes, to identify anomalies and provide real-time alerts to healthcare providers. Despite the accelerated adoption of telehealth during the pandemic, the scalability of AI-driven remote monitoring systems is still limited due to challenges associated with data security, regulatory compliance, and reimbursement policies [ 66 ]. The fight against AI underuse The challenges of technological underuse - in particular, the underuse of AI systems for health - have recommended lawmakers to complement the hard tools of the law with the means of soft law, such as policies, guidelines, recommendations, and opinions of public agencies. Rather than a symptom of weakness, the soft tools of the law can be understood as an interface between the top-down instructions of the regulator, i.e., lawmakers and administrative agencies, and the interests of (some or all of) the stakeholders [ 67 ]. The US regulation of software as a medical device (‘SaMD’) illustrates this mix of hard and soft law with the powers of the Federal Drugs Administration [ 16 ]. It is up to the FDA to examine applications, develop policies, publish guidance, or ask for feedback. This mix of soft and hard law is at work also with the EU agency, EMA. The European Medical Agency shall evaluate applications for marketing authorisation, monitor the safety of medicines across their lifecycle, facilitate development and access to medicines, and provide information to healthcare professionals and patients. Much as occurs with further fields of legal regulation vis-à-vis the dynamics of technological innovation, e.g., AI drones in civil aviation law [ 68 ], the soft tools of the law appear particularly fruitful to enforce, strengthen, clarify, or stimulate the adoption of the top-down provisions of the regulator. In certain cases, e.g., development of standards, soft law provides for methods of coordination and cooperation to further define the content of the top-down provisions of the legislator. The engagement with some or all stakeholders with their feedback represents in many cases the only way in which legislators and public agencies can tackle the hurdles of technological innovation. Such stakeholders may include software engineers and computer scientists, patients and clinicians, academia and the industry, such as manufacturers and distributors, up to health technology assessments groups, the media and the public at large. In the US, the FDA has set up patient outreach newsrooms and med watch, training modules and education programs, or networks with experts, e.g., regulatory associations. By interacting with health providers and educators, academia and the market, professionals and patients, the aim is to acquire reviews and contributions to reports, to learn about advancements of science and technology, and to inform stakeholders about policies, or to respond to requests related to the FDA authority [ 69 ]. Likewise, in EU law, the European Commission relaunched the ehealth stakeholder group initiative in July 2020, which refers to “all umbrella organisations/associations with a European outreach, representing the following sectors/groups: the health tech industry, patients, healthcare professionals and the research “community [ 70 ].” The intent is to "support the Commission in the development of actions for the digital transformation of health and care in the EU,” by providing advice and expertise, in particular, as regards the topics set out in the communication on enabling the digital transformation of health and care in the EU Digital Single Market, from April 2018. Such topics comprise some main drivers of AI underuse, such as health data interoperability and record exchange formats for digital health services through AI and “other cross cutting aspects linked to the digital transformation of health and care, such as financing and investment proposals and enabling “technologies [ 70 ].” In May 2022, the European Commission presented the proposal of the European Health Data Space, or EHDS, which covers a part, although important, of this challenge on the design and functioning of e-health record (EHR) systems through a mandatory self-certification scheme for such EHR systems. Further initiatives such as the Medicines and Healthcare Products Regulatory Agency in UK, [ 71 ] the Health Sciences Authority of Singapore, [ 72 ] and the Department of Health of the Australian Government illustrate a similar approach against the underuse of AI. For example, the ‘Stakeholder Engagement Framework’ of the Australian Department of Health, provides for five principles of engagement that should help the law tackling cases of technological underuse through a clear understanding of the aims of the engagement, its inclusiveness, timeliness, transparency, and respectfulness, including expertise, perspectives, and needs of all stakeholders [ 73 ]. Five levels of engagement, that is, from simple information to consultation, involvement, collaboration, and delegation of legal powers to stakeholders, complement traditional top-down approaches of legislation [ 73 ]. Such forms of flexibility through different levels of engagement should allow the regulator to properly address cases of AI underuse that depend either on professional reluctance, or on public distrust and misapprehension, or on the difficulties to insert such AI systems into the organization and workflows of the health sector. Yet, there is another formidable hurdle that legal flexibility and methods of cooperation and collaboration shall address in the fight against technological underuse, namely, public bureaucracy. The more legal systems rely on methods of coordination and cooperation to tackle the challenges of technological innovation, the more such methods of coordination and cooperation demand a new role of public agencies and authorities. Consider the engagement of stakeholders for tackling cases of technological underuse in medicine and health, the feedback that public agencies and authorities should have through such forms of engagement about the current state of the art, or the development of standards and metrics for the assessment of technologies that entails confrontation with developers and private companies. All these cases show that the role of authorities and public guardians is not only to enforce the top-down commands of lawmakers, but rather, to work together with all relevant stakeholders, such as private companies, non-governmental organizations, or the public at large, to find out solutions for an agile implementation of AI systems from labs to society. This approach often means, however, a change of mentality that, as many public agencies admit, can be as difficult as regulating the technology [ 16 ]. Traditional notions of legal compliance make this change of mentality even harder [ 74 ]. The old assumption, according to which either legal agents are compliant, or they are not, hardly fits the challenges of AI underuse and the corresponding policies of engagement, collaboration, and coordination under scrutiny in this section. Rather than 0s and 1s, between compliance and non-compliance, focus should be on more nuanced assessments that distinguish between ideal, sub-ideal and non-compliant statuses of legal agents [ 75 ]; or, between ‘good’, ‘ok’, or ‘bad’ compliance [ 76 ]; down to more fine-grained views that distinguish between average compliance, reasonably high compliance, very high compliance, and full compliance [ 77 ]. The binary alternative of compliance or non-compliance does not provide any useful information for the assessment and improvement of such institutional initiatives, as the soft law of the FDA, the ehealth stakeholder group initiative of the European Commission, or the ‘Stakeholder Engagement Framework’ of the Australian Department of Health. In addition, compared to other regulatory systems in society, such as ethics and social norms, technology and the forces of the market, there is a paradox unique to the law. Legislators and initiatives of the public sector can be the cause of, or the solution for AI underuses in today’s human societies. On the one hand, many drivers of technological underuse do not depend on the law, but the law aims to govern them, on the other hand, the law can hinder technological innovation with its own provisions. Several examples of EU law in the regulation of e-money, drones, and now with the ‘high-risk uses’ of AI illustrate this risk, or impasse [ 18 ]. What is the state-of-the-art in health law? Success stories This paper has stressed the pros and cons of AI in the health sector, its benefits and normative challenges. The threats not only regard misuses and overuses of AI, e.g., infringement of individual and group privacy, but also underuse of AI that depends on lack of infrastructures, connectivity standards, or the proverbial opacity and resistance of bureaucracy. How to strike a fair balance between affordances and constraints of AI is admittedly no easy task. Some success stories luckily illustrate how this balance is however feasible. For example, consider the developmental roadmap and validation of an AI service for health, i.e., CURATE.AI and foundational technology of IDentif.AI. The idea was to implement an AI system for “combination treatments where drugs administered together can interact with each other, which is often the case” [ 78 ]. The project moved from lab to ward, i.e., the National University of Singapore and its hospital, in a striking short amount of time in 2020. The project included engagement with the Medical Devices Branch of the Health Sciences Authority (HSA) in Singapore, that is, the regulator for risk classifications associated with such a device, since the very beginning of the project. For each trial and subsequent discussion for submission, rapid and informative responses and active engagement from HSA regulatory team members resulted in efficient turnaround times for trial initiation. Feedback from HSA ultimately resulted in a positive outcome for a refractory oncology patient [ 78 ]. A sustained track record of engagement with HSA made the whole process smooth, from lab to ward, playing a key role in helping a clear process flow to be developed for downstream guidance requests. Remarkably, some of these success stories come also from low- or medium-income countries: AI systems that predict birth asphyxia in children by scrutinising the birth cry of a child via mobile phones in Nigeria [ 79 ]; AI apps that offer guidance and recommendations to nurses and paramedic personnel in India and sub-Saharan Africa [ 80 ]; or AI that detects water contamination [ 81 ]; or control dengue fever transmission [ 82 ]; or predicts Ebola outbreaks [ 83 ]. Such success stories show how making good through AI systems is feasible in the health sector. Human ingenuity provides for the means to implement in those settings technologies that are already used or have been developed in high-income countries, although within the specificities of each nation under scrutiny. The flexibility of the law that many western institutions have adopted through methods of coordination and cooperation - as illustrated above in the previous Section 5 - can thus address issues of technological underuse in developing countries, fleshing out which specific threats of AI underuse should be prioritized through initiatives that can be scaled up, or down, through the modularization of the projects [ 67 ]. This approach also fits high-income countries that have no experience of co-regulatory approaches, for example, Italy [ 16 ]. AI or respectively its workforce ML in cancer research offers novel perspectives such as ML clustering (with age or other groups) to find novel biomarkers [ 84 ]. In this domain, success stories include the use of ML enabling the largest Brain Tumor Study To-Date [ 85 ], as well as deep patient studies and derivates [ 86 ]. Some attempts indicate that AI-driven language systems can be used to correct misconceptions, to inform patients on cancer [ 87 ], and may be of further use in clinical scenarios [ 88 ]. A further perspective deals with opportunities regarding sustainability of XAI [ 89 ]. Success stories shall inspire models of governance that consider persisting crucial differences between countries and jurisdictions, between cultures and social norms. It is likely that striking differences among medical sectors, e.g., the opportunity costs of radiology vis-à-vis the opportunity costs of research in bacteria, should be expected in, say, Germany, France, or Spain. This conjecture rests, on the one hand, on traditional distinctions among medical sectors: we already stressed that, in medical diagnostic investigation and clinical decision-making, for example, AI systems provide for high quality services at low costs[ 32 ]. On the other hand, we should take into account clear differences among countries and their health services, for example, between the regional health services in Northern and Southern Italy. The result is that no single answer exists for the opportunity costs that follow ‘the’ underuse of AI in the health sector. Such opportunity costs will depend on the fields and types of AI systems under scrutiny, as well as on countries and jurisdictions taken into account. These constraints shall not make us overlook the inspirational source of all success stories and the general trend of technological regulation around the globe with a significant regulatory convergence between several health agencies. From Australia to the EU, Singapore, UK, or USA, public authorities have been setting up mechanisms of coordination, cooperation and co-regulation to cope with the challenges of technological innovation and risks of AI underuse. Some of these coordination mechanisms and methods of cooperation were at work with the success stories of this section. The next step of the analysis is to reflect on the overall achievement of such policies and initiatives against the underuse of AI for health. Policy shortcomings We already illustrated the array of initiatives and policies of public agencies and national regulators that aim to prevent the risk of underusing AI systems for medicine and health. The premise of this paper was that the analysis on the underuse of AI for medicine and health does not revolve around whether such underuse exists, but rather, how much it costs to human societies. To corroborate the assumption, there is no need for personal experiences in hospitals and national health services around the world. Although generalizations must be avoided, according to the warnings of the previous section 6 on success stories, national and international institutions admit policy shortcomings in the fight against AI underuse. This paper already mentioned the 2020 press release of the European Parliament on the ‘threat of AI underuse.’ In the wording of the EU institution, “underuse of AI is considered as a major threat: missed opportunities for the EU could mean poor implementation of major programmes, such as the EU Green Deal, losing competitive advantage towards other parts of the world, economic stagnation and poorer possibilities for people.” Similar threats of AI underuse have been underscored by further international organizations, such as the OECD, or ITU and WHO [ 9 ]. In the Tokyo 2019 AI Principles of the G20, the intergovernmental forum recommended a proactive approach of governments and public institutions to the risks of technological underuse. According to Art. 3.2(a) of the AI Principles, a proactive approach means that “governments should promote a policy environment that supports an agile transition from the research and development stage to the deployment and operation stage for trustworthy AI systems.” Another very important but often hugely underestimated aspect is a technological effect: for certain tasks, algorithms can achieve performance beyond human levels, however, unfortunately, the most powerful AI methods suffer from the fact that, on the one hand, it is difficult to explain why a certain result was achieved. On the other hand, they lack robustness. In plain language, this means that the most powerful AI models are very sensitive to even small changes and perturbations. Thus, small perturbations in the input data can have a dramatic impact on the output and lead to completely different results - which of course can have fatal consequences in medicine and nursing - there is a lack of trustworthiness. This is precisely the reason that has led to an increased demand for trustworthy AI [ 90 ]. The lack of trustworthiness in this kind of technology strengthens the reasons why the affordances of AI systems are often missed due to wrong reasons, such as popular beliefs and mistrust on ‘black boxes’ technologies that encourage professional reluctance, bureaucratic resistance, or simply crazy conspiracy theories [ 91 ]. In the sensitive domain of medicine, traceability, transparency and interpretability are required. Some legal systems have already come a long way, with explainability now even mandatory due to legal requirements - for example, in the European Union [ 92 ]. Arguably, AI systems shall be made more robust and combined with explainability. Combining statistical machine learning with knowledge representations can help make AI models to be more robust. Certain tasks benefit from the inclusion of humans in the loop. Although not necessarily, but they often bring experience, domain knowledge, and conceptual understanding to the AI pipeline [ 92 ]. This may entail a virtuous circle. While including humans in the loop eases the liability burden of the legal system, the ‘why’ in many application areas is often more important than a pure classification result. Explainability and robustness can thus promote reliability and trust, and guarantee that humans remain in control when using AI systems for prediction, prevention, and personalization in medicine. The overall aim should be to complement human intelligence with artificial intelligence, i.e., the mission of the enormously growing field of Trustworthy AI [ 93 ]. The ‘agile transition’ from labs to society faces, however, formidable obstacles. The lower standards in products and services that follow as a result with the redundancy or inefficiency of some of such services have a cost. Consider the hours spent in unnecessary waiting times in hospitals or health facilities with the often-superfluous transport costs for patients, families and assistance staff, etc. We already noted above in the introduction that the evaluation of these costs is no easy task due to traditional hurdles of econometrics as well as the novelty of the phenomenon. Lack of standards and metrics does not only make it difficult to assess the opportunity costs of AI, but also the costs of its overuses or misuses. The footprint of AI, for example, is controversial as regards energy costs, carbon emissions [ 94 ], and the metrics of AI systems are optimized for, e.g., efficiency through model training [ 95 ]. In the health sector, such indexes, as the health-related quality of life (HRQOL), or the quality-adjusted life-year (QALF), enjoy a certain consensus, still, it remains unclear about the amount of the opportunity costs for the underuse of AI, also but not only, for medicine and digital health [ 96 ]. As of this writing, it is noteworthy that such international organizations and institutions, as the OCSE, the G20, or the European Parliament have not released any assessment on the opportunity costs that follow the underuses of AI under scrutiny in their own documents. Work on the AI underuse in the health sector suggests that in some countries, e.g., Italy, which invests around 9% of its gross domestic product (GDP) in the public health sector, the opportunity costs of technological underuse may amount to 1% up to 2% of the Italian GDP [ 16 ]. The figure includes the optimization of services, quality of standards, and the ‘shadow prices’ of people’s useless waiting lists and movements, traffic congestion, pollution, etc. The estimate also includes the costs for the modernization of public agencies and continuing education and formation of the personnel, as well as such trends as the increase in life expectancy, the spread of chronic or endemic diseases, and the greater familiarity that the new generations have with technological devices. AI systems look particularly promising to tackle these (social, legal, demographic, technological, etc.) trends with the full array of applications for medicine and health examined throughout this paper. The empirical assessment of the opportunity costs of AI is critical to determine public policies. We already noted that, especially dealing with medium- or low-income countries, or high-income countries with few or no experience of co-regulatory models, such as Italy, the first step is to specify what threats of AI underuse should be prioritized through initiatives that can be scaled up through the modularization of the projects. Waiting for further work on the opportunity costs of AI in the health sector, we should not overlook, however, the ways in which current initiatives and public policies against the underuse of AI can be ameliorated, thus diminishing the costs of such underuse. This is another territory of work and research on the underuse of AI that legal theory and its corresponding models of legal governance have scarcely explored. The law can be a cause of, or a solution for technological underuses. Scholars have extensively debated new forms of legal governance and regulation for the challenges of AI and other emerging technologies, but rarely on how underuses of AI should be tackled through the means of binary governance [ 97 ], linked democracy [ 98 ], legal experimentation by derogation and by open access [ 99 ], co-regulation [ 100 ], algorithmic regulation [ 101 ], and more. The next Section 8 intends to fill this gap in discussions about the normative challenges of AI and its governance, proposing some recommendations to improve current policies and initiatives against the underuse of AI in the health sector. Empirical research on the underuse of technology and the quantification of its costs should be complemented with policy recommendations as to how to reduce such opportunity costs. New recommendations for an old paradox All the recommendations in this section regard the specificities of soft law at work in complex digital ecosystems. The use of coordination mechanisms and methods of collaboration by governments and public agencies, as examined above in this paper, seems in fact indispensable in the fight against AI underuses. Drivers of technological underuse, such as public distrust and business diffidence, can hardly be addressed only on the basis of the top-down instructions of the regulator. The response of most public health agencies to the threat of AI underuse has thus intended complementing the binding rules of the law with mechanisms of coordination, methods of cooperation, etc. It is notable that, in EU law, specific norms are devoted to the functioning of such coordination mechanisms and the corresponding flexibility of the legal system. For example, in the data protection regulation, the GDPR has aimed to attain such aims of coordination with Articles 5, 60, 61, 75(4) and 97(2)(b); in the Chips Act from 2022, specific coordination mechanisms are set up with Art. 1(1)(c), Art. 14(4), and Art. 23(4), in accordance with the objectives of the Act (i.e., 1.4.2.3, at 63); etc. This mix of soft law and hard law in the governance of complex digital ecosystems determines the extent to which the law may augment or decrease cases of technological underuse. Consider the success stories of this paper. What they have in common is not only the digital dimension of what ended well in all those stories, but rather, how the different actors of such stories exploited and benefited from the opportunities opened up by AI systems interacting on the internet although in an intricate social context. The constraints and affordances of the digital dimension of data-driven technologies, such as the use of AI systems, affect all stakeholders, from patients and clinicians to software engineers and computer scientists, manufacturers and distributors, health technology assessments groups and the public at large. Such impact of technology on the health sector also affects the way in which lawmakers and public agencies have conceived the governance of this sector. As declared repeatedly by the FDA in its 2020s reports, the impact of AI and further emerging technologies makes it necessary modernizing the organization of such agencies, as the FDA, fragmented in watertight scientific review processes [ 16 ]. Initiatives of engagement and coordination have similarly had to be reengineered. As shown by the focus groups set up by both ITU and WHO in their 2020-2022 AI for Health (AI4H) project, [ 102 ] traditional research on bacteria and dental issues, diabetes and endoscopies, radiology and malaria, has been complemented with the new data-driven scenarios of today’s biomedicine and AI that span across all traditional fields of medicine and health: analysis on software life cycle and data requirements, best practices and evaluation considerations, scale up and adoption of AI technologies, down to the assessment of AI applications and platforms. The paper stressed that this digital component of the underuses of AI for medicine and health not only impacts the traditional organization of public agencies, but also the way in which the law should address the challenges of technology. The law should be ‘flexible’ especially when dealing with cases of technological underuse. The analysis has so far cast light on three different ways of legal flexibility, given by the means of soft law, different levels of engagement with stakeholders and methods of cooperation, and legislations that shall not hinder technological innovation nor require over-frequent revision to tackle the progress of technology. What all the variables of the analysis illustrate is, so to speak, the strength of soft law. Although not enforced by the threat of physical and pecuniary sanctions, as occurs with the tools of hard law, soft law can be effective. A certain degree of compliance by the legal actors involved in efforts of cooperation, coordination, etc., thus determines the success of a certain policy, e.g., the current fight against the AI underuse. The binary alternative of compliance or non-compliance, which makes of course sense when dealing with the top-down rules of hard law, does not provide, however, any useful information for the assessment and improvement of current institutional initiatives against the underuse of AI for health. A subtler, ‘more flexible’ approach to legal compliance in cases of AI underuse, also but not only in the health sector, shall provide more nuanced assessments than the traditional stance, according to which either the legal agent is compliant, or not. Rather than 0s and 1s, subtler forms of evaluating the status of adherence to the regulatory provisions of the law are indispensable when using the soft tools of the law, or complementing the top-down commands of lawmakers with more flexible ways of legal coregulation. Such assessment regards also but not only the fight against the AI underuse for medicine and health. Consider the ‘popularity’ of the registration mechanisms for data altruism that have been set up with the Data Governance Act in EU law, as well as the ‘soundness’ of technological solutions for the principle of data protection by design, and by default, in several jurisdictions that endorse the principle [ 103 ]. The binary alternative of compliance or non-compliance does not help us improving current institutional initiatives that hinge on methods of coordination and cooperation since the alternative, or 0, or 1, does not clarify the nuances of such engagement. The principle of legal certainty that fits the hard tools of the law hardly adapts to the evaluation of the tools of soft law. This is particularly true when assessing the current fight against AI underuses for medicine and health. The development of new legal standards, e.g., the assessment of legal compliance in cases of AI underuse, goes of course hand-in-hand with the development of new technical standards in medicine and health [ 104 ]. We noted that many fields lack standards and metrics not only for the assessment of the opportunity costs of AI, but also for the costs of its overuse or misuse. This lack of standards may trigger a vicious circle because the lack of standards is another crucial driver of technological underuse. Therefore, the development of new legal standards and technical standards is not only necessary but should be conceived of as the two sides of the same coin. On the one hand, such standards should provide thresholds for the assessment of the opportunity costs that follow the underuse of AI, and whether and to what extent current policies against the underuse of AI can be deemed as successful; on the other hand, new legal and technical standards for the use of AI systems can dramatically increase the protection of human rights in the health sector. This paper has mentioned manifold AI systems for predictive diagnosis, targeted prevention, or the personalization of medical services, as sound examples of how such uses of AI can enhance “a high level of human health protection” and the “right of access to preventive health care and the right to benefit from medical treatment,” according to the wording of Art. 35 of the European Union’s Charter of Fundamental Rights (CFRs). The aim should not only be to protect the right to health against misuses and overuses of technology, but moreover, to strengthen today’s standards of protection through the increasing use of trustworthy AI systems for health.

Funding Open access funding provided by Medical University of Graz. Funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – 520382567. Parts of this work have also received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement No. 826078 (Feature Cloud). This publication reflects only the authors' view, and the European Commission is not responsible for any use that may be made of the information it contains. Parts of this work have also been funded by the Austrian Science Fund (FWF), Project: P-32554 (Explainable Artificial Intelligence). Availability of data and materials Not applicable. Declarations Ethics approval Not applicable. Consent to participate Not applicable. Consent for publication Not applicable. Conflict of interest The authors declare no conflict of interest. The following abbreviations are used in this manuscript: Artificial Intelligence Capital Expenditure Charter of Fundamental Rights Clinical Decision Support Systems Decision Support System Explainable Artificial Intelligence International Covenant on Economic, Social, and Cultural Rights Machine Learning Natural Language Processing Robotic Process Automation Universal Declaration

CC BY

no

2024-01-16 23:41:58

Health Technol (Berl). 2024 Dec 12; 14(1):1-14

oa_package/d1/71/PMC10788319.tar.gz

PMC10788320

37991561

Introduction CVD is one of the leading causes of death and morbidity worldwide accounting for 34% (20.2 million) of deaths in 2022 with an estimated 500 million active cases [ 1 ]. CVD is associated with increased levels of inflammation, especially in the vascular endothelium. This inflammation may contribute to the progression of CVD and cause myocardial and vascular damage [ 2 ]. Accurate, earlier diagnosis of CVD allows for swifter intervention and treatment which may help lower the global healthcare burden and mortality rates associated with CVD. Biomarkers play a key role in this. Some markers for inflammation such as high sensitivity c-reactive protein (hs-CRP) have previously been studied in the context of CVD [ 3 ]. It has been shown, however, that hs-CRP may underestimate inflammation and therefore have lower predictive power than that required for diagnosis or prediction of CVD onset [ 4 ]. This highlights the need for biomarkers of CVD that have high predictive power and the ability to discriminate between CVD cases and healthy controls. Previous studies have identified N -acetyl neuraminic acid (sialic acid, Neu5Ac) as both a potential biomarker for CVD such as heart failure, and risk of future cardiovascular events such as heart attack and stroke [ 5 ]. Neu5Ac concentrations were shown to be clearly elevated in plasma taken from patients with CVD versus plasma taken from healthy controls. Neu5Ac is a monosaccharide with a nine-carbon backbone that is generally located as the terminating unit of N- and O -glycans. In turn, these glycans form parts of glycoproteins and other glycoconjugates [ 6 ]. Neu5Ac is one of a family of over 50 sialic acids. The most abundant sialic acids present in humans are Neu5Ac and Neu5,9Ac 2 [ 7 ] (Fig. 1 ), with Neu5,9Ac 2 present in quantities around 100–200 times less than Neu5Ac. Neu5Ac is ubiquitous in the body, while Neu5,9Ac 2 has been identified in biological fluids (blood, urine, saliva), in the brain, lungs, kidneys, intestines, and pancreas [ 8 , 9 ]. Neu5Ac has many functions, primarily as a receptor mask or determinant which aids in cell-cell recognition and immune response, generally through the mechanism of receptor binding [ 10 ]. The carboxylic acid functional group conveys an overall negative charge to the cell surface and the endothelium which aids in cell repulsion [ 6 ] and the prevention of cell aggregation, especially among erythrocytes [ 11 ]. Neu5Ac and Neu5,9Ac 2 expression on the surface of glycoproteins improves their circulating half-life in the blood [ 12 ]. These sialic acids have also been shown to have an anti-inflammatory effect. This takes place through various mechanisms such as the reduction of recruitment of leukocytes and the suppression of specific immunogenic proteins such as interleukins [ 13 , 14 ]. This can help to protect cells from damage during an inflammatory state but can also be utilised by cancer cells to protect against the immune system [ 15 – 17 ]. Further to this, inflammation is associated with the acute-phase response, with a marked increase or decrease in specific acute-phase proteins during inflammation [ 18 ]. Acute-phase proteins are generally highly sialylated and as such would contribute to increased sialylation levels in the blood during an inflammatory state, such as that associated with CVD [ 19 ]. Neu5,9Ac 2 concentrations may also appear upregulated due to a potential reduction in plasma esterase activity during an inflammatory state. Hubbard et al. reported significant associations between elevated inflammatory markers hs-CRP, interleukin-6 (IL-6), tumour necrosis factor alpha (TNF-alpha), and reduced plasma esterase activity [ 20 ]. This reduced plasma esterase activity may reduce the conversion of Neu5,9Ac 2 to Neu5Ac hence resulting in elevated concentrations of Neu5,9Ac 2 during an inflammatory state. Previous research in this area has utilised assays which suffer from poor specificity for sialic acid which can lead to inaccurate results [ 21 ]. Quantitative analysis of these sialic acid derivatives as biomarkers for CVD requires a sensitive and specific analytical technique with a sufficiently low limit of detection. This is because Neu5Ac is present in sufficiently large quantities (average 45.49 mg/100 mL in healthy controls), with Neu5,9Ac 2 being present in quantities 100–200 times less (average 0.29 mg/100 mL in healthy controls) [ 22 ]. Ideally, the assay of choice will also analyse total sialic acid and not just sialic acids present on N -glycans. Labelling of sialic acids with 1,2-diamino-4,5-methylenedioxybenzene dihydrochloride (DMB) followed by ultra-high performance liquid chromatography (UHPLC) analysis was chosen in this instance as it allows for the analysis of multiple types of sialic acid from all sources found in plasma in the same assay while exhibiting high specificity for them and a low limit of detection (< 0.01 nmol) and quantitation (< 0.04 nmol) [ 23 , 24 ]. Like all quantitative techniques, the DMB assay requires quantitative standards for effective and accurate quantitation of sialic acids. Neu5Ac is commercially available in sufficiently large quantities but Neu5,9Ac 2 is not, and therefore must be chemically synthesised. For the accurate quantitation of Neu5,9Ac 2 in plasma samples, we have previously synthesised Neu5,9Ac 2 and analysed the standard using quantitative nuclear magnetic resonance (QNMR) techniques [ 22 ]. A small-scale pilot study was designed herein to measure Neu5Ac and Neu5,9Ac 2 concentrations in plasma samples from both healthy controls (n = 30) and CVD cohorts (n = 30). Analysis was carried out to determine if the elevation of concentrations of Neu5Ac and Neu5,9Ac 2 between the healthy and disease cohorts was statistically significant. ROC curves were prepared to determine the sensitivity and specificity of each marker (Neu5Ac, Neu5,9Ac 2 and a combined marker of Neu5Ac + Neu5,9Ac 2 ) for the prediction of CVD. The same analysis was carried out for hs-CRP, which was measured in all samples, to allow for a comparison to a more well-established marker of inflammation and CVD. Hs-CRP was also combined with each of the sialic acids biomarkers as well as the combined Neu5Ac + Neu5,9Ac 2 marker to determine any affect on the predictive power.

Materials and methods Study population 30 Plasma samples from patients (16 female; 14 male) with an average age of 65 ± 22 with CVD were selected and purchased from the BioIVT biobank along with 30 age and sex matched healthy controls with an average age of 60 ± 13. Samples were chosen from volunteers that had one or multiple diagnosed CVDs including: hypertension, hypercholesterolemia, atrial fibrillation, congestive heart failure, coronary artery disease, but no other health conditions that could otherwise affect plasma sialic acid concentration such as type-2 diabetes [ 25 ], arthritis [ 26 ] or chronic obstructive pulmonary disorder (COPD) [ 27 ]. These conditions were chosen based on a literature search for health conditions associated with elevated plasma concentrations of sialic acid. Full details of the cohort can be found in Appendix 1 of the supplementary information. Analytical methods Analysis of sialic acids was carried out using the DMB method [ 22 , 24 , 28 ]. Two samples were required per analysis, one to quantify Neu5Ac and one to quantify Neu5,9Ac 2 . Release of sialic acids and DMB labelling of the samples was achieved using LudgerTagTM DMB Sialic Acid (LT-KDMB-96). 5 μL of each sample was added in triplicate to a 96-well plate. Each sample was subjected to acid release with 25 μL of 2 M acetic acid. The samples were vortexed and centrifuged (RCF 1677) followed by incubation at 80 o C for 2 h. The samples were cooled to room temperature before 5 μL of each released sample was transferred to a new 96-well plate. To this, 20 μL DMB labelling solution was added. The samples were vortexed and briefly centrifuged (RCF 1677) followed by incubation for 3 h at 50 o C. The reaction was quenched by addition of water to make-up the volume to 500 μL. Neu5Ac samples were then subjected to a 1 in 10 dilution, Neu5,9Ac 2 samples were not. All work was carried out using a Hamilton STARlet Liquid Handling Robot, apart from the initial dispensing of the samples into the 96-well plate. Fetuin derived from fetal calf serum (GCP-Fet-50U), an A2G2S2 [ 29 ] glycopeptide (BQ-GPEP-A2G2S2-10U) both from Ludger Ltd. and Visucon-F frozen control plasma from Affinity Biologics were utilised as system suitability standards. These standards were subjected to the same release and labelling conditions as stated above for the samples containing Neu5Ac. One nmol standards of Neu5Ac and Neu5,9Ac 2 [ 22 ] were also labelled using DMB (Scheme 1 ). 20 μL of labelling solution (3.5 mg DMB dye, 2.2 mL mercaptoethanol (1.4 M), 20 mg sodium dithionite) was added to each standard. The samples were vortexed and centrifuged (RCF 1677) before incubation for 3 h at 50 o C. The labelling reaction was quenched with water to bring the final volume to 500 μL. Standard curves were prepared for each standard with points: 0.01, 0.02, 0.1, 0.2, 0.5, and 1.0 nmol. The labelled sialic acids were analysed by LC-FLD. 5 μL of sample was injected into a U3000 UHPLC equipped with a fluorescence detector (λ ex = 373 nm, λ em = 448 nm, Thermo, UK) and separated on a C18 LudgerSepUR2 column (10 cm x 2.1 mm, 1.9 μm particles) at 30 o C. For Neu5Ac analysis an isocratic solvent system (7:9:84 MeOH:ACN:H 2 O) was used. For Neu5,9Ac 2 analysis a gradient solvent system was used of 7:6:87 MeOH:ACN:H 2 O for 6.5 min followed by 6:9:85 MeOH:ACN:H 2 O for 11.5 min. Analysis of hs-CRP was carried out using an Invitrogen CRP human ELISA kit purchased from Thermofisher. Each sample was analysed according to the manufacturer instructions. Statistical analysis Data are presented as mean ± standard deviation. p = 0.05 was used as the threshold for statistical significance. Differences in mean values were estimated using a two-sided t-test. Analysis was performed using R version 4.1.1 (RRID: SCR_001905) [ 30 ]. ROC curves were prepared using a Support Vector Classifier model (RRID: SCR_019053) [ 31 ] using the individual traits (Neu5Ac and Neu5,9Ac 2 concentrations) as predictors, followed by a combination of both. The model was trained on 70% of the data and tested on the remaining 30% after data was standardised such that it followed a normal distribution. The model performance was evaluated by a five-fold cross validation. Predictive power was evaluated using AUC, information on sensitivity and specificity was also obtained. No outliers were identified. The outcomes were whether a given person had a diagnosed CVD (CVD cohort) or whether they had no diagnosed CVD (control cohort).

Results In this study, the suitability of both Neu5Ac and Neu5,9Ac 2 as novel diagnostic markers of CVD was investigated. The DMB assay employed for the analysis of Neu5Ac and Neu5,9Ac 2 exhibited low levels of inter and intra-assay variation (< 10%). Neu5Ac and Neu5,9Ac 2 were detected in each sample (Fig. 2 ). The concentrations in all samples met the criteria to overcome the limit of detection and limit of quantitation (three times and nine times signal to noise ratio respectively). The cohort characteristics are shown in Table 1 . The statistically significant elevation of Neu5Ac between healthy controls and disease patient cohorts (45.19 ± 8.46 versus 63.55 ± 17.49; P < 0.001) (Fig. 3 ) in this study is supported by, and aids in confirming, the findings of previous work. The presence of atherosclerosis, hypertension, coronary heart disease and heart failure have all previously been associated with increased Neu5Ac concentrations [ 5 ]. Neu5,9Ac 2 has not been previously investigated as a marker in the context of CVD. A statistically significant elevation in Neu5,9Ac 2 concentration between healthy controls and CVD is reported here (0.32 ± 0.06 versus 0.40 ± 0.19; P < 0.04) (Fig. 3 ). ROC analysis showed that plasma Neu5Ac and Neu5,9Ac 2 concentrations can be used to discriminate between CVD cases and healthy controls and thus their potential utility as biomarkers has been highlighted. The AUC, or ability to distinguish between the CVD cases and healthy controls, was calculated for plasma concentrations of Neu5Ac (0.86), Neu5,9Ac 2 (0.71) as well as Neu5Ac + Neu5,9Ac 2 as a combined marker (0.93) (Fig. 4 ). Neu5Ac was shown to have good predictive power for the CVD cases versus healthy controls. Neu5,9Ac 2 performed less well in this regard with a lower AUC value. The improved predictive power of Neu5Ac over Neu5,9Ac 2 indicated that it may be a better predictor of CVD. Interestingly, the combined marker for Neu5Ac + Neu5,9Ac 2 showed a marked improvement in terms of AUC when compared to either of the individual markers. This appears to indicate that while Neu5,9Ac 2 may not have high utility as a marker by itself, it offers an improvement on a previously utilised biomarker for CVD. Neu5Ac exhibited high specificity (0.81) and sensitivity (0.82), Neu5,9Ac 2 exhibited equivalent specificity (0.82) but very low sensitivity (0.44). This data shows that both markers have a low false positive rate, however Neu5,9Ac 2 poorly discriminates the positive results (CVD cases) and as such would not be a good marker in this context. Neu5Ac, however, performs very well in this cohort in all aspects. The combined marker Neu5Ac + Neu5,9Ac 2 is interesting in that it offers both higher sensitivity (0.87) than that of Neu5Ac, but also higher specificity (0.90). This data is summarised in Table 2 . Comparing the three markers overall shows that Neu5Ac is a good potential biomarker for CVD which backs up previous research carried out in this area [ 5 ]. Neu5,9Ac 2 however did not perform as well in the capacity as a biomarker. This could be due to the low concentrations of Neu5,9Ac 2 in plasma compared to Neu5Ac. Neu5,9Ac 2 did offer some benefit, however, in that when combined with Neu5Ac, it was found that this combined biomarker offered a good improvement in terms of AUC, sensitivity and specificity over the previously established marker of Neu5Ac. Further to this, hs-CRP was analysed in both the CVD and healthy control cohorts. This was performed to allow for a direct comparison of the markers detailed in this work against a well-established inflammatory marker. The mean values for each cohort were determined (6.21 ± 15.25 versus 1.85 ± 2.37) (Table 2 ) and the dataset was subjected to the same statistical analysis as the sialic acid biomarkers discussed above. The mean hs-CRP values were not significantly different between the CVD patients and healthy controls (Fig. 5 ). ROC analysis was performed to determine what predictive power, if any, hs-CRP had for discriminating between CVD cases and healthy controls in this cohort. The AUC obtained was 0.50 ± 0.14 which is a very poor result, showing no ability for hs-CRP to be utilised for the discrimination of CVD cases from healthy controls beyond random chance for this particular cohort. Comparing this more well-established marker to the markers outlined in this paper highlights the potential utility of Neu5Ac and Neu5,9Ac 2 as biomarkers. This is especially highlighted when comparing the AUC of hs-CRP to that of the combination marker Neu5Ac/Neu5,9Ac 2 (0.50 ± 0.14 versus 0.93 ± 0.10) (Fig. 6 ). Table 3 details the AUC values obtained when combining the hs-CRP with each of the sialic acid markers and the combined biomarker. Interestingly, despite the poor performance of hs-CRP as an individual marker for CVD, it increased the AUC when combined with the individual sialic acids and the combined marker. The combined biomarker reaching an AUC of 0.97 ± 0.05 with very high sensitivity and specificity.

Discussion Inflammation is a component of CVD that can result in myocardial and endothelial damage [ 32 , 33 ]. An increase in inflammation related to CVD can lead to an acute-phase response which is characterised by upregulation of specific proteins such as alpha-1-acid glycoprotein, alpha-1-antitrypsin, fibrinogen, alpha-2-macroglobulin and hemopexin [ 34 ]. A variety of sialylated glycans decorate the surface of these proteins. A2G2S2 is the main glycan present in human plasma and is present in large quantities on alpha-1-antitrypsin and hemopexin. Alpha-1-acid glycoprotein has been reported to account for nearly all highly sialylated N -glycan species in plasma circulation [ 19 ]. These glycoproteins contain large quantities of sialic acids and an increase in the concentrations of these proteins during an acute-phase response would account somewhat for elevated Neu5Ac and Neu5,9Ac 2 concentrations in plasma. Further to this, elevated sialic acid concentrations might mean that the sialic acid can act as a substrate for the resialylation of low-density lipoprotein (LDL) and erythrocytes. Desialylated LDL and erythroctyes have been found to aggregate more than unmodified variants thus leading to the build-up of atherosclerotic plaques [ 35 , 36 ]. This is perhaps supported by evidence of an increase in the activity of sialyltransferase during an inflammatory state, which is perhaps an attempt to resialylate these structures and prevent cardiovascular damage occurring [ 37 ]. On the other hand, downregulation of the activity of plasma esterases has been observed with increased levels of inflammation [ 20 ]. This may reduce the quantity of acetylated sialic acid derivatives in plasma that are cleared by the activity of these enzymes, leading to the observation of higher than usual concentrations of these derivatives during an inflammatory state. Hs-CRP is an acute-phase protein, and, as such, an increase in the protein concentration in plasma samples from patients with higher levels of inflammation would be expected. Hs-CRP is generally present in very low levels in individuals with no inflammation but rises quickly during an inflammatory state [ 38 ]. While hs-CRP is used as a measure of the acute phase response by healthcare organisations, some authors reviewing data from large-scale studies have called into question the utility of hs-CRP due to high variation in concentrations and low AUC value when it comes to predictive power [ 39 , 40 ]. However, hs-CRP does appear to show some utility as a marker when combined with sialic acid markers. To conclude, Neu5Ac and a combined Neu5Ac + Neu5,9Ac 2 biomarker may have potential as biomarkers for CVD, possibly with some utility beyond currently utilised biomarkers such as hs-CRP. The sialic acid markers may be improved by combining with hs-CRP despite the poor performance of hs-CRP as an individual marker. It would be worthwhile to perform a follow-up study with a larger cohort size where the samples are sourced from a more well-controlled setting. Ideally, this would be from a single or family of well-described clinical sites with good clinical oversight to ensure the robustness of the sample collection with access to metadata for each sample. This would allow for the findings outlined here to be challenged and potentially reinforced. Including samples from patients with co-morbidities, one of the most common being diabetes, would also be valuable to determine the utility of these biomarkers in a wider, more representative, clinical setting. This would be particularly valuable if the co-morbidities also affect concentrations of plasma sialic acids independently of CVD and allow for the determination of whether these biomarkers can be utilised in a wider range of patients. Strengths and limitations This study showed strength in the higher diversity of biomarkers that were studied compared to similar studies investigating sialic acids in the context of CVD. The standards used were also of high purity offering excellent quantitation. This was further backed up by the utilisation of a highly specific and sensitive method allowing for the accurate quantitation of sialic acid species in plasma without concern of interferences. Extremely small quantities of material could be detected with this method, allowing for effective quantitation of Neu5,9Ac 2 which has not previously been reported. The study was limited by the sample size which was small and will have resulted in an underpowered study, potentially explaining the borderline significance (P < 0.04) of plasma Neu5,9Ac 2 concentrations between CVD cases and healthy controls. This may also have been explained by the relatively low concentrations of Neu5,9Ac 2 in human plasma and the associated difficulty with measurement of low concentrations of analytes related to signal-to-noise ratio. The AUC value for hs-CRP may have been affected by the small samples size also highlighting the need for a larger future study. This was only a pilot study, however, and as such a higher-powered future study with more samples would help to supplement this preliminary data. Further to this, samples were selected only from patients with one or more CVDs, but no other co-morbid health conditions. As such, this may limit the utility of the biomarker and it would be useful to test these biomarkers in patients with CVD and co-morbid conditions such as diabetes.

Conclusions Neu5Ac and Neu5,9Ac 2 concentrations were determined in samples from 30 healthy controls and 30 patients with diagnosed CVD. Neu5,9Ac 2 could be detected and quantified with high precision even when present in very small quantities compared to Neu5Ac. Statistically significant elevations of concentrations for both Neu5Ac and Neu5,9Ac 2 were observed. Neu5Ac was shown to be a good marker for the discrimination of these CVD patients from the healthy controls, Neu5,9Ac 2 however suffered from low sensitivity and as such poor prediction of CVD cases. Interestingly, despite the poor performance of Neu5,9Ac 2 as an individual marker, a combination marker of Neu5Ac + Neu5,9Ac 2 offered improved predictive power over Neu5Ac alone. The markers identified here offer an improvement, in terms of AUC value, over currently utilised biomarkers for CVD. Neu5Ac, as shown in previous research, could be used as a potential biomarker for CVD diagnosis. This research also indicates that the addition of Neu5,9Ac 2 is valuable resulting in improved predictive power for CVD diagnosis and Neu5Ac Neu5,9Ac 2 could act as markers for the presence of CVD with excellent predictive power. Comparison to hs-CRP also highlighted the potential utility of the markers shown in this paper, with hs-CRP offering very poor predictive power compared to Neu5Ac and Neu5,9Ac 2 . Hs-CRP may offer some utility when combined with the sialic acid markers, offering a slight increase to AUC values. Future studies will be of benefit to solidify the results obtained in this research and to determine the utility of Neu5Ac as a biomarker in patients with CVD and co-morbid conditions. The outcome of this would have a great impact on the utility of the marker in a clinical setting where patients often present with multiple co-morbidities.

Cardiovascular disease (CVD) is a group of health conditions affecting the heart and vascular system with very high prevalence and mortality rates. The presence of CVD is characterised by high levels of inflammation which have previously been associated with increased plasma concentrations of N -acetyl neuraminic acid (Neu5Ac). While Neu5Ac has been studied in the context of CVD, Neu5,9Ac 2 has not, despite being the second most abundant sialic acid in human plasma. A small-scale pilot study of thirty plasma samples from patients with diagnosed CVD, and thirty age and sex-matched healthy controls, was designed to gain insight into sialic acids as biomarkers for CVD and potential future areas of study. Each sample was assayed for Neu5Ac and Neu5,9Ac 2 concentrations. Mean Neu5Ac and Neu5,9Ac 2 concentrations were significantly elevated in patients with CVD compared to healthy controls (Neu5Ac: P < 0.001; Neu5,9Ac 2 : P < 0.04). Receiver operator curve (ROC) analysis indicated that both Neu5Ac and Neu5,9Ac 2 have reasonable predictive power for the presence of CVD (Neu5Ac AUC: 0.86; Neu5,9Ac 2 AUC: 0.71). However, while Neu5Ac had both good sensitivity (0.82) and specificity (0.81), Neu5,9Ac 2 had equivalent specificity (0.81) but very poor sensitivity (0.44). A combination marker of Neu5Ac + Neu5,9Ac 2 showed improvement over Neu5Ac alone in terms of predictive power (AUC: 0.93), sensitivity (0.87), and specificity (0.90). Comparison to a known inflammatory marker, high sensitivity c-reactive protein (hs-CRP: P-value: NS, ROC:0.50) was carried out, showing that both Neu5Ac and Neu5,9Ac 2 outperformed this marker. Further to this, hs-CRP values were combined with the three different sialic acid markers to determine any effect on the AUC values. A slight improvement in AUC was noted for each of the combinations, with Neu5Ac + Neu5,9Ac 2 + hs-CRP giving the best AUC of 0.97 overall. Thus, Neu5Ac would appear to offer good potential as a predictive marker for the presence of CVD, which the addition of Neu5,9Ac 2 predictive power improves, with further improvement seen by the addition of hs-CRP. Supplementary Information The online version contains supplementary material available at 10.1007/s10719-023-10138-3. Keywords

Electronic supplementary material Below is the link to the electronic supplementary material.

Author contributions D.I.R.S, G.K and H.M.I.O designed the study, won funding for the programme, and supervised the study. J.C. and D.T. carried out the selection of samples with the assistance of M.M., G.K., D.I.R.S. and H.M.I.O. D.T. and M.M. were responsible for the ordering and initial processing of purchased samples. J.C. designed and carried out the analysis with the assistance of C.B. and R.G. G.K. was responsible for statistical analysis, G.EH. performed ROC analysis and modelling. J.C. wrote the first draft of the main manuscript text and prepared Figs. 1 and 2 , scheme 1 and all tables. G.EH. prepared all other figures. All authors reviewed the manuscript and J.C prepared the final draft for submission. Funding Financial support from the Medical Research Council and Ludger Ltd (MR/P015786/1) to Jack Cheeseman, is gratefully acknowledged. We would also like to thank the Irish Research Council for supporting the study (Enterprise Partnership Scheme Project EPSPG/2019/511). Data Availability The metadata for the patient cohorts in this study can be found in the supplementary information. Declarations Competing interests The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: The authors based at Ludger Ltd work in commercializing analytical assays for use in the field of glycomics and the analysis of biopharmaceuticals. The remaining authors do not have competing interests to declare. Ethical approval The samples used in this study were purchased from the BioIVT biobank. All BioIVT biobank’s activities are carried out in accordance with applicable laws, regulations, and ordinances. All of the BioIVT biobank samples are collected with informed consent of each subject. More information related to the ethics, informed consent and quality assurance at BioIVT can be found at the following website: https://bioivt.com/about/quality-assurance .

CC BY

no

2024-01-16 23:41:58

Glycoconj J. 2023 Nov 22; 40(6):645-654

oa_package/2c/14/PMC10788320.tar.gz

PMC10788321

38221600

Introduction The cantilever fixed partial denture (CFPD) has been defined as a fixed restoration that has one or more abutments at one end, with the other end unsupported [ 1 ]. In principle, CFPDs can replace any tooth in the dental arch, but are considered particularly useful for avoiding a removable partial denture in patients with distal edentulism [ 2 ] or for establishing the minimum acceptable number of occlusal units in the shortened dental arch concept [ 3 ]. Favorable outcomes can be expected for CFPDs that have at least two abutment teeth and do not replace more than one tooth [ 3 – 5 ]. The survival rate of CFPDs with two abutment teeth that were vital at the time of cementation and had at least two-thirds residual alveolar bone was comparable to that of conventional endabutment FPDs [ 6 ]. Even in the age of implants, CFPDs can be useful because there are still patients who cannot afford implant treatment or for whom implant treatment is not an option or is too complicated due to insufficient bone or other reasons [ 7 ]. In addition, restoration of edentulous ridges adjacent to implants using cantilevers may be considered for esthetic or economic reasons or to avoid greater augmentation effort without negatively affecting peri-implant health or increasing the risk of mechanical complications [ 8 – 11 ]. A recent review specified that the use of CFPDs on implants did not have a negative impact on prosthesis survival or success or marginal bone loss [ 12 ]. The unique biomechanics, with eccentric forces acting on both restorations and abutments, place high demands on the design of CFPDs and the properties of the materials used in their fabrication. This involves ensuring that the CFPDs are durable over the long term under the forces acting in the oral cavity. Here, metal-ceramic CFPDs represent the therapeutic standard [ 11 , 13 , 14 ]. Biological, economic, and aesthetic considerations have led to the increasing use of zirconia in place of traditional metal ceramics [ 15 ]. Early clinical experience with mostly completely veneered zirconia CFPDs on teeth [ 13 , 16 ] and implants [ 14 , 17 – 19 ] indicates that the success of restorations on teeth is mainly limited by chipping [ 13 , 16 ], whereas implant-supported CFPDs also experience framework fractures [ 14 , 18 ]. At the same time, in vitro studies cast doubt on whether the load-bearing capacity of completely veneered zirconia CFPDs is sufficient for molar replacement, despite numerous design modifications involving reinforced zirconia frameworks [ 20 – 23 ]. Recent developments have focused on monolithic full-contour zirconia restorations [ 24 ], also for CFPDs [ 25 ]. The monolithic design has two major advantages for CFPDs. Firstly, the veneer is no longer a potential site of initial damage under load [ 26 , 27 ], which should reduce the risk of chipping [ 28 ]. Secondly, the space that would be taken up proportionally by the much weaker veneering ceramic can be used entirely for the high-strength zirconia, which should maximize the load-bearing capacity of ceramic CFPDs [ 29 ]. The acceptance of such monolithic restorations depends not only on the strength of the material but also on esthetic criteria. For this reason, early attempts were made to improve the translucency of the available high-strength 3-mol yttria-stabilized zirconia polycrystal (3Y-TZP) materials. This was achieved with the 2nd generation zirconia by reducing the amount of alumina in 3Y-TZP and by increasing the sintering temperature [ 30 ]. The result was acceptable esthetics for the posterior region while maintaining maximum flexural strength. In subsequent generations, the yttria content was increased to 4 mol% (4 mol% partially stabilized zirconia, 4Y-PSZ) and 5 mol% (5Y-PSZ) to further improve translucency [ 30 ]. The increased yttria content improves translucency by increasing the mean grain size and proportionally stabilizing the cubic phase, which has an isotropic crystal structure and thus more uniform light scattering compared to the tetragonal phase. However, because the cubic phase does not have the property of transformation toughening, the flexural strength values of these zirconia materials is lower [ 30 ]. Nowadays, there are also multilayer materials combining several types of zirconia in one milling disk (translucent layers located in the area of the incisal edges/cusp tips) and thus exhibiting a strength/translucency gradient [ 31 ]. However, for monolithic zirconia single crowns and FPDs made of 2nd generation zirconia, it has been shown that a facial veneer in the anterior and premolar region can achieve high patient satisfaction with regard to restoration esthetics without measurably increasing the restorations’ risk of technical complications [ 28 , 32 ]. Such a design variant would also be interesting for zirconia CFPDs, especially since it can be assumed that the critical stresses during loading of the pontic occur mainly on the upper site of the restorations [ 5 , 25 , 33 ]. Accordingly, the aim of this in vitro study was to compare the initial damage and failure loads after artificial aging and under different loading conditions (axial/oblique) of monolithic, partially (facially), or completely veneered posterior 2nd generation zirconia CFPDs attached to either cobalt–chromium (CoCr) or natural abutment teeth or implants. Axially loaded completely veneered CoCr CFPDs attached to CoCr abutment teeth served as the control. The null hypothesis was that there would be no differences between the different CFPD design, loading, and abutment groups.

Material and methods CFPDs were designed to replace a lower first molar by means of a distal cantilever in the shape of a premolar (mesio-distal dimension: 8 mm) retained by 2 splinted complete crowns attached to the first and second premolar. A mandibular typodont (type ANA-4, Frasaco, Tettnang, Germany) was used as the anatomical basis for the experiments. The prepared typodont teeth (see below) were then replicated in CoCr (Remanium GM800, Dentaurum, Ispringen, Germany) and served as abutments in the test models. In addition to testing CFPDs on the CoCr tooth replicas, CFPDs attached to implants (tissue level implants, SP, RN, SLA, Roxolid, diameter 4.1 mm, length 10 mm, Straumann, Basel, Switzerland) or natural teeth were also tested. Three different designs (monolithic, partially veneered, completely veneered) for zirconia CFPDs were tested with abutment tooth replicas. Zirconia CFPDs supported by implants or natural teeth were only investigated with a partially veneered design. As a control, the completely veneered design was also tested with a CoCr framework. All groups were exposed to axial loading on the pontic during aging and fracture tests. Additional tests with oblique loading (30° tilt) on the pontic were conducted for monolithic and partially veneered CFPDs supported by abutment tooth replicas (Fig. 1 ). Based on a previous study of similar restorations [ 22 ], a sample size of n = 8 per group was considered adequate to detect statistically significant differences with adequate power. The CFPDs in the different test groups were designed to be congruent with respect to their external geometry (Fig. 2 ). Design of CFPDs based on the prepared typodont teeth The typodont abutment teeth were prepared with an axial and occlusal reduction of 1.5 mm, a 0.5-mm deep chamfer finishing line, rounded edges, and a total occlusal convergence of the axial walls of 6° using a paralleling device. A type-IV gypsum master cast (GC Fujirock-EP, GC Europe, Leuven, Belgium) of the situation was digitized by use of a laboratory scanner (D800, 3Shape, Copenhagen, Denmark). First, a completely veneered CFPD was designed (Dental Designer, 3Shape) using the situation with unprepared abutment teeth as a wax-up scan. The framework had a minimum layer thickness of 0.8 mm and supported the veneer anatomically (Fig. 2 ). The connectors were set to minimum cross-sections of 9 mm 2 between the abutments and 12 mm 2 for the cantilever (Table 1 ). Second, by combining framework and veneer using 3D manipulation software (Geomagic Design X, 3D Systems, Moerfelden-Walldorf, Germany), a monolithic CFPD with identical outer geometry was created (Fig. 2 ). Third, the monolithic design was reduced on the buccal side of the crown retainers by 0.7 mm to provide space for a partial veneering (Fig. 2 ). Therefore, the area of the veneering window was marked in the software (Geomagic Design X, 3D Systems) and reduced by the appropriate amount. At the boundaries, the transition areas were given a profile with a radius (0.4 mm radius → 0.5 mm radius in the scaled situation during milling) corresponding to the smallest milling tool (1 mm in diameter) used on this surface. Design of partially veneered CFPDs supported by implants The implants were used with standard abutments (RN synOcta Cementable Abutment, height 5.5 mm; Straumann, fixation torque 35 Ncm). The digitized geometry of an implant with standard abutment was placed at each abutment tooth position of the partially veneered design described above. The implant axes were oriented vertically, and their spatial position was chosen such that the implant neck center was identical to the center of the respective abutment tooth margin line in horizontal and vertical direction. Inner surfaces of crowns designed on the implants with abutments were complemented by the existing outer CFPD geometry (Fig. 2 ). Design of partially veneered CFPD supported by natural teeth Natural premolar teeth were implemented in a gypsum model such that they resembled the typodont situation as closely as possible, i.e., parallel tooth axes and a tooth center distance of 7.5 mm. After preparation under microscopic control, tooth dimensions did not deviate from those of the typodont teeth situation by more than 0.5 mm in any spatial direction. Natural teeth in the study were allowed to have small defects that did not affect the pulp system. After caries excavation and prior to preparation, such defects were filled with a composite resin (Rebilda DC, VOCO, Cuxhaven, Germany) using a total-etch adhesive technique (Primer and Adhesive, Optibond FL, Kerr, Kloten, Switzerland). For the use of natural teeth, a positive ethics vote was available (S-034/2010), and tooth donors signed an informed consent form. Until their use, the teeth were stored in 1% chloramine-T solution. For CFPDs on natural abutment teeth, the veneer of the completely veneered design (based on the typodont teeth) was milled from wax and positioned as congruent as possible over the prepared natural teeth using a paralleling device. Missing contours up to the preparation margins were completed with wax. This situation was digitized and used as a wax-up scan for the CFPD design. Partial reduction of the crown retainers on the buccal side by 0.7 mm was done individually at the end. Standardization of loading site The cantilever pontic was modified such that the inner ridges of the cusps were planar and featured angles of ±30° to the horizontal direction and the mesial-distal axis as rotational axis (Fig. 3 ). This enabled a standardized loading of all CFPDs. CFPD fabrication Zirconia frameworks were centrally milled (CNC 500 milling unit, 3M Oral Care, Seefeld, Germany) from translucent (2nd generation) 3Y-TZP (Lava Plus Multi L, 3M Oral Care), monochromatically dyed (A4) by immersion for 2 min in an appropriate dyeing solution (Lava Plus Dyeing Liquid, 3M Oral Care), dried for 2 h at room temperature, and sintered at 1450 °C (Lava Furnace 200, 3M Oral Care). Monolithic zirconia frameworks subsequently received two glaze firings (VITA AKZENT GLAZE SPRAY, VITA Zahnfabrik, Bad Säckingen, Germany). For completely veneered zirconia CFPDs, the zirconia frameworks were overpressed (IPS e.max ZirPress HT A4, Ivoclar Vivadent, Schaan, Lichtenstein) according to the manufacturer’s instructions for the veneering ceramic. The 3D-designed veneer was milled in wax (LAWAX, 3M Oral Care) to serve as a space holder during the overpressing procedure. After liner firing (ZirLiner Liquid Build Up Allround and Zir Liner Clear, Ivoclar Vivadent), the wax-milled veneer was positioned on the zirconia framework and manually supplemented with wax at the basal/cervical aspects using a putty silicone negative of the monolithic CFPD as a mold. Two glaze firings (IPS e.max Ceram Glaze Paste) completed the fabrication process. Partially veneered zirconia CFPDs were finalized using the layering technique (VITA VM 9, VITA Zahnfabrik) and two subsequent glaze firings (AKZENT GLAZE SPRAY, VITA Zahnfabrik). Completely veneered metal-ceramic CFPDs served as the control. Therefore, the anatomically reduced framework of the respective all-ceramic group was milled from wax and cast with CoCr alloy (Remanium Star, Dentaurum). The veneering was carried out analogously to the completely veneered zirconia CFPDs using leucite-containing pressable ceramics (IPS InLine PoM A4, Ivoclar Vivadent) and finalized with two glaze firings (IPS Ivocolor Glaze Paste, Ivoclar Vivadent). All CFPDs were checked for marginal and internal fit and adjusted manually if necessary. Model fabrication and cementation Roots of CoCr abutment tooth replicas as well as natural teeth were coated with a heat-shrink tubing (HIS-A 12/4-PO-X-BK, HellermannTyton, Tornesch, Germany) to achieve realistic tooth mobility during the tests. The shrink tubing was cut off 2 mm below the apical end of the root and the opening filled with polyvinylsiloxane (Flexitime Correct Flow, Kulzer, Hanau, Germany) (Fig. 3 ). Using molds resembling the negative shape of the occlusal surface and a paralleling device, abutment tooth replicas, implants, or natural teeth provisionally fixed in the respective CFPD were embedded in acrylic resin (Technovit 4071, Kulzer) in a metal specimen holder in the planned position and orientation (Fig. 3 ). For CFPD cementation, the intaglio surfaces of the zirconia crown retainers were alumina-particle abraded at 0.1 MPa pressure (Alustral 50 μm, Omnident Dental-Handelsgesellschaft, Rodgau Nieder-Roden, Germany). CoCr crown retainers and CoCr abutment teeth as well as implant abutments were alumina-particle abraded with 0.2 MPa (Alustral 50 μm). Subsequently, CFPDs, CoCr abutment teeth, and implants were steam cleaned and thoroughly dried. Natural abutment teeth were cleaned with polishing paste (Zircate, Dentsply Sirona, Bensheim, Germany), rinsed with water, and lightly dried with oil-free air. Cementation was performed in a universal testing machine (Z005, Zwick/Roell, Ulm, Germany) with self-adhesive resin cement (RelyX-Unicem Automix 2 A3, 3M Oral Care) at 400 N applied for 180 s centrally between the abutment teeth. After storage for 24 h at 100% humidity and 37 °C in an incubator (Heraeus Functionline Heating Oven, Thermo Fisher Scientific, Waltham, MA, USA), CFPDs were examined under a stereo light microscope (Stemi SR, Zeiss Microscopy, Oberkochen, Germany; 8× magnification) for damage during cementation such as fractures, cracks, or chipping. Artificial aging and failure testing All CFPDs were artificially aged using 10,000 thermocycles (bath temperatures 6.5 °C and 60 °C, Thermocycler TC 1, SD Mechatronik, Feldkirchen-Westerham, Germany) and 1.2 million chewing cycles (CS-4.8, SD Mechatronik) with a force magnitude of 108 N. During chewing simulation, samples were immersed in deionized water and a steel ball (Ø 6 mm) served as antagonist. For each group, loading conditions during aging resembled those used later on during the fracture tests. After artificial aging, CFPDs were inspected again for possible damage such as fractures, cracks, chipping, or decementation at up to 200× magnification (Stemi SR, Zeiss Microscopy). Failure testing was performed in a universal testing device (Z005, Zwick/Roell) at a feed rate of 0.5 mm/min. Forces were applied with a steel ball (Ø 6 mm) 3 mm from the distal end of the pontic (Fig. 3 ). With axial loading, the test force was applied via contact on both cusps (Fig. 3 ). For oblique loading, samples were fixated with 30° tilt such that the loaded cusp was oriented horizontally (Fig. 3 ), and the force application point was 1.3 mm below the cusp tip. The end of the failure test was defined as when the test force decreased to less than 30% of the previous maximum value or damage equivalent to clinical failure occurred. During the failure tests, body-borne sound signals were recorded (20 kHz sampling rate) to help identify damage prior to CFPD failure (Fig. 4 ). A damage event was given for an interim drop in test force coinciding with a high sound signal exceeding 75% of the maximum magnitude recorded during the complete test. Forces at failure and the first damage event (initial damage) were recorded. If no pre-failure event occurred, initial CDFP damage coincided with CFPD failure. In case of initial damage or failure during artificial aging, a force of 108 N (force magnitude during chewing simulation) was associated with initial damage and/or failure. Failure modes and fractography All tested CFPDs were examined by light microscopy and classified according to their failure modes. Representative specimens were fractographically examined by light (Stemi SR, Zeiss Microscopy) and scanning electron microscopy (SEM) using a field emission scanning electron microscope (Auriga 40, Carl Zeiss Microscopy; acceleration voltage: 1.5 kV, working distance: 3–6 mm) to identify fracture origin and crack propagation and thus get information about possible causes for the respective fracture. The fractographic examination was performed by the manufacturer of the ceramic framework material (3M Oral Care). Statistical evaluation Test forces at failure and at initial damage were analyzed separately using one-way analysis of variance (ANOVA), and Tukey honest significant difference post hoc tests were used for the pairwise comparisons (2-sided α = 0.05).

Results Test forces corresponding with failure and initial damage are listed in Table 2 and shown in Fig. 5 . Failure load Mean test forces at failure ranged between 392 N for axially loaded partially veneered zirconia CFPDs on natural teeth and 1181 N for axially loaded monolithic zirconia CFPDs on CoCr abutment teeth. The mean failure load of the axially loaded monolithic zirconia CFPDs was not statistically significantly different ( p = 0.777) from that of the control group, which had a mean failure load of 1042 N. These two groups also had significantly higher mean failure loads than all other groups in the test ( p < 0.001). Compared to axial loading, oblique force application led to statistically significantly ( p < 0.001) lower fracture forces for monolithic zirconia CFPDs (460 N) and a slightly higher (no significant effect, p = 0.819) mean failure load for partially veneered zirconia CFPDs (599 N compared to 468 N). The failures of all tested CFPD could be classified with eight different fracture modes (Fig. 6 ). The most common fracture was through the connector between the two crown retainers. Half of the test groups exclusively or predominantly showed this failure mode. Axially loaded monolithic or partially veneered groups on CoCr abutment teeth were particularly affected. For obliquely loaded CFPDs, however, a shift of the failure pattern towards a breakout of the retainer walls was observed. Also different were implant-supported CFPDs, completely veneered CFPDs (controls), and CFPDs on natural teeth: implant-supported CFPDs fractured through the pontic connector, controls failed exclusively due to excessive chipping of the veneering ceramics, and all natural tooth–supported CFPDs failed due to fractures of the abutment teeth. Initial damage Initial damage before failure, i.e., crack formation or small chippings within the veneer, was registered for completely or partially veneered CFPDs. No initial damage events before the final fracture were observed in monolithic restorations. Mean test forces at initial damage reached from 361 N for axially loaded partially veneered zirconia CFPDs supported by natural abutment teeth to 1181 N for axially loaded monolithic zirconia CFPDs supported by CoCr abutment teeth. Axially loaded monolithic zirconia CFPDs differed significantly from control CFPDs ( p < 0.001) with the second highest mean forces corresponding with initial damage (654 N), and from all other test groups ( p < 0.001) showing initial damage below 500 N for most samples. SEM analysis and fractography Analysis of SEM images of selected samples revealed three main types of detectable failure causes of the ceramic materials in this study: (1) process-related, (2) design-related, and (3) aging-related. In many samples, milling marks (Fig. 7 ) and small chippings (Fig. 8 ) on the zirconia framework surface caused by the milling process, and milling dust deposits as well as resulting superficial pore formation (Fig. 9 ) along the zirconia surface were found to be process-related flaws that could be identified as origins of fracture. Design-related fractures originated from edges of the zirconia frameworks where stress concentrations could occur. Such “edge effects” were observed predominantly in partially veneered CFPDs in the area of the sharp-edged boundary of the anatomical reduction for the ceramic veneer (Fig. 10 ). Aging-related damage origins were detected for the control group (completely veneered CFPDs with CoCr frame) where veneer chipping started at the loading site in the area of pre-damage caused by “Hertzian compression” (Fig. 11 ).

Discussion This study tested the hypothesis that there would be no differences between the loads at failure and initial damage between three-unit CFPDs for the replacement of a premolar-size tooth as a function of restoration design (monolithic/partially veneered/completely veneered zirconia or completely veneered CoCr), loading (axial/oblique), and abutment type (CoCr teeth/natural teeth/implants). The hypothesis had to be partially rejected. Highest failure loads for CFPDs were associated with axial loading and resiliently embedded CoCr abutment teeth and occurred for monolithic zirconia CFPDs and completely veneered restorations with a CoCr framework. All other test groups differing in loading/support conditions and design, showed significantly lower fracture resistance. This result was to be expected as the monolithic zirconia CFPDs had the largest dimension of the zirconia framework compared to the partially and completely veneered zirconia restorations, and it is known that the loading capacity of a ceramic restoration depends not only on the flexural strength but also on geometric parameters such as wall thickness [ 34 ]. For CFPDs with a CoCr framework, the metal only deformed plastically until large parts of the veneering got lost, thus leading to a high fracture resistance of the entire restoration [ 35 ]. For a dentist, however, an initial damage may already be a failure of a restoration. Restorations of all test groups but monolithic zirconia CFPDs experienced initial damage in the form of cracking in the veneering ceramic at loads about half of the respective failure loads. Knowledge of this initial damage allows the results to be better transferred to a clinical context where veneer chipping is a common complication of tooth- or implant-supported CFPDs [ 36 , 37 ]. Even if initial damage (chipping, cracks) does not result in immediate clinical restoration failure, it can influence restoration failure over time [ 38 ]. Related to maximum achievable bite forces of 800 N and 600 N for young adult males and females [ 39 ] and chewing forces at about 40% of the maximum bite force [ 40 ], the results of the current study suggest that only monolithic zirconia CFPDs and veneered restorations with a CoCr framework will provide clinically acceptable failure loads. This finding is somewhat put into perspective when considering the forces that occur in an older age group, which corresponds more to a prosthetic patient population. Completely dentate patients with an average age of 70.2 years were found to exhibit mean maximum bite forces of 377 N in the first molar region [ 41 ]. Furthermore, there are indications that for eccentric forces on cantilevers self-inhibition mechanisms of the masticatory system may limit the maximally exerted force. Lundgren and Laurell reported that the maximum individual bite force was 150 N in patients treated with cross-arch FPDs with unilateral posterior cantilevers when the occlusal load was actively focused on the cantilever [ 42 ]. This may suggest that, during clinical function, CFPDs are not subjected to bite forces as high as those previously described. This assumption is indirectly supported by the fact that in clinical studies of tooth-supported zirconia CFPDs, no framework fractures have been observed [ 13 , 16 ]. The situation is somewhat different for implants, where the feedback mechanisms described above are less likely to be effective. For example, over a period of 10 years, framework fracture was identified as the most frequent cause of failure for both CFPDs and end-abutment FPDs on implants in posterior dentitions [ 14 ]. For many years, optimization of all-ceramic CFPDs is researched. Gabbert et al. [ 20 ] tested completely veneered zirconia CFPDs (12 mm 2 connector cross section) replacing a premolar-sized molar and found mean fracture forces between 603 N and 703 N after aging for axially loads applied to the pontic. Reinforcement of the zirconia frameworks with an additional shoulder had no positive effect on the fracture loads. Fractures were usually located at the distal wall of the distal crown retainer, and no fractures of the connectors were observed. In a later study, Ohlmann et al. [ 21 ] tested zirconia CFPDs with similar configuration and different reinforcement modifications: Highest fracture loads were measured with zirconia frameworks reinforced at the oral wall of the distal abutment by a cervical shoulder (2.0 mm or 3.0 mm high, 1.0 mm wide), whereas a general thickening of the walls of the distal abutment (from 0.7 to 0.8 mm) or an isolated thickening of the occlusal surface (from 0.7 to 1.0 mm) resulted in only a slight increase in fracture load. Overall, with mean fracture loads ranging from 346 N to a maximum of 548 N, none of the tested groups achieved a load-bearing capacity justifying a recommendation for clinical use [ 21 ]. A further investigation dealing with additional zirconia framework reinforcements compared CFPDs differing in the wall thickness on the distal retainer crown [ 22 ]. With increased framework thickness of the distal crown (1 mm wall thickness) or application of an occlusal reinforcement (2 mm wide and 1 mm deep notch at the central fissure of the distal abutment tooth) corresponding mean fracture loads lay between 529 N and 590 N [ 22 ]. Regardless of the type of reinforcement, the predominant failure mode was still a partial breakout of the distal wall of the terminal abutment crown. The authors concluded that the fracture loads observed were not sufficient to recommend zirconia CFPDs without reservation for posterior tooth replacement [ 22 ]. With the current study, the monolithic design allocates all the space created by the tooth preparation to the zirconia framework, resulting in maximum reinforcement of the restorations and rather high fracture loads above 940 N (mean value 1181 N). This increased fracture resistance of the monolithic CFPD was to some extent expected and is consistent with previous mathematical estimates, including the fracture pattern in the area of the connection between the crown retainers [ 29 ]. Zhang et al. [ 29 ] used topology optimization and extended finite element method to propose an optimized design of posterior veneered zirconia CFPDs leading to higher fracture loads. For this purpose, they created a model of an all-porcelain CFPD that was iteratively modified by replacing porcelain elements with zirconia elements until crack initiation no longer occurred under a simulated vertical load of 250 N on the pontic and the abutment teeth. They found that especially reinforcement in the occlusal embrasures reduced the maximum principal stresses in porcelain ceramic CFPDs and that, when the occlusal surface was completely reinforced, the region of maximum tensile stress shifted to (i) the cervical embrasure of the connector between the abutment teeth and (ii) to the margin (mesial and distal) of the near-cantilever crown retainer [ 29 ]. What was initially surprising was the low fracture loads of the facially veneered CFPDs in our study compared to the monolithic CFPDs. Considering the common assumption that tensile forces occur mainly in the occlusal area when the pontic is loaded [ 25 , 33 ], it was not expected that a facial veneer would reduce the fracture load of the restorations to such an extent (i.e., by more than half). In this context, analysis of the fractured CFPDs using SEM provided further insights. In particular, the sharp edge of the veneering window proved to be very disadvantageous as it was found to be a site of predilection for fracture of the restorations in the area of the mesial connector, acting as a site of possible stress concentration [ 43 ]. In addition, material defects or fabrication or post-treatment defects were other origins of fractures of the ceramic CFPDs [ 26 , 44 ] and, in the case of metal-ceramic restorations, deterioration of the ceramic veneer in the occlusal contact area [ 45 ]. As derived from 3D finite elements analysis, highest cantilever prostheses’ displacement and functional stresses can be produced (i) when a lateral loading direction of the pontic is chosen and (ii) when only the pontic is loaded [ 5 ]. Accordingly, oblique loading had an additional negative effect on the maximum load to failure of monolithic CFPDs in the present study. In this respect, the results of the monolithic CFPDs are consistent with those of another in vitro study of zirconia CFPDs, which showed that oblique loading of the pontic reduced the failure load by half compared to axial loading of the pontic [ 46 ]. For the monolithic CFPDs in the present study, this was accompanied by a change in failure mode from fracture in the mesial connector area to breakout of one or more crown walls. Interestingly, oblique loading did not reduce the fracture load of the partially veneered zirconia CFPDs, which could be due to the fact that the changed force vector moved the previous weak point (sharp-edged edge of the veneering window) out of the area of maximum tensile stress. From a clinical point of view, the results suggest that dynamic occlusion on cantilever pontics should be avoided through consistent functional occlusal design and occlusal adjustment measures. Implant-supported CFPD are biomechanically a completely different system than CFPDs on resilient abutment teeth. As known from FEA, without periodontal resilience, the maximum principal stress can be expected at the top of the pontic connector just distal to the terminal abutment and the minimum principal stress at the bottom of the connector [ 33 ]. Accordingly, all but one of the implant-supported CFPDs failed in this region by fracture through the pontic connector. With natural abutment teeth, no restoration failed, but the teeth fractured at rather low mean forces (392 N). Since extracted teeth are likely to be damaged during extracting, maximum loads with sound and vital teeth will likely be higher especially when considering clinical observations that tooth fractures can be expected in only about 3% of the CFPDs [ 47 ]. The results are also in contrast with the results of the study of Naumann et al. who tested a group of zirconia CFPDs on natural abutment teeth in a test setup very similar to that used in the present study [ 23 ]. Here, decementation rather than tooth fracture was the most common cause of failure, but mean failure loads (411 N) were in the same range as those found in the present study. The use of artificial abutment teeth can have a tremendous effect on dental restorations. However, this is the case for thin-walled restorations and a load case where the fracture starts near the load application site [ 48 ]. For example, in minimally invasive zirconia crowns, the deflection magnitude of the thin occlusal zirconia layer will depend on the stiffness of the underlying structures, i.e., the fracture resistance will increase with increasing stiffness of the supporting structures (cement layer, enamel, dentin). For FDPs, which typically fracture far from the loading site, the abutment material is of minor importance and the deformation of the entire restoration, which is strongly influenced by the abutment resilience, is critical. This has been shown, for example, in FE analyses that complemented in vitro tests of inlay-retained FDPs [ 49 ]: in this publication, varying material parameters for abutment teeth and cement, as well as abutment tooth resilience, demonstrated that the simulation of tooth mobility in in vitro tests has a significant effect on fracture resistance. For all abutment materials except acrylic resin, no significantly different strains and stresses were found in the fracture-relevant connector areas compared to the situation with natural teeth. This will be even more true for the stresses and strains within the unsupported cantilever element of the CFPDs. Since tooth mobility was simulated, the fractures that occurred in the connector between the two abutment teeth were also not considerably influenced by the abutment tooth material. Therefore, it was assumed that the fracture forces for CFPDs found in this investigation should be in the same range as those found in a clinical setting. It is a limitation of the study that zirconia CFPDs supported by implants or natural teeth were only investigated with a partially veneered design. However, no different results would be expected for monolithic CFPDs on natural teeth, since the failure of the restorations in the test was due to fracture of the abutment teeth. The case of implant-supported monolithic CFPDs might be different. Here, the complete omission of a veneer could also lead to an increase in fracture load. The “loading direction” factor was also not varied for all restoration groups, so no conclusions can be drawn about the fracture load of completely veneered zirconia CFPDs on CoCr tooth replicas and implant-supported CFPDs under oblique loading. For CFPDs on natural teeth, the previously made assumption that the teeth fail first in the test also appears to be valid for the variation of the load case. Another limitation is that, despite all standardization efforts, different connector diameters were used in the various groups. This does not refer to the difference between monolithic and completely veneered CFPDs. This is a logical consequence of the standardized external geometry, which led to veneered parts being replaced by monolithic zirconia. Rather, it concerns the partially veneered CFPDs on teeth and implants, whose different connection geometry to the supporting structure (tooth or implant) has slightly affected the connector area and thus influenced the absolute comparability of the results. Furthermore, two different veneering ceramics were used, one press ceramic and one layering ceramic. Both ceramics were recommended by the zirconia manufacturer and have a coefficient of thermal expansion matched to the zirconia material. With the press ceramic, the standardized geometry of the completely veneered CFPD could be implemented one-to-one by using a wax-milled space holder in the shape of the veneer. In contrast, the layering ceramic was used for the partially veneered CFPDs to vestibularly veneer them in a practice-oriented procedure. This may be a limitation of the study results in that full veneers are often layered by hand in daily practice, restricting the study findings for completely veneered CFPDs to overpressed restorations. This could be important, as it has been shown in the past that layered restorations have a higher fracture resistance than overpressed restorations when anatomically designed frameworks and comparable materials were used [ 50 ]. It is important to note that although the tested material is a zirconia with increased translucency, it is still a 3Y-TZP. It must therefore be distinguished from materials with increased yttria content. Future research approaches could therefore focus not only on design improvements of (partially) veneered zirconia CFPDs but also on material alternatives for the monolithic fabrication of zirconia CFPDs.

Conclusions Monolithic zirconia CFPDs may be considered a viable alternative to completely veneered metal-ceramic CFPDs with CoCr frameworks in terms of fracture load. Oblique loading of the cantilever pontic drastically reduced the fracture load of monolithic zirconia CFPDs and should therefore be avoided in a clinical scenario. Design flaws negatively affected the fracture load of the partially veneered zirconia CFPDs by promoting stress concentrations under axial loading due to the presence of sharp edges around the veneering windows.

Objectives The aim of this study was to compare failure load and initial damage in monolithic, partially veneered, and completely veneered (translucent) zirconia cantilevered fixed partial dentures (CFPDs), as well as completely veneered metal-ceramic CFPDs under different support and loading configurations. Materials and methods Eight test groups with anatomically congruent CFPDs ( n = 8/group) were fabricated, differing in CFPD material/support structure/loading direction (load applied via steel ball (Ø 6 mm) 3 mm from the distal end of the pontic for axial loading with a 2-point contact on the inner cusp ridges of the buccal and oral cusps and 1.3 mm below the oral cusp tip for 30° oblique loading): (1) monolithic zirconia/CoCr abutment teeth/axial, (2) monolithic zirconia/CoCr abutment teeth/oblique, (3) partially veneered zirconia/CoCr abutment teeth/axial, (4) partially veneered zirconia/CoCr abutment teeth/oblique, (5) completely veneered zirconia/CoCr abutment teeth/axial, (6) completely veneered CoCr/CoCr abutment teeth/axial (control group), (7) partially veneered zirconia/implants/axial, and (8) partially veneered zirconia/natural teeth/axial. Restorations were artificially aged before failure testing. Statistical analysis was conducted using one-way ANOVA and Tukey post hoc tests. Results Mean failure loads ranged from 392 N (group 8) to 1181 N (group 1). Axially loaded monolithic zirconia CFPDs (group 1) and controls (group 6) showed significantly higher failure loads. Oblique loading significantly reduced failure loads for monolithic zirconia CFPDs (group 2). Initial damage was observed in all groups except monolithic zirconia groups, and fractography revealed design flaws (sharp edges at the occlusal boundary of the veneering window) in partially veneered zirconia CFPDs. Conclusions Monolithic zirconia CFPDs might be a viable alternative to completely veneered CoCr CFPDs in terms of fracture load. However, oblique loading of monolithic zirconia CFPDs should be avoided in clinical scenarios. Design improvements are required for partially veneered zirconia CFPDs to enhance their load-bearing capacity. Clinical relevance Monolithic zirconia may represent a viable all-ceramic alternative to the established metal-ceramic option for CFPD fabrication. However, in daily clinical practice, careful occlusal adjustment and regular monitoring should ensure that oblique loading of the cantilever is avoided. Keywords Open Access funding enabled and organized by Projekt DEAL.

Acknowledgements The authors would like to thank the company 3M Oral Care for providing the ceramic materials and the adhesive luting cement used in this study and Mr. Robert Schnagl (3M Oral Care) for the scanning electron microscopic fracture analysis. We would also like to thank Mr. Philipp Doebert (3M Oral Care), who competently supported the project as dental technology contact. Author contribution W.B. and S.R.: conceptualization, methodology, data curation, supervision, formal analysis, data interpretation, and writing—original draft preparation; P.R.: resources, data interpretation, and writing—review and editing; P.B.: data acquisition (conducted the experiments), data interpretation, and writing—review and editing. All authors read and approved the final manuscript. Funding Open Access funding enabled and organized by Projekt DEAL. The study was self-funded by the Department of Prosthetic Dentistry of the University of Heidelberg. The ceramic materials and adhesives used in this study were provided free of charge by the manufacturer (3M Oral Care). The CFPD frameworks were milled free of charge by the manufacturer of the ceramic material (3M Oral Care). Declarations Competing interests The authors declare no competing interests. Ethics approval and consent to participate The use of human teeth for in vitro testing in this study was approved by the Ethics Committee of the Medical Faculty of the University of Heidelberg (S-034/2010), and tooth donors provided written informed consent. Conflict of interest The authors declare no competing interests. CoCr Cobalt-chromium, SD standard deviation, Min minimum, Max maximum Different superscript letters in the same column indicate statistically significant results between the test groups separately for the test forces at initial damage (lower case letters) and failure (upper case letters)

CC BY

no

2024-01-16 23:41:59

Clin Oral Investig. 2024 Jan 15; 28(1):94

oa_package/ab/e2/PMC10788321.tar.gz

PMC10788322

38221585

Introduction Climatic change is a continuous process that has happened since the first stable global climate system was formed and the first living place on Earth was developed (IPCC, 2013 ). Governments have agreed to move towards better sustainability through the Paris climate change agreement and the 2030 Agenda for Sustainable Development. Most recently, numerous nations made commitments to achieve net-zero greenhouse gas (GHG) emissions by the middle of the century. To achieve these commitments, long-term goals must be defined in order to balance present needs with those of the future. These goals serve as a solid foundation for tracking the progress made by each government. With increasing poverty and a lack of economic resources in developing countries compared to developed countries, developing countries are most vulnerable to climate change. Most developing countries suffer from immense challenges due to climate extremes such as heat waves, torrential rain, freezing temperatures, tropical cyclones, and floods. This is especially true in countries that have a high population density, limited resources, a reliance on non-renewable energy sources, and little ability to respond to these environmental threats (Rahman, 2018 ). Coastal zones are considered one of the areas with the densest populations and essential infrastructure due to their complex socio-economic systems. The majority of the world's population lives in these zones (Nelson, 2018 ). They occupy 20% of the land area and house 40% of the world's population (Rangel-Buitrago et al., 2020 ). Despite this significance, these zones are particularly vulnerable to global environmental changes, such as climate change. Climate change will pose threats to different components of coastal zones. One of the most serious threats posed by climate change in these zones is sea-level rise. Where about 7% of the world's population lives in flood-prone areas (Li et al., 2009 ). Increasing sea levels will lead to saltwater intrusion on land, coastal erosion, and exacerbated the effect of coastal storms (Woodruff et al., 2018 ). This leads to significant socio-economic effects such as population displacement, loss of properties, recession of economic and industrial activities, loss of coastal habitats, and loss of tourism, the recreation, and transportation functions (Torresan et al., 2008 ). In addition, increased temperatures will result in more heat waves, which will have an impact on human and ecosystem health, putting humans, animals, and plants in danger of illness and even death (CoastAdapt, 2017 ). Preserving coastal zones from the effects of climate change is a major challenge due to changing coastal conditions. In the past, when it came to development-related consequences on the coast, planners and managers would often create concrete barriers and shields to safeguard human settlements (French, 2004 ). However, due to variances in local geomorphic properties and changes in such qualities over time, these adaptation systems have often proven ineffective in the long run. The notion of 'integrated management' was born out of the necessity for long-term solutions to coastal management difficulties as well as a better understanding of dynamic coastal processes (McFadden, 2007 ). Measuring coastal vulnerability is essential in planning and managing coastal zones (Baučić et al., 2019 ). The vulnerability was described as “a nation's ability to cope with the repercussions of accelerated sea-level rise and other coastal effects of global climate change” (IPCC, 1992 ). Vulnerability assessment attempts to categorize coastlines into units with comparable features, establish the form of vulnerability (e.g., erosion vs. inundation), and offer a rating of probable coastline alterations. These assessments are often stated quantitatively, which frequently takes into account five categories (very low, low, medium, high, and very high) to simplify complicated and interacting factors that are utilized to influence coastal management. Coastal vulnerability to Sea Level Rise (SLR), wave erosion, and human effects has been the focus of several assessments (VanZomeren & Aeevedo-Mackey, 2019 ). Therefore, vulnerability analysis is a reasonable approach for assessing such crises because the information provided may be useful for policymakers to aid in the adaptation processes. Coastal zones can be protected from climate change impacts in two ways: mitigation and adaptation. Mitigation is an intervention to reduce GHG emissions sources or improve GHG sinks (IPCC, 2001 ). In comparison, adaptation is the adjustment process to the actual or expected climate and its effects to moderate harm or take advantage of beneficial opportunities (IPCC, 2014 ). It is determined according to the country’s economic position as well as the results of a vulnerability assessment based on the geographical location, the types of climate change impacts, and the intensity of negative environmental consequences. Hard engineering protection and soft engineering protection are the two forms of coastal adaptation management. Hard engineering protection is man-made structures that protect the shoreline from extreme and destructive natural processes. These structures help safeguard coasts by absorbing wave energy and reducing erosion and floods. These structures are costly, short-term solutions that frequently have a harmful influence on the environment, might have negative consequences farther down the coast, and diminish the ability of the coastline to respond naturally to changing conditions (Luo et al., 2015 ). Appendix 1 illustrates the hard adaptation structure used in coastal zones. On the other hand, soft engineering protection works with nature rather than against it to safeguard the shoreline. It employs ecological principles and practices, resulting in a lower negative influence on the natural environment. It is less expensive to develop and maintain than hard engineering projects and offers more long-term, sustainable solutions (Bongarts Lebbe et al., 2021 ). Appendix 2 presents the soft adaptation structure used in coastal zones. Therefore, this paper presents the development of a decision-making framework for assessing climate change impacts on coastal vulnerability and adopting a climate change adaptation protection system to coastal zones, overcoming existing limitations of currently utilized protections such as downdraft erosion and disrupting natural processes. Remote sensing (RS) time series are the main part of the tracking system for climate variability and changes (Allard, 2017 ). RS and Geographic Information System (GIS) might be combined to accomplish these goals. GIS provides a platform for organizing thematic maps in several forms (e.g., raster or vector data) and is used to execute logical and mathematical calculations during vulnerability assessments. GIS is recognized as a decision-support system that integrates geographically referenced data in a problem-solving context (Cowen, 1990 ). It combines many types of data into accessible formats, evaluates and analyzes data, and provides descriptive and predictive modeling of various scenarios. Mitigating the consequences of disasters such as climate change necessitates real-time access to disaster-related information. The susceptibility of coastal zones may be recognized using environmental satellites by monitoring the present situation—before, during, and after a catastrophe. GIS methods provide an appropriate framework for integrating and evaluating the various data sources needed for climate change monitoring. A decision support system (DSS) is used to aid in decision-making, judgment, and action. A DSS, besides GIS and RS, is responsible for filtering, and analyzing massive amounts of data, generating precise information that may be used to solve problems and make choices. It is critical to describe the temporal and spatial evolution of climate change vulnerability and its driving factors in order to cope with it (Jiang et al., 2021 ). The Analytic Hierarchy Process (AHP) is one of the MCDA techniques to detect the importance and relative weight. Several MCDA approaches have been used to assess coastal vulnerability due to environmental and human hazards. It is a decision-making tool that is used to provide a reasonable framework for making an informed decision, considering various indicators or criteria while prioritizing and selecting (quantifying) choices (Saaty, 1977 ). Bagheri et al. ( 2021 ) used AHP for coastal city hazard management, and Maanan et al. ( 2018 ) assigned AHP and principal component analysis (PCA) to help them assess coastal vulnerability to environmental hazards. While Boulomytis et al. ( 2015 ) utilized Compromise Programming (CP) to evaluate the watershed’s vulnerability to floods in a future scenario, the AHP is the most commonly used and broadest process among the MCDA techniques (Bagheri, et al., 2013 ; Yannis et al., 2020 ).

Results AHP model results According to the research parameters, weights were allocated to various indicators based on experts’ assessments. They were computed for each of the three parameters’ indicators. A consistency check is required since weights are only valid if they are acquired from consistent or close consistent matrices (Bagheri et al., 2021 ). The consistency of the judging matrix is determined by examining the total Consistency Ratio (CR) it is based on the values of the greatest eigenvalue (λmax) and the Consistency Index (CI). The results showed that the CR is less than 0.1, as listed in Table 3 . This indicates that the matrixes are reasonably consistent and the errors are acceptable. Decision-makers can use the indicators weights for further calculations and analysis. CCIA model results The CCIA model was tested with datasets spanning 30 years. The impacts of the study parameters and each indicator in the four assessment parameters were grouped into five classes, from very low to very high, for areas damaged by climate change. According to the results shown in Table 4 , the research hypothesis is valid, as the results of the proposed model showed that there is a variation in the degree of vulnerability in the study area. Based on the simulated datasets, the results present the sensitivity analysis of climate change impacts on the overall model. Six scenarios are devised to examine the most critical parameters of climate change affecting the study area. These scenarios are as follows: 1) a base-run scenario able to study the model performance under equal weight indicators without dividing them by its parameters, considering equal weight for all; 2) a base-run scenario able to study the model performance under equal weight parameters, considering the relative weights of indicators that were determined using AHP; 3) a topographical structure scenario; 4) a shoreline scenario; 5) a meteorological data scenario; and 6) a geology of a city scenario. Finally, the most critical scenario is compared to the general strategic plan of Al Alamein New City to recommend the critical places that must be adapted to climatic changes. Figure 6 illustrates the output of the overall model according to the six simulated scenarios. The estimated vulnerability degrees of six scenarios are shown in Table 4 . The outcomes of six scenarios show that the shoreline is where the consequences of climate change are most noticeable. In the first scenario, the moderately sensitive area is the largest area. Furthermore, the second largest region is made up of highly sensitive places. The highly susceptible area occupies the majority of the resulting vulnerable areas and is dispersed along the shoreline in the second scenario. The second-greatest area, however, is filled by moderately sensitive areas. According to the third scenario, which places the most emphasis on topography, there are four levels of susceptibility, ranging from low to very high. The sensitive zones of medium and high degree make up the majority of the city's vulnerable area. According to the fourth scenario, which focuses on geology, the largest area in danger, which accounts for the majority of the land, is highly vulnerable. The fifth scenario, which places the most emphasis on metrological data, is very similar to the effect of topography, where the largest inferred areas of vulnerability are medium and high vulnerabilities. The last scenario, which focuses on shoreline, has the same effect as the topography and metrological data scenarios. However, as indicated in Table 4 and Fig. 6 , the three scenarios differ in terms of vulnerability degrees and their locations.

Discussion According to Table 4 , Scenario 2 has the highest risky areas compared to scenario 1, estimated at 1.12 km 2 . This scenario considers the actual weights of the indicators based on experts’ judgments. Because of the variable nature of indicators, it is impossible to balance the relative weights of all variables, and hence the conclusions derived from experts’ experiences are more accurate. By comparing the results obtained from the geology parameter scenario (Scenario 4) and the equal weights parameters scenario (Scenario 2), it came to a consensus on the locations of the region's high and highly susceptible spots. Refereeing to the results of the four sub-model parameters priorities (i.e., Scenarios 3, 4, 5, and 6) the most critical parameter of climate change that causes highly vulnerable areas in the coastal zone is the region’s geology, which constitutes the largest risky area of 1.7 km 2 . As a result, when developing new urban communities in coastal cities, it is critical to first preserve the integrity of the region's geology while also considering other issues. This finding agrees with Culshaw and Price ( 2011 ), who claim that geology is the most essential component in urban development and regeneration. Geologic hazards are responsible for significant loss of life and property destruction. As a result, studying a region’s geology helps to manage geologically based hazards that affect people and institutions as they interact with their built and natural environments. This result is an answer to the research question about the degree of climate change indicators effects on vulnerable area formation, where each indicator has its effects according to its values and importance. The second parameter affected by climate change is the topographical structure of the region, which constitutes the second-largest risky area at 0.81 km 2 . The results are normal outcomes since coastal areas with high slopes and low elevation are considered great danger areas. This result asserts that the steep slopes of coastal locations are regarded as one of the primary reasons that put these areas in great danger, such as storm surge flooding and inundation (McInnes et al., 2006 ; Torresan et al., 2008 ). Low elevation values are also associated with very risky locations that are regarded as most vulnerable to inundation and flooding due to their proximity to or below sea level. On the other hand, areas of high elevation have a strong potential to resist flooding caused by increasing sea levels and storm surges (Mani Murali et al., 2013 ). The third one affected by climate change and its effect on the development of new communities is the metrological data parameter, which constitutes 0.7 km 2 risky areas. This impact resulted from Al Alamein New City being an inactive city without significant human, industrial, or economic activity. As a result, it faces modest problems in temperature, sea level pressure, precipitation, dew point temperature, and wind during the simulation time. Furthermore, this weak significance is due to the fact that the smaller the scale, the more sea levels and vulnerability to coastal disasters is determined by variables other than climate (Bongarts Lebbe et al., 2021 ). The degradation of the shoreline is the least important characteristic considered when building new urban communities since it covers the smallest risky area and is estimated to be 0.65 km 2 . Despite the critical relevance of the magnitude of the shoreline region's deterioration, this deterioration is the consequence of both the influence of metrological data on the region and the impact of human activities. If climate conditions are regulated while limiting the effect of human activities, shoreline deterioration will be reduced. The shoreline is considered one of the dynamic landforms that constitute the coastal environment, it changes due to accretion and erosion processes caused by sea level rise, wave action, and sediment transportation (Kaliraj et al., 2015 ). In addition, erosion and accretion are closely related to climate change, as are the nature of the geological formations on the coastline and the movement of waves (El-Said, 2020 ). According to its general strategic plan, Al Alamein New City was built to rejuvenate unused locations to attract investors, employees, and residents. tourist attractions, archaeological areas, industrial sites (Hamra port, Petrojet site, and solid waste facility), restricted sites, and utility stations are all existing uses inside the city boundaries, as shown in Fig. 7 . However, the city’s general strategic plan is intended to include multiple sustainable uses of the land and existing uses to achieve the city’s strategic objectives, as shown in Fig. 8 . By comparing the general strategic plan with the city’s coastal vulnerable locations and taking into consideration the prohibited activities in the coastal zone under Water Resources and Irrigation Law No. 147 of 2021, a buffer zone was created using a GIS environment at a distance of 200 m from the shoreline to the land to detect the uses located in the prohibited area as well as those located in highly vulnerable sites due to climatic changes. There are five strategic locations at risk: 1) Hamra port; 2) future expansion areas that may include a tourist and residential extension; 3) tourist activities in the North Coastal Road that include commercial services, conference centers, parks, museums, and sports activities to serve the tourist villages scattered along the coast to prolong the tourist season; 4) areas for special uses such as the establishment of a zone for technological industries in Al Alamein; and 5) tourism activities in East Marina Al Alamein as well. As a result of the hazards of climate change, the creation of these communities on the shoreline must be prohibited, necessitating an adequate adaptation plan to safeguard the city from numerous harmful consequences. According to research results, there is a need to include adaptation and mitigation measures in the planning of communities to safeguard the city against the destructive impacts of climate change and reduce future vulnerability along the shoreline. Figure 9 depicts the areas in Al Alamein new city, which are designated as zones 1, 2, and 3. These zones should have priority in protection because they contain important strategic locations, namely Hamra Mina and future expansion areas for tourist and residential communities. The breakwater hard protection system is the defensive device used to defend Al Alamein New City’s beaches. Because natural processes such as waves and artificial barriers around the shoreline interact complexly over time, relying on this system alone to safeguard the shoreline from climate change impacts, particularly SLR, has a negative long-term impact. The conclusion drawn from evaluating how these barriers affect the most vulnerable zones indicated earlier is that sand moves around the downdrift headland of breakwater barriers due to ongoing wave motion. As shown in Fig. 9 , barriers that significantly reduce littoral drift, alongshore sediment movement and beach material erosion will also induce sediment retention behind the barriers. There will be more erosion and accretion regions along the shoreline due to the waves’ continued interactions with the barriers and the change in their intensity over time, which raises the danger of vulnerable areas (Barton & Brown, 2019 ). This is consistent with the results of El-Masry et al. ( 2022 ) study, which found that the existing adaptive measures in the El Hammam—EL Alamein area are insufficient to deal with future climate change, SLR, and other natural risks like storms and tsunamis. Therefore, this paper recommends developing a hybrid adaptation system that uses a combination of soft protection and hard infrastructure to provide effective and long-term adaptation to the effects of climate change in the study area. It has the ability to defend the shoreline against SLR, erosion, and other negative effects of climate change. In addition, it is more cost-effective in the long run than hard infrastructure alone (Bongarts Lebbe et al., 2021 ; Sutton-Grier et al., 2015 ). Choosing a combination of protection systems requires considering the balance of recreational access, a lifetime of systems, and budget constraints. According to the advantages and disadvantages of the artificial and natural protection systems for the beaches listed in Appendices 1 and 2, a hybrid adaptation system can be recommended for the study area. The most viable options for protecting the Al Alamein New City's shoreline are using a combination of breakwaters as hard engineering protection and afforestation of coastal dunes as soft engineering protection. Breakwaters absorb wave energy during storm events, reducing the consequences of storm surge and coastal erosion. The chosen natural protection can overcome the erosion and accretion caused by these barriers. Coastal dunes have the potential to protect cities from storm surges and wave attacks. They offer a natural particulate barrier that erodes during storms, dissipating wave energy. Planted sand dunes help to improve the dune's capacity to resist erosion.

Conclusions Spatial vulnerability maps due to climate change impacts for Al Alamein New City have been generated according to the developed decision-making framework. The proposed framework overcomes the problem of recognizing the destructive consequences of climate change on coastal communities due to its ability to evaluate the studied parameters with their indicators in the formulation of risky vulnerable zones from a holistic perspective. The parameters include meteorological data, topographical structure, engineering geology, and shoreline. Vulnerable areas susceptible to these parameters are identified along the study coastline. These areas are located in three portions of the study area: Sidi Abd El-Rahman, Al Alamein City, and Tel Al-Eis. The risks involved range from modest to very high. Based on the relative relevance of each vulnerable parameter, the geology of the region is the most essential parameter influenced by climate change due to its propensity to form a highly sensitive area in the coastal zone. As a result, developing new urban settlements in coastal areas demands first conserving the integrity of the region's geology while also taking into account other challenges, such as coastal topography, meteorological data, and shoreline. Although the proposed framework’s vulnerable areas range from low to very high, they must be carefully equipped to deal with future climatic changes and be suitable for the displacement of the neighboring region's citizens. To adapt to the long-term effects of climate change on coastal zones in a sustainable manner, a combination of soft and hard protective systems should be considered. The paper recommends a hybrid adaptation system for the examined area. It combines breakwaters as hard engineering protection and afforestation of coastal dunes as soft engineering protection to overcome current limitations of presently utilized protections such as downdraft erosion and interrupting natural processes.

The world is currently confronting one of its biggest environmental challenges: combating climate change. Coastal zones are one of the areas thought to be most sensitive to current and future climate change threats. The paper integrates Remote Sensing (RS), Geographic Information System (GIS) techniques, and Multi-Criteria Decision Analysis (MCDA) to detect vulnerable areas from climate change impacts in coastal zones in order to recommend adaptation systems in new coastal zones that can withstand various climatic changes. The proposed decision-making framework was developed in three phases: 1) climate data collection and processing; 2) Coastal Climate Impact Assessment (CCIA) model development; and 3) implementation and adaptation system selection. The climate data collection and processing phase involves determining the most significant climate change parameters and their indicators that affect coastal zone stability, extracting climatic data indicators from different climate database sources, and prioritizing the selected indicators. The indicators’ weights were estimated using the Analytical Hierarchy Process (AHP) through a questionnaire survey shared with experts in climate change impacts. A CCIA model development phase involves the formulation of the proposed model using GIS technique to discover the vulnerable areas according to the most dominant impact. The implementation and adaptation system selection phase involves the application of the framework to Al-Alamein New City in Egypt. A sensitivity analysis was conducted to measure the behavior of several climate change parameters to identify the most critical parameter for climate change in Al-Alamein New City. The results showed that the geology of the region is the most crucial component influenced by climate change. It is capable of producing a very sensitive area in the coastal zone while also taking other factors into account. When creating new urban neighborhoods, the erosion of the shoreline is the least important factor to consider. This is because coastal deterioration is caused by both the influence of metrological data on the region and the impact of human activity. Shoreline deterioration will be reduced if climate conditions are maintained while limiting the impact of human activities. To adapt to the long-term effects of climate change on coastal zones, a combination of soft and hard protection systems should be considered. Keywords Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB).

Literature review Several research efforts have been devoted to assessing coastal vulnerability to climate change impacts using different methodologies. Satta ( 2014 ) established an Index-based method for the integrated analysis of coastal risk to multiple hazards that considers the impacts of SLR, storm variability, and human-induced forcing. The author identified population increase and tourism development as the most significant human-induced causes, with coastal erosion, coastal flooding, and seawater intrusion as the most significant natural hazards in Mediterranean coastal zones. The study produced vulnerability and risk maps for every single hazard and multiple hazards .Sanuy et al. ( 2018 ) applied the Bayesian network-based approach to evaluate coastal risk by predicting risks at the receptor scale, which were converted into impacts through vulnerability relations. The proposed framework has helped analyze storm-induced risks and strategies significantly. Haugen et al. ( 2018 ) proposed a generic framework to monitor climate change impacts on historic buildings and their interiors to build a data-driven decision-making process. The study asserted that effective monitoring could be achieved by systematically using more detailed images. Sahoo and Bhaskaran ( 2018 ) developed a study to investigate the physical, environmental, social, and economic impacts on coastal vulnerability associated with tropical cyclones. Weather parameters, along with storm surge height and onshore inundation, were used to estimate the Physical Vulnerability Index (PVI) using the GIS-based approach. El-Shahat et al. ( 2021 ) evaluated the vulnerability of African coasts to SLR. The authors applied the Coastal Vulnerability Index (CVI) method using GIS and the RS approach. The study has customized CVI to include seventeen parameters that represent the physical, social, and economic features of coastal zones. Vieira et al. ( 2021 ) used GIS and RS to assess coastal erosion from climate change, taking into consideration seven factors: elevation, geomorphology, geology, land cover, anthropogenic activities, distance to the shoreline, and maximum tidal range. The authors considered the CVI method for vulnerability to erosion evaluation. Rahmawan et al. ( 2022 ) examined six parameters (geomorphology, elevation and slope, land use, rigid structure, and coastline changes) to evaluate coastal vulnerability using GIS-based and scoring assessments. The coastline change is quantified based on the Modified Normalized Difference Water Index (MNDWI). The scored parameters are then analyzed using CVI. Palacios-Abrantes et al. ( 2022 ) developed the Nature Futures Framework (NFF) by the Task Force on Scenarios and Models of the Intergovernmental Science-Policy Platform on Biodiversity and Ecosystem Services as a heuristic method. The authors concluded that a variety of different viewpoints on people's attitudes toward the environment might lead to decision-making that is adaptable and resilient in dealing with the effects of climate change on social-ecological systems. El-Masry et al. ( 2022 ) investigated the potential effects of climate change and sea-level rise on the vulnerability of the Mediterranean coastal of the El Hammam—El Alamein. The assessment of the area's and coastal tourism's vulnerability to climate change and SLR through the development of a digital elevation model (DEM) and inundation models, as well as the assessment of temperature change using the tourist climate index. According to the study, most seaside resorts will be submerged. Roy et al. ( 2023 ) investigated the vulnerability of the potential impact of climate change on SLR and coastal habitats based on the CVI method. Their study characterizes the physical setting, including geomorphology (G), sea level change (SLC), coastal slope (CS), relative sea-level change (RSLC), mean wave height (MWH), mean tide range (MTR), shoreline change rate (SCR), land use and human activities (LU), and population (P). The study concluded that vulnerability assessment can help the decision-maker for consider the most appropriate development strategies to maintain the sustainable development of coastal ecology. Prior literature efforts analyzed coastal zone vulnerability to climate change. However, such efforts did not consider the results of hard protection systems on the dynamic relationship between climate change indicators in their natural state (Cabana et al., 2023 ; Gargiulo et al., 2020 ; Wei et al., 2021 ). Also, previous research studies lack the representation of weights using MCDA approaches such as the AHP for climate change indicators to evaluate, rank, and define the relative weights and relevance of each climatic criterion (Allipour Birgani et al., 2022 ; Thirumurthy et al., 2022 ). Most aggregated climate-change vulnerability scores using the classic MCDM approach such as a weighted sum method (WSM) for key indicators (Kim & Chung, 2013 ). The spatial influence of all indicators affecting the shoreline, local weather, topographical structure representing the earth's form, and zone geology in defining the vulnerable areas due to climatic changes was not considered. The vulnerability of coastal zones to be suitable for the construction of sustainable new communities was not studied as a result of the effect of each parameter separately, and the comparison between these results and alternative scenarios for other parameters was not taken into consideration. Previous research studies did not investigate the sensitivity of relative weights and degrees of importance of variables in transforming the form, area, and locations of coastal zone vulnerability (Azab, 2022 ). The hypothesis of this paper is that climate change impacts have negative effects on any coastal zones, and these effects are represented in threatened areas with different degrees of vulnerability. To achieve the desired goal and verify the paper hypothesis, this research proposes a decision-making framework capable of evaluating the impact of climatic change on the vulnerability of coastal zones. This aids in adopting a protection system for climate change in coastal communities. The proposed framework contributes to resolving the issue of recognizing the harmful impacts of climate change on coastal zones and maintaining their long-term survival. Decision-makers will benefit from this framework and its findings as they choose the most suitable solutions to climate change adaptation. The framework combines RS, GIS, and MCDA analysis to plan new sustainable coastal communities that can resist various climatic changes. Considering the above clarification of climate change problems in coastal zones, systems to address them, and the gaps in previous studies, this paper attempt, through the proposed framework, to answer the question, do the climate change indicators have the same effect in the formation of vulnerability areas of their different degrees? Proposed decision-making framework The developed framework helps government agencies and/or contractors integrate RS, GIS, and MCDA. It has great potential for coastal vulnerability assessment to provide an accurate indication of climate change's impact on the dynamic environmental systems of coastal zones. This leads to choosing the most suitable coastal adaptation system to resist climate change impacts and reduce coastal vulnerability to current and projected climate changes. RS and GIS are important for generating baseline data on land cover shifts in coastal zones. They are used to capture, explore, manage, and model all data forms that are geographically referenced. The integration of RS and GIS techniques helps to generate a descriptive and predictive Coastal Climate Impact Assessment (CCIA) model for more accurate monitoring and decision-making. The proposed framework’s application follows a procedure consisting of three phases: climate data collection and processing, CCIA model development, and implementation and adaptation system selection (see Fig. 1 ). Climate data collection and processing Identification of the most significant climate change indicators that impact the stability of coastal zones is performed in three steps. The first step concerns identifying the main indicators of climate change according to the literature. To evaluate the effect of SLR in coastal zones, the sea level pressure, the atmospheric pressure at sea level, is used to indicate the sea level. The behavior of sea level changes by changing the value of sea level pressure as the inverse barometer effect, as the higher the pressure, the lower the sea level, and vice versa. (Wunsch & Stammer, 1997 ). Marzouk et al. ( 2021 ) identified the most affected climate change indicators on the suitability of coastal zones, which are divided into four main effective groups: meteorological, topographical structures (Earth Shape), engineering geology, and shoreline parameters (see Table 1 ). The second step is extracting data for each identified indicator based on the studied area of the research from various climate database sources, such as RS sensors and climate stations. RS is able to sense, map, and monitor changes in the coastline because they are recognized as main sources of trustworthy and durable data for all circumstances needing spatial data at a variety of temporal and geographical scales. The Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data of Global Digital Elevation Model (GDEM) at a horizontal spatial resolution of 30 m is used to create detailed maps of topographical characteristics such as coastal slope and coastal regional elevation. A geologic map is used to study engineering geology, or the composition of the earth’s components. Finally, climatic data are gathered from two climate databases based on daily observations. The final step involves developing a scaling questionnaire and decision-making model from climate change indicators to evaluate, rank, and define each indicator’s relative weights and importance for performing a realistic representation of weights. The AHP model has been developed as an MCDA technique for pairwise comparison to determine priorities and relative weights of indicators. Analytic Hierarchy Process (AHP) model The relative importance of the indicators and their weights must be determined to integrate all of the parameters and their indicators into a single spatial model. This is deemed important to determine the vulnerability areas due to climate change in coastal zones to promote a suitable development for new communities and determine the most influencing parameter. This paper uses the AHP as one of the MCDA techniques for prioritizing the various indictor and ranking them as per their weights. This is according to a questionnaire survey focusing on expert knowledge about the impacts of climate change and weather conditions to obtain a preference degree for each impact of climate change according to the following procedure: Data collection using questionnaire experts survey The AHP technique represents a problem and then develops priority scales for gauging qualitative performance using numeric scale calibration (Saaty, 1977 ). The first stage in the AHP approach is to break down the problem into smaller components, starting with the goal, which is the climate change impacts assessment, which is decomposed into research parameters, and finally the indicators. The four main parameters of this paper with 11 indicators are meteorological data; topographical structure (earth shape); engineering geology; and shoreline suitability (see Fig. 2 ). The questionnaire was distributed and collected within a two-week timeframe. The questionnaire is divided into four sections: basic information, introduction, informed consent form, and questions about the level of importance of various factors (i.e., the importance degree of Factor 1 (F1) when compared to Factor 2 (F2) according to A Saaty’s scale) (Saaty, 1985 ). The scale is used to form pairwise comparison matrices at all levels. For example, the pairwise comparison matrix in the first level includes meteorological data, topographical structure (earth shape), engineering geology, and shoreline suitability. The expert choice software estimates the weights of the parameters’ indicators. The questionnaires were administered online and face-to-face to a diverse group of climate specialists. Fifteen Egyptian experts and specialists have more than 15 years of experience, according to their responses from various governmental agencies and authorities, such as the National Authority for Remote Sensing and Space Sciences (NARSS), Ministry of Environment, Egyptian Environmental Affairs Agency (EEAA), National Research Center, Faculty of Urban and Regional Planning, Egyptian Meteorological Authority (EMA), National Institute of Oceanography and Fisheries, Ministry of Water Resources and Irrigation, and Institute of National Planning (INP). Coastal climate impact assessment model development A CCIA model is developed using RS and GIS techniques. A more in-depth explanation of the CCIA model may be found somewhere (Azab, 2022 ; Marzouk et al., 2021 ). The CCIA model has been developed according to four steps: Step 1: Modeling the problem This step is very important in modeling the climate indicators data, which aims to specify all aspects of the problem, such as defining the objectives, time horizon, and decision variables. The study period is taken to be 30 years, according to the World Meteorological Organization ( 2017 ). It is divided into equal periods (i.e., 5 years for each period) starting in 1988 and ending in 2018 to monitor and predict changes to the climate. Step 2: Formulation problem sub-models This step aims to convert the study problem into sub-models and perform the necessary processing of each climate change indicator using a GIS environment. Each sub-model is manipulated in the GIS environment before being integrated into a single model. To run in the model, all sub-models are turned into raster datasets. Each layer or point must be turned into a raster dataset before being integrated into a single model. For example, Fig. 3 depicts the shoreline sub-model from the CCIA model. It includes two sub-models: erosion and accretion. The output is the rate of shoreline change (positive for accretion and negative for erosion from land). Step 3: Reclassify datasets The third stage is to reclassify the values of each sub-model of a sequence of input vector data to a common scale, assigning maximum numbers to the most harmful characteristics and lower values to the least harmful attributes. Step 4: Weighted overlay The fourth step is a weighted overlay analysis of all sub-models in the map combination according to the importance degree of each sub-mode. This enables the computation of a multiple-criteria analysis between many imagery rasters and the control of the effect of various indicators in the suitability model. Each sub-model has a proportional impact on the total assessment based on how important it is. All sub-models are equally weighted at 100%. For the importance degree of each sub-model, the relative weights of sub-models obtained from the AHP approach are incorporated into a GIS to generate the vulnerable areas or hotspots most susceptible to climate change. Also, this paper uses equal weights for all sub-models as a base scenario to examine each parameter’s effects that constitute the vulnerable areas. The procedure for formulating and analyzing the proposed CCIA model is depicted in Fig. 4 . Implementation and adaptation system selection This phase is concerned with controlling the consequences of climate change and adapting the ecological system in response to real or anticipated climatic stimuli and their effects or implications. Validation of the developed model’s performance is necessary to build confidence in the proposed CCIA model. In this phase, a real-life case study is worked out to demonstrate the use of the proposed CCIA model. A sensitivity analysis is conducted to choose the most critical parameters affected by climate change that have the potential to form the most vulnerable areas. It provides an MCDA model for describing the uncertainty linked to the effects of climate change indicators. Sensitive Analysis is a quantitative model to assist decision-makers in their decision-making processes. An appropriate adaptation system should be chosen to develop new sustainable communities in coastal zones. The challenge of coastal adaptation to climate change necessitates establishing a primary goal that can influence decision-making. The basic goal of the challenge is to define the most critical parameter influenced by climate change, which has the potential to weaken coastal communities if not taken into account. The investigation of the affected degree of climate change parameters to detect the riskiest areas in the coastal zone is achieved through testing the sensitivity of critical parameters. It is conducted by considering different weights for the parameters and their indicators, forming different scenarios. These scenarios are simulated using the CCIA model. Six scenarios are devised to examine the behavior of each indicator and parameter in the final output. Comparative analysis is performed using a GIS environment to explore the vulnerable strategic locations in the study area, which match the weak spots inferred from the case study. After conducting these tests, decision-makers can decide on the best scenario for the riskiest areas. Finally, it suggests an adaptation system for the study area towards climate change. CCIA model implementation The proposed CCIA model is conducted in Al Alamein New City in northern Egypt. Al Alamein New City is located on the North Coast between latitudes of 30° 49′ 48.25" N and longitudes of 28° 57′ 18.07" E. It is located within the geographical limits of the governorate of Marsa Matrouh. It covers an area of approximately 227.65 km 2 . It is being developed to be a residential and tourist city. In addition, it contributes to reducing the problem of unemployment by providing 1.5 million job opportunities (Associated Consultants, 2015 ). It is divided into three divisions based on the administrative borders established by the Central Agency for Public Mobilization and Statistics (CAPMAS), as seen in Fig. 5 . The first sector is Sidi Abd El-Rahman, which is located in the city's west and accounts for around 6.5% of the overall area. The second sector is Al Alamein City, which is located in the city’s east and accounts for approximately 65.2% of the overall area. The last component is Tel Al-Eis, which is positioned between the two preceding sections and accounts for around 28.3% of the city’s total area. The strategic objectives for developing this new city are as follows: 1) increasing national and regional income; 2) reducing population density in existing cities; 3) offering diverse job opportunities; and 4) creating a livable city where the residents will have an adequate standard of living (e.g., research, logistics, etc.). As for the implementation of the CCIA model, all indicator datasets have been created and included in the developed CCIA model as a raster layer for each indicator. For shoreline spatial data, seven multispectral Landsat images were downloaded. Pre-processing and processing have been conducted on the extracted images using ENVI 5.1 software. The mosaicking of images was not necessary since the entire study area fell within a single scene. After classification, the shoreline for each image has been extracted, and erosion and accretion have been calculated using ESRI ArcGIS 10.4.1 software. For topographical structure, slope and elevation have been estimated and mapped from ASTER based DEM of Al Alamein New City with 30 m spatial resolution. The data on meteorological indicators: surface air temperature, precipitation, sea level pressure, and dew point temperature, have been extracted from the databases of the Marsa Matrouh climate station from January 1988 to December 2018. The data extracted from climate stations is displayed in the ArcGIS environment in the form of a point for each year, with its value and coordinates. Because climate change is monitored every 5 years in this study, each period for each indicator is represented by only five points. Different methods can be used for the interpolation values between the available points, whether deterministic or stochastic methods, such as the Thiessen (Voronoi) polygons, the nearest neighbor, the Kriging, and the spline methods. Choosing one method over the other depends on the distribution of sample points and the parameter being studied. For the Thiessen (Voronoi) polygons, it is considered the simplest method, which uses only one point to interpolate the nearest sample location without considering all other points used to interpolate the unknown point. Also, the nearest neighbor method is utilized to determine the nearest neighbor index, which relies on the average distance from each point to its nearest neighboring point (Sibson, 1981 ). This method can be used with a small number of input variables, but for a large number of variables, the method struggles to produce the output of the interpolated points (Childs, 2004 ). Kriging is a geo-statistical interpolation method that uses the spatial continuity of the data, relies on the spatial only without considering the actual values, and does not use any of the point values, owing to interpolated values being lower or higher than actual values (Arun, 2013 ). The spline method depends on the interpolation process of a two-dimensional minimum curvature spline technique. It is useful for reducing the surface curvature, which results in a smooth surface (Hutchinson, 1988 ). It cannot be used when the spatial characteristics of sample points are close together and have extraordinary contrasts in their values because this method utilizes slope calculations (change over distance) to figure out the final raster surface (Childs, 2004 ). While the Inverse Distance Weighted (IDW) interpolation method is an exact method to determine the depth values at un-sampled points by averaging the values of sample data points weighted by an inverse function of the distance from the point of interest to the sampled points (Li & Heap, 2008 ), It does not need to make any assumptions about the input data. It helps to interpolate dense, evenly spaced cell points, such as flat areas with cliffs. Also, it is considered that if the distance between the points increases, the influence of the cell will decrease on the output value (Childs, 2004 ; Huang et al., 2011 ). It is used to interpolate different cell values of many phenomena, such as soil, geology, oceanography, and meteorology (Papari & Petkov, 2009 ). Therefore, to predict the different values around these points in this research, IDW interpolation is used. According to the location of sensors yearly, weather points have been signed, and the points between these points for each period have been interpolated using the Inverse Distance Weighted (IDW) method. Using the boundaries of Al Alamein new city, the weather maps have been developed inside the city boundaries only as raster images. For wind speed and direction, Al Alamein new city boundaries are located between four climate stations, which are Al-Alamin, Marsa Matruh, El-Dabaa, and Siwa (as listed in Table 2 ). The data from four stations has been extracted every 5 years in net CDF file format and then mapped. The IDW technique was used to interpolate the values of wind speed and direction between the four points each time. The interpolated raster for each layer has been approximated for the values of speed and direction. By choosing to create a fishnet based on the resultant IDW, a grid with regular points has been created, where each point in this grid has a speed value and the location represents its direction. Using the boundaries of Al Alamein new city, the wind maps have been developed inside the city boundaries only as raster images. For the composition of the earth, the geological map of the city has been extracted from the shapefile of Egypt’s geology. This shapefile is converted into a raster map and inserted into the proposed CCIA model.

Appendix 1 Table 5 . Appendix 2 Table 6 . Author contribution Conceptualization: Mohamed Marzouk and Shimaa Azab; Methodology: Mohamed Marzouk and Shimaa Azab; Validation: Mohamed Marzouk and Shimaa Azab; Formal analysis: Shimaa Azab; Investigation: Shimaa Azab; Writing—original draft preparation: Shimaa Azab; Writing—review and editing: Mohamed Marzouk; Visualization: Shimaa Azab; Supervision: Mohamed Marzouk; Project administration: Mohamed Marzouk. Funding Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB). Data availability Not applicable. Declarations Ethical approval Not applicable. Consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:59

Environ Monit Assess. 2024 Jan 15; 196(2):147

oa_package/69/e1/PMC10788322.tar.gz

PMC10788323

0

Introduction The simultaneous capture of two bound atomic electrons followed by the emission of two neutrinos plus X-rays or Auger electrons, known as two-neutrino double-electron capture (2 ECEC), is a nuclear process allowed in the Standard Model. Compared to the two-neutrino double-beta ( ) decay, the simultaneous emission of two electrons and two anti-neutrinos, 2 ECEC processes have lower probabilities due to the smaller phase space, therefore experimentally, they are much more challenging to observe. The first direct observation of 2 ECEC was made only in 2018 by the XENON1T experiment with Xe [ 1 ]. Previously, indications of 2 ECEC were found in geochemical measurements with Ba and Ba [ 2 ] and in a large proportional counter experiment with Kr [ 3 ]. The lepton number violating counterpart of 2 ECEC, the neutrinoless double-electron capture (0 ECEC), in which no neutrinos are emitted, is also predicted [ 4 ]. This process must be accompanied by the emission of at least another particle to ensure energy and momentum conservation. Different modes can be considered in which 0 ECEC is associated with the emission of different particles like pairs, one or two photons, or one internal conversion electron [ 5 , 6 ]. In analogy with the neutrinoless double-beta ( ) decay, the 0 ECEC violates the lepton number symmetry by two units and implies that neutrinos have a Majorana mass component [ 7 ]. Although the sensitivity of 0 ECEC processes to the Majorana neutrino mass is estimated to be many orders of magnitude lower than that of the decay, the interest in 0 ECEC is theoretically motivated by the possibility of resonant enhancement when the parent nucleus and an excited state of the daughter nucleus are energetically degenerate [ 4 , 6 – 9 ]. In this case, the half-life of 0 ECEC processes becomes comparable to that of decays. Experimental searches for 0 ECEC have been performed by double- decay experiments, even though with less sensitivity compared to the search for decay [ 6 ]. The GERmanium Detector Array ( Gerda ) experiment, whose main goal was to search for the decay of Ge [ 10 , 11 ], operated enriched high purity germanium detectors in a liquid argon (LAr) cryostat, which naturally contains the Ar isotope with an isotopic abundance of 0.33%. Ar can undergo 2 ECEC to the ground state of S [ 12 ]. The corresponding lepton number violating process, 0 ECEC, may occur via the simplest radiative mode 1 The Ar nucleus captures one electron each from its K- and L-shells and turns into S. Two X-rays are emitted, with energies E keV, and E keV, corresponding to the capture of the electrons from the K- and the L-shell, respectively. Given the available energy of the decay Q keV [ 14 ], the corresponding energy for the ray is E keV. Resonance enhancement of the process is not possible for Ar [ 6 ]. In the light neutrino exchange scenario, assuming a Majorana mass of 0.1 eV, the half-life of Ar 0 ECEC is predicted in the order of year, with calculations based on the quasiparticle random-phase approximation (QRPA) [ 13 ]. Experimental searches for 0 ECEC of Ar have been performed since the early stages of the Gerda experiment [ 15 ]. The most stringent limit to date on the Ar 0 ECEC half-life is T year (90% CI), established in Phase I of the Gerda experiment [ 16 ]. More recently, this process has been searched with the DEAP detector [ 17 ], although with less sensitivity than Gerda Phase I. In this paper, we report on the search for the 429.88 keV line from the Ar 0 ECEC with the whole Gerda data, accumulated for a total live time of 3.08 year during Gerda Phase II and 1.26 year during Gerda Phase I.

Analysis methods The energy region used to set a limit on the half-life of 0 ECEC of Ar is defined between 410 and 450 keV (±20 keV around the energy of 429.88 keV, as indicated by the orange band in Fig. 1 ). Given the high statistics in this energy region, data are used in a binned form, with a 1 keV binning. It was checked that the binning choice did not impact the analysis results. In this energy region, the dominant backgrounds are the decay of Ar and the decay of Ge. Subdominant contributions to the background are, in order of importance, the K decays in LAr, the K, Pb, and Bi decays in structural materials. The sum of these contributions in the analysis window can be approximated by a linear distribution, as seen in Fig. 1 . The signal is modeled with a Gaussian peak centered at the energy and with the width given by the detector energy resolution ( = FWHM/2.355). Uncertainties on the energy scale are parametrized by a shift of the signal peak compared to the nominal energy. A simultaneous fit is performed on the eight data sets listed in Table 1 by adopting the following binned likelihood: where the number of events in each bin is Poisson distributed, and the likelihood is given by the product of the Poisson probabilities for all bins i and data sets d . The likelihood depends on the half-life of the investigated process, which is a common parameter among the eight data sets and is the only parameter of interest, and on some nuisance parameters that are data set specific and affect both the signal and background distributions. Gaussian pull terms are introduced in the likelihood to constrain some of the nuisance parameters. Finally, denotes the number of observed events in the data set d and bin i , and is the expectation value for the same data set and bin. The latter is given by the sum of the signal and background in that bin: . The number of signal events is given by the integral of the signal distribution for the data set d in the bin i . This is a Gaussian distribution centered at , where E is the energy of 429.88 keV and the energy bias for the data set d calculated for the same energy, and with the width given by the detector energy resolution evaluated for the same data set and at the same energy. The total number of signal events in a data set d is related to the half-life through the relation: where is the Avogadro constant, is the molar mass of argon (35.968 g/mol), is the mass of LAr in the simulations from which the detection efficiencies are extracted (the product is given in Table 1 for each analysis data set), is the abundance of Ar in ultra-pure natural Argon (0.334%) [ 35 ], and is the live time of the experiment. The total efficiency for the Phase II data sets is given by the product , where is the detection efficiency, the X-rays survival probability (both discussed in Sect. 5 ), and is the efficiency of the LAr veto cut. The latter was estimated to be (97.7±0.1)% for the pre-upgrade data and (98.2±0.1)% for the post-upgrade data [ 11 ]. The total efficiency of the Phase I data sets equals the detection efficiency , because no LAr veto cut was available in Gerda Phase I. Analogously, the number of background events is given by the integral of the background distribution for the data set d in the bin i . The background distribution is a linear function that depends on two parameters, the normalization and the slope, both data set-specific. We verified that the first-order polynomial function describes the data in this energy region well and that a second-order polynomial function does not fit the data better. In modeling the background of Phase I data, an additional Gaussian distribution is used to describe the full energy deposition of the 433.9 keV ray from Ag, which lies in the energy region of the analysis. Contamination from Ag was observed in the screening measurements, and all the three expected lines from Ag were observed in Gerda Phase I data [ 16 , 36 ]. The origin of the Ag contamination in Gerda Phase I was found in the signal cables [ 36 ], which were exchanged in Gerda Phase II [ 18 ]. In addition, none of these lines was observed in Gerda Phase II data after the LAr veto cut. The decay of Ag proceeds through a cascade of three equally probably rays at energies of 433.9 keV, 614.3 keV, and 722.9 keV. Therefore, even if any Ag contamination were still present in Gerda Phase II, the LAr veto cut would likely discard the corresponding events. In total, the fit has 42 floating parameters, 22 describing the signal peak ( , , ), 4 10 for the linear background of Phase II data sets, 6 for the linear background of Phase I data sets, 3 parameters for the number of Ag events in Phase I data sets, plus one common parameter to all data sets . The latter is constrained to positive values. Gaussian pull terms in the likelihood given in Eq. ( 3 ) constrain some of the nuisance parameters, namely the efficiency , the energy bias , and the energy resolution around their central value and uncertainty. All the other nuisance parameters are free and unconstrained, and their uncertainties are propagated into the result by profiling. To set a lower limit on the half-life of the investigated process, we use a modified frequentist approach, namely the method [ 37 ]. The latter was found to be a more appropriate choice in the case of an experiment with low sensitivity or, in different words, a background-dominated experiment [ 37 ]. Compared to a pure frequentist approach, the exclusion region does not assure the correct coverage and often results in an over-coverage, thus a more conservative result. The profile likelihood ratio test statistic is used for the calculation. Asymptotic distributions of the test statistic and the Asimov data set are used [ 38 ]. The statistics in each bin is high enough for this assumption to be valid.

Results The best fit, defined as the minimum of the profiled likelihood ratio, yields = 0, i.e. we do not observe any signal events from 0 ECEC. Data from the five Gerda Phase II analysis data sets and in the energy region of the analysis are shown in Fig. 4 together with the best-fit model and the residuals normalized to the expected statistical fluctuations of the bins. The 90% CL limit on the half-life is obtained by scanning the observed over different values of and finding the value for which = 0.1. For Gerda Phase II data only, this gives year. The 90% CL sensitivity of the Gerda Phase II experiment, i.e. the median expectation under the no signal hypothesis, is obtained analogously by scanning the expected over different values of and finding the value for which = 0.1. The latter gives year. The analysis of the combined Gerda Phase I and Phase II data gives a 90% CL sensitivity of year and an observed lower limit of year. Figure 5 shows the scan of the observed and expected over a range of values of obtained in the analysis of the combined Gerda Phase I and Phase II data. Systematic uncertainties on the efficiency , the energy bias , and the energy resolution are identified as primary sources of systematic uncertainties and included in the likelihood through nuisance parameters constrained by Gaussian pull terms as explained in Sect. 6 . Their overall effect on the limit derived in Sect. 7 is estimated to be 2%. Potential systematic uncertainties related to the fit model, particularly the background distribution, are also investigated. First, the assumption of a linear distribution is compared to a more general second-order polynomial distribution. This has a negligible impact on the result. The presence of additional structures in the background is also investigated. As discussed in Sect. 6 , a line from Ag, very close to the expected signal energy, is included in the background model of the Phase I data sets, as in previous analysis [ 16 ]. A possible systematic uncertainty due to the above line in Phase II data is investigated by introducing it in the background model. This would worsen our result of a 2%.

Conclusions In this work, we searched for the 429.88 keV line from the Ar 0 ECEC using the final total exposure of the Gerda Phase II experiment, combined with the Gerda Phase I exposure. No signal was observed, and a lower limit on the half-life of this process was derived, yielding year (90% CL). This is the most stringent limit on the half-life of the Ar 0 ECEC. This work shows that the potential of the Gerda experiment in investigating physics beyond the Standard Model extends further than the search for the decay of Ge (see also [ 39 , 40 ]). Even if the sensitivity is many orders of magnitude below the theoretical expectation for this process, to our knowledge, the Gerda experiment was, to date, the only experiment with the capability to search for the 0 ECEC of Ar with competitive sensitivities. The Gerda sensitivity is limited by the physical background from Ar and Ge decays in the energy region where the peak is expected, which is, for instance, orders of magnitude higher than the background in the region of interest for the Ge decay. An additional limiting factor is the low detection efficiency since the ray is emitted in the LAr and must be detected in one of the germanium detectors. Only rays emitted in the proximity of the detector array contribute to the total efficiency as discussed in Sect. 5 (see Fig. 3 ). Among the planned future experiments, the Large Enriched Germanium Experiment for Neutrinoless- Decay (LEGEND) experiment can extend the search for the 0 ECEC of Ar to higher sensitivity. In the first phase of the project, LEGEND-200 will deploy about 200 kg of germanium detectors. This is more than a factor of four compared to the Gerda detector mass and will imply a higher detection efficiency to the ray emitted in this process. On the other hand, the background in the energy region where the peak is expected should be comparable to the Gerda background, largely dominated by the Ar decay. Still, an improvement in the current sensitivity is foreseen. LEGEND-1000 will deploy about 1 ton of germanium detectors, implying an even higher detection efficiency to the ray emitted in this process. In addition, using underground Ar instead of atmospheric Ar is intended. This is depleted of Ar, which is the main background contribution in this search. A significant improvement in the sensitivity is therefore expected. To our knowledge, no other planned experiment has competitive sensitivity to LEGEND in the search for 0 ECEC of Ar.

The GERmanium Detector Array ( Gerda ) experiment operated enriched high-purity germanium detectors in a liquid argon cryostat, which contains 0.33% of Ar, a candidate isotope for the two-neutrino double-electron capture (2 ECEC) and therefore for the neutrinoless double-electron capture (0 ECEC). If detected, this process would give evidence of lepton number violation and the Majorana nature of neutrinos. In the radiative 0 ECEC of Ar, a monochromatic photon is emitted with an energy of 429.88 keV, which may be detected by the Gerda germanium detectors. We searched for the Ar 0 ECEC with Gerda data, with a total live time of 4.34 year (3.08 year accumulated during Gerda Phase II and 1.26 year during Gerda Phase I). No signal was found and a 90% CL lower limit on the half-life of this process was established year. Supplementary Information The online version contains supplementary material available at 10.1140/epjc/s10052-023-12280-6.

The GERDA experiment The Gerda experiment was located at the Laboratori Nazionali del Gran Sasso (LNGS) of INFN, in Italy [ 10 , 18 , 19 ], where a rock overburden of 3500 m water equivalent reduces the flux of cosmic muons by six orders of magnitude [ 10 ]. High-purity germanium (HPGe) detectors, isotopically enriched in Ge, were operated inside a 64 m LAr cryostat [ 20 ]. In the second phase of the experiment, 10 coaxial (including 3 detectors with natural isotopic abundance) and 30 Broad Energy Germanium (BEGe) detectors were used [ 18 ]. After an upgrade in May 2018, the three natural coaxial detectors were removed, and 5 additional inverted coaxial (IC) detectors were installed [ 11 ]. Detectors were mounted on 7 strings, and each string was placed inside a nylon cylinder to limit the collection of radioactive potassium ions on the detector surfaces [ 21 ]. The LAr volume around the detectors was instrumented with a curtain of wavelength-shifting fibers connected to silicon photo-multipliers (SiPM) and 16 cryogenic photo-multiplier tubes (PMTs) to detect scintillation light in the LAr [ 18 , 22 ]. During the upgrade, the geometrical coverage of the fibers was improved, more SiPM channels were added, and their radiopurity increased [ 11 ]. The cryostat was surrounded by a water tank containing 590 m of pure water, equipped with PMTs to detect the Cherenkov light of residual cosmic muons reaching the detector site. The instrumented water tank formed, together with scintillator panels on the top of the experiment, the muon veto system [ 23 ]. Data selection The Gerda Phase II data taking started in December 2015; it was shortly interrupted in the Summer of 2018 for the upgrade of the setup and lasted until November 2019. The total collected data used to search for the 429.88 keV line from the 0 ECEC of Ar corresponds to a live time of 3.08 year, divided into 1.91 year before the upgrade and 1.17 year after the upgrade. Due to the different detector properties, e.g. energy resolution and efficiency, and the changes in the detector configuration during the upgrade, data were split into 5 data sets, namely pre-upgrade BEGe, pre-upgrade Coax, post-upgrade BEGe, post-upgrade Coax, and post-upgrade IC. The Coax detectors were excluded from the analysis since they have a low duty factor due to their unstable operation in Gerda Phase II and made up a minimal amount of the exposure. Data have been processed following the procedures and digital signal processing algorithms described in [ 24 ]. The energy of an event is reconstructed using a zero-area-cusp filter [ 25 ]. Events must pass several quality cuts based on the flatness of the baseline, polarity, and time structure of the pulse to reject non-physical events. The acceptance efficiency of physical events by quality cuts is larger than 99.9% [ 11 ]. Events preceded by a trigger in the muon-veto system within 10 s are also discarded, with negligible induced dead time (<0.01%) [ 11 ]. The experimental signature used to search for Ar 0 ECEC in the Gerda data corresponds to the full energy deposition of the ray in one germanium detector. Neglecting the energy deposition of the two X-rays, no coincident energy deposition is expected, neither in the other germanium detectors nor the LAr. Consequently, the detector anti-coincidence cut and the LAr veto cut were also applied. The energy of the two X-rays is low enough that, even if they reached the germanium detector surface, they could not penetrate the 1–2 mm dead layer and, therefore, not be detected by the germanium detector. Nevertheless, since they deposit their energy in the LAr, they could be seen by the LAr instrumentation and trigger the LAr veto. The corresponding event would escape the data selection. This effect is considered in the total detection efficiency, as will be explained in Sect. 5 . The LAr veto cut reduces the background in the region of interest of this analysis by a factor of , as can be seen in Fig. 1 . In this energy region, Ar decay dominates up to the endpoint at 565 keV, while decay is the second dominant contribution. The Pulse Shape Discrimination (PSD) cut, successfully employed in the search for decay [ 26 ], is unsuitable for this analysis and, therefore, not used. In fact, rays mostly result in multiple separated energy depositions in the germanium detector, i.e. multi-site events, in contrast to the single-site events produced in the decay. In addition, the performances of the PSD cut at the energy of interest of this analysis are poorly known. Consequently, part of the data excluded in the decay analysis from BEGe and IC data sets because of the PSD cut was instead included here. We combine the analysis of Gerda Phase II data with that of Gerda Phase I data reported in [ 16 ]. The Gerda Phase I data taking started in November 2011 and lasted until May 2013. The total collected data used for searching for 0 ECEC of Ar corresponded to a live time of 1.26 year and was divided into three data sets, namely Coax, BEGe, and Coax. More details on the data processing and selection of these three data sets can be found in [ 16 ]. It has to be noticed that the instrumentation of the LAr volume is a unique feature of Gerda Phase II and that no LAr veto cut was available in Gerda Phase I. Energy resolution and energy scale The energy calibration of the Gerda germanium detectors was performed during dedicated weekly calibration runs in which the germanium detectors were exposed to three Th sources [ 27 ]. All calibration data were combined as detailed in [ 27 ] to determine the energy scale and resolution throughout the experiment. This work uses the effective resolution curves calculated for the five analysis data sets [ 28 ]. The resolution curves are evaluated at the Ar 0 ECEC energy of 429.88 keV. The energy resolution in full width at half maximum (FWHM) and their uncertainties are summarized in Table 1 . The uncertainty on the FWHM is calculated assuming the same relative uncertainty as for the FWHM at the of the Ge decay ( = 2039 keV). This was calculated in [ 27 ] as exposure-weighted standard deviation. The picture might be different at low energy, and the results obtained for the decay peak at 2039 keV might not be valid for the 0 ECEC peak at 429.88 keV. In fact, the lowest energy peak used to determine the resolution curves above is the 583 keV Tl peak, above the energy region of interest in this analysis. To cross-check the energy resolution at the energy of interest, we use the results of the special low-energy calibration performed at the end of the Gerda data taking. This calibration run aimed to study the energy scale and stability at low energy. The energy threshold was set to 100 keV (while it was 400 keV during regular calibration runs), allowing to extend the energy range in which the resolution curve is calculated to about 238 keV, the energy of the first Pb peak usable for the calibration. We use the peak at 583 keV as a proxy for the 0 ECEC peak, being the closest in energy. We should note that also the topology of the events for the two peaks is the same. In both cases, it is a full energy deposition of the energy in one germanium detector, with the ray starting in the surrounding of the detector array. We calculate the residuals on the FWHM as the difference between the FWHM extracted in the special low-energy calibration and the value obtained evaluating the resolution curves above at 583 keV. The residuals for each detector are shown in a histogram at the left-handed side of Fig. 2 . We find no systematic deviation of the FWHM at this energy compared to the resolution curves. The RMS of the residuals is 0.049 keV, with only one detector with a larger residual of 0.2 keV. 2 The monitoring of the energy scale for the decay search was performed using the single escape peak of Tl at 2103 keV, which is typically used as a proxy for the decay peak at . The residuals between the peak position after energy calibration and the nominal energy value were evaluated over time, giving a mean energy bias of 0.07 keV with an average uncertainty of 0.17 keV [ 27 ]. To cross-check the energy bias at the energy of interest, we use the results of the special low energy calibration run and the 583 keV peak as a proxy for the 0 ECEC peak again. We calculate the residuals on the peak position as the difference between the nominal energy value and the energy value extracted from the special low-energy calibration. The residuals for each detector are shown in a histogram at the right-hand side of Fig. 2 . We find a mean energy bias of 0.03 keV with a RMS among detectors of 0.084 keV. This is below the estimated bias uncertainty of 0.17 keV for the decay peak at . It should be noted that these biases are well below the binning of 1 keV used in the analysis. The effect is therefore expected to be marginal. In this work, we adopt a mean energy bias of 0 keV with an uncertainty of 0.1 keV for all the five analysis data sets. Detection efficiency The detection efficiency is defined as the probability that a 429.88 keV ray entirely deposits its energy inside a single germanium detector. This was determined via Monte Carlo simulations with the Geant4 -based MaGe framework [ 29 , 30 ]. In total, rays with an energy of 429.88 keV were generated in a cylindrical volume of LAr, with a radius of 1.5 m and a height of 2.5 m, around the detector array. This corresponds to a net volume of LAr, after taking into account the volume occupied by the germanium detectors and structural materials, of 17.657 m . The corresponding LAr mass, given the LAr density of 1385 kg/m , is 24,459 kg. The contribution from rays originating from outside this volume to the detection efficiency is negligible, as shown in Fig. 3 . The projected distribution of vertices from which the simulated rays originate is shown in blue for all the events resulting in an energy deposition in the germanium detectors and black for the events resulting in the deposition of the entire 429.88 keV energy in one germanium detector. Only the last contribute to the detection efficiency, defined for each data set as the ratio between the number of events in which the full energy is deposited in one germanium detector in the specific data set and the number of initially simulated events. The number of simulated events is high enough that the statistical uncertainties on these quantities are negligible. Detector active volume and the status of each detector over the whole data taking are considered in the simulation, as detailed in [ 31 ]. The dominant systematic uncertainty on the detection efficiency comes from the detector active volume uncertainty. This is estimated by varying the detector dead layer in the simulation by , where is the dead layer uncertainty, and evaluating the impact on the efficiency. Typical sizes of the detector dead layers are 1–2 mm known with a typical uncertainty of 5–30 % [ 32 ]. The corresponding systematic uncertainty on the detection efficiency is 3% for BEGe detectors, 4% for Coax detectors, and 1% for IC detectors. The detection efficiencies multiplied by the mass of LAr in the simulation volume, together with their uncertainties, are summarized in Table 1 for the different data sets. The two X-rays that are emitted in the process being searched for are neglected in the simulations. As anticipated in Sect. 3 , their energy deposition in LAr could trigger the LAr veto. To account for this possibility, the survival probability of the two X-rays to the LAr veto cut is evaluated and combined with the detection efficiency. We use the Gerda photon detection probability map developed in [ 33 ] to estimate the probability p ( x , y , z ) to detect scintillation light for each simulated event starting at position ( x , y , z ) and corresponding to a full energy deposition. From this probability, the number of photons n produced by the two X-rays of total energy = (2.47 + 0.23) keV is obtained: where 28.12 is the number of photons produced for an energy deposition of 1 keV expected in the Gerda LAr [ 33 ]. The probability P that the corresponding event survives the LAr veto cut is the Poisson probability . 3 The mean survival probability is obtained by averaging the survival probabilities of the events corresponding to a full energy deposition and results in . Thus, the data selection discards almost 5% of the events due to the X-rays depositing their energy in LAr. The calculation of the survival probability assumes that the two X-rays deposit all the energy at the exact point where the ray is emitted. This assumption is considered valid since the attenuation length for a 3 keV X-ray was estimated to be about 42 m [ 34 ], negligible compared to the mm binning of the photon detection probability map. The main systematic uncertainty on the mean survival probability comes from the photon detection probability map. The uncertainties on this probability map given in [ 33 ] result in a 0.5% systematic uncertainty on the survival probability. Finally, we should note that the photon detection probability map assumes the pre-upgrade configuration of the LAr instrumentation [ 33 ]. This means the model does not include the inner fiber shroud installed during the upgrade to improve the light detection efficiency near the germanium detectors [ 11 ]. Therefore, a customized LAr veto cut was applied to select the post-upgrade data used in this work: the SiPM channels corresponding to the inner fiber shroud are not considered to build the LAr veto condition. This way, the X-rays survival probability obtained with the pre-upgrade photon detection probability map is extended to the post-upgrade data sets. Supplementary Information Below is the link to the electronic supplementary material.

Acknowledgements The Gerda experiment is supported financially by the German Federal Ministry for Education and Research (BMBF), the German Research Foundation (DFG), the Italian Istituto Nazionale di Fisica Nucleare (INFN), the Max Planck Society (MPG), the Polish National Science Centre (NCN, Grant number UMO-2020/37/B/ST2/03905), the Polish Ministry of Science and Higher Education (MNiSW, Grant number DIR/WK/2018/08), the Russian Foundation for Basic Research, and the Swiss National Science Foundation (SNF). This project has received funding/support from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie Grant agreements no 690575 and no 674896. This work was supported by the Science and Technology Facilities Council, part of the UK Research and Innovation (Grant no. ST/T004169/1). The institutions acknowledge also internal financial support. The Gerda collaboration thanks the directors and the staff of the LNGS for their continuous strong support of the Gerda experiment. Data Availability Statement This manuscript has associated data in a data repository. [Authors’ comment: The data shown in Figs. 1 and 4 is available in ASCII format as Supplemental Material [ 41 ].]

CC BY

no

2024-01-16 23:41:59

Eur Phys J C Part Fields. 2024 Jan 14; 84(1):34

oa_package/cc/8d/PMC10788323.tar.gz

PMC10788324

38221592

Background Glues are liquid embolics applied in endovascular embolizations since approximately 50 years: first cyanoacrylate application in peripheral bleeding embolizations was described by Dotter in 1975 [ 5 ]. Since then, the role of glue in the embolization field has grown and nowadays, thanks also to imaging and microcatheters developments, glues are a consolidated part of interventional radiology toolbox, especially in its endovascular use. Seldomly, glue has been applied also via percutaneous technique, mainly in elective treatments: neck tumors [ 1 ], arterio-venous malformations [ 16 ], varices [ 4 , 19 ], type-II endoleaks [ 18 , 22 ] and aneurysms [ 11 ]. Regarding pseudoaneurysms (PAs) embolization, only case reports and few case-series have been published [ 9 , 10 , 21 ] with this approach. This multicentric study aims now to report on PAs percutaneous embolization with glue ( N -butyl cyanoacrylate) in a cohort of patients, analyzing safety and effectiveness.

Methods This is a retrospective analysis of patients with PA treated in five centers from 2020 to 2023. Electronic radiological and medical records were reviewed. All patients signed a written informed consent, if clinical conditions allowed. The local ethical committee approved the study. Inclusion criteria were: PA detection at preoperative contrast-enhanced Computed Tomography (CT), anamnesis of trauma or inflammatory disease, glue embolization, percutaneous approach first, imaging follow-up almost at one week. Exclusion criteria were: true aneurysms, additional injection of intrasaccular embolics (thrombin, cohesive liquids, coils, etc.). PA diagnosis was based on CT findings and anamnestic data. All patients were hemodynamically stable at the time of treatment with unruptured PAs. Lesions features, procedural details, coagulation status and follow-up data were analyzed. Procedure All interventions were performed by a single operator in the angio-suite or in CT (Figs. 1 and 2 ) immediately after diagnosis of PA. After CT evaluation (Fig. 2 A), a diagnostic digital-subtracted arteriography (DSA) was acquired before the percutaneous approach to confirm PA occurrence and assess lesions features in terms of morphology and vascular supply (Fig. 2 B). PAs were treated via direct puncture under image guidance (Ultrasound (US), fluoroscopy, ConeBeamCT, CT or a combination) as first choice; mandrinated needles with Quincke tip were adopted, different calipers and lengths according to operator preference and lesion site. In case of US approach (Fig. 2 C-D), the needle tip could be scrubbed with a scalpel to improve its sonographic visibility [ 8 ], based on operator preference. After local anesthesia, PA was punctured attempting to cross the largest amount of soft tissues in order to mantain the needle stable during the subsequent maneuvers; needle stylet was removed, blood gushing out from needle hub and imaging confirmed correct needle tip positioning (Fig. 2 E); iodinated contrast agent was injected to opacify the PA, except in case of procedures performed under sole US guidance; then, needle lumen was flushed with glucosate 10% and finally glue ( N -butyl cyanoacrylate, Glubran2-GEM®, Italy) mixed with Lipiodol (Guerbet®, France) was continuously injected (Fig. 2 F) until PA cavity fullfilment (Fig. 2 G), using slip tip syringes; needle was so retracted still injecting glue to seal the percutaneous tract and preventing eventual iatrogenic bleeding. A DSA was acquired to detect glue effects (Fig. 2 H); in case of partial exclusion of the PA, other percutaneous attempts were conducted; if percutaneous puncture failed or was technically unfeasible because of impaired imaging visualization related to glue-air-Lipiodol artifacts, a standard endovascular approach was applied (Fig. 3 ). The anesthesiological protocol depended on institution facilities and operators choice. Technical success was considered as complete PA embolization at final imaging with sole percutaneous strategy, without need for additional endovascular embolization. Clinical success was intended as PA resolution within one week follow-up with stabilization or restored clinical conditions. Complications were evaluated according to the CIRSE classification [ 6 ].

Results Sample features The study sample included 54 patients (19 females, 35 males; mean age: 63.8 years, range 18–91) affected by PAs (Table 1 ). Lesions etiology was mainly traumatic, in 31 subjects (57.4%), the others being inflammatory (24.1%) or spontaneous (18.5%); 39 (72.2%) were symptomatic, presenting with pain and/or pulsatile mass. PAs presented a mean diameter of 19.3 mm (range: 7–30) measured at preoperative CT. Concerning lesions sites, PA were localized in abdomen (visceral: bowel 7, hepatic 5, gastric 4, splenic 2, pancreatic 1, renal 1; wall 6), thorax (visceral: peripheral pulmonary artery 2; wall 3) and limbs (lower: femoralis 10, geniculate 4, anterior tibial 3, gluteal 2, peroneal 1; upper: brachial 3). In 9 patients (16.6%) the coagulation function was impaired: INR > 1.5 in 5, platelets < 50,000/mmc in 3 and both INR > 1.5 and platelets < 50,000/mmc in 1. Twenty-six patients (48.1%) uptook antiplatelets/anticoagulation therapy at the time of embolization: 7 assumed oral anticoagulants (2 Fondaparinux, 3 Apixaban, 1 Warfarin, 1 Rivaroxaban), 9 heparin at prophylactic dosage, 9 antiplatelets (4 Clopidogrel, 5 acetilsalicid acid, 1 Clopidogrel plus Acetilsalicid acid), 1 heparin at prophylactic dosage and acetilsalicid acid; when the underlying pathology allowed, these therapies were suspended or shifted to heparin. Patients with PAs caused by inflammatory and traumatic etiologies were under antibiotic regimen at the time of treatment. Procedural details The procedural outcomes are summarized in Table 2 . The rationale of adopting a percutaneous strategy, over a standard endovascular, was because of: endovascular unfavourable conditions, as tortuosity, thin collateral feeders and spasm in 28 (51.8%); operator preference related to superficial locations in 20 (38.9%); no PA detection at DSA in 5 (9.3%). In 9 cases (16.6%) the percutaneous approach followed previous treatments failure (external compression) performed in other centers. The glue-Lipiodol dilution most adopted was 1:1, in 26 cases (48.1%). The maximum amount of glue adopted for a single procedure was 2 vials (1 ml/vial). Needles caliper was 22G in 17 (31.5%), 20G in 14 (25.9%), 18G in 8 (14.8%), 17G in 8 (14.8%) and 21G in 7 (13%), depending on operator preference; needle lenght varied (5 to 20 cm) according to PA location deepness. The image-guidance modality was most often US combined with fluoroscopy in 21 cases (38%). In 36 patients (66.7%) embolization was performed with a single puncture, in 16 (29.6%) with two punctures and in 2 (3.7%) with three. Regarding anesthesiological management, procedures were mainly performed under sole local anesthesia (28 patients; 51.8%). Embolization outcome Clinical success was obtained in all patients (100%). Technical success occurred in 51 cases (94.4%); in 3 patients an additional endovascular embolization with glue (2) or coils (1) was needed to completely seal the PA. Complications were registered in 8 patients (14.8%): 7 non target embolization because of glue migration and one post-embolization syndrome (PES), all without clinical sequelae neither prolonged recovery (grade I); PES required additional anti-inflammatory and anti-piretic therapy without increased prolonged hospital stay.

Discussion In this retrospective analysis percutaneous glue embolization of PAs was safe and effective; the most common technical approach consisted in a US-fluoroscopic guided puncture using a 22G needle and injecting a 1:1 glue-Lipiodol mixture. PAs were located in thoraco-abdominal districts, both visceral and parietal, and limbs. The lesion etiology was mainly traumatic; the rate of spontaneous lesions was also elevated compared to the general population, but this should be related to the large number of subjects uptaking antiplatelets/anticoagulation therapy in the study population. Until now only few data are available in literature regarding this technique for the treatment of PAs. The largest series has been published by Del Corso et al. [ 3 ] on 91 patients; however all PAs were superficial and located in the limbs, mainly involving the femoral artery. Instead Carriero et al. [ 2 ] and Gorsi at al. [ 10 ] embolized abdominal visceral PAs by direct puncture using cyanoacrylate in 12 and 21 patients respectively, both obtaining 100% technical success. Another embolic applied in percutaneous PAs embolization is thrombin; compared to glue, multiple papers have already reported its successful application, mainly in superficial districts and limbs under US guidance [ 12 – 15 , 20 ]. Glue could so represent an alternative to thrombin: especially in case of fluoroscopy/ConeBeam CT guided procedures, where thrombin is poorly controllable, while glue mixed with Lipiodol presents the advantage of being visible [ 7 ]. Compared to the existing literature, this paper confirms safety and effectiveness of percutaneous glue PA embolization but this is one of the largest literature sample on this topic; a wide range of PA locations and etiology, as well as different image guidance techniques have been included, reflecting the everyday practice. From a technical point of view, syringes with luer slip tip should be preferred over luer lock tip to minimize the risk of needle displacement during connection to the needle hub; samely, glue-Lipiodol mixture should be prepared in advance and ready to inject when PA is punctured. Compared to endovascular embolization, larger amounts of glue with low Lipiodol dilution and faster injection rates are adopted to fulfill directly the PA lumen, not needing glue distalization; for this reason no adjunctive procedures were performed in this sample, however some authors [ 17 ] suggested the application of endovascular balloon catheters temporary inflation close to PA neck to prevent distal migration, especially in femoral PAs percutaneous embolization. This approach proof to be safe and effective also in patients with impaired coagulation function, glue embolic properties not being influenced; furthermore, the percutaneous tract also is sealed with glue, minimizing the risk of iatrogenic bleeding. The complications rate was 14.8%, however it was mainly related to non target embolizations without any clinical sequela, in accordance with the technical approach: being glue injection directly performed into the target, some amount migrated to the vessels outsourcing from PA; a possible solution to reduce this event could be the application of low glue:Lipiodol dilution (1:1 or 1:0.5) to prevent unvoluntary distalization. Regarding image guidance, US and fluoroscopy allow real time monitoring of glue injection and this is a relevant advantage compared to CT and ConeBeamCT. Finally, even if not analyzed, this approach seems cost-effective: compared to endovascular, it would allow to reduce procedural time and costs related to microcatheters and embolics. Main limitations of this paper are: first, the number of patients is still limited with miscellaneous PA etiology and locations, so larger and homogeneous prospective studies are needed to identify which patients will benefit the most from this approach; the technique adopted was not standardized in terms of image guidance and glue dilution because these parameters changed according to operators preference and lesions site; furthermore, a certain learning curve is required because operators should be skilled both in percutaneous interventions and glue handling; PA presented a maximum diameter of 30 mm (mean 19.3 mm), no evaluations can be provided for larger lesions; finally, follow-up is short and mid-long term results are required to verify the long-time effectiveness.

Conclusion In this study, PAs embolization with glue via percutaneous direct puncture was safe and effective with a low rate of minor complications.

Background This retrospective multicentric study aims to report on technical safety and effectiveness of pseudoaneurysms embolization with glue ( N -butyl cyanoacrylate) adopting a percutaneous direct puncture approach. Results Fifty-four patients data were collected from five centers. All patients at the time of treatment presented with unruptured PAs and were hemodynamically stable. True aneurysms and lesions treated with embolics other than glue were excluded. Pseudoaneurysms diagnosis was based on CT and anamnestic data; initial investigation with digital-subtracted arteriography was acquired in all cases; then, percutaneous embolizations were performed in the angio-suite (ultrasound, fluoroscopy, ConeBeam CT guidance) or in CT. Technical success was considered as complete pseudoaneurysm embolization at final imaging with sole percutaneous strategy, without need for additional endovascular embolization. Clinical success was intended as pseudoaneurysm resolution within one week follow-up with stabilization or restored clinical conditions. Pseudoaneurysms origins were traumatic (57.4%), inflammatory (24.1%) or spontaneous (18.5%); 39 patients (72.2%) were symptomatic, presenting with pain and/or pulsatile mass. Mean lesions diameter was 19.3 mm (range: 7–30); pseudoaneurysms were located in abdomen (48.1%), limbs (42.6%) and thorax (9.3%). Coagulation function was impaired in 16.6% and 48.1% was under antiplatelets/anticoagulation therapy. In 16.6% the percutaneous approach followed previous treatments failure. The image-guidance modality for percutaneous puncture was most often ultrasound combined with fluoroscopy (38%). Clinical success was obtained in all patients while technical success occurred in 94.4% because 3 patients required an additional endovascular embolization. Complications were registered in 14.8%, all of low grade without clinical sequelae neither prolonged recovery (7 non target embolizations, 1 post-embolization syndrome). Conclusions In this study, pseudoaneurysms embolization with glue via percutaneous direct puncture was safe and effective with a low rate of minor complications. Keywords

Abbreviations Cardiovascular and Interventional Radiology Society of Europe Computed Tomography ConeBeam Computed Tomography Digital subtraction angiography Gauge Index Normalized Ratio Maximum Intensity Projection Pseudoaneurysm Post-embolization syndrome Ultrasound Acknowledgements None. Authors’ contributions Conception and design: Giurazza F, Loffroy R, Niola R; Provision of study materials and patients: Giurazza F, Pane F, Corvino R, Ierardi AM, Lucatelli P, Sironi S, Carrafiello G, Marra P, Loffroy R, Niola R; Collection and assembly of data: Giurazza F; Data analysis and interpretation: Giurazza F, Ierardi M, Loffroy R; Manuscript writing: Giurazza F, Niola R; Final approval of manuscript: All authors. Funding None. Availability of data and materials The dataset supporting the conclusions of this article is available at Cardarelli Hospital RIS-PACS system; for any questions, please contact the corresponding author. Declarations Ethics approval and consent to participate All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and national research committee and with the 1964 Helsinki Declaration and its later amendments or comparable ethical standards; informed consent to participate was obtained from all individual patients included in the study. The institutional ethic committee approved the retrospective data analysis and study pubblication. Consent for publication Written informed consent for publication was obtained from all individual patients included in the study. Competing interests All authors declare no financial and non-financial competing interests.

CC BY

no

2024-01-16 23:41:59

CVIR Endovasc. 2024 Jan 15; 7:11

oa_package/9e/f4/PMC10788324.tar.gz

PMC10788325

38221572

Background Esophageal epiphrenic diverticulum is a rare disease. The estimated incidence of epiphrenic diverticula is approximately 1:500,000/year [ 1 ]. Esophageal motility disorders, such as achalasia, are often comorbidly associated with epiphrenic diverticula [ 2 ]. Surgery is generally considered to be indicated for symptomatic esophageal epiphrenic diverticula [ 1 ]. However, no unified opinions on the details of surgery and approach are currently available. In addition to resection of the esophageal diverticulum, esophageal myotomy and fundoplication may be performed depending on esophageal motility. In recent years, remarkable advances in minimally invasive surgery have been made. As a result, the choice of laparoscopy, thoracoscopy, or a combination of both in the surgical management of this disease has become controversial. In this study, we report the successful results of thoracoscopic esophageal diverticulectomy and laparoscopic fundoplication for a case of a large epiphrenic diverticulum with diverticular erosion due to reflux esophagitis.

Discussion We encountered a rare case of a giant epiphrenic diverticulum associated with gastroesophageal reflux disease. We performed thoracoscopic diverticulectomy and laparoscopic fundoplication for the hiatal hernia and gastroesophageal reflux disease, while avoiding esophagomyotomy due to absence of any evidence of underlying esophageal motility disorders, and obtained excellent surgical results. Epiphrenic esophageal diverticulum is a rare condition, and the indication for surgery depends on symptoms. Conservative management and radiologic or endoscopic follow-up are acceptable in cases of asymptomatic or minimally symptomatic epiphrenic diverticulum. Surgery is generally indicated in cases of severe dysphagia, regurgitation, or retention of contrast material during esophagography, and the risk of aspiration pneumonia is considered high [ 4 ]. On the other hand, some clinicians suggest that surgery should be performed for all epiphrenic diverticula, regardless of the presence or absence of symptoms [ 5 ]. In this case, the patient visited our hospital with worsening symptoms of eructation. Since a large esophageal diverticulum was thought to cause the symptoms, we decided that surgery was indicated. After surgery, the patient’s symptoms improved, and the absence of signs of impaired transit and reflux on the esophagogram resulted in a favorable outcome. Small esophageal diverticula tend to shrink with myotomy only, and resection may be unnecessary [ 6 , 10 ]. On the other hand, large diverticula require resection [ 1 , 6 ]. Traditionally, diverticulectomy is performed by open thoracotomy (especially left open thoracotomy), which is associated with a high complication rate and postoperative mortality [ 8 , 9 , 11 ]. In recent years, however, with the development of minimally invasive surgery, use of thoracoscopy, laparoscopy, or both procedures have been widely reported. Minimally invasive surgery has been reported to have a lower perioperative mortality rate than open surgery [ 12 ]. Laparoscopic surgery allows adequate observation of the esophagogastric junction when performing a myotomy or a fundoplication. The advent of the endo stapler, which allows resection of the diverticular neck parallel to the esophageal axis, has popularized laparoscopic surgery. However, the distance between the hiatal hernia and diverticulum, size of the diverticulum, and adhesions between the diverticular wall and pleura limit the sole use of laparoscopy for diverticulectomy [ 1 , 12 – 15 ]. Thoracoscopic surgery, conversely, provides excellent visualization and maneuverability for the dissection of large diverticula and diverticula with surrounding areas of inflammation [ 16 ]. Some authors have suggested that a transthoracic resection (open or thoracoscopic) should be followed by a laparoscopic myotomy and fundoplication for large diverticula [ 1 ]. According to Achim et al., diverticula should be approached by thoracoscopy for cases where the diverticular oral end is > 5 cm away from the esophagogastric junction and by laparoscopy only for patients with diverticula that are < 5 cm away [ 6 ]. However, few reports of combined laparoscopic and thoracoscopic surgery have been made. Brandeis et al. reported a case series of minimally invasive surgery for epiphrenic diverticula. Among the 189 patients who underwent minimally invasive surgery, only 12 (6.3%) patients underwent the combined laparoscopy and thoracoscopy [ 17 ]. In this case, the patient had a large diverticulum, which had a length of 8 cm and posed a challenge for resection solely by laparoscopy. The thoracoscopic approach facilitated a safe and reliable resection of the diverticulum. After diverticulectomy, suture failure from the staple line is the most severe complication. Addition of adventitia and muscular layer sutures along the staple line has been reported to decrease suture failure rate. In clinical practice, a number of clinicians routinely use this approach [ 16 ]. In this case, endo staplers were used for diverticulectomy. Thereafter, adventitia and muscular layer sutures were added along the entire staple line length without evidence of postoperative suture failure. Epiphrenic diverticula are often associated with esophageal motility disorders [ 1 ]. Therefore, some clinicians suggest that esophagomyotomy should be performed as a primary procedure. A review of several previous reports shows that suture failure and recurrence rates are higher when esophageal myotomy is not performed [ 1 , 6 ]. On the other hand, several clinicians have recommended the use of elective myotomy [ 4 , 7 – 9 ]. The significance of myotomy is to decrease intraesophageal pressure, and its use in patients without esophageal motility disorders or increased LES pressure might be questionable. Additionally, it may promote postoperative reflux esophagitis [ 4 ]. In this case, high-resolution manometry showed no esophageal motility disorders; therefore, myotomy was not performed. As a result, the patient had a good outcome with no postoperative complications, such as suture failure. Reflux esophagitis is rarely associated with epiphrenic diverticulum, which often develops in association with esophageal motility disorders. A search of the PubMed database did not reveal any case reports of epiphrenic diverticulum with reflux esophagitis. In this case, preoperative upper gastrointestinal endoscopy revealed multiple erosions and scars within the diverticulum. Upper gastrointestinal endoscopy also revealed a hiatal hernia, Barrett’s esophagus, and esophageal erosions contiguous with the SCJ, indicative of the presence of gastric acid reflux. Therefore, the erosions and scars observed in the diverticulum were thought to be associated with the reflux esophagitis. The anti-reflux mechanism was considered more impaired during the diverticulectomy procedure; therefore, we decided that anti-reflux surgery was desirable and performed laparoscopic Dor fundoplication. Partial fundoplication was chosen to prevent the risk of postoperative dysphagia. The patient reported no postoperative symptoms, suggesting that the appropriate procedures were performed.

Conclusions In this report, we described a rare case of a large epiphrenic diverticulum with associated reflux esophagitis. A good surgical outcome was achieved by thoracoscopic resection of the diverticulum and laparoscopic Dor fundoplication.

Background Surgery is indicated for symptomatic epiphrenic esophageal diverticula. Based on the features of a case, thoracoscopic or laparoscopic approaches may be used. Epiphrenic diverticula are often associated with esophageal motility disorders, but cases of reflux esophagitis have rarely been reported. In this report, we describe a case of an epiphrenic esophageal diverticulum with reflux esophagitis, which was successfully treated by thoracoscopic diverticulectomy and laparoscopic fundoplication. Case presentation A 69-year-old man visited the hospital with a chief complaint of eructation and hiccup. Upper gastrointestinal endoscopy revealed a diverticulum in the left wall of the esophagus, which was 37–45 cm distal to the incisors. High-resolution manometry (HRM) showed no esophageal motility disorders. Due to the large size of the diverticulum, a thoracoscopic resection of the esophageal diverticulum was performed. Additionally, the patient had reflux esophagitis due to a hiatal hernia. The anti-reflux mechanism would be more impaired during the diverticulectomy; therefore, we decided that anti-reflux surgery should be performed simultaneously. Thoracoscopic esophageal diverticulectomy and laparoscopic Dor fundoplication were performed. The patient had an uncomplicated postoperative course and was discharged on the tenth operative day. He has been symptom-free without acid secretion inhibitors for 21 months after the surgery. Conclusions We described a rare case of a large epiphrenic diverticulum with reflux esophagitis. A good surgical outcome was achieved by thoracoscopic resection of the diverticulum and laparoscopic Dor fundoplication. Keywords

Case presentation Chief complaints Eructation and hiccup. History of present illness A 69-year-old man was diagnosed with esophageal diverticulum 35 years ago but had been followed up without treatment. Five years ago, he visited a clinic with an exacerbation of his chief complaint. Although he was prescribed esomeprazole, his symptoms did not improve. Therefore, he was referred to our hospital for surgery. History of past illness Internal hemorrhoids. Laboratory examinations BUN was elevated at 25.3 mg/dL, and HBsAg was positive. No other abnormalities in blood counts or biochemical tests were reported. Imaging Upper gastrointestinal endoscopy revealed a diverticulum in the left wall of the esophagus, which was 37–45 cm distal to the incisors. After the contents of the diverticulum were drained into the stomach by changing body position, multiple erosions and scars were observed within the diverticulum. In addition, the endoscopic examination also revealed the presence of a hiatal hernia, Barrett’s esophagus, and esophageal erosions contiguous with the squamocolumnar junction (SCJ), indicative of gastric acid reflux (Fig. 1 ). A barium esophagogram and CT scan showed a large diverticulum with fluid accumulation on the left side of the esophagus (Fig. 2 ). High-resolution manometry Preoperative high-resolution manometry showed no esophageal motility disorders. Operation The patient had a large esophageal diverticulum with reflux esophagitis and no esophageal motility disorders. Therefore, we decided to perform diverticulectomy and Dor fundoplication. The diverticulum extended 8 cm above the diaphragm. A transthoracic approach to the resection of the diverticulum was considered to be most appropriate. After induction of general anesthesia, the patient was placed in the prone position, and five trocars were inserted into the left thoracic cavity (Fig. 3 ). Using CO 2 gas at 10 mmHg and 6 mmHg (after a lung collapse), a pneumothorax was induced [ 3 ]. A preoperative 3D-CT of the diverticulum was used to simulate the surgical procedure, and this simulation contributed to the success of the surgery (Fig. 2 c). Thoracoscopic observation revealed that the diverticulum was located on the left side of the esophagus and was 8 cm long at the base. We inserted an endoscope into the stomach just prior to resection of the diverticulum, to prevent stenosis of the esophageal lumen during the surgery. We used the SigniaTM stapling system (Covidien Japan, Tokyo) with purple staplers. This stapler can be mechanically adjusted in any direction, so that the direction of the staplers can be easily aligned with that of the diverticular resection (Fig. 4 ).Two 45-mm staplers and one 60-mm stapler were used to resect the diverticulum from the cranial end towards the caudal end. The overlying sutures of muscles and adventitia were added over the stapler line along its entire length using 3-0 PDSII (Fig. 4 ). Thereafter, the patient was placed in a supine position, and five trocars and a Nathanson retractor were inserted into the abdominal cavity (Fig. 3 ). The esophageal hiatus was sutured appropriately, and Dor fundoplication was performed using a laparoscopic approach. Postoperative course The patient had no postoperative complications. On the second postoperative day, an esophagogram revealed that the contrast medium could flow freely into the stomach without stagnation and did not reflux into the esophagus (Fig. 5 ). The patient resumed oral intake on the third postoperative day and was discharged from the hospital on the tenth postoperative day. As of 21 months after surgery, the patient was not taking any acid secretion inhibitors and had no symptoms, such as reflux and dysphagia. Endoscopic findings showed no evidence of reflux esophagitis or recurrence of the diverticulum (Fig. 6 ).

Abbreviations High-resolution manometry Squamocolumnar junction Lower esophageal sphincter Acknowledgements The authors used an English Language Service (International Medical Information Center, Tokyo, Japan) for editing the language in the manuscript. Author contributions YU and SO have made substantial contributions to the concept and design of the case report. TA, KH, NA, and MS conceived of the study, participated in its design and coordination, and helped to draft the manuscript. All authors have read and approved the submission of the final manuscript. Funding Not applicable. Availability of data and materials The data used in this study are available from the corresponding author upon reasonable request. Declarations Ethics and approval and consent to participate This report was written in accordance with our institution’s policies, and the need for ethics board approval was waived. Consent for publication A written informed consent was obtained directly from the patient. Competing interests The authors declare that they do not have any competing interests regarding the publication of this article.

CC BY

no

2024-01-16 23:41:59

Surg Case Rep. 2024 Jan 15; 10:17

oa_package/19/40/PMC10788325.tar.gz

PMC10788326

38219249

Introduction Audiological disorders, including deafness and tinnitus, represent significant challenges to individuals and healthcare systems worldwide [ 1 ]. These conditions can lead to profound and often debilitating consequences, affecting auditory perception and overall quality of life. Deafness, complete loss of hearing, and tinnitus, the perception of phantom sounds are major health concerns [ 2 ]. Identifying auditory disorders early is key to personalized treatment and better outcomes [ 3 ]. Traditional methods for diagnosing hearing disorders are limited due to subjective assessments and variability in interpretation. As a result, there is a growing interest in leveraging advanced neuroimaging techniques and machine learning approaches to provide more objective and accurate means of diagnosis. Integrating Multi-View Brain Network data with state-of-the-art machine learning algorithms shows promising results [ 4 ]. This organ is the product of successful evolution, allowing us to perceive, understand, and interact with the world in ways that surpass all other species [ 5 ]. The brain’s intricate network of neurons, synapses, and signaling pathways directs human cognition and behavior [ 6 ]. The brain networks are like a massive communication network, with interconnected areas constantly transmitting signals [ 7 ]. Neural networks sustain our consciousness, emotions, and memories, creating unique human experiences. These networks intricately weave together to create a tapestry that sets us apart as unique individuals [ 8 ]. Advanced technologies are crucial for neuroscience to comprehend the intricacies of neural networks, such as functional magnetic resonance imaging (fMRI) and EEG, to peek into the real-time dynamics of brain activity [ 9 ]. Brain network analysis tools reveal how our cognitive and emotional faculties work. The study of connections between neurons and humans provides insight into consciousness, self-determination, and human nature [ 10 ]. Deafness is a sensory impairment that exerts a considerable impact on the organization of human brain networks, leading to neuroplasticity [ 11 ]. Changes in sensory input or experiences can trigger the brain’s reorganization ability. This is especially true for individuals who experience hearing loss, such as those who are deaf. As a result, they often possess superior visual processing skills, including improved visual acuity, motion detection, and spatial abilities [ 12 ]. Changes in connectivity among different brain regions, including the left hemisphere responsible for sign language processing, occur in patients with hearing loss due to the reorganization of brain networks [ 13 ]. Acknowledging sign language as a legitimate language significantly influences the brain regions that manage language processing. The cognitive system responsible for short-term memory and information manipulation is often enhanced in deaf individuals [ 14 ]. Hearing loss can have a significant impact on social and emotional processing. It is widely acknowledged that people with normal hearing process emotional signals differently than those with hearing impairment [ 15 ]. Understanding neural adaptations and developing targeted interventions and assistive technologies are essential to improve the quality of life for individuals with hearing impairments [ 16 ]. Tinnitus is a common medical condition that causes individuals to experience sounds in their ears even when there is no external auditory stimulus. This condition affects millions worldwide and should not be taken lightly [ 17 ]. Tinnitus has many factors that interact in complex ways, leading to its cause. Although the cause may not always be clear, it is believed that alterations in brain networks could play a role in its development [ 18 ]. Any changes within the auditory system will undoubtedly affect its ability to hear and process sound effectively [ 19 ]. Machine learning in audiology is indispensable for the Automated Diagnosis of auditory disorders through Multi-View Brain Networks. With their remarkable ability to process vast amounts of data and identify intricate patterns, Machine Learning algorithms are highly effective in accurately and efficiently classifying auditory disorders [ 20 ]. The primary objective of this study is to create a system that can improve the diagnosis of auditory disorders such as deafness and tinnitus. This is achieved by analyzing the complex relationships between different brain regions using a clustering coefficient based on triangle motifs. The feature reduction technique PCA is utilized to manage the high-dimensional nature of brain connectivity data. The study employs ensemble learning techniques to produce accurate and reliable predictive models. The main contributions of this work can be summarized as follows: Use machine learning for early and accurate diagnosis of deafness and tinnitus to enhance patient outcomes and quality of life. Apply Multi-View data and combine different views in EEG data and 10 ROI to achieve better diagnosis results. Build a Robust model by leveraging the strengths of ensemble learning algorithms. Assess the effectiveness of classification findings by evaluating proposed models using a variety of metrics. Compare our findings results with existing state-of-the-art approaches. The remaining sections of this paper are arranged in the following: Sect. 2 provides a comprehensive overview of pertinent research concerning identifying auditory disorders through Multi-View Brain Networks. Section 3 covers the proposed research methodology, which includes the creation of multi-view brain networks, exploratory data analysis, and data visualization. In section 5 , the details of our experimental hypothesis modeling and setup are outlined. The experimental findings and results are detailed in Sect. 6 . Section 7 gives the work conclusion and recommendations for future work that can be implemented to achieve positive outcomes.

Proposed research methodology Data insights The multi-layer brain network dataset was collected by Sun Yat-sen University [ 41 ]. The dataset includes three distinct groups, consisting of 51 deaf cases, 54 with tinnitus, and 42 healthy individuals. The study thoroughly analyzed their respective brain network function. Resting-state EEG data can provide valuable information on neural processes. Extracting multi-layered brain network datasets from this data helps us understand how the brain functions. Complex datasets require advanced analysis methods, including varied subjects, electrodes, and frequency bands. These data examine the brain network dynamics in individuals with deafness, tinnitus, and those without hearing problems using 70 electrodes. Data acquisition and preprocessing are explained. The dataset features nine different frequency bands, namely Delta, Theta, Alpha1, Alpha2, Beta1, Beta2, Beta3, Gamma1, and Gamma2. Pearson’s correlation coefficients calculate the interconnections of the electrodes for each frequency band. Based on EEG data, the network highlights significant disparities in neural networks between the three subject types. The dataset provides vital insights into the characteristics of brain networks in deafness and tinnitus patients compared to normal controls. Exploratory data analysis and visualization Exploratory data analysis (EDA) is crucial in data science as it helps understand data patterns and gain insights. Visualizations and plots play a significant role in making complex data more accessible. In this EDA, we have employed various visualization techniques to elucidate the multi-layer brain network datasets obtained from EEG data. Graphical representations and data visualizations are essential for transforming intricate numerical data into easily understandable graphics. During this EDA process, we have tailored these visualizations to meet the unique requirements and characteristics of the multi-layer brain network datasets. Histograms display how connection strengths are distributed within brain networks, such as the ’alpha1’ and ’alpha2’ connections in the normal category. These histograms group connection strengths into bins to reveal whether the networks consist predominantly of weak or strong connections. This visualizes the connections in the brain, where varying cell sizes indicate strength and color intensity represents patterns like clusters. Adjacency matrices offer a comprehensive network topology view by revealing how nodes are interconnected. The clustering coefficient, based on triangle motifs or transitivity, measures how tightly knit a network is [ 42 ]. It is calculated as shown in Eq. 1 : where is the clustering coefficient of node i , is the number of triangles node i is part of and is the degree of node i . A node’s local clustering (per class) coefficient measures how well its neighbors are connected. It is calculated as shown in Eq. 2 : where is the local clustering coefficient of node i , is the number of edges between the neighbors of node i and is the degree of node i . Figure 1 shows the clustering coefficients for the frequency band of the three cases and their corresponding averages. Provides a visual representation of clustering coefficients in EEG data that can be used as features to understand neural dynamics in auditory disorders better [ 43 ]. The clustering coefficient measures how tightly connected nodes are in a network, reflecting the degree of local connectivity in the EEG data. Across all scan channels, the ”Deafness” case tends to have slightly higher clustering coefficients compared to the ”Normal” and ”Tinnitus” cases. Additionally, the ”delta” scan channel exhibits the highest clustering coefficients among all cases. Furthermore, scan channels ”alpha1” and ”alpha2” demonstrate relatively high clustering coefficients for all diagnosis cases. The clustering coefficients obtained from the EEG scan data for the three cases of auditory diagnosis have ignited a strong motivation to explore machine learning for classification. These coefficients provide a unique perspective on local connectivity patterns within the EEG data, offering insights into the intricate relationships between different scan channels and diagnostic outcomes. The higher clustering coefficients associated with the ”Deafness” case across various scan channels hint at distinctive network properties that may indicate auditory disorders. This observation sparks curiosity about the underlying neural dynamics and its potential role in auditory conditions. Moreover, the clustering coefficients shed light on the complex interplay between brain regions and their connectivity patterns. This intricate network of connections can be harnessed as valuable features for machine learning models to classify auditory diagnosis cases. Variations in clustering coefficients between different scan channels highlight the potential for discriminative features that capture the essence of each diagnostic category. We aim to create accurate machine learning models that classify individuals by auditory diagnosis for early detection to get accurate treatment.

Experimental findings and results In audiology, it can be challenging to differentiate individuals who suffer from hearing loss or tinnitus from normal cases. It demands careful identification and comprehensive analysis, which cannot be underestimated. Patients suffering from auditory disorders experience a significant reduction in their quality of life. Therefore, it is essential to develop accurate and efficient classification methods. This study presents a novel approach that utilizes Multi-view Brain Networks data and various ensemble learning technologies. Analyzing brain connectivity data obtained from EEG measurements can be difficult due to the many dimensions involved. This study effectively overcomes the challenge by implementing PCA. PCA can decrease the number of dimensions of features but also enhance the accuracy of the diagnosis process [ 66 ]. Classification results Table 3 provides a comprehensive overview of the performance of each of the four ensemble learning models per class. As an instance, the Extra Trees model indicates CVA rates of 89.58% for individuals with deafness and 86.96% for those tinnitus cases. These metrics reflect the model’s capability to diagnose the respective classes accurately. Moreover, the recall values, which measure the model’s accuracy in correctly identifying instances from each class, are closely aligned with the precision values. When evaluating the performance of different models, the F1-score is a useful metric that considers both precision and recall. Specifically, it measures how well a model balances correct classifications by minimizing false positives and negatives. For instance, the Gradient Boosting model has an F1-score of 88.22 for the deafness class, indicating its ability to achieve this balance effectively. Table 4 presents the evaluation results of the proposed models for classifying individuals with auditory disorders. The Extra Trees model demonstrated a balanced performance with accuracy, precision, recall, and F1-score of approximately 89.5%, indicating consistent accuracy across different classes. Moreover, CatBoost had slightly lower precision and recall values within the range of 89%. However, the Random Forest model outperformed the other models with accuracy and F1-score of 89.76%, accompanied by corresponding precision and recall values of 89.74%. This shows a solid ability to minimize false positives and false negatives. Although Gradient Boosting had slightly lower metrics, it maintained competitive precision, recall, and F1-score scores above 89%. Also, Table 4 displays the performance metrics that evaluate the effectiveness of the proposed models in handling the classification of auditory disorders. This analysis gives a complete overview of their performance. Interestingly, Extra Trees had the highest Kappa and MCC values, reaching over 0.84. This highlights its strong ability to capture the underlying patterns accurately. Extra trees consistently displayed low values for zero-one loss and hamming loss, with a score of 0.10476, highlighting its ability to minimize misclassifications across multiple classes. The Random Forest algorithm demonstrated exceptional performance, achieving Kappa and MCC values above 0.83. This highlights its ability to classify data and minimize prediction errors accurately. Even though CatBoost and Gradient Boosting had slightly lower metrics than Extra Trees and Random Forest, they still had impressive Kappa and MCC values, exceeding 0.79. Furthermore, their zero-one loss and hamming loss results were slightly elevated but fell within an acceptable range. Based on the results, it is clear that ensemble learning models were successful in multiple performance metrics. In particular, Extra Trees and Random Forest demonstrated exceptional performance across all metrics, proving their ability to classify individuals with auditory disorders while accurately minimizing classification errors. We have included mean and variance values, developing our model performance evaluation. Mean values provide an important direction measure, showing the average performance across different cross-validation folds. For instance, the mean classification results range from 89.02 to 89.75%, demonstrating our models’ consistency. Variance values provide insights into the spread or variability of the results, highlighting the stability of our proposed methodology. Low variance values, such as 0.00015, suggest a narrow distribution of performance metrics, increasing the robustness and reliability of our classification models. This accurate examination of mean and variance values enhances the clarity and completeness of our evaluation, contributing to a more slight understanding of our models’ performance characteristics. Proposed study findings and discussion To accurately diagnose auditory disorders, evaluating the effectiveness of different methodologies is essential. Recent advancements in this field have the potential to significantly improve diagnostic and therapeutic approaches, ultimately leading to better patient care. Pei-Zhen et al. [ 25 ] conducted a study using a Random Forest algorithm-based classification model to enhance the precision of auditory disorder diagnoses. We performed a comparative analysis of our proposed methodology and the findings presented by Pei-Zhen et al. in their study. Our results are shown in Fig. 2 , where the bars and error indicators illustrate the superiority of our approach in terms of both performance and stability. Furthermore, our proposed model outperformed Pei-Zhen et al.’s study, indicating significant improvement in classification accuracy. This promising result could lead to the development of refined diagnostic frameworks and more effective therapeutic interventions for auditory disorders. Table 5 compares the proposed approach and Pei-Zhen et al.’s method, indicating that the difference is significant with a p -value of 0.0001 and a test statistic of 210. There is a significant difference in performance between the two methods, with a test statistic of 210 and a p -value of 0.0001. Our approach outperforms Pei-Zhen et al.’s method significantly. The test statistic reflects the magnitude of this difference, and the low p-value indicates a high level of significance, further supporting the superiority of our approach. Table 6 provides the statistical significance of proposed different classifiers used in diagnosing auditory disorders with Multi-View Brain Network data. It presents the outcomes of the Wilcoxon signed-rank test, which includes test statistics, p -values, and corresponding significance classifications. This information sheds light on the effectiveness of each classifier and their differences in the diagnosis of auditory disorders. The comparison between the ”Random Forest” classifier and other classifiers such as ”Extra Trees”, ”Gradient Boosting”, and ”CatBoost” showed statistically significant differences in accuracy. The test statistic was 22.5000 with a p -value of 0.0296 for ”Extra Trees”, 0.001 with a p -value of 0.0039 for ”Gradient Boosting”, and 12.200 with a p -value of 0.0002 for ”CatBoost”. These results highlight the significant differences in diagnostic performance between ”Random Forest” and each of these classifiers, emphasizing the critical nature of classifier selection in influencing diagnostic accuracy. On the other hand, the comparison between ”Extra Trees” and ”Gradient Boosting” resulted in a ”Not significant” outcome with a test statistic of 48.0000 and a p -value of 1.0000. This implies a lack of statistically discernible differences in accuracy between these two classifiers, suggesting a certain level of similarity in their diagnostic performance. The comparisons between ”Extra Trees” and ”CatBoost” (test statistic = 4.0000, p -value = 0.0156) and ”Gradient Boosting” and ”CatBoost” (test statistic = 8.0000, p -value = 0.0010) produced ”Significant” results, indicating significant disparities in diagnostic accuracy between these classifier pairs. These findings highlight the important impact of classifier selection on diagnostic outcomes, as the choice between ”Extra Trees” and ”CatBoost” or ”Gradient Boosting” and ”CatBoost” significantly influences the overall diagnostic performance in auditory disorders. In closing, the outcomes of the Wilcoxon signed-rank test provide a slight understanding of the statistical significance underpinning the comparative performance of diverse classifiers in the diagnosis of auditory disorders using Multi-View Brain Network data. The discerned ”Significant” disparities in accuracy, presented in the comparisons involving Random Forest with Extra Trees, Gradient Boosting, and CatBoost, highlight the significant differences in performing Random Forest compared to these classifiers. Conversely, the observed ”Not Significant” result for the Extra Trees vs. Gradient Boosting pairing explains the absence of a discernible difference in accuracy, emphasizing the similarity in their performance. The additional ”Significant” findings for the Extra Trees vs. CatBoost and Gradient Boosting vs. CatBoost comparisons highlight the essential role of classifier selection in auditory disorder diagnosis. These results highlight the significance of robust statistical analyses in guiding the selection of the most effective models, presenting a promising avenue for advancing early detection and personalized treatment strategies and contributing to reducing the quality of life for individuals with auditory disorders.

Proposed study findings and discussion To accurately diagnose auditory disorders, evaluating the effectiveness of different methodologies is essential. Recent advancements in this field have the potential to significantly improve diagnostic and therapeutic approaches, ultimately leading to better patient care. Pei-Zhen et al. [ 25 ] conducted a study using a Random Forest algorithm-based classification model to enhance the precision of auditory disorder diagnoses. We performed a comparative analysis of our proposed methodology and the findings presented by Pei-Zhen et al. in their study. Our results are shown in Fig. 2 , where the bars and error indicators illustrate the superiority of our approach in terms of both performance and stability. Furthermore, our proposed model outperformed Pei-Zhen et al.’s study, indicating significant improvement in classification accuracy. This promising result could lead to the development of refined diagnostic frameworks and more effective therapeutic interventions for auditory disorders. Table 5 compares the proposed approach and Pei-Zhen et al.’s method, indicating that the difference is significant with a p -value of 0.0001 and a test statistic of 210. There is a significant difference in performance between the two methods, with a test statistic of 210 and a p -value of 0.0001. Our approach outperforms Pei-Zhen et al.’s method significantly. The test statistic reflects the magnitude of this difference, and the low p-value indicates a high level of significance, further supporting the superiority of our approach. Table 6 provides the statistical significance of proposed different classifiers used in diagnosing auditory disorders with Multi-View Brain Network data. It presents the outcomes of the Wilcoxon signed-rank test, which includes test statistics, p -values, and corresponding significance classifications. This information sheds light on the effectiveness of each classifier and their differences in the diagnosis of auditory disorders. The comparison between the ”Random Forest” classifier and other classifiers such as ”Extra Trees”, ”Gradient Boosting”, and ”CatBoost” showed statistically significant differences in accuracy. The test statistic was 22.5000 with a p -value of 0.0296 for ”Extra Trees”, 0.001 with a p -value of 0.0039 for ”Gradient Boosting”, and 12.200 with a p -value of 0.0002 for ”CatBoost”. These results highlight the significant differences in diagnostic performance between ”Random Forest” and each of these classifiers, emphasizing the critical nature of classifier selection in influencing diagnostic accuracy. On the other hand, the comparison between ”Extra Trees” and ”Gradient Boosting” resulted in a ”Not significant” outcome with a test statistic of 48.0000 and a p -value of 1.0000. This implies a lack of statistically discernible differences in accuracy between these two classifiers, suggesting a certain level of similarity in their diagnostic performance. The comparisons between ”Extra Trees” and ”CatBoost” (test statistic = 4.0000, p -value = 0.0156) and ”Gradient Boosting” and ”CatBoost” (test statistic = 8.0000, p -value = 0.0010) produced ”Significant” results, indicating significant disparities in diagnostic accuracy between these classifier pairs. These findings highlight the important impact of classifier selection on diagnostic outcomes, as the choice between ”Extra Trees” and ”CatBoost” or ”Gradient Boosting” and ”CatBoost” significantly influences the overall diagnostic performance in auditory disorders. In closing, the outcomes of the Wilcoxon signed-rank test provide a slight understanding of the statistical significance underpinning the comparative performance of diverse classifiers in the diagnosis of auditory disorders using Multi-View Brain Network data. The discerned ”Significant” disparities in accuracy, presented in the comparisons involving Random Forest with Extra Trees, Gradient Boosting, and CatBoost, highlight the significant differences in performing Random Forest compared to these classifiers. Conversely, the observed ”Not Significant” result for the Extra Trees vs. Gradient Boosting pairing explains the absence of a discernible difference in accuracy, emphasizing the similarity in their performance. The additional ”Significant” findings for the Extra Trees vs. CatBoost and Gradient Boosting vs. CatBoost comparisons highlight the essential role of classifier selection in auditory disorder diagnosis. These results highlight the significance of robust statistical analyses in guiding the selection of the most effective models, presenting a promising avenue for advancing early detection and personalized treatment strategies and contributing to reducing the quality of life for individuals with auditory disorders.

Conclusion and future scope The main aim of this study is to address the essential challenges associated with differentiating individuals with hearing impairments, such as deafness and tinnitus, from those with normal hearing. The study has contributed significantly to the field by presenting an improved approach for accurately classifying auditory disorders. We achieved this by using Multi-View Brain Network data and leveraging advanced machine learning algorithms, including Random Forest, Extra Trees, Gradient Boosting, and CatBoost. Our research findings have revealed that our proposed model performs exceptionally well across multiple evaluation metrics, such as accuracy, precision, recall, and F1-score. Notably, the Random Forest model has shown outstanding accuracy, precision, and F1-score values, highlighting its effectiveness in differentiating between different subject groups. These promising results have the potential to revolutionize the early detection and personalized treatment of auditory disorders, leading to better patient outcomes and an enhanced quality of life. Our study investigates the performance of different classifiers in diagnosing auditory disorders and compares them using the Wilcoxon signed-rank test. This statistical analysis emphasizes the importance of selecting the appropriate classifier for accurate diagnosis. The significant differences in accuracy between various classifiers highlight the critical need to choose the right model for maximizing diagnostic accuracy. In the future, more research can be conducted to improve this work by including other data sources, such as neuroimaging data or genetic markers, which can increase the model’s predictive power. Testing the model’s reliability among different populations is also recommended to ensure its effectiveness in various clinical settings. As deep learning and neuroscience advance, there is a great opportunity to refine and enhance the model’s methods, which can lead to significant progress in audiology and auditory disorder diagnosis.

In the field of audiology, achieving accurate discrimination of auditory impairments remains a formidable challenge. Conditions such as deafness and tinnitus exert a substantial impact on patients’ overall quality of life, emphasizing the urgent need for precise and efficient classification methods. This study introduces an innovative approach, utilizing Multi-View Brain Network data acquired from three distinct cohorts: 51 deaf patients, 54 with tinnitus, and 42 normal controls. Electroencephalogram (EEG) recording data were meticulously collected, focusing on 70 electrodes attached to an end-to-end key with 10 regions of interest (ROI). This data is synergistically integrated with machine learning algorithms. To tackle the inherently high-dimensional nature of brain connectivity data, principal component analysis (PCA) is employed for feature reduction, enhancing interpretability. The proposed approach undergoes evaluation using ensemble learning techniques, including Random Forest, Extra Trees, Gradient Boosting, and CatBoost. The performance of the proposed models is scrutinized across a comprehensive set of metrics, encompassing cross-validation accuracy (CVA), precision, recall, F1-score, Kappa, and Matthews correlation coefficient (MCC). The proposed models demonstrate statistical significance and effectively diagnose auditory disorders, contributing to early detection and personalized treatment, thereby enhancing patient outcomes and quality of life. Notably, they exhibit reliability and robustness, characterized by high Kappa and MCC values. This research represents a significant advancement in the intersection of audiology, neuroimaging, and machine learning, with transformative implications for clinical practice and care. Keywords

Related work In recent years, the intersection of audiology, neuroimaging, and machine learning has prompted various investigations to advance the understanding and diagnosis of auditory disorders. A survey of relevant literature reveals several pertinent studies that contribute to developing similar methodologies. Chen et al. [ 21 ] explored the application of fMRI data in characterizing functional brain network alterations associated with tinnitus. Their study underscored the utility of resting-state fMRI data in identifying distinctive connectivity patterns within the auditory networks of tinnitus patients. Smith and Jones [ 22 ] conducted an extensive review of neuroimaging studies focused on deafness-related plasticity in the auditory cortex. Their synthesis highlighted the remarkable capacity of the brain to reorganize neural pathways in response to auditory deficits, a phenomenon contributing to the establishment of novel diagnostic frameworks. In machine learning, Li et al. [ 23 ] explored the efficacy of Support Vector Machines in classifying individuals with tinnitus based on their neuroimaging profiles. Their results demonstrated promising classification accuracy, motivating further exploration of diverse machine learning techniques as we undertake in this study. Moreover, Johnson et al. [ 24 ] intersects with our methodology by utilizing Multi-View Brain Network data to differentiate neurological disorders. While not confined to auditory disorders, their successful integration of multi-view data offers a model for a multi-modal approach. EEG data were used in [ 25 ] to build a brain networks model and detect functional connectivity patterns for individuals with auditory disorders. By analyzing functional connectivity in brain networks, another study [ 26 ] sought to differentiate between prelingually deaf infants with and without cochlear implants. Using a novel method for dividing regions of interest (ROIs), the study obtained significant enhancements in classification accuracy. In addition, machine learning has been used to predict normal and pathological phenotypes from large-scale human brain networks by comparing various brain network kernels for classification purposes [ 27 ]. Moreover, functional near-infrared spectroscopy (fNIRS) and machine learning have been used to differentiate individuals with and without tinnitus, with significant differences between tinnitus patients and controls in resting-state measures of connectivity and evoked responses [ 28 ]. While fMRI studies have examined brain activation in tinnitus patients, cognitive control, and default mode networks may be involved in non-auditory aspects of the disorder [ 29 ]. The structure of the human cerebral cortex can be estimated using intrinsic functional connectivity. Using resting-state functional connectivity magnetic resonance imaging (MRI), the configuration of networks in the human cerebrum was investigated [ 30 ]. Local networks confined to the sensory and motor cortices and distributed networks of association regions were discovered. Functional connectivity within the sensory and motor cortices followed topographic representations across adjacent areas, whereas connectivity patterns in the association cortex frequently exhibited abrupt network boundary transitions [ 31 ]. According to another study, three interdependent architectural gradients underline the organization of intrinsic functional connectivity in the human cerebral cortex. These gradients correlated with external versus internal information sources, content representation versus attentional modulation, and central versus peripheral brain regions [ 32 ]. In addition, intrinsic functional connectivity MRI was used to compare rodent and human cortico-hippocampal connectivity. The results demonstrated preferential connectivity of sensory cortical networks in rats, as opposed to association cortical networks in humans [ 33 ]. Using machine learning techniques, human brain networks can be analyzed [ 34 ]. These techniques employ various algorithms, including K-Nearest Neighbor, Support Vector Machine, and Artificial Neural Network, to analyze brain images and identify connectivity patterns [ 35 ]. Machine learning methods can also be used to develop fMRI network inference methods, such as BrainNET, which quantify the contributions of various brain regions [ 36 ]. Deep learning techniques, such as Graph AuTo-Encoding (GATE), have been devised to characterize the population distribution of brain graphs and infer their relationships with human characteristics [ 37 ]. In addition, deep learning methods have been applied to classify brain networks for detecting Alzheimer’s disease (AD) [ 38 ]. Several statistical and machine learning link selection methods have been evaluated for brain functional networks, resulting in better utilization of network representations. Multimodal neuroimaging can present valuable information in the diagnosis of dementia. However, the small size of complete multi-modal data limits the ability of representation learning. In Ref. [ 39 ], the authors proposed a novel framework for the AD diagnosis called Multimodal-Representation-Learning and Adversarial Hypergraph-Fusion. This framework combines distribution-based GraphGAN and CNN-based GraphAE to extract features in the representation space. An adversarial strategy is utilized in modal fusion to improve the accuracy of AD detection. Results obtained on the ADNI dataset show that prior information can help enhance discrimination of representation learning. Also, adding more modalities can improve the detection performance. In Ref. [ 40 ], the authors proposed a novel Consistent Perception Generative Adversarial Network (CPGAN) for semi-supervised stroke lesion segmentation. The proposed CPGAN can reduce the reliance on fully labeled samples. Specifically, a Similarity Connection Module (SCM) was designed to capture the information of multi-scale features. The proposed SCM can selectively aggregate the features at each position by a weighted sum. An assistant network was constructed using a consistent perception strategy to improve meaningful feature representation learning to enhance brain stroke lesion prediction accuracy for unlabeled data. They employed the assistant network and the discriminator to decide whether the segmentation results were real or fake. The CPGAN was evaluated on the Anatomical Tracings of Lesions After Stroke (ATLAS). The experimental results demonstrated that the proposed network achieves superior segmentation performance. Ensemble learning classifiers Identifying auditory impairments is challenging but crucial to improving quality of life. Efficient classification methods are needed to distinguish between affected and healthy individuals. Ensemble learning classifiers in audiology combine multiple models to improve classification accuracy and reduce variability and noise in auditory data [ 44 , 45 ]. To effectively address auditory impairments, classification methods must be able to identify and understand subtle patterns and relationships within the data. Ensemble learning classifiers inherently capture diversity by incorporating distinct base classifiers, each specialized in recognizing specific patterns within the auditory features [ 46 ]. By using techniques such as bagging, boosting, and stacking, ensemble classifiers can better utilize the complex nature of auditory data [ 47 ]. Ensemble learning classifiers can accurately diagnose and treat auditory impairments using multiple models, providing better care for those with auditory disorders. Given an annotated dataset , where represents the input features and is the corresponding target label to build the ensemble. Extra Trees classifier The Extra Trees classifier is known for effectively capturing complex patterns within intricate datasets [ 48 ]. Furthermore, brain networks provide a comprehensive display of functional connectivity across various neural regions. The Extra Trees classifier is a random forest algorithm that seeks to improve diversity and generalization by creating a group of decision trees [ 49 ]. It uses the input features and their corresponding target labels to build the ensemble, which involves training several decision trees. A single decision tree in the Extra Trees ensemble is grown using a random subset of features and training samples [ 50 ]. The splitting process aims to reduce the variance and overfitting inherent in individual trees [ 49 ]. The prediction of the t th tree is represented by , which is based on the input feature vector x provided. The ensemble prediction for an input x is calculated by averaging the predictions of all trees as presented in Eq. 3 : The variable T represents the total number of trees in the ensemble. To create a prediction model, several decision trees are generated by training them in various random subsets of the data. The final prediction is made by combining the predictions of these trees. The randomness in the training process helps increase the model’s diversity and prevents overfitting [ 51 ]. This algorithm is simple and works as an ensemble. The algorithm’s simplicity and ensemble approach make it effective for various classification tasks, including diagnosing auditory disorders such as deafness, tinnitus, and normal auditory function. By enhancing the diversity among decision trees and considering the multi-view aspect of brain networks, the Extra Trees classifier holds the potential to enhance the accuracy and reliability of complex medical classifications. The combination of Extra Trees can lead to promising results in diagnosis. CatBoosting classifier The CatBoosting classifier is a tool that can identify complex patterns within various datasets [ 52 ]. One area where it has been particularly useful is in analyzing the multi-view nature of brain networks. This allows for a detailed understanding of how different parts of the brain are connected and function together. The CatBoosting classifier is designed to improve an objective function by creating a sequence of decision trees. It achieves this by repeatedly building an ensemble of decision trees. The main function for CatBoosting can be expressed as in Eq. 4 : where represents the model parameters, is the ensemble prediction for input , L is the loss function measuring the discrepancy between predicted and true labels, and is a regularization term for the t -th decision tree . Each decision tree is constructed by recursively partitioning the feature space based on binary splits. The splitting process minimizes the loss function by determining optimal threshold values for each feature. Furthermore, CatBoosting uses a category-specific enhancement mechanism to effectively handle categorical characteristics [ 53 ]. When building trees with CatBoost, it is crucial to follow a step-by-step approach. Each new tree should be trained to learn the residuals of the previous ensemble’s predictions. This is done through gradient boosting, where new trees are added to reduce the gradient of the loss function. Combining the CatBoosting classifier with the multi-view approach can create a more accurate diagnostic model for identifying auditory disorders. Gradient Boosting classifier The Gradient Boosting algorithm is a powerful classifier that can detect complex patterns in complicated datasets [ 54 ]. Combined with the multi-view analysis of brain networks, it comprehensively represents functional connectivity across different brain regions. Gradient Boosting combines weak learners to create a powerful predictive model [ 55 ]. The objective is to find an additive model as denoted in the Eq. 5 : where F ( x ) is the final prediction for input x , T is the total number of weak learners, is the weight assigned to weak learner which represents the output of the t th weak learner. Gradient Boosting minimizes the difference between predicted and actual labels using a loss function as denoted L ( y , F ( x )). This is done by adding weak learners to the ensemble in an iterative manner. In each iteration, a new weak learner denoted , is trained to estimate the negative gradient of the loss function related to the predictions of the current ensemble. This negative gradient is represented by the Eq. 6 : The weight is determined by minimizing the loss function when the new weak learner is presented as in Eq. 7 : where represents the ensemble’s prediction up to the -th iteration. The ensemble prediction is then updated as in Eq. 8 : Gradient Boosting uses weak learners and adjusts their weights to create an accurate predictive model. This method combines the strengths of multiple learners for more accurate results [ 56 ]. To diagnose auditory disorders effectively, use gradient boosting and multi-view network data in the proposed model. Random Forest classifier The Random Forest algorithm is a trustworthy machine learning technique that uses ensemble learning to achieve accurate classifications [ 57 ]. This classifier is essential for managing large and complex data, which has been widely used in various fields [ 50 , 58 , 59 ]. The Random Forest classifier is an ensemble learning technique that aims to improve the predictive accuracy and robustness of individual decision trees [ 60 ]. The proposed algorithm constructs an ensemble of decision trees to make predictions. In a Random Forest ensemble, each decision tree is built by dividing the feature space using binary splits. To introduce randomness, only a subset of features is considered for each split during training. The output of a decision tree for a given input feature vector is denoted as . The ensemble prediction for input x is obtained by aggregating the predictions of all trees as presented in Eq. 3 . The Random Forest algorithm mitigates the overfitting often associated with individual decision trees. The ensemble approach leverages the diverse perspectives of individual trees to improve generalization performance. By introducing randomness in the feature selection process and aggregating the predictions, Random Forest balances bias and variance, leading to robust and accurate predictions. Experimental modeling hypothesis Developing a robust classification model to differentiate between individuals with auditory disorders requires the integration of Multi-View Brain Networks. These complex neural connectivity patterns captured by Multi-View Brain Networks provide a rich source of information for the classification process. Therefore, leveraging these multi-dimensional brain network representations can potentially build an effective and accurate classification model. This model can distinguish between different auditory diagnosis classes and underlying neural mechanisms associated with these conditions. This can be a significant step in enabling early diagnosis and treatment of auditory disorders. Evaluation metrics To rigorously assess the performance and efficacy of our classification model, we employ a comprehensive set of evaluation metrics, each offering unique insights into its capabilities. These metrics include: Accuracy: The main measure for performance by computing the proportion of instances that are correctly classified and provides an overall model performance. Cross-validation accuracy (CVA): An important metric that measures the performance of a model across multiple iterations of cross-validation. It ensures that the classification model is consistent and can make accurate predictions across diverse data. Precision: A measure that quantifies the proportion of true positive predictions relative to the total positive predictions when evaluating the model’s ability to make accurate positive classifications. Recall: The appropriate measure to evaluates the model’s ability to identify all actual positive instances by calculating the proportion of true positives identified correctly. F1-score: A measure balances precision and recall by considering false positives and negatives, providing a robust measure of accuracy. Kappa: A measure provides a reliable measure of classification understanding between the model’s predictions and actual labels. MCC: A useful metric for imbalanced data. It measures the correlation between a model’s predictions and the actual labels. Zero-one loss: A measures for the ratio of misclassified instances and highlights the severity of misclassifications. Hamming loss: A measures computes the proportion of incorrectly assigned labels, providing insight into multi-class classification accuracy. We use Multi-View Brain Networks to classify auditory diagnostic cases through a hypothesis-based methodology. By incorporating essential metrics and diverse evaluation techniques, we aim to conduct a thorough assessment of the classification model’s effectiveness by leveraging a comprehensive suite of evaluation metrics (Table 1 ). Proposed modeling: experimental setup To create accurate classification models for individuals with auditory disorders, a comprehensive experimental setup was designed to ensure reliable and meaningful results. This involved fine-tuning the hyperparameters and optimizing the preprocessing steps to enhance the performance of the ensemble learning model. The study used extra trees, random forest, gradient boosting, and CatBoost due to their proven efficacy in complex classification tasks. Due to their intrinsic capacity to capture complex relationships within the data, these models were considered suitable for diagnosing auditory disorders from Multi-View Brain Networks data. An essential step in the experimental process was to preprocess the raw data, improving its suitability for model training and evaluation. PCA was used to reduce the dimensionality of brain connectivity data effectively. This technique reduces feature space while retaining influential patterns. The data were then divided into training and testing subsets for model evaluation. Normalization was applied as an essential preprocessing step to ensure all features were on the same scale. This minimized the influence of varying scales on model performance and accelerated the convergence of iterative optimization algorithms. As a result, the training process was more efficient and effective. Table 2 shows the selected hyperparameters for each model after tuning. Wilcoxon signed-rank test The Wilcoxon signed-rank test is a non-parametric statistical test used to determine whether there are statistically significant differences between two related or paired groups or conditions [ 61 ]. It is widely used for experiment evaluation purposes such as in [ 62 – 64 ]. This test is particularly useful when the assumptions of normality and equal variances still need to be met or when dealing with ordinal or non-normally distributed data. The Wilcoxon signed-rank test compares two sets of related or paired observations [ 65 ]. These paired observations can represent measurements taken before and after an intervention or any other related data points. Hyperparameters were fine-tuned through grid search and cross-validation technique ( K =5 folds) to optimize performance and improve generalization while mitigating overfitting. The combined approach of preprocessing, hyperparameter tuning, and ensemble learning was tested. Models trained on preprocessed data with optimized hyperparameters showed better performance in classifying individuals with auditory disorders.

Acknowledgements We appreciate the editorial board and reviewers feedback. Author contributions M.A. directed the research and planned the hypothesis as the corresponding author. M.A. and Y.A. meticulously conducted the experiments and produced the figures. Both authors actively contributed to the composition and editing of the manuscript. Y.A. actively contributed to experimental work, gathering data, creating figures, and revising manuscripts. E.M. and E.A. contributed to developing and improving the manuscript by collecting data, defining the EDA hypothesis, and improving the writing and editing article organization. Funding Open access funding provided by The Science, Technology & Innovation Funding Authority (STDF) in cooperation with The Egyptian Knowledge Bank (EKB). Science and Technology Development Fund (STDF Egypt). Availability of data and materials The authors declare the source code and any additional materials are accessible upon request. Declarations Ethics approval and consent to participate The proposed data used in our experiments are publicly accessible data that are free permission to use [ 41 ]. So, there is no need for ethics approval or consent to participate. Consent for publication Not applicable. Competing interests The authors assert no contending interests.

CC BY

no

2024-01-16 23:41:59

Brain Inform. 2024 Jan 14; 11(1):3

oa_package/ce/cd/PMC10788326.tar.gz

PMC10788327

38219269

Background Physical activity (PA) is described as 'any body movement produced by skeletal muscles that requires energy expenditure', including activities performed at work, play, housework, travel, and recreation [ 1 ]. It provides health benefits, including prevention and treatment of many conditions such as osteoarthritis, low back pain, hypertension, stroke, obesity, diabetes, and mental health disorders (i.e., distress, anxiety and symptoms of depression both in healthy and ill adults) [ 1 , 2 ]. There is also evidence that the risk of frailty might be reduced by modifying levels of PA [ 3 ]. Almost 30% of adults do not follow PA recommendations [ 4 ]. Failure to meet PA recommendations is likely influenced by a modern society that promotes long-term sitting during free time, work, and commuting. Sedentary behaviours (SB) are associated with metabolic diseases (e.g., type 2 diabetes mellitus), cancer, musculoskeletal conditions, cardiorespiratory diseases, possibly increasing mortality, especially among individuals with poor socioeconomic status [ 5 , 6 ]. The World Health Organization has promoted the Global Action Plan on Physical Activity 2018–2030 [ 1 ], intending to improve PA by 15% by 2030. The plan includes 20 policy actions to ensure equal opportunities and enable the environment to be physically active, including digital interventions such as mobile apps, remote counselling, and wearable devices (WDs). WDs for physical activity tracking, are electronic non-invasive monitoring devices, mainly in the form of wrist devices that enable the tracking of PA metrics (e.g., the number of steps taken, energy expenditure, time spent sleeping and time spent in different activities levels) [ 7 , 8 ]. In particular, WDs might serve as a useful tool to be included in broader programmes to increase PA, stress management, physical and mental quality of life, and reduce SB and weight [ 9 , 10 ]. The global fitness tracker market has been valued at around US$ 40 billion in 2022, up from US$ 36 billion in 2020, and is expected to expand to US$ 46 billion in 2023 globally ( https://market.us/report/fitness-tracker-market/#overview ). The market value is forecast to reach US$ 187 billion by 2032 ( https://market.us/report/fitness-tracker-market/#overview ). Increasing market growth has occurred concomitantly with research, as several systematic reviews (SRs) have been published on the efficacy of WDs. To date, contrasting results are reported when activity was measured as SB or light PA [ 11 – 13 ]. One umbrella review systematically analysed secondary studies on the effectiveness of WDs on PA levels [ 14 ], retrieving systematic reviews published until April 2021, with WDs as a key intervention. There is an opportunity for an updated umbrella review, encompassing the latest systematic reviews on WDs, either alone or combined with other interventions. Furthermore, the aforementioned recent umbrella review [ 14 ] did not investigate SB, providing room for improvement for evaluating the efficacy of WDs, considering that SB is widely studied alongside physical activity. This umbrella review aims to summarise the available SRs regarding the efficacy of WDs use on increased PA levels and reducing SB in adults. Research Question The research question was: does using WDs increase PA levels and reduce SB in adults (aged ≥ 18 years)?

Methods We conducted an umbrella review of SRs in accordance with the Cochrane Handbook’s chapter on overviews of reviews and the Joanna Briggs Institute Manual for Evidence Synthesis [ 15 , 16 ]. We followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) [ 17 ] for the flow chart and the Preferred Reporting Items for Overviews of Reviews (PRIOR) [ 18 , 19 ], as reporting checklist (Additional file 1 : Supplemental File 1). The review protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO) database (CRD42022339140). Summary of methods and deviations from the protocol are reported in Additional file 1 : Supplemental File 2.

Results Study Inclusion The search identified 278 publications from databases, 3426 from registers, and 15 through other sources. Of the 85 SRs screened by the full text, 34 were excluded (Additional file 1 : Supplemental File 5) and 51 SRs [ 11 , 13 , 34 – 81 ] were included (Fig. 2 ). Characteristics of Studies Included The included SRs were published in 37 journals between the dates of 2007 to 2022. Of the 51 included SRs, 74.5% incorporated a meta-analysis. The majority of SRs included only RCTs (80.4%). The characteristics of the SRs are described in Table 1 and details in Additional file 1 : Supplemental File 6. The SRs included a median of 17 primary studies and a median of 2,355 participants per SR; across reviews there were 302 unique primary studies. Overall, 22 SRs (43%) included people with pathologies (e.g., orthopaedic, rheumatological, neurological, cardiometabolic, and tumours), 19 (37%) involved a mixed population, five (10%) focused exclusively on obese or overweight people, and five (10%) older adults. Most of the 51 SRs assessed steps per day outcome ( n = 33), followed by minutes of MVPA per week ( n = 21), composite outcome ( n = 17), and minutes SB per day ( n = 9). In 48 SRs (94%), there were multi-component interventions, whereas in two SRs, data were unavailable. Overall, in the SRs with multi-component intervention, the intervention was a combination of WDs as the main component of the intervention, or as part of a multimodal intervention consisting of various elements (e.g., use of a wearable device and a diary to record step count with feedback from a facilitator, or use of the wearable device and telephone support). Across the 48 SRs, the most frequent brands of WDs were Fitbit ( n = 21), Jawbone ( n = 12 SRs), and Polar ( n = 11 SRs)-no data on brands of worn devices were provided in four SRs. Methodological Quality The evaluation with the AMSTAR 2 checklist outlines that confidence in the results of 51 SRs (72.5%) was rated as ‘critically low confidence’, 11 (21.6%) as ‘low confidence’, and three (5.9%) as ‘moderate confidence’. The primary critical weaknesses corresponded to not providing a list of excluded studies with a justification of the reasons ( n = 44), not using a comprehensive literature search strategy ( n = 29), and not justifying the choice of meta-analysis as an appropriate tool for the statistical combination of results ( n = 19). The most frequent flaws of non-critical weaknesses were not reporting the sources of funding of the studies included in the SRs ( n = 46), not justifying the choice of the design of the studies included in the SRs ( n = 41), and not performing the extraction data by at least two independent authors ( n = 3) (Fig. 3 ). AMSTAR 2 assessments for each SR are reported in Additional file 1 : Supplemental File 7. Systematic Reviews Without Meta-Analysis We found sparse effects that favoured WD from primary studies included in the 13 SRs without meta-analysis. The most reported outcomes were PA (generically and inconsistently defined) and SB. On average, the proportion of trials reporting statistically significant results were 56% (95%CI 0.23%-0.81%) and 32% (95%CI 0.11%-0.69%), respectively (Additional file 1 : Supplemental File 8). Systematic Reviews with Meta-Analysis Overlapping Of the 57 meta-analyses included in the 38 SRs, four meta-analyses from two SRs [ 43 , 71 ] were not considered, because of unclear reporting of primary studies and measure of effect. At the outcome level, we found a slight overlap of citation of primary studies with a CCA of 3.87% for steps per day, 3.12% for minutes of MVPA per week, 4.06% for minutes of SB per day and 2.68% for composite measurements (Fig. 4 ). Similar overlap was reported in subgroup analysis for population, except for steps per day in obese/overweight people and mixed populations, where we found moderate overlap. Overlap was reported in the subgroup of the population at the outcome level (Additional file 1 : Supplemental File 9). Efficacy Results Of the 53 meta-analyses considered for the analysis, most favored intervention using WDs with low-to-moderate certainty of evidence (CoE) ( n = 43, 81.1%). The remaining 10 found no differences between WD and comparators. In Additional file 1 : Supplemental File 10, we report all the effect sizes with CoE assessments and AMSTAR 2 ratings, whereas in Additional file 1 : Supplemental File 11, we report bubble plots linking CoE with the direction of effect overall and stratified for each outcome, underlining concordances and discordances in the direction of the effects. Physical Activity Steps Per Day Seventeen out of 21 meta-analyses (median 679 [IQR 298.7–1474.5] participants) favoured interventions using WDs with low-to-moderate CoE, whereas four reported no differences. At the population level, considering meta-analyses with moderate CCA, we found concordance on the superiority of WDs in all meta-analyses on mixed ( n = 6, CCA moderate 6.98%) and obese/overweight ( n = 3, CCA moderate 9.62%) populations. In contrast, discordance was found in older adults ( n = 2 favour intervention, n = 1 no differences) and people with pathologies ( n = 6 favour interventions, n = 3 no differences). These discordant meta-analyses presented a slight overlap in conditions (older adults 3.85%, pathologies 3.13%). Minutes of Moderate to Vigorous Physical Activity Per Week Eleven out of 12 meta-analyses (median 1,206 [IQR 519–1665] participants) favoured interventions using WDs with low-to-moderate CoE, whereas one found no differences. At the population level, we found concordance on the superiority of WDs in all meta-analyses on obese/overweight ( n = 3) and people with pathologies ( n = 3), whereas discordances were found in a mixed population ( n = 4 favour intervention, 1 no difference). The only meta-analysis on older adults found superiority of WDs over controls. The overlap at the population level was slight (mixed population 3.48%, people with pathologies 0%, obese/overweight people 2.5%). Composite Measurements Thirteen out of 14 meta-analyses (median = 1356 [IQR 867–1435] participants) favoured interventions using WDs with low-to-moderate CoE, whereas one found no difference between groups. At the population level, we found concordances in the superiority of WDs in all the meta-analyses on mixed populations ( n = 7) and people with pathologies ( n = 6). The overlap at the population level was slight (mixed population 4.56%, people with pathologies 3.06%). The only meta-analysis on older adults found no differences between groups. No meta-analyses on obese/overweight people were reported by SRs. Minutes of Sedentary Behaviour Per Day Two out of six meta-analyses (median = 1189 [IQR 288.5–2797] participants) favoured interventions using WDs with low-to-moderate CoE, whereas four found no differences. At the population level, we found discordances in a mixed population ( n = 2 favour intervention, n = 3 no difference) with a slight overlap (4.3%). The only meta-analysis on the older adults found no differences between groups. No meta-analysis on obese/overweight people and people with pathologies were found across SRs. Certainty of Evidence Of the 53 meta-analyses, 29 were rated as moderate CoE ( n = 12 steps per day, n = 7 MPVA, n = 5 SB, n = 5 composite measurements), 21 as low CoE ( n = 8 steps per day, n = 3 MPVA, n = 9 composite measurements, n = 1 SB), and for three, the overall assessment was not possible ( n = 1 steps per day, n = 2 MPVA). Reasons for downgrading were mainly due to serious ( n = 19) and very serious ( n = 12) limitations of methodological quality of SRs, inconsistency (I 2 > 75%) ( n = 23), and risk of bias at the trial level ( n = 22). Most meta-analyses ( n = 47) involved more than 200 participants, indicating precise effect sizes (Additional file 1 : Supplemental File 10). Clinical Relevance and Overall Interpretation of PA and SB In Figs. 5 and 6 , we plotted MDs of all meta-analyses for steps per day and MVPA according to population categories, whereas in Additional file 1 : Supplemental File 12 we plotted SB and composite measurements. Overall, WDs may increase PA by a median of 1,312.23 (IQR 627–1854) steps per day and 57.8 (IQR 37.7 to 107.3) minutes of MVPA per week and may reduce minutes of SB by a median of -27.76 (IQR -41.28 to 9.9) per day. Clinical relevance was found definitive for 15% and ‘probable to possible’ for 48% of SRs for steps per day; definitive for 25%, and ‘probable to possible’ for 67% of SRs for minutes of MVPA per week; and definitive for 16.7% and probable for 33% of SRs for minutes of SB per day (Additional file 1 : Supplemental File 12).

Discussion Main Findings Overall, our results were consistent across different PA outcomes (steps per day, MVPA, and PA as composite outcome) in almost all the SRs, while few discordances in SB were found. With low to moderate CoE and ‘possible to definitive clinical relevance’, using WDs may increase PA by a median of 1,312.23 (IQR 627–1854) steps per day, and a median of 57.8 (IQR 37.7 to 107.3) minutes per week of MVPA compared to passive controls. Sparse results were found for minutes of SB per day (two out of six SRs favoured intervention; median of -27.76 min [IQR -41.28 to 9.9]) with imprecise confidence intervals (i.e., reaching clinical relevance but not statistically significantly different). When we explored the results across different population, SRs reported consistent results on the efficacy of WDs on PA in mixed population (steps per day and composite outcomes) in populations with pathologies (MVPA and composite outcomes) and for obese or overweight populations (steps per day and MVPA). Among older adults, results were inconsistent across outcomes (discordance on steps per day, positive findings on MVPA, no difference on SB), and emerged from small samples in few SRs (< 200 participants for steps per day and MVPA). Generally, the results are comparable to those detected by a previous umbrella review which reported that the use of activity trackers improved PA except for older adults [ 14 ]. However, the SB outcome was not covered by this review [ 14 ]. In our umbrella review, we found some uncertainty for PA in pathologies and older adults subgroups and for SB in mixed and older adults subgroups. Considering the subgroup with pathologies ( n = 23 SRs), we found poorly informative SRs with large confidence intervals in quantitative meta-analyses ( n = 16 SRs) and variability of effects in SRs with qualitative synthesis in approximately one-third of all SRs ( n = 7 SRs). On one hand, some populations, such as orthopaedic patients (osteoarthritis, low back pain), may benefit from physical activity, in terms of locomotor function, balance and strength, whereas other frail populations, with reduced airflow and cardiac capacity, could encounter barriers when trying to increase PA [ 82 ]. However, a recently published SR [ 83 ], which included 38 studies on a population with chronic airway diseases, confirmed the positive results in improving PA when using WDs. Considering the subgroup of older adults, we found different effects compared to the previous umbrella review of Ferguson et al. [ 14 ], which found positive effects on steps per day in this population. The difference might be explained by the absence of a SR that we included [ 57 ]. This SR [ 57 ] did not reach statistical significance for steps per day, contributing to our inconsistent findings. For older adults, increasing their PA levels through the use of WDs might be hampered by difficulty using new technologies, such as activity monitors [ 50 ]. This point aligns with Franssen and colleagues [ 61 ], who reported that younger peoples’ use of wearable trackers was associated with a significantly greater increase in PA. Evidence suggests that engagement in eHealth interventions is reduced among older adults with a lower level of formal education, limited computer experience, and poorer cognition [ 84 ]. Strategies to overcome this barrier could consist of introducing scheduled follow-up assessments by meetings and telephone consultations [ 85 ], in light of evidence of poor communication with health professionals and lack of feedback and human support as hindrances to the acceptance and usability of digital technologies in older adults [ 86 , 87 ]. All SRs with meta-analysis included multicomponent interventions in which WD was the tool used in implementing the activity-based approach. After sub-analysing the types of interventions investigated among SRs, we found that WDs could be more effective when associated with feedback, coaching, or motivational interventions, rather than as a stand-alone intervention, even across populations with pathologies. For example, Laranjo et al. [ 65 ] have demonstrated that the intervention might be more effective if it includes text conveying motivational messages or personalisation (e.g., personalized goal setting, contents and feedback). Furthermore, in keeping with the prior report, interventions might be more effective if based on theories of self-regulation and if interactive functions are adopted to engage users in behaviour change [ 88 ]. For example, artificial intelligence chatbots show great promise in promoting healthy lifestyles and physical activity [ 89 ], based on several features such as understanding user background, establishing persuasive conversations, and providing input according to behavioural outcomes [ 90 ]. Optimal follow-up timing was unclear in most SRs, suggesting a gap in research regarding the most effective length of intervention. The relevance of follow-up length may be related to poor long-term behavioural changes needed with use of WDs, possibly due to the waning initial novelty of WD interventions [ 91 ]. More than half of the people who buy a wearable activity device stop using it, of which 1/3 stop within the first months [ 38 ]. Research and Clinical Practice Implications Our findings suggest that using WDs to promote PA may be effective in populations with and without diseases. However, careful attention should be paid before transferring these results into clinical practice. The CoE was low in less than half of the meta-analysis, meaning that the true effect might be different from the estimated effect. Second, inconsistent results emerged regarding the efficacy of WDs for PA in pathologies and older adults’ subgroups and for SB in mixed and older adults’ subgroups. Third, the follow-up timing was unclear, thus not providing information on the best length of intervention. Fourth, the age of the samples included in the SRs was not always reported, limiting the comparison across SRs. Fifth, the SRs included different types of WDs: each device may have a different accuracy in measuring PA (sensitivity or specificity), which could affect the overall estimation of the effects. Finally, the Hawthorne effect may have influenced the results of the studies analysed by the SRs. For instance, the awareness of being part of a physical activity study could have prompted participants in the control group to increase their activity levels. [ 53 ]. As implications for research, the efficacy of WDs should be further investigated, especially in the long-term and on SB. Future studies are needed to investigate the effectiveness of WDs among older adults and if additional components (e.g., telephone follow-up) might improve WDs' effect in this population. It would be helpful to investigate more in-depth in all populations if and which additional intervention components can increase the effect size with repercussions on PA. Despite these issues, our findings are important for healthcare professionals, who may consider WDs to improve people's health and well-being. For example, there are clear dose–response associations between increasing step counts and decreasing mortality, with 1,000 more steps per day associated with a 15% lower risk in older men [ 92 ] and 6% in younger men and women [ 93 ]. In addition, the NAVIGATOR study [ 94 ], which includes 9,000 individuals with high cardiovascular risk or impaired glucose tolerance, showed that for every increase of 2000 steps per day, the risk of developing cardiovascular problems decreased by 10%, the risk of developing type 2 diabetes mellitus (T2DM) by 5.5%, and the metabolic syndrome risk score was reduced by 0.29 [ 94 – 96 ]. A recent large cohort study on 81,717 participants showed that an increase of 20 min in daily MVPA is associated with a reduction in hospitalization ranging from 3.8% for colon polyps, 14.1% for pneumonia, 19.8% for gallbladder disease, 22.7% for urinary tract infection, and 23% for diabetes [ 97 ] Strengths and Limitations To our knowledge, this is the most comprehensive umbrella review investigating the effectiveness of WDs on PA levels. We set a population limit for adults aged over 18 years, allowing a wide breadth of assessment. A comprehensive hand search detected additional reports for inclusion. We considered different outcomes when assessing PA that allowed us to detect most of the sensitive measures for evaluating the efficacy of WDs. We analysed overlap among SRs, both at the outcome and population level, discovering a slight overlap among most comparisons. This further corroborates our conclusions on the potential benefits of using WDs to improve PA. There are limitations to this umbrella review. Although the decision to include a wide range of populations may provide a broader understanding of the role of WDs, the clinical heterogeneity included in the SRs (e.g., age, type of intervention and comparison, how to use the device, outcome measures), suggests difficulty in comparison and summarization of results, and produces statistical heterogeneity [ 98 ]. In addition, we did not investigate the specific effect of a WD in the presence of multi-component interventions. We adopted the compared interventions as they were reported by the original SR authors, relying on their eligibility criteria and analyses. We cannot exclude bias in the conduct of SRs, although we attempted to reduce this by assessing the methodological quality of SRs. However, we cannot rule out that these factors may have affected the significance of the results.

Conclusions Our results indicate that the adoption of WDs represents a beneficial approach to enhance physical activity across diverse population groups, although there is some level of uncertainty, particularly in the subgroups of individuals with pathologies and older adults, as well as in the case of sedentary behaviour within mixed and older adults' subgroups. Further research is required to enhance the precision of the effects within all population subgroups.

Background Several systematic reviews (SRs), with and without meta-analyses, have investigated the use of wearable devices to improve physical activity, and there is a need for frequent and updated syntheses on the topic. Objective We aimed to evaluate whether using wearable devices increased physical activity and reduced sedentary behaviour in adults. Methods We conducted an umbrella review searching PubMed, Cumulative Index to Nursing and Allied Health Literature, the Cochrane Library, MedRxiv, Rxiv and bioRxiv databases up to February 5th, 2023. We included all SRs that evaluated the efficacy of interventions when wearable devices were used to measure physical activity in adults aged over 18 years. The primary outcomes were physical activity and sedentary behaviour measured as the number of steps per day, minutes of moderate to vigorous physical activity (MVPA) per week, and minutes of sedentary behaviour (SB) per day. We assessed the methodological quality of each SR using the Assessment of Multiple Systematic Reviews, version 2 (AMSTAR 2) and the certainty of evidence of each outcome measure using the GRADE (Grading of Recommendations, Assessment, Development, and Evaluations). We interpreted the results using a decision-making framework examining the clinical relevance and the concordances or discordances of the SR effect size. Results Fifty-one SRs were included, of which 38 included meta-analyses (302 unique primary studies). Of the included SRs, 72.5% were rated as ‘critically low methodological quality’. Overall, with a slight overlap of primary studies (corrected cover area: 3.87% for steps per day, 3.12% for MVPA, 4.06% for SB) and low-to-moderate certainty of the evidence, the use of WDs may increase PA by a median of 1,312.23 (IQR 627–1854) steps per day and 57.8 (IQR 37.7 to 107.3) minutes per week of MVPA. Uncertainty is present for PA in pathologies and older adults subgroups and for SB in mixed and older adults subgroups (large confidence intervals). Conclusions Our findings suggest that the use of WDs may increase physical activity in middle-aged adults. Further studies are needed to investigate the effects of using WDs on specific subgroups (such as pathologies and older adults) in different follow-up lengths, and the role of other intervention components. Supplementary Information The online version contains supplementary material available at 10.1186/s40798-024-00678-9. Key Points There is a moderate certainty of evidence in our umbrella review, which included 51 systematic reviews (of which 38 included meta-analyses with 302 unique primary studies); Available evidence suggests that using wearable devices may effectively increase physical activity across different population in number of steps per day and in minutes spent from moderate to vigorous physical activity per week; Results on efficacy of wearable devices on minutes of sedentary behaviour per day are inconsistent. Supplementary Information The online version contains supplementary material available at 10.1186/s40798-024-00678-9. Keywords

Eligibility Criteria Types of Interventions We included systematic reviews that investigated the use of WDs to improve PA levels. WDs included devices such as accelerometers, pedometers, Electronic Activity Monitor Systems (EAMSs), or global positioning systems (GPS). We included the use of WDs when it was the only component of the intervention or when it was included in a multi-component intervention. Control groups included active, passive, or no interventions, as originally described by SR authors in their eligibility criteria. Passive interventions were defined as those minimal interventions related to PA (e.g., PA educational booklets, PA and dietary counselling), standard of care (e.g., routine outpatient follow-up, standard medical advice) or wait list assignment. Active interventions were considered the same intervention of the intervention group but delivered without WDs, with WDs but blinded, or with another intervention to promote PA. Types of Outcome Measures The primary outcomes were PA level and SB. Physical activity was measured objectively in terms of the number of steps per day, minutes of moderate to vigorous physical activity (MVPA) per week, and/or any composite measurements (e.g., metabolic equivalent for a task [MET], min/week, intensity, time spent walking), whereas SB was objectively measured by minutes per day. Types of Studies In accordance with Cochrane’s definition, all SRs of primary studies (e.g., randomized controlled trials [RCT]) with or without meta-analysis were included [ 20 ]. No restrictions on language and publication date were applied. Figure 1 summarized the eligibility criteria. Search Strategy Two independent authors (JL, CM) launched the search strategy (Additional file 1 : Supplemental File 2) on June 10th, 2022, and updated it on February 5th, 2023, to include the most updated evidence through the following databases: PubMed, Cumulative Index to Nursing and Allied Health Literature, Cochrane Database of Systematic Reviews including the Database of Abstracts of Reviews of Effects (DARE). Both free and MeSH (Medical Subject Headings) terms were used. In addition, a free search was also performed through scientific websites (MedRxiv, Rxiv and bioRxiv databases) adapting the search strategy provided for other databases. If a published scientific version was available in a journal article, we prioritized it. We also checked references of included studies to include other potential reviews. Study Screening and Selection Records retrieved were processed through EndNote X8.2 (Clarivate, Philadelphia) to eliminate duplicates and then uploaded onto the Rayyan website [ 21 ] for selection. Afterwards, two independent researchers (JL, CM) screened records, applying the eligibility criteria to titles and abstracts. Potential eligible records were retrieved to read the full text and determine the final inclusion. A third author (GR) was consulted to reach a consensus in cases of disagreement between reviewers. We evaluated the agreement in the screening process of full-text by Cohen’s kappa statistics resulting in 0.83 (interquartile range, IQR 0.75 – 0.91), indicating a near-perfect agreement [ 22 ]. Data Collection Two independent researchers (SG, SB) extracted the data exactly as they were reported from the original SR using a standardised Microsoft® Excel® 2019 MSO spreadsheet. The extracted data included: characteristics of the SRs (title, year of publication, first author, journal, study design, objective, population analysed, outcome studied) and characteristics of the primary studies included in each review (number and typology of studies, population inclusion criteria, intervention, control, brands of WDs). We extracted mean difference (MD) or standardized mean difference (SMD) for quantitative results related to PA, expressed as continuous outcomes. To summarize the effect estimates and the certainty of the evidence, when additional controls were available, data were extracted on the following a priori-defined list: (1) passive control and (2) other active intervention. Either the shortest available follow-up data or the available measurements reported for the meta-analyses were used since the aim of this umbrella review was to assess immediate effects of receiving an intervention. In cases of missing information, the corresponding authors of SRs were contacted. Disagreements in the data collection process were resolved by either a consensus process or consultation with a third author (GR). Data Synthesis We presented the summary of evidence without re-analysing outcome data. Data were extracted as they were reported in the included SRs (with and without meta-analysis) and then reformatted and presented in text, tables, and figures. We described review characteristics such as eligibility criteria to ensure that SRs are investigating similar clinical questions. We grouped SRs into four categories according to the following population: (a) studies on mixed populations, including SRs on adults in general, healthy adults, or mixed populations of healthy and overweight/obese adults or adults with cardiovascular risk factors; (b) studies on populations with pathologies, e.g., cardiometabolic, pulmonary or orthopaedic diseases; (c) studies on older adults, including SRs on adults over 55, or 60, or 65 years old; or (d) studies on overweight and obese populations. For SRs without meta-analyses, we calculated and then summarized by plotting the percentage of primary studies that found positive findings over the total number of primary studies reporting the outcome (i.e., statistically significant difference in favour of WD). For SRs with meta-analyses, the lists of the primary studies included in each SR with meta-analyses were collated and cross-referenced in a matrix of evidence tables to ascertain the degree of overlap between SRs for each treatment comparison of PA outcome. The “corrected covered area” (CCA) was calculated to quantify the degree of overlap between reviews at both the outcome and population levels. To interpret the results providing context for clinical implications, we followed the decision tree that Hennessy et al. 2021 [ 23 ] proposed. We used a conceptual framework presenting results by outcome and population subgroups [ 24 , 25 ]. In order to visualize findings for steps per day, minutes of MVPA per week, minutes of SB per day and composite outcomes of PA, we showed in a forest plot the effect size of each meta-analysis without calculating the overall pooled estimate. We then created a visual map of the scientific evidence based on bubble plots to display the information of each review as a bubble according to the direction of effect and Certainty of Evidence (CoE) assessment [ 26 ], to quickly examine the concordance or discordance of results [ 27 ]. Discordances were explained in the case of SRs with similar PICO questions, including the same trials (i.e., moderate, high, very high overlapping) [ 24 , 25 , 28 ]. Assessment of Methodological Quality The methodological quality was assessed using “A MeaSurement Tool to Assess systematic Reviews 2 tool” (AMSTAR 2) [ 29 ] by two independent researchers (AT, SB). This tool allows for a reproducible critical evaluation of SRs of RCTs and non-randomized studies of interventions (NRSI) in terms of an overall assessment of the reliability of the results included in the SRs (Additional file 1 : Supplemental File 3). Certainty of the Evidence Two independent researchers (SG, SB) used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach to evaluate the Certainty of Evidence (CoE) of the SRs, adopting the algorithm from Pollock and colleagues [ 30 ] for PA, and separately assessing each population category. In this algorithm, each SR starts with a ranking of high certainty and can be downgraded for severe methodological concerns (Additional file 1 : Supplemental File 3) [ 30 ]. Clinical Relevance and Overall Interpretation of PA We adopted the effect size of the main representative SR (i.e., highest number of participants, most updated and highest methodological quality) assessing patients with mixed populations [ 31 ] (e.g., pathologies, older adults, obese or overweight people), as a measure of clinical relevance between WD and controls. Accordingly, we imputed 1,235 daily steps, 48.5 min weekly of MVPA, and 9.9 min daily less of SB, as minimal important difference between group interventions. Clinical relevance was interpreted considering the categories proposed by Man-Son-Hing et al. 2002 (e.g., definite, probable, possible, definitely not) [ 32 ]. Subgroup populations order of publication year was plotted, including the effect size of all meta-analyses in mean differences, to give an overall interpretation of steps per day, minutes of MVPA per week and minutes of SB. When effect sizes were reported in SMD, we first searched if back-translations were already reported by SRs; otherwise, we back-translated them using the standard deviation of the control of the RCT with the highest number of participants of each meta-analysis [ 33 ]. In Additional file 1 : Supplemental File 4, all details for the interpretation of clinical relevance and back-translation are reported. Supplementary Information

Abbreviations Assessment of multiple systematic reviews, version 2 Corrected covered area Certainty of evidence Chronic obstructive pulmonary disease Database of abstracts of reviews of effects Electronic activity monitor systems Global positioning system Grading of recommendations, assessment, development, and evaluations InterQuartile range Mean difference Medical subject headings Metabolic equivalent Moderate to vigorous physical activity Non-randomized studies of interventions Physical activity Overviews of reviews Preferred reporting items for systematic reviews and meta-analyses Prospective register of systematic reviews Randomized controlled trials Sedentary behaviour Standardized mean difference Systematic reviews Wearable devices Acknowledgements None. Author Contributions C.M, G.R, J.L, S.Bar, S.G contributed to conceptualization, methodology, data curation, formal analysis, and writing—original draft. A.P, A.DI,A.T, P.P, S.Bat, G.C, C.C contributed to data curation, writing—review & editing. All authors read and approved the final version of the manuscript. Funding This work was supported and funded by the Italian Ministry of Health - "Ricerca Corrente”. The APC was funded by Italian Ministry of Health - “Ricerca Corrente”.The funding source had no controlling role in the study design, data collection, analysis and interpretation or writing. Availability of Data and Materials The full dataset is freely available online in OSF ( https://osf.io/udptm/ ), a secure online repository for research data. Declarations Ethics Approval and Consent to Participate Not applicable. Consent for Publication Not applicable. Competing interests The authors declare that they have no competing interests.

CC BY

no

2024-01-16 23:41:59

Sports Med Open. 2024 Jan 14; 10:9

oa_package/c9/04/PMC10788327.tar.gz

PMC10788328

38221549

Background Nowadays, clear aligners (CAs) have gained widespread acceptance for treating malocclusion due to their aesthetic and comfortable features [ 1 , 2 ]. However, long-distance space closure in cases of premolar extraction with aligners can be challenging for orthodontists, owing to the occurrence of the “roller coaster” effect. This phenomenon involves the loss of anchorage, resulting in the distal tipping of anterior teeth and mesial tipping of posterior teeth [ 3 ]. While poor patient compliance and protocol design issues can contribute to this phenomenon [ 4 , 5 ], the more crucial factors are insufficient material stiffness and uncertain biomechanical mechanisms [ 3 , 6 ]. The stiffness of CA material is equivalent to, or even less than, the nickel–titanium archwire in fixed appliances [ 7 – 9 ]. As a result, flexible aligner material is not suitable for teeth movement at all stages due to their susceptibility to deformation, especially for space closure. Additionally, CAs wrap around the entire crown and apply forces to multiple points or surfaces of the teeth, making the biomechanical mechanism more complex than that of fixed appliances [ 6 ]. Therefore, reinforcing the anchorage of posterior teeth and clarifying its biomechanical mechanisms during space closure is a critical clinical problem that must be addressed in clear aligner technology. In recent years, scientists and clinicians have conducted numerous studies attempting to address this issue by means of protocol design, ancillary devices and material modifications. One approach is to consider anchorage preparation of posterior teeth [ 10 ], as well as torque compensation of the anterior teeth [ 11 ] during the design phase for cases involving extraction. While helpful, this approach causes reciprocal movement of teeth, which may burden periodontal tissue. Furthermore, the value of this anchorage design lacks standardization, and achieving the desired outcome is uncertain due to various influencing factors in clear aligner tooth movement. Auxiliary appliances include attachments, power arms and mini-implants. The use of ancillary devices can increase the complexity of clinical operations and chairside time. Besides, excessive use of attachments increases the difficulty of aligner insertion and removal. The power arm, while capable of controlling the point of force application closer to the center of resistance (CR), has a strong foreign body sensation and poor esthetics. Moreover, neither the attachments nor the power arms can be bonded to veneers and crowns. Although mini-implants can reinforce posterior tooth anchorage, there are multiple complications of implant fracture, implant loosening, soft tissue inflammation and root contact. Additionally, attempts to improve material properties of the clear aligner diaphragm through blending modification [ 12 ] and multi-layer structures [ 13 ] have had some success in enhancing force delivery and decreasing stress relaxation, but balancing stiffness and resilience remains a challenge. Furthermore, overall diaphragm material change cannot alter local stress distribution in the posterior area, so the mesial tipping of posterior teeth cannot be prevented precisely. As a result, to date, there has not been a satisfactory method to minimize anchorage loss of posterior teeth during the extraction space closure. Studies have demonstrated a strong correlation between aligner thickness and the forces delivered by it, indicating that thicker diaphragms transmit greater forces [ 14 , 15 ]. However, increasing overall thickness changes the mechanical properties of the entire aligner and fails to address local stress distribution in the posterior teeth region. What is more, the process of vacuum thermoforming aligners creates an uneven thickness of the aligner [ 16 , 17 ]; thus, the thickness of the diaphragm is not indicative of the thickness of the posterior part. Uneven thickness of thermoformed aligners has an impact on the stress distribution [ 15 ]. Closing the extraction space with CAs is akin to closing it on a nickel–titanium archwire, which is too flexible and contributes to the “roller coaster” effect. Ideally, the posterior segment of the archwire should be more rigid than the anterior segment [ 18 , 19 ], so that the tipping of anchor teeth during space closure could be well avoided. Building on this idea, we hypothesize that by thickening the local thickness of the posterior region of CAs to improve the stiffness, the force distribution of the posterior teeth during the anterior retraction could be changed so that the mesial tipping of the posterior teeth would be reduced. Through a three-dimensional finite element (FE) model and an experimental model, we investigated the biomechanical changes resulting from partially thickened CAs, providing new insights for advancing clear aligner therapy.

Materials and methods Finite element model study A healthy adult with well-aligned complete dentition was selected as the subject for this study. Based on cone-beam computed tomography (CBCT) data, the three-dimensional model of alveolar bone and the maxillary dentition with the extraction of the first premolars were reconstructed using Mimics Research 21.0 (Materialize, Leuven, Belgium) and Geomagic Studio 2016 (3D systems, Rock Hill, SC, USA). The periodontal ligament (PDL) was obtained by making an external offset from the root surface, and the thickness was set at 0.3mm [ 20 ]. Vertical rectangular attachments (3mm height, 2mm width, 1mm thickness) were set on the buccal surface of canines, the second premolars and molars. The crowns and attachments were extended outward to model CAs with a uniform thickness of 0.75 mm. All the components were meshed with HyperMesh 14.0 (Altair, Troy, Mich, USA) and then imported into ABAQUS 2016 (Dassault SIMULIA, Providence, RI, USA) to be assembled into the three-dimensional finite element model. As shown in Fig. 1 A-E, the FE model consisted of alveolar bone, PDL, teeth with attachments and CA. The number of nodes and elements for each component of the model are shown in Table 1 . All the components were regarded as linear elastic, isotropic and homogeneous materials in our study based on previous studies [ 21 , 22 ]. The mechanical properties of all components are shown in Table 2 . The properties of teeth, attachments, PDL and alveolar bone were obtained from the literature [ 23 – 26 ], and the properties of CA was provided by Angelalign Inc. In this study, we proposed a new aligner scheme by thickening the marginal region of the aligner with a width of 1.5 mm by an additional 0.5 mm, hence creating an “enhanced structure.” The FE model was divided into a control group without the enhanced structure and an enhanced structure group. The parameters and properties of the enhanced structure are shown in Tables 1 and 2 . In a preliminary study, different variations of enhanced structures were tested, including buccal interproximal space between the posterior teeth (Additional file 1 : Fig. 1), palatal cervical line of posterior teeth (Additional file 1 : Fig. 2) and both the buccal and palatal sides of the posterior teeth (Fig. 1 F–H). The results (Additional file 1 : Table 1–3) indicated that the enhanced structure on both buccal and palatal sides was most effective; and therefore, this type of enhanced structure was selected for further analysis. The interactions of these parts in FE model were treated carefully to simulate the real situation. The external surface of each tooth’s PDL was connected directly to the corresponding alveoli through the same nodes in the FE model, so it is between the inner surface of PDL and the root of the tooth. Fixed boundary condition was set to the upper transverse section of the maxillary bone. The designed treatment plan involved 0.3mm retraction of anterior teeth without torque compensation of the anterior teeth and anchorage preparation of the posterior teeth. The aligner with designed teeth movement was assembled on the dentition and their interaction was simulated by numerical algorithm of surface-to-surface contact with a friction coefficient of 0.3. The interaction effects, including the deformation and stress of the aligner and the force and moment generated by the aligner and acting on the teeth were solved by the finite element analysis synchronously. A local coordinate system (LCSYS) was established for each tooth, represented by the X, Y, and Z axes (Fig. 1 I -J). The X-axis represented the mesial-distal direction, with the mesial direction being positive. The Y-axis represented the buccal-lingual direction, with the lingual orientation being positive. The Z-axis represented the long-axis direction of the tooth, with the positive direction toward the apex. The initial forces (F, g) and moments (M, gmm) components in the LCSYS of each tooth were measured, referring to CR, and the distance (d, mm) was calculated by dividing M by F. To prevent repetition, only the right side of the maxillary dentition was presented here, as the biomechanical situation on both sides of the dentition was essentially symmetrical and the results on the left side were similar to those on the right side. Von Mises stress distribution of clear aligner and force distribution of maxillary dentition were analyzed. Experimental study The physical experiments were conducted with a specially developed mechanical testing device, which can measure the forces and moments in the three-dimensional direction of each tooth [ 27 – 30 ]. This device consisted of 12 high-precision Force/Torque sensors (Nano 17-E, ATI Industrial Automation, Apex, NC, USA), 12 isolated 3D-printed resin teeth (Object 30 OrthoDesk, Stratasys Ltd, MN, USA), and a set of data acquisition and processing software (Fig. 2 A). Each resin tooth of the maxillary dentition was separately attached to a sensor using three fixed screws. The method of the experimental study is presented in Fig. 2 B. The designed treatment plan was the same as the FE model. The maxillary dentition from the FE model was 3D-printed (Objet30 Pro, Objet Ltd., Rehovot, Israel) for the production of the corresponding CA for the control group by thermoforming with 0.75 mm diaphragms (Young’s Modulus 1000 MPa, Angelalign Technology Inc., Shanghai, China). Each of the six CAs of the control group was then worn on the experimental device, and the initial force and moment in the three-dimensional direction were collected as the results of the control group. After completing the testing, each of the six CAs in the control group required further processing to create aligners with enhanced structures. The dimensions of the enhanced structure of the experimental model were identical to those of the FE model simulation. For the enhanced structure group in the experimental study, the margins of the thickened area were initially drawn on both the buccal and lingual sides of the aligner at the posterior teeth to determine the length and width. The length was determined separately on the buccal and palatal sides according to the anatomical features, and the width was 1.5 mm for both. Subsequently, a syringe was used to uniformly injected light-curing glue (Angelalign Technology Inc., Shanghai, China) with Young’s modulus of 1980 MPa within the thickened region and then irradiated for 10 s with a dental LED curing light to form an enhanced structure with a thickness of 0.5 mm (Fig. 2 C-D). The fabricated enhanced structures were measured to verify their dimensions as described in the Additional file 1 : Fig. 3. Finally, the enhanced structure group was measured for forces and moments following the same approach as the control group. The LCSYS of each tooth was defined in the same way as in the finite element model study. The initial forces (F) and moments (M) components in the LCSYS of each tooth were measured six times. The Shapiro–Wilk test was used to verify the normal distribution of data. Statistical analysis was performed using paired t tests or Wilcoxon signed-rank test for F and M in GraphPad Prism 9.0 (Dotmatics, USA). P < 0.05 was considered statistically significant.

Results F, M and d in the mesial-distal direction of the maxillary dentition During the retraction of anterior teeth, all F and M values of different directions recorded by both methods were reported in the supplementary information (Additional file 1 : 3–6). When closing the extraction space of the first premolar, the posterior teeth are primarily applied forces and moments in the mesial-distal directions. As shown in Fig. 3 , the second premolar had positive F, M and d values and oriented toward the mesial. In the finite element simulation, the magnitudes of Fx and Mx decreased in the enhanced structure group compared to the control group, indicating a downward trend in mesial tipping movement of the second premolar. The decline of d in the enhanced structure group demonstrated that the resultant force was closer to the CR. In the experimental study, the decreasing trend of Fx magnitude was more noticeable ( P < 0.01, 95%CI -26.38 to -10.92), but the drop in Mx values was not statistically significant ( P > 0.05, 95%CI -106.00 to 65.75). Both the first and second molars were applied mesial forces during the anterior retraction. In both the FE model and the experimental model, the Fx values for the molars in the enhanced structure group were higher than those in the control group and showed statistical significance ( P < 0.01, 95%CI of the first molar 7.12 to 26.85, 95%CI of the second molar 7.91 to 22.79) in the experimental model. For the canine, the Fx and Mx values were negative, suggesting that the canine was applied a distal force during retraction. In the FE model, after partial thickening of the aligner at the posterior teeth, both Fx and Mx values increased, representing an increase in the retraction force on the canine. The decrease in the d value illustrated that the force was closer to the CR. Similarly, in the experimental model, the enhanced structure group exhibited higher Fx and Mx values compared to the control group but did not show statistical significance ( P > 0.05, 95%CI for Fx of canine −42.53 to 5.47, 95%CI for Mx of canine −4.75 × 10 2 to 1.16 × 10 2 ). In summary, the finite element model and the experimental model showed similar trends. Changing the local thickness of the CA at the posterior teeth improved the biomechanics during retraction to prevent tipping movement of the posterior teeth. Force distribution of maxillary dentition As illustrated in Fig. 4 , forces were predominantly concentrated on the anterior teeth and the distal side of the second premolar during anterior retraction. With the implementation of the enhanced structure of the posterior teeth, forces applied on the cervical third of the palatal surface of the second premolar and first molar relatively increased (red arrow), while forces applied on the occlusal third of the second premolar relatively decreased (black arrow). Consequently, for the second premolar, the force distribution was closer to the CR, reducing the risk of anchorage loss and increasing the likelihood of bodily movement of the tooth. Von-mises stress distribution of clear aligner The stress distribution of CA is illustrated in Fig. 5 . In the control group, it can be seen that during the extraction space closure, the areas of higher stress were mainly located at the anterior teeth, the extraction space and the second premolar. After thickening of the buccal and palatal gingival sides of the posterior teeth, stress in the interproximal spaces between the posterior teeth corresponding to the thickened areas increased (red arrow), while the stress in the occlusal side of the second premolar and the first molar decreased (grey arrow). Since the enhanced structure was located at the gingival margin of the clear aligner, the stress was distributed closer to the CR of the premolar and molars, resulting in a reduction in the tipping movement of these teeth.

Discussion Anchorage control is critical for extracted patients with incisor protrusion or severe crowding. Loss of anchorage can be disastrous in these cases where strong anchorage is needed, so reinforcing anchorage of posterior teeth should be carried out throughout the orthodontic process. Our study proposed a new method of anchorage reinforcement in clear aligner therapy, which is to enhance the local stiffness by increasing the local thickness of CA, so that the CA can balance both the flexibility and the local stiffness, resulting in a reduction of mesial tipping of the posterior teeth. This new method is applicable during the entire process of orthodontics and has no impact on the design of the treatment staging. Localized thickening of the aligners can be performed during the fabrication phase without clinical operation and does not occupy chairside time. This approach is non-invasive and can be used in any case that requires reinforcement of anchorage. At the same time, it does not conflict with other ways of reinforcement anchorage and accordingly can be employed in combination with other alternatives. Consequently, this new approach of anchorage reinforcement has potential for clinical applications. In this study, we observed that after partial thickening of the aligner at the posterior segment, the FE and experimental models both showed an increase in force magnitude on the molars, while a decrease in force magnitude was observed on the second premolar. Simultaneously, the force distribution in the posterior segment shifted toward the gingival side, closer to the CR of the tooth. This performance after partial thickening was more in accordance with optimal principles of biomechanics. Studies have shown that the area of PDL differs from tooth to tooth [ 31 ] and the size of a tooth’s PDL directly influences its resistance to movement and anchorage value [ 7 ]. The second premolar, being a single rooted tooth, has the smallest anchorage value among the posterior teeth and is located adjacent to the extraction space, and the tooth has a natural tendency to drift toward the extraction space where there is less resistance; therefore, it is very important to control the anchorage of the second premolar. The results of this study demonstrated that by thickening the aligners, the force exerted on the second premolar decreased by 18.65 g; while, the force distribution was closer to the CR of the tooth. This adjustment effectively reduces the mesial tipping of the posterior teeth and decreases the anchorage loss when closing the extraction space. Conversely, molars, being multi-rooted teeth, have anchorage values (53.3 mm 2 ) that are twice as large as that of premolars (25.4 mm 2 ) [ 32 ] and can withstand greater forces without mesial movement relative to premolars. While the first and second molars did experience an average increase in forces of 16.99 g and 15.35 g, respectively, due to the addition of the enhanced structure, this increased anchorage value effectively equipped the molars with the capacity to resist unwanted movement. The aforementioned phenomenon is beneficial for extraction cases to control the position of anchor teeth and partly avoid the “roller coaster effect”. In fixed appliances, the sliding mechanics of space closure requires sufficiently stiff wires to prevent archwire bending, which can lead to the tipping of anchor teeth and the “roller coaster effect”. Burstone had proposed the idea that the posterior segment of the archwire needed to be more rigid than the anterior segment [ 18 ]. Drawing inspiration from this concept, we sought to enhance the stiffness of the posterior segment of CAs to facilitate space closure. Both the FE and experimental models demonstrated that this enhancement effectively reduced Fx and Mx values on the second premolar, and shifted force distribution closer to the gingival direction. There were two possible reasons for this outcome. Firstly, CAs exert orthodontic forces through elastic restoration. The counter diagram (Fig. 5 ) of the control group for FE model showed that the stress in the posterior segment of the aligner was mainly concentrated in the interproximal spaces of posterior teeth. Accordingly, when thickening the interproximal spaces of posterior teeth, the tensile stress generated in this section can be reinforced. Since the enhanced structure was located at the distal of the second premolar, it resulted in a rise in the distal force magnitude on the second premolar, reducing the tendency of mesial tipping of the second premolar and the possibility of anchorage loss. Secondly, the thickened site, positioned at the cervical third of the aligners, increased force near the gingival margin, shifting the action point of the resultant force toward the cervical, closer to the CR. There are a variety of methods used to perform CA biomechanical studies, such as finite element analysis [ 33 ], photoelastic stress analysis [ 34 ] and micro-sensor [ 30 ]. In our study, two models, namely finite element analysis and experimental apparatus, were used to verify that partial thickening of CAs could indeed reduce the mesial tipping of anchor teeth in the process of anterior tooth retraction. Each model had its own advantages and complemented the other. Three-dimensional finite element analysis is commonly used to study the biomechanics of clear aligners, with the advantage of simulating the PDL as an anatomical structure [ 35 ]. However, in the FE model, the thickness of the aligners was even and the aligner was uniformly stretched when deformation occurred, unlike in actual clear aligners. Actual CAs are produced by vacuum thermoforming technology and do not have a uniform thickness on different tooth surfaces [ 16 ]. Therefore, they are not evenly tensile during deformation. To address this, our experimental model utilized realistic CAs to study the forces exerted on each tooth, compensating for the limitations of the FE model. Although the experimental model lacked in vitro PDL representation, it effectively explored the biomechanics in orthodontic appliances with uneven thickness. Admittedly, due to the inherent differences between these two models, a direct comparison of measured values between the two models was not available. We observed the same trends of the variation in force magnitude of the two models separately after adding the enhanced structure, and thus drew the corresponding conclusions. Therefore, these two approaches complemented each other and collectively demonstrated a tendency for the enhanced structure to reinforce the anchorage of posterior teeth. Orthodontic forces are classified as light, moderate and heavy forces based on their magnitude. The use of light but lasting force is preferred in clinical practice as it allows for rapid tooth movement while minimizing root resorption. Excessive orthodontic force can lead to ischemic necrosis and hyalinization in the PDL, leading to undermining resorption and slowed tooth movement [ 7 ]. Our results from the FE model indicated a maximum force of approximately 250 g, which exceeds the range of light forces. This is due to the fact that the FE model simulates the initial forces generated by the aligners; while, the force values of the aligners decay rapidly by half or more after initial insertion due to stress relaxation [ 36 ]. Consequently, stable orthodontic forces exerted by CAs are considered light forces. Despite the initially higher stress value, studies have shown that CAs have a lower risk of root resorption than fixed appliances due to the stress relaxation and intermittent loading method [ 37 , 38 ]. In the experimental model, the measured F and M values were twice as large as those in the FE model due to the absence of the PDL as a physiological structure in the experimental apparatus. The PDL has a cushioning effect on forces applied to teeth, accommodating forces exerted on the crown [ 39 ]. In contrast, the experimental model is mechanically connected to the sensor by screws, which are much stiffer than the PDL. Therefore, this inherent difference between the models leads the experimental setup to record much larger force values than the FE model that incorporates the PDL. Moment is generated by a force acting at a certain distance, causing the tooth to have a tendency to rotate around the CR. Quantitatively, it is the multiplication of the force and the perpendicular distance from the point of force application to the CR. As a result, the error in moment value is magnified compared to the error in force value, which is noticeable in the experimental model, and a similar phenomenon has been reported in the previous study [ 27 ]. This is likely due to the fixed mechanical connection between the sensor and the resin tooth of the experimental device, and the absence of the PDL, which is where this experimental model still needs to be improved. Therefore, in our study, we observed the distance change through the FE model (Figs. 3 C, 4 and 5 ) instead of calculating it directly in the experimental model.

Conclusions Our study, by both the FE model and the experimental model, indicated that the biomechanics of the clear aligner during space closure was optimized by locally thickening the posterior segment of the aligner to form the enhanced structure. The enhanced structure resulted in a decrease in the force magnitude on the second premolar and allowed the force distribution closer to the CR, therefore reducing the mesial tipping of anchor tooth and mitigating the roller coaster effect related to orthodontic extraction cases. Consequently, our study provided a new dimension for anchorage reinforcement in clear aligner therapy.

Background Mesial tipping of posterior teeth occurs frequently during space closure with clear aligners (CAs). In this study, we proposed a new modification of CA by localized thickening of the aligner to form the enhanced structure and investigate its biomechanical effect during anterior retraction. Methods Two methods were employed in this study. First, a finite element (FE) model was constructed, which included alveolar bone, the first premolars extracted maxillary dentition, periodontal ligaments (PDL), attachments and aligners. The second method involved an experimental model—a measuring device using multi-axis transducers and vacuum thermoforming aligners. Two groups were formed: (1) The control group used common CAs and (2) the enhanced structure group used partially thickened CAs. Results FE model revealed that the enhanced structure improved the biomechanics during anterior retraction. Specifically, the second premolar, which had a smaller PDL area, experienced a smaller protraction force and moment, making it less likely to tip mesially. In the same vein, the molars could resist movement due to their larger PDL area even though they were applied larger forces. The resultant force of the posterior tooth was closer to the center of resistance, reducing the tipping moment. The canine was applied a larger retraction force and moment, resulting in sufficient retraction of anterior teeth. The experimental model demonstrated a similar trend in force variation as the FE model. Conclusions Enhanced structure allowed force distribution more in accordance with optimal principles of biomechanics during the extraction space closure while permitting less mesial tipping and anchorage loss of posterior teeth and better retraction of anterior teeth. Thus, enhanced structure alleviated the roller coaster effect associated with extraction cases and offered a new possibility for anchorage reinforcement in clear aligner therapy. Supplementary Information The online version contains supplementary material available at 10.1186/s40510-023-00502-2. Keywords

Limitations Although finite element analysis is one of the best ways to analyze biomechanics delivered by orthodontic appliances, it still has its limitations in accurately simulating the true oral environment (like body temperature and saliva), and it cannot simulate the continuous application of orthodontic forces. Moreover, the experimental model, despite using real CAs, is difficult to mimic the PDL in an in vitro setting. Therefore, further optimization in model construction is needed in future studies. Additionally, the efficacy of this enhanced structure proposed in our study requires validation through clinical applications. Nevertheless, our study presents an innovative approach to address the “roller coaster phenomenon” and reinforce the anchorage of posterior teeth when closing the extraction space using CAs. Supplementary Information

Abbreviations Clear aligner Center of resistance Cone-beam computed tomography Periodontal ligament Finite element Local coordinate system Force Moment Distance Acknowledgements We thank Lei Huang and Xianbo Xu for their assistance with this study. Author contributions XJ and XT contributed to methodology, data curation, data analysis and original draft writing. VL performed data curation and draft review & editing. YZ provided resources, software and professional technical support. JS involved in data curation. XH contributed to conceptualization, funding acquisition, draft review & editing and supervision. All authors read and approved the final manuscript. Funding This study was supported by EA Medical Center Device Technologies Co., Ltd. (Grant number 20H925), West China Hospital of Stomatology (Grant number RD-03–202001), Ministry of Science and Technology of the People's Republic of China (Grant number 2018FY101003), Dazhou Science and Technology Bureau (Grant Number 2020CDDZ-14) and Chongqing Municipal Education Commission (Grant number KJCX202001). Availability of data and materials Data will be made available on reasonable request. Declarations Ethics approval and consent to participate This study was approved by the ethics committee of the West China Hospital of Stomatology, Sichuan University (WCHSIRB-D-2020–319). Consent for publication Not applicable. Competing interests The author YZ from EA Medical Instruments Company Limited (i.e., EA) mainly carries out technical support, which has no association with the products and business of EA. The other authors declare that they have no competing interests.

CC BY

no

2024-01-16 23:41:59

Prog Orthod. 2024 Jan 15; 25:3

oa_package/f3/c7/PMC10788328.tar.gz

PMC10788329

38225950

Introduction Rice is one of the important staple crops globally, and hybrid rice technology has greatly increased rice production, ensuring global food security. Along with the progression of society and the enhancement of living standards, there has been a gradual increase in the demand for rice quality (Li et al. 2023 ). Rice quality is also an important and complex trait, including eating and cooking quality, milling quality, apparent quality, and nutritional quality. The quality of rice directly impacts its commercial value and palatability. Improving the quality of hybrid rice has always been the goal pursued by breeders. Over the years, both traditional and modern molecular breeding techniques have been employed to enhance the quality of hybrid rice steadily (Tian et al. 2009 ; Zhang et al. 2016 ). However, the quality traits of rice are susceptible to environmental factors such as light, temperature, and humidity, resulting in variations (Liu et al. 2013 ; Lu et al. 2022 ). Therefore, how to mitigate the impact of environmental factors on rice quality and enhance its stability is also one of the concerns in improving hybrid rice. Phenotype plasticity refers to the ability of the same genotype to produce different phenotypes in different environments (Sultan 2000 ), reflecting the relationship between organisms and their environment, which is widely present in plants (Bradshaw 1965 ). Phenotype plasticity is related to the adaptability and stability of plants (Chevin et al. 2013 ; Finlay and Wilkinson 1963 ). From an evolutionary perspective, varieties with high phenotype plasticity exhibit stronger adaptability to the environment (Des Marais et al. 2013 ; Bonamour et al. 2019 ). However, in the context of crop production, plants with lower phenotype plasticity exhibit greater stability. Consequently, implementing techniques to decrease the phenotype plasticity of crops and enhance their stability becomes crucial in enabling the expression of desired traits across a broader range of locations. Phenotype plasticity is under genetic control and can be targeted for artificial improvement in crop breeding (Gage et al. 2017 ). To achieve this goal, based on scientific quantification methods for plasticity, efforts have been made to study the genetic architecture of crop plasticity and analyze the QTLs across various crops (Wang et al. 2015a ; Kadam et al. 2017 ; Jin et al. 2023 ). Exploratory studies on rice have revealed the genetic structure and potential QTLs underlying the plasticity of yield-related traits (Kikuchi et al. 2017 ; Mu et al. 2022 ). Although the precise functions of these QTLs remain unclear, they provide new insights and methods for artificially selecting and improving crop phenotype plasticity. Quality-related traits are influenced by the environment, exhibiting phenotype plasticity. However, previous studies on phenotype plasticity in rice have primarily focused on yield-related traits, but there has been limited research on the patterns and genetic architecture of phenotype plasticity in quality traits, which are crucial for breeding advancements. Enhancing rice quality stability has implications for enhancing the potential and commercial value of rice varieties. Therefore, understanding the patterns of phenotype plasticity and genetic structure of quality-related traits in rice will provide better references for breeding improvements. Parents selection is crucial in hybrid rice breeding (Chen et al. 2019 ). However, the challenge of selecting ideal parental materials from a large population to induce strong heterosis is significant. Therefore, breeders use the combining ability to assess the breeding value of parental materials in hybrid production (Sprague and Tatum 1942 ). By identifying the combining ability of parents in phenotype plasticity, breeders can predict the performance of hybrid combinations, thereby enhancing the efficiency and stability of hybrid rice production (Abd El-Aty et al. 2022 ). The large-scale phenotypic analysis is an important foundation for plasticity research. In this study, a total of 141 hybrid rice combinations were obtained from 7 TGMS lines and 25 restorer lines. These combinations were planted in five locations in Southern China in the 2020 summer season and arranged 3 to 5 intermittent sowings in each trial location. Seven quality traits and their phenotype plasticity were investigated, including amylose content (AC), alkali spreading value (ASV), gel consistency (GC), chalkiness degree (CD), percentage of grains with chalkiness (PGWC), transparency (TP), and milled rice ratio (MRR). We analyzed the combining ability of phenotype plasticity for 32 parental materials and elucidated the genetic structure of phenotype plasticity for quality-related traits. Genetic effects and candidate gene analyses were conducted on the identified QTLs. The statistical results suggest that delaying sowing date is beneficial for enhancing rice quality in the Yangtze River basin. We utilized a model to evaluate the phenotype plasticity of each hybrid and identify TGMS and restorer lines that exhibit improved quality stability in hybrid rice breeding. Furthermore, our research uncovered the genetic basis of phenotypic plasticity in rice quality traits and discovered QTLs associated with quality plasticity while predicting candidate genes. These findings offer theoretical guidance for determining the optimal sowing date of high-quality rice and enhancing the quality stability of hybrid rice.

Materials and methods Plant materials and field experiments In this study, a total of 141 hybrid rice combinations were obtained by crossing 7 TGMS lines with 25 restorer lines. The TGMS lines include Jing4155S (J4155S), LongZhen36S (LZ36S), HuaYue468S (HY468S), ZhenXiangS (ZXS), HuaXuan302S (HX302S), LongKe638S (LK638S), and HuaWei338S (HW338S). The restorer lines include HuaHui8612 (HH8612), HuaHui7810 (HH7810), HuaHui3135 (HH3135), HuaHui7503 (HH7503), TaiSi (TS), LongKeSiMiao13 (LKSM13), HuaHui8012 (HH8012), HuaHui2646 (HH2646), HuaHui6210 (HH6210), HuaHui3748 (HH3748), DiZhan (DZ), HuaHui8037 (HH8037), HuaHui5106 (HH5106), HuaHui8549 (HH8549), YuZhan (YZ), HuaHui8700 (HH8700), HuaHui8954(HH8954), HuaHui4153 (HH4153), R1206, R1212, R1377, R1988, WuShanSiMiao (WSSM534), HuaZhan (HZ), and Jin4 (J4). All parental lines are elite parents in hybrid rice production and sourced from the Core Germplasm Bank of Commercial Hybrid Rice Breeding in Yuan Longping High-tech Agriculture Co., Ltd. (Supplementary Table S1 , S2 ). The field experiments were conducted in the summer season of 2020 in five locations including SC-GH (Guanghan, Sichuan Province, 104°25′ E, 30°99′ N), HN-LS (Lingshui, Hainan Province, 109°45′ E, 18°22′ N), HN-CS (Changsha, Hunan Province, 112°59′ E, 28°12′ N), HB-EZ (Ezhou, Hubei Province, 114°52′ E, 30°23′ N), and AH-HF (Hefei, Anhui Province, 117°17′ E, 31°52′ N). Among the five trial locations, HN-LS is situated in the south China rice cropping region, SC-GH is located in the upper Yangtze River rice cropping region, while AH-HF, HN-CS, and HB-EZ are situated in the middle and lower Yangtze River rice cropping region. In SC-GH, three intermittent sowings were arranged from April 1 to May 1, with a 10-day interval between sowings. In HB-EZ and AH-HF, four intermittent sowings were arranged from April 15 to June 1, with a 15-day interval between sowings. In HN-CS, five intermittent sowings were arranged from April 10 to June 10, with a 15-day interval between sowings. In HN-LS, four intermittent sowings were arranged from June 15 to July 30, with a 15-day interval between sowings (Supplementary Table S3 ). Seedlings of 5-leaf age were transplanted. Each material was planted in a five-row plot with eight individuals in each row at a spacing of 20 cm × 26.5 cm density in the field and standard management practices throughout the growing period. At maturity, nine uniform plants in the middle of each plot were harvested. Parental DNA extraction and whole genome sequencing After germination of parental seeds, young leaves were collected and immediately flash-frozen in liquid nitrogen, and then the samples were stored at − 80°C for future use. DNA extraction was performed using the FastPure Plant DNA Isolation Mini Kit (Vazyme, Jiangsu, China). The concentration of the extracted DNA was evaluated using a NanoDrop spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA) and a Qubit 3.0 fluorometer (Life Technologies, Carlsbad, CA, USA). To assess the purity and integrity of the DNA, 1% agarose gel electrophoresis was conducted. For library preparation, a short-read library with a DNA-fragment insert size of 200–400 bp was generated using 1 μg of genomic DNA. The library preparation was carried out following the manufacturer’s instructions using a library preparation kit compatible with DNBSEQ-T7 (BGI, Shenzhen, China). Subsequently, paired-end (PE) sequencing was performed on a DNBSEQ-T7 platform using the PE 150 model. Measurement of quality trait A minimum of 200 g seeds from each accession were used for measuring the seven quality traits, including AC, ASV, GC, CD, PGWC, TP, and MRR. The CD, PGWC, and TP were measured according to NY/T 2334–2013 (Chinese Ministry of Agriculture Standards). CD, PGWC, and TP were measured with a Microtek Scan Wizard EZ scanner and rice quality analyzer SC-E software (Hangzhou Wanshen Detection Technology Co., Ltd., Hangzhou, China). The AC, GC, and ASV were measured according to Chinese National Standards GB/T 15683–2008(GB/T 22294–2008 and NY/T 83–2017), respectively. Measurement of overall quality and meteorological condition To evaluate the overall quality of the hybrids, we followed the quality standards for cooking rice varieties set by the Chinese Ministry of Agriculture and Rural Affairs (NY/T 593–2021). Here, a record is specified as the quality performance of a specific hybrid, which has been planted at a trial location on a particular sowing date. We assessed the quality of all 2774 records in our experiment by applying the standard to each of them. In order to analyze the correlation between quality characteristics and meteorological conditions, we have also summarized the meteorological data collected at different times within each day into the following meteorological factors: minimum temperature, maximum temperature, average temperature, day/night temperature difference, surface solar radiation, accumulated rainfall, and number of rainy days (Cheng and Zhu 1998 ). In our analysis, we consider a day with more than 1 mm of rainfall as a rainy day. Spearman correlation was conducted between each quality trait and meteorological factors using the SciPy library in Python. Analysis of phenotype plasticity According to the reaction norm model, the phenotypic record of the th hybrid line observed under the th environment can be modeled as follows: where is the mean value of the trait, is the main effect of the th line, and is the main effect of the th environment. represents the interaction term (G-by-E) between th hybrid line (genotype) and th environment. is the error term. Here, we introduced the Finlay–Wilkinson regression (FW) as the representation of phenotype plasticity. FW reorganized Eq. 1 into follows: If we consider as the independent variable of the function , then the slope of this function indicates the expected trait variation ( ) in traits with changes in unit environmental effects ( ), capable of indicating the phenotype plasticity. To calculate the plasticity of quality traits at different locations and sowing stages, we selected the hybrid lines with multi-location field trials and at least three intermittent sowings per location from the 141 hybrids. We used the FW package implemented in R by Lian et al. (Lian and De Los Campos 2016 ) and applied the Bayesian method for regression. Analysis of combining ability Combining ability analysis is a statistical technique employed in plant breeding to assess the genetic potential of parents and predict the performance of their hybrids. The analysis considers both general combining ability (GCA), which measures the average performance of a parent across different crosses, and specific combining ability (SCA), which measures the performance of a specific parent in specific crosses. Since there are no extra replications for each sowing management in our experiment, we could not properly evaluate the SCA for each hybrid. For convenience, we simply treated the SCA as 0. Consequently, the phenotype plasticity of hybrid , which has parents i and j , can be modeled as follows: where is the mean value of the phenotype plasticity, is the error term, and represents the breeding value of phenotype plasticity for th parent. Since the distribution of has a mean value of , the sum of the GCAs of all parents is 0 (i.e., ). We used the sommer package (Covarrubias-Pazaran 2016 ) in R to calculate the GCA in a mixed model, in which the parents and crosses are considered as random effects. If there are no significant additive effects detected from the parents for the plasticity of hybrid traits, the model will indicate a GCA of 0, which will be later displayed in the paper. Analysis of BLUP The analysis for various traits of each location is conducted by fitting a linear mixed model and computing best linear unbiased predictors (BLUPs) with the lme4 R package (Bates et al. 2015 ): where Y represents the phenotypic records, the parentheses indicate random effects, “1|” denotes groups, and “:” refers to interactions. LINE indicates to the hybrid lines, and ENV indicates to different sowing stages or different locations. We utilized the fitted random effects of hybrid lines as the representation of the overall genetic effects of each genotype for these traits while eliminating the environmental effects. Variant detection and annotation The sequencing data was aligned to the Oryza sativa Nipponbare reference genome (IRGSP1.0) using BWA mem (Li 2013 ). PCR and optical duplicates were removed using Picard software (Broad Institute). SNP and InDel detection were performed using GATK HaplotypeCaller (McKenna et al. 2010 ). To minimize false positives, SNPs that did not meet the following criteria were filtered out: QUAL < 30.0, QD < 2.0, SOR > 3.0, FS > 60.0, MQ < 40.0, MQRankSum < 12.5, or ReadPosRankSum < 8.0. Similarly, InDels that did not meet the following criteria were filtered out: QUAL < 30.0, QD < 2.0, FS > 200.0, MQ < 40.0, or ReadPosRankSum < 20.0. SNPs and InDels were annotated using the software SnpEff (Cingolani et al. 2012 ) based on their impact on genes. The annotation categorized them into different types, such as synonymous mutations, non-synonymous mutations, frameshift mutations, and others. SnpEff utilized Ensembl release 41 as the gene annotation version for this analysis. Genome-wide association analysis The GEMMA (Zhou and Stephens 2014 ) software was utilized to conduct Genome-Wide Association Studies (GWAS) for BLUP and FW values of different traits and locations, with a mixed linear model (MLM) fitted. The kinship matrix was used as a random effect, while the first three principal components from principal component analysis (PCA) were included as fixed effects in the MLM. PCA analysis was performed using the smartPCA program implemented in the Eigensoft package (Patterson et al. 2006 ). The significance threshold for GWAS analysis was determined using the Bonferroni correction method, which means dividing 0.05 by the number of SNPs (n) in the analysis. Here, the threshold was set at p < 5.32 × 10 −8 (i.e., − log10( p ) < 7.27). The entire genome was partitioned into LD blocks based on an LD (linkage disequilibrium) threshold of r 2 = 0.6 using the gpart R package (Kim et al. 2019 ). LD blocks with a distance of less than 1 Mb are merged. LD blocks that contain significant SNPs were considered as candidate QTLs.

Results Effect of sowing date and meteorological factors on quality of hybrid rice The rice quality traits of the hybrids collected from 5 trial locations, with multiple intermittent sowing arrangements, were measured, and the distributions of the trait values are illustrated in Fig. 1 a and Supplementary Table S4 . Nearly all quality traits in these rice cropping regions exhibited significant correlations with the sowing date, suggesting a general contribution of sowing date to rice quality trait variation (Supplementary Table S5 ). As sowing dates were delayed, we observed the mean of AC increasing by 2.37–3.97% in AH-HF, 3.82–5.12% in HN-CS, 4.37–5.32% in HB-EZ, and 0.77–0.89% in SC-GH, while decreasing 0.32–0.75% in HN-LS. In contrast, the mean of CD decreased by 1.31–6.2 in AH-HF, 2.71–5.03 in HN-CS, 2.85–7.82 in HB-EZ, and 2.34–2.07 in SC-GH, while increasing by 0.84–1.89 in HN-LS in tandem with the delay in sowing dates (Fig. 1 b–g). We assessed the overall quality of the hybrids by referring to the Chinese Ministry of Agriculture and Rural Affairs’s quality standard for cooking rice variety (NY/T 593–2021). Out of the 2774 records examined, 872 of them exhibited high-quality with grade 3 or superior performance. These high-quality records are sourced from 124 hybrid varieties and shown in all five trial locations and all sowing dates. For all trial locations except HN-LS, the improvement of rice quality correlates with the postponement of sowing dates, with a greater number of records achieving quality grade 3 or higher grades (Fig. 1 h–l). To assess the meteorological variation more precisely, we decomposed it into the following meteorological factors: minimum temperature, maximum temperature, average temperature, day/night temperature difference, surface solar radiation, accumulated rainfall, and number of rainy days (Cheng and Zhu 1998 ) (Supplementary Fig. S1 ). Existing studies have shown that the meteorological factors during the grain filling stage, particularly the initial 15 days following full heading, have a considerable influence on rice quality (Wu et al. 2016 ; Yan et al. 2021 ). Thus, we employed the mean values of the aforementioned meteorological factors during this time period for further evaluation . We performed an analysis to investigate the correlation between meteorological factors and quality traits. Due to the non-normal distribution of the quality traits data, we computed Spearman correlation coefficients between meteorological factors and these quality traits (Fig. 2 ). Nearly all quality traits displayed a significant correlation with multiple meteorological factors in each trial location. The correlation trend of each meteorological factor with each quality trait differed across different trial locations. More specifically, at HN-LS, the minimum, maximum, and average temperatures exhibit significant positive and negative correlations with AC and GC, respectively. At the other four locations, there were significant negative and positive correlations between these meteorological factors and AC and GC, respectively. Likewise, at HN-LS, surface solar radiation was positively correlated with AC and negatively correlated with GC. However, at the other four trial locations, contrary correlation trends were observed. In terms of rainy days and accumulated rainfall, our research suggests that increased rainy days are potentially associated with lower levels of AC and higher levels of CD. The significant impact of accumulated rainfall on quality traits was observed, but a more general pattern was not observed in our experiment. Moreover, we considered the rice quality grade as a trait and investigated its correlation with meteorological factors as a reference. For trial locations in the rice cropping region of the upper, middle, and lower reaches of the Yangtze River (i.e., SC-GH, AH-HF, HN-CS, and HB-EZ), there exists a significant correlation between better rice quality and the minimum, maximum, and average temperatures. General combining ability analysis for phenotype plasticity of quality traits In this study, we evaluated the stability of quality traits for 141 hybrid combinations across five trial locations using the Finlay-Wilkinson regression (FW), a measurement of phenotypic plasticity (Supplementary Table S6 ) (Fig. 3 ). The combining ability for phenotypic plasticity was conducted to identify potential parents capable of producing hybrid combinations with stable quality traits. Figure 4 shows the GCA value of the hybrid parents in relation to the plasticity of the seven quality traits. Notably, parents with a low GCA value for plasticity are more likely to derive hybrid combinations with stable quality performance. Three TGMS lines (ZXS, HX302S, and HY468S) and nine restorer lines (HH8012, HH7503, DZ, LKSM13, HH2646, HH5106, HH8549, YZ, and WSSM534) consistently exhibited comprehensive low GCA values (≤ 0) for plasticity across more than three trial locations. It indicates their potential to consistently produce hybrid combinations with stable quality traits across different cropping regions and varied sowing management. Interestingly, parents with superior quality tend to exhibit low GCA values for plasticity. The 12 parental lines with low plasticity GCA value are all of high quality and have been widely used in high-quality hybrid rice development. We analyzed the quality statistics from yield testing trials of nationally approved hybrid rice varieties derived from all tested parental lines from 2016 to 2022. Notably, the hybrid rice varieties derived from parents with low plasticity GCA values exhibited a higher rate of high quality compared to those derived from parents with high plasticity GCA values (Supplementary Table S7 ). Genome-wide association analysis for phenotype plasticity and BLUP measurement of quality traits The genome-wide association study (GWAS) was conducted independently for the plasticity of each quality trait across different trial locations. This analysis revealed a total of 13 QTLs associated with grain quality plasticity (Fig. 5 a–b; Table 1 ). Among these, four QTLs were detected in two trial locations, two QTLs were detected in three trial locations, and the remaining seven QTLs were only detected in a single trial location. These findings underscore the complex genetic basis of plasticity for quality traits in different regions. In the case of AC plasticity, we identified six QTLs were identified on chromosomes 3, 9, 10, 11, and 12. Among these QTLs, four were identified in both HN-CS and AN-HF, while two were exclusively identified in HN-CS. As for CD plasticity, we detected five QTLs were detected on chromosomes 2, 3, 4, 7, and 8. Notably, all QTLs were identified in a single trial location, either HB-EZ or AN-HF. We identified QTL6 on chromosome 6 as being associated with the plasticity of two key taste quality traits (ASV and GC). The effect of QTL6 on ASV plasticity was detected in both AH-HF and SC-GH, while its effect on GC plasticity was observed exclusively in HB-EZ. These findings suggest that the plasticity of ASV and GC may share a similar genetic basis. A single QTL, QTL7, located on chromosome 7, was found to be associated with the plasticity of MRR. This association was observed across three trial locations including HB-EZ, HN-CS, and SC-GH. Furthermore, we employed the best linear unbiased prediction (BLUP) approach to mitigate the impact of non-genetic factors and calculate the genetic effect values for these quality traits (Supplementary Table S8 ). We also conducted a GWAS for the BLUP measurement. This analysis identified a total of 15 QTLs for the BLUP measurement of seven quality traits across five trial locations (Fig. 5 c–d, Supplementary Table S9 ). Among these QTLs, seven QTLs (QTL2, QTL3, QTL6, QTL7, QTL11, QTL12, and QTL13) were also identified in the previous GWAS of plasticity (Supplementary Table S10 ). These overlapping QTLs, identified in both GWAS results, could play a crucial role in regulating quality traits and responding to the quality plasticity observed in diverse cropping environments. Analysis of QTLs genetic effects and prediction of candidate genes We evaluated the genetic effects of all 13 plasticity QTLs (Fig. 6 a–f, Fig. S2 ). Among these, two QTLs (QTL6 and QTL7) were detected in three trial locations, thereby establishing them as stable major QTLs for rice quality plasticity. As these two QTLs also demonstrated an effect on the BLUP measurement, further investigation into the genetic effects of these QTLs could lay a foundation for molecular breeding strategies aimed at enhancing both the quality and stability of hybrid rice. The QTL6 was found to be associated with ASV and GC. The leading SNP (Chr6:6,722,905) of QTL6 had a − log10 ( p value) of 7.83 in AH-HF, 13.61 in HB-EZ and 31.03 in SC-GH. Three genotypes of the leading SNP of QTL6 were identified. A significance analysis revealed that CC genotypes showed significantly lower GC plasticity compared to CT and TT genotypes in HB-EZ. It is worth noting that the direction of the genetic effects of the identical ASV plasticity allele varied across different cropping environments. In the AH-HF, CC genotypes exhibited significantly lower ASV plasticity compared to varieties with CT and TT genotypes. Conversely, in the SC-GH, CC genotypes showed significantly higher ASV plasticity compared to varieties with CT and TT genotypes (Fig. 6 a–c). The QTL7 was associated with MRR. The leading SNP (Chr7:24,665,290) of QTL7 had a − log10 ( p value) of 15.69 in HN-LS, 9.93 in HB-EZ, and 7.67 in SC-GH. The leading SNP alleles exhibited significantly different MRR plasticity, as shown in Fig. 6 d–f. Notably, the AA genotypes displayed the lowest MRR plasticity value. Within the associated genomic region of the QTL6 and QTL7, a total of 6 and 18 annotated genes were found, respectively. Among these genes, ALK and GL7 , known for their functional annotations in controlling rice quality, were pinpointed as candidate genes for further analysis in rice quality plasticity. The ALK gene, located within QTL6, is a key gene controlling rice gelatinization temperature which encodes the soluble starch synthase II-3. The GL7/GW7 gene, located within QTL7, encodes a protein that is homologous to the LONGIFOLIA proteins found in Arabidopsis thaliana. This protein is known to regulate cell elongation, thereby affecting grain length and grain shape.

Discussion Rice quality is one of the important goals in hybrid rice breeding improvement. The formation of rice grain quality is under genetic control and also influenced by the environment. Numerous studies have been undertaken to explore the impact of sowing dates on rice quality. However, due to differences in location, tested varieties, climatic conditions, and other factors, the outcomes of these studies lack consistency. Yao et al. observed that a delay in the sowing date leads to an improvement in the appearance quality of rice, but results in a decrease in the eating and cooking quality (Yao et al. 2011 ). Wang et al. found that a delay in the sowing date causes the CD, PGWC, AC, and GC to exhibit a declining trend (Wenting et al. 2021 ). In this study, we analyzed the rice quality across multiple sowing dates in five trial regions located in Southern China and gathered the meteorological data for each region. It was observed that in the Yangtze River basin trials, the amount of high-quality rice (≥ grade 3 set by the Chinese Ministry of Agriculture) increases as the sowing date is delayed. The different sowing date essentially influences the weather conditions in rice growing. Previous studies indicated that an inappropriate sowing date will expose rice to unfavorable climatic conditions during the grain filling stage, resulting in a decline in rice quality (Cheng and Zhong 2001 ). It has been proved that light, temperature, and rainfall during the grain filling stage are the pivotal climatic factors influencing rice quality (Resurreccion et al. 1977 ; Cheng and Zhu 1998 ). This study quantified and analyzed several crucial meteorological factors during the grain filling stage to assess their relationship with rice quality and endeavored to improve the precision of selecting the appropriate sowing date in high-quality hybrid rice production. Deng et al. found that the average daily temperature ranges of 22–27 °C in the grain filling stage are recommended to achieve high grain yield and quality for irrigated rice in the Yangtze River basin (Deng et al. 2015 ). Our results suggest that lower average daily temperature can contribute to improving the quality of rice, supporting the conclusion from Deng et al. In addition, we found that the impact of meteorological factors on rice quality exhibited varied trends among trial locations, and even a single factor may affect multiple quality traits in contrasting directions, resulting in a complex effect on overall rice quality. For instance, surface solar radiation was positively correlated with AC and negatively correlated with GC at HN-LS, while contrary correlation trends were observed at the other four trial locations. This difference may be due to the fact that HN-LS is located in the South China rice cropping region, with persistent high temperatures and abundant rainfall in the dry season, and has distinct climate conditions compared to the other trial locations within the Yangtze River basin. Simultaneously, the decomposed meteorological factors are unavoidably related in different ways. For instance, rainfall consistently results in a more significant temperature decrease within AH-HF, HN-CS, and HB-EZ in the middle and lower reaches of the Yangtze River basin in comparison to SC-GH and HN-LS. This discrepancy can be attributed to the lower altitude and the prevalence of hot and sunny weather in the summer. Concurrently, rain can reduce solar radiation, so SC-GH in the upper reaches of the Yangtze River has low solar radiation due to cloudy and rainy weather. As the quality of rice is the result of the combined effects of multiple meteorological factors, further analysis and modeling based on more experiments are necessary. It is crucial to develop varieties with stable phenotypes that are less sensitive to environmental changes. This will help mitigate the adverse effects of frequent extreme climate events on rice yield and quality, guaranteeing the production of high-quality rice. This study employs phenotypic plasticity measurement to evaluate the stability of different quality traits in hybrid rice combinations. Combining ability analysis is a useful approach to selecting ideal parent lines for hybrid rice breeding (Chen et al. 2019 ). Through the analysis of combining ability in phenotype plasticity, we discovered that the TGMS lines ZXS, HX302S, and HY468S and the restorer lines HH8012, HH7503, DZ, LKSM13, HH2646, HH5106, HH8549, YZ, and WSSM534 showed low GCA effects on five or more quality traits plasticity across three rice cropping regions; they could be recommended for utilization in rice hybrid breeding programs to improve the stability of rice quality. A noteworthy result was found that parents with superior quality tend to exhibit low GCA values for plasticity and have more potential to develop a hybrid with optimal quality stability. ZXS is an indica TGMS line with high-quality approved in Hunan province in 2020. Its milled rice rate is 63.2%, chalky grains rate is 7%, chalkiness degree is 1.3%, amylose content is 16.1%, gel consistency is 60 mm, alkali digestion value is 7.0, and transparency is grade 1 (China Rice Data Center). A total of ten new hybrid rice varieties of ZXS have been nationally or provincially approved and certificated, and their quality has all reached the grade 2 or grade 1. The quality of ZhenLiangYouYuZhan (ZXS/YZ), a nationally approved new hybrid rice variety of ZXS, met grade-1 in the new variety regional trial and was awarded the Gold Award for Taste Evaluation of High-Quality Rice at the 4th National High-Quality Rice Competition in 2023. In our experimental field settings, the ZXS/YZ combination showed high-quality with grade 3 or superior performance in 12 out of 19 records, and these records occurred in all locations except for HB-EZ. To uncover the genetic basis of phenotype plasticity in rice quality traits, GWAS for the quality plasticity was performed and identified 13 plastic QTLs. In line with previous studies (Zan and Carlborg 2020 ), we determined that plasticity is polygenic and exhibits a variable genetic basis for rice quality traits across regions. The genetic effects of plastic QTL can change in different locations. For instance, the QTL6 is a multi-effect QTL associated with the plasticity of ASV and GC, where genetic effects on ASV differ among trial locations. We also identified seven plastic QTLs overlapped in the GWAS of BLUP measurement of quality traits, indicating these QTLs could regulate the quality traits and respond to the plasticity of quality traits. This further validates previous research that there may be a connection between the genetic regulation of traits and plasticity (Zan and Carlborg 2020 ; Jin et al. 2023 ). In the present study, we endeavored to identify the candidate genes of two major plasticity QTLs for rice quality. The region of these two major QTLs contains 24 annotated genes and 2 of them ( ALK and GL7 ) are known to regulate the rice quality. ALK is the key gene controlling rice gelatinization temperature, which is closely associated with the eating and cooking quality in rice (Gao et al. 2011 ). According to previous studies, there are three main alleles of ALK , including ALK a , ALK b , and ALK c , with ALK c controlling high gelatinization temperature, and ALK a and ALK b controlling low gelatinization temperature ( Chen et al. 2020 ; Huang et al. 2021 ). In this study, we identified two alleles of ALK including ALK b and ALK c in our parental rice accessions. Interestingly, we found that the leading SNP genotype of QTL6 was completely linked with ALK . Specifically, the CC genotype of the leading SNP of QTL6 was linked with ALK c while the TT genotype was linked with ALK b (Fig. 6 g–h). Sequence analysis showed that among the above low plasticity GCA parents, all carried ALK b except HH7503 and HH5106, which carried ALK c . To some extent, this result explains although rice varieties with low to medium GT are preferred by consumers, those carrying ALK c alleles that regulate high GT are still commonly employed in rice breeding, potentially due to their plasticity levels. The other plasticity candidate gene, the GL7 locus, plays a significant role in grain size diversity and has been utilized in rice breeding. Wang et al. demonstrated that the copy number variation (CNV) at the GL7 locus led to the difference in the rice grain size (Wang et al. 2015b ). Whole genome sequencing analysis of parental rice accessions revealed that the CNV at GL7 was linked with the leading SNP of QTL7. The varieties containing the CNV at GL7 carry the AA genotype of the leading SNP of QTL7 and the remainder carry the GG genotype (Fig. 6 i). Rice varieties with multi-copy at GL7 locus exhibit lower plasticity in terms of milled rice ratio, which can be attributed to their long-grain phenotype. Further studies will be necessary to determine the effect of these candidate genes on phenotypic plasticity and their pleiotropic effects on quality traits and plasticity. These QTLs and candidate genes have the potential to contribute to improving rice quality and maintaining the stability of quality in breeding.

Improving quality is an essential goal of rice breeding and production. However, rice quality is not solely determined by genotype, but is also influenced by the environment. Phenotype plasticity refers to the ability of a given genotype to produce different phenotypes under different environmental conditions, which can be a representation of the stability of traits. Seven quality traits of 141 hybrid combinations, deriving from the test-crossing of 7 thermosensitive genic male sterile (TGMS) and 25 restorer lines, were evaluated at 5 trial sites with intermittent sowing of three to five in Southern China. In the Yangtze River Basin, it was observed that delaying the sowing time of hybrid rice combinations leads to an improvement in their overall quality. Twelve parents were identified to have lower plasticity general combing ability (GCA) values with increased ability to produce hybrids with a more stable quality. The parents with superior quality tend to exhibit lower GCA values for plasticity. The genome-wide association study (GWAS) identified 13 and 15 quantitative trait loci (QTLs) associated with phenotype plasticity and BLUP measurement, respectively. Notably, seven QTLs simultaneously affected both phenotype plasticity and BLUP measurement. Two cloned rice quality genes, ALK and GL7 , may be involved in controlling the plasticity of quality traits in hybrid rice. The direction of the genetic effect of the QTL6 ( ALK ) on alkali spreading value (ASV) plasticity varies in different cropping environments. This study provides novel insights into the dynamic genetic basis of quality traits in response to different cropping regions, cultivation practices, and changing climates. These findings establish a foundation for precise breeding and production of stable and high-quality rice. Supplementary Information The online version contains supplementary material available at 10.1007/s11032-024-01442-3. Keywords

Supplementary Information Below is the link to the electronic supplementary material.

Author contribution YZY, XML, and KW conceived and designed experiments. RRC, DXL, and JF contributed to data analysis and the first draft of the manuscript. RRC and YJW contributed to data collection. CJF, PQ, XWZ, ZBS, KH, LL, and WZ were in charge of filed management. YZY, XML, and KW supervised and complemented the manuscript writing. All authors read and approved the final manuscript. Funding This work was supported by grants from National Natural Science Foundation of China (U19A2032), Science and Technology Innovation Program of Hunan (2021NK1001, 2022NK1010, 2023NK1010), and Shenzhen Science and Technology Project (202320D314). Data availability All data generated or analyzed during this study are included in this published article and its supplementary information files (Supplementary Table S1 , S2 , S3 , S4 , S5 , S6 , S7 , S8 , S9 , S10 ; Fig. S1 , S2 ). Declarations Ethics approval and consent to participate Not applicable. Consent for publication Not applicable. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:59

Mol Breed. 2024 Jan 15; 44(1):4

oa_package/cc/31/PMC10788329.tar.gz

PMC10788330

38221524

Introduction Developmental deformations of the skull in infants, especially those resulting from premature fusion of cranial sutures, are an area of a multidisciplinary interest 1 , 2 . Skull deformities and associated developmental problems constitute a hallmark of single suture synostoses, from which nonsyndromic sagittal craniosynostosis (NSC) is the most common type, estimated for 2–3:10,000 live births 1 , 3 . Premature sagittal synostosis leads to an elongation of the skull in sagittal axis with a simultaneous constriction in transverse axis, often with compensatory changes, such as frontal and/or occipital bossing, or a retroorbital depression 4 . This cranial deformation affects the child's social and emotional development, making sagittal synostosis not only a surgical problem, often requiring an additional neurologic investigation and psychosocial support 1 , 5 , 6 . Children with NSC have normal IQ but in 7–37% they present verbal or language problems, visuospatial deficits and other cognitive delays 7 , 8 . The incidence of epilepsy among children with craniosynostosis is estimated at 5%, with such independent risk factors as brain compression, obstructive sleep apnea and hydrocephalus 9 . An electroencephalographic interest in craniosynostosis dates back to 60’s of the XXth century. Few studies from that period focused mainly on the search for epileptiform discharges in EEG 10 – 12 and hypothesized a predisposition to epilepsy in children with a craniosynostosis. The introduction of quantitative EEG (QEEG) protocols enabled the computational analysis of EEG records 13 . It is based on objective parameters, such as the amplitude, recorded in each channel, and coherence, a measure of phase synchrony between EEG signals 14 , 15 . The amplitudes and coherence indices have been studied in patients at various age 16 – 18 and have been implemented in the diagnostics of neurologic and psychiatric diseases 19 , 20 as well as in the assessment of functional connectivity within the cerebral hemisphere (intrahemispheric coherence, HCoh) or between the hemispheres (interhemispheric coherence, ICoh) 21 , 22 . HCoh may vary in conduction-affecting disorders (i.e., mild head injury has been found to raise its value), whilst dementia or Alzheimer’s disease is characterized by its decrease 23 . In turn, high ICoh values characterize the areas of well-developed connections 24 . Recent EEG studies on pediatric population revealed the usefulness of QEEG parameters in predicting language development, also in patients with craniosynostosis 25 – 27 . Clinical rationale for the study The functional assessment of children with NSC has recently become one of the postulates of international collaborations aimed at patient stratification and clinical staging 28 . So far, the electroencephalographic assessment of children with craniosynostosis has been only a sporadic element of the management, not bringing significant changes to the diagnostic procedure and surgical technique. However, an increasing interest in the QEEG protocols in a developmental and clinical research has created the need of an objective, quantitative evaluation of the observed EEG phenomena, but to date any study was focused on children with craniosynostosis. The primary goal of our study was to investigate the selected QEEG parameters (peak-to-peak amplitudes, HCoh and ICoh) in children with NSC compared to a group of healthy children in order to assess a relation between the cranial deformation and the electrical activity and connectivity in the brain. A rationale for using coherence and amplitude values was that a premature fusion of the sagittal suture, resulting in local skull constriction, may affect the power and hemispheric connectivity in children with NSC. These parameters can be easily obtained in every hospital setting, they are also reproducible and comparable with other papers. The secondary goal was to evaluate the effect of surgical treatment on the EEG records in patients with NSC.

Materials and methods Study design This retrospective study was performed on EEG records of patients with NSC treated in our institution between January 2018 and June 2020. The EEG was registered as an element of the diagnostic protocol of NSC in our institution and an informed consent was obtained from legal guardians of all patients. The study protocol was approved by the Bioethics Committee of Medical University of Warsaw (decision number AKBE/110/2021), and abides by the 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Population The group of children with NSC aged 1–18 months included 25 subjects (22 M, 3F). In all children a typical skull deformation (scaphocephaly) was diagnosed and a total or partial fusion of sagittal suture was evidenced by radiology. In none of the patients genetic aberrations, metabolic and immune system disorders, or a family history of craniosynostosis were reported. Mean age of the NSC group was 9.62 months, median age 6.83 months, standard deviation 4.83. An age-matched control group was designed (mean age 9.42 months, median age 8.00 months, standard deviation 4.81). The control group included 25 infants (21 M, 4F) with normocephalic skulls and normal EEG records. In none of the controls seizures were observed and none of the patients received pharmacologic treatment. Method In the study we retrospectively analyzed the QEEG data from preoperative examinations (group A, 25 cases) and compared them with postoperative records performed 3 months after the surgery (group B, 23 patients). Two patients were lost to follow-up. All patients were operated on with the same technique (strap craniectomies) by the same neurosurgical team. All surgeries were considered successful, with no complications. In 1-year follow up there was no need to repeat surgery in any patient. The obtained data were finally compared with a control group (group C). Group A was also compared for age (below and above 6 months), pattern of sagittal suture fusion and differences in skull shape. The pattern of sagittal suture fusion included the indication of the fused 1/3 portion of the suture (A-anterior, M-middle, P-posterior, or a combination of above if more than one portion was fused). For the differences in skull shape a classification including 5 distinct types was used (dolichocephaly, leptocephaly, clinocephaly, bathrocephaly, sphenocephaly), as in the paper by Di Rocco et al. 29 . QEEG procedure Recordings The EEG was performed during physiologic sleep (NREM phase 2), in a quiet room with dim lighting. The examination was carried out with the use of Elmiko DigiTrack v. 14 device. QEEG data were obtained from 19 scalp electrodes placed according to 10–20 system. The impedance of each electrode was maintained below 20 kΩ. The sampling rate was 250 Hz, and the filtering range was 0.5–70 Hz. A unipolar reference was used. QEEG signal preprocessing 50 Hz notch filtering was performed to remove power frequency interference. The high-pass filter, with the − 3 dB cutoff the frequency of 0.5 Hz and the low-pass filter with the − 3 dB cutoff frequency of 70 Hz were selected. Physiological artifacts were removed using the software provided by the manufacturer. QEEG parameters calculation Power spectra for each lead were obtained with the Fast Fourier Transformation algorithm. The measured parameters included the amplitude, interhemispheric (ICoh) and intrahemispheric (HCoh) coherence indices calculated from an artifact-free epoch of 2 s duration. The amplitudes were measured at 8 points for each side of the head: O1, P3, T5, C3, T3, F3, F7, Fp1 for the left hemisphere and O2, P4, T6, C4, T4, F4, F8, Fp2 for the right hemisphere. The amplitude was defined as the maximum peak-to-peak deviation of the signal within the epoch. The coherence was defined by equation: Coh = (Sxy) 2/(Sxx × Syy), where Sxy, Sxx and Syy were cross-spectrum estimates of leads x and y, respectively. The HCoh values were computed for 24 intrahemispheric electrode pairs: O1–P3, O1–T5, P3–C3, P3–T5, P3–T3, C3–F3, C3–T3, C3–F7, F3–Fp1, T5–T3, T3–F7, F7–Fp1 for the left hemisphere and O2–P4, O2–T6, P4–C4, P4–T6, P4–T4, C4–F4, C4–T4, C4–F8, F4–Fp2, T6–T4, T4–F8, F8–Fp2 for right hemisphere. The ICoh values were computed for 12 interhemispheric electrode pairs: O1–O2, P3–P4, C3–C4, F3–F4, T5–T6, T3–T4, F7–F8 and Fp1–Fp2. Statistical analysis The calculated parameters were compared between the NSC group (A) and the control group (C) with the directional non-parametric Mann–Whitney rank-sum test using z-scores. The alternative hypothesis (control group parameters higher than in NSC group) was accepted if the results were considered significant ( p < 0.05). In an additional analysis the age groups were compared with Mann–Whitney rank-sum test. Multiple comparisons in relation to the pattern of sagittal suture fusion and the type of cranial deformity were performed using non-parametric ANOVA (Kruskal–Wallis test) with Bonferroni correction in order to avoid false significant results (an adjusted p value for the pattern and shape groups was set at p = 0.008). The effect of surgery on QEEG was investigated by the comparison of parameters calculated for groups A and B (pre- and postoperatively). A non-parametric Wilcoxon signed-rank test was used for the analysis with p value < 0.05 considered significant. The statistical analysis was performed with TIBCO Data Science/Statistica software by StatSoft Europe, version 13.3 PL for Microsoft Windows 10 Pro.

Results NSC group versus control group in terms of amplitudes and coherence The amplitude was calculated for all leads. The amplitude values are presented in Table 1 . Significantly lower amplitudes were found in leads O1, O2, P3, P4, T5 and C4 in the group of children with NSC (Table 1 , Fig. 1 ). In terms of interhemispheric coherence, no differences were observed between groups A and C (Table 2 ). However, significant differences in intrahemispheric coherence were found, mainly in the occipital, posterior parietal and posterior temporal areas (Table 2 , Fig. 2 ). Amplitudes and coherence in NSC children in relation to age, pattern of sagittal suture fusion and skull shape Age In the studied population there were 12 children at the age below 6 months and 13 children above this age (4 children were older than 12 months). Mean amplitudes in most leads were higher in children older than 6 months (differences not statistically significant) but the only significant difference was observed in O1 lead ( p = 0.0377, z = 2.0776) in favor of younger children. No differences were found in terms of interhemispheric coherence. In terms of HCoh, the differences were significant for P3–T3 ( p = 0.0324, z = 2.1396) and C3–T3 ( p = 0.0438, z = 2.0156). In both cases, higher values were noted in children at the age below 6 months. The amplitudes and coherence in children older than 12 months were not statistically different comparing to individuals aged 6–12 months. Pattern of sagittal suture fusion Four distinct pattern of sagittal suture fusion were identified: M (5 cases), MP (9 cases), AMP (10 cases) and AM (1 case). In the comparison of patterns, no significant differences in amplitudes were found. The highest ICoh values were observed for the MP pattern, but only in O1–O2 the values were significantly different from other patterns ( p = 0.0060, H = 10.2261). There were no differences in intrahemispheric coherence. Skull shape Four types of cranial deformity were identified: sphenocephaly (10 cases), clinocephaly (8 cases), bathrocephaly (4 cases) and dolichocephaly (3 cases). The highest amplitude values were observed in dolichocephaly (not significant). No differences in terms of amplitudes and coherence indices were observed among the skull shapes. The differences depending on age and pattern of sagittal suture fusion are presented in Fig. 3 . Amplitudes and coherence in NSC patients before and after surgery In terms of amplitudes, only the T6 lead showed a significant difference between groups A and B ( p = 0.0244, z = 2.2507). In the remaining leads, no significant differences were registered before and after the surgery (Table 1 ). Postoperative ICoh values were lower than preoperative (Table 2 ), but the only significant difference was in F7–F8 ( p = 0.0244, z = 2.2507). In terms of intrahemispheric coherence, no significant differences were found between groups A and B (Table 2 ), with the exception of HCoh for P4–C4, which was higher after the surgery ( p = 0.0157, z = 2.4156).

Discussion Electroencephalography in patients with craniosynostosis provides an interesting information about the brain functioning, even many years after the diagnosis 30 . Nevertheless, EEG is not typically included in the standard diagnosis of premature suture fusion, except in a few cases of syndromic craniosynostosis 31 . To date, no EEG pattern has been recognized as characteristic for craniosynostosis, despite the objective differences in brain structure resulting from this disease 32 . However, despite the implemented treatment, findings on cognitive functioning in NSC children still indicate considerable problems 1 , 33 , refocusing an interest of researchers on detailed EEG assessment with the use of objective parameters. As shown by the results, children with NSC, despite normal EEG records and lack of epileptiform graphoelements, present some differences in the electroencephalographic profile comparing to normocephalic children. This is evident both in the amplitudes calculated bilaterally in occipital and posterior parietal areas, but also in HCoh, reflecting the condition of intrahemispheric connections. The distribution of significantly lower amplitudes in children with NSC corresponds to typical areas of cranial constriction in posterior parietal and occipital regions and may be related to local brain compression, which was postulated by other authors 34 . Amplitude values in NSC patients were not specific for the child’s age, although a trend to increase with age was observed. Despite various types cranial deformation and different patterns of sagittal suture fusion, the amplitude values did not differ significantly across our NSC population. Single differences observed in the age subgroups may suggest subtle alterations in occipital areas in NSC children comparing to healthy subjects. Surgical procedure did not change the mean amplitudes in children with NSC. In our population, the only change observed was a decrease in the mean amplitude in right posterior temporal area. Nevertheless, significant differences in occipital, parietal and left posterior temporal areas were still observed postoperatively (Fig. 1 ). In addition, the noted differences in C3 and T3 leads do not seem to be related to surgery. Concluding, despite showing some individual differences, we do not see significant variation in amplitudes between age groups, pattern of sagittal suture fusion or skull shape. The situation is different with intrahemispheric coherence, which showed marked differences between the group of children with NSC and the control group. In the NSC group, the differences in connectivity clearly concerned the occipital, posteroparietal, posterotemporal and, to a lesser extent, centro-frontal areas, indicating the regions of significantly smaller intrahemispheric connectivity. Interestingly, the implemented treatment did not significantly improve the intrahemispheric coherence in children with NSC and after surgery they still presented lower Hcoh values than the control group (Fig. 2 ). It seems that surgery allows some improvement of HCoh in the posterior temporoparietal and centro-temporal areas, but the decreased values still remain bilaterally in the area of the occipitotemporal junction, which might be considered as a microfunctional substrate of reported neurocognitive decline. An additional observation is the persistence of reduced HCoh values in frontotemporal and centro-frontal areas, which may be related to problems with speech development observed in children with NSC even after surgical treatment 35 . Interhemispheric coherence did not differ between groups A and C, suggesting that NSC children do not differ from normocephalic children in terms of interhemispheric connectivity. It is worth emphasizing that surgery does not affect the ICoh and even contributes to a slight decrease in connections between the posterior frontal areas. It seems, however, that both this observation and the decrease of HCoh in frontal regions may reveal a problem that is possibly not sufficiently addressed in the surgical technique. Classic surgical methods (inverted "pi", strap craniectomies, barrel-stave osteotomies) and minimally invasive techniques (endoscopy-assisted suturectomies or spring-assisted surgery) do not primarily focus on the frontal and fronto-basal regions. Leaving these areas intact may promote further local constriction which to some extent may be associated with a decrease in the number of associations in frontal areas. In our opinion, however, these results, along with the existing clinical evidence, do not support the concept of an extension of the surgical field to these regions only due to the QEEG findings. Strengths and limitations of the study The most important advantage of this work is a novel insight into the EEG recording in NSC children. In this paper we characterize not only the values of amplitudes in individual leads, but also the values of intrahemispheric and interhemispheric coherence. It was also possible to relate these findings to normocephalic children and to compare the impact of surgery on the calculated parameters. However, this study has several limitations. The QEEG data were obtained from 19 scalp electrodes placed according to 10–20 system, which may be burdened with inaccuracies and spatial aliasing 13 . The system, however, is still widely used in research and in clinical practice ensuring comparability and repetitiveness 36 . In our opinion the findings should be validated in a setting based on higher resolution EEG techniques. The second limitation is the relatively small number of subjects, which means that the obtained data are rather illustrative and do not constitute clear guidelines in the diagnosis of patients with NSC. A larger number of children would make the observed trends credible. The third limitation is the fact that the presented QEEG data were not correlated with the neurodevelopmental data, mainly due to the age of our patients. Language assessment tests (such as Bayley scale version III) are less reliable in infants than in older children. In addition, the third version of Bayley scale was not yet validated in Polish population at the time of the study, therefore an electroencephalographic-developmental correlation is still one of our goals in the future.

Conclusions Children with NSC have their own unique EEG profile. The differences in relation to normocephalic individuals are visible mainly in the occipital, posterior parietal and posterior temporal regions. NSC patients achieve lower values of amplitudes and intrahemispheric coherence there, but interhemispheric connectivity seems not to be affected comparing to normocephalic subjects. Surgical treatment does not change the EEG profile of NSC children—differences in amplitudes in the occipital, posterior parietal and posterior temporal regions are still visible after surgery. The operation improves intrahemispheric connectivity (associations), but there is a marked difference in the anterior temporofrontal and centro-frontal areas, which may be related to speech development difficulties and other neurocognitive disorders being a potential goal of future therapies targeted at coherence improvement (i.ex. neurofeedback). Further investigation based on higher resolution EEG techniques is needed in order to validate these preliminary findings. A clinical correlation with speech and neurocognitive delay might promote the use of QEEG as a prescreening method of early diagnostics of future neurodevelopmental and, potentially, mental disorders.

Despite the undertaken treatment, children with nonsyndromic sagittal craniosynostosis (NSC) are burdened with problems with speech development, visuospatial and other cognitive deficits. The electroencephalographic assessment has not influenced the diagnostics and treatment strategy of craniosynostosis so far but the introduction of quantitative EEG (QEEG) protocols renewed an interest in the functional aspect of this disease. In this study we retrospectively assessed the QEEG records of 25 children with NSC aged 1–18 months (mean age 9.62 months) before and after surgery. In each case, the amplitude, interhemispheric (ICoh) and intrahemispheric (HCoh) coherence indices were calculated. Obtained data were compared to age-matched control group of 25 normocephalic children. Children with NSC presented significantly lower values of amplitudes and intrahemispheric coherence in occipital, posterior parietal and posterior temporal regions than normocephalic children. The values of amplitudes, ICoh and HCoh in pre- and postoperative QEEG records mostly remained unchanged, with a slight improvement in HCoh in centro-parietal area. These findings suggest that NSC children present their own QEEG profile. The operative treatment improves an intrahemispheric connectivity, but there still exists a significant difference in the occipitotemporal, frontotemporal and centro-frontal areas, which may be considered as a functional substrate of reported speech and neurocognitive problems. QEEG findings in nonsyndromic sagittal craniosynostosis. Subject terms

Author contributions T.S.—Protocol development, Data collection, Data analysis, Figures preparation, Manuscript writing J.S.—Data collection, Manuscript writing Both authors read and approved the final version of the manuscript. Funding This publication was prepared without any external source of funding. Data availability The data belong to Bogdanowicz Memorial Hospital for Children in Warsaw and are not available to share unless in the form included in the manuscript and supplementary materials. For additional data requests please contact Dr. Tymon Skadorwa at [email protected]. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:59

Sci Rep. 2024 Jan 14; 14:1301

oa_package/34/ad/PMC10788330.tar.gz

PMC10788331

38221525

Autosomal dominant episodic ataxia type 2 (EA2) is caused by variants in CACNA1A . We examined a 20-year-old male with EA symptoms from a Japanese family with hereditary EA. Cerebellar atrophy was not evident, but single photon emission computed tomography showed cerebellar hypoperfusion. We identified a novel nonsynonymous variant in CACNA1A , NM_001127222.2:c.1805T>G (p.Leu602Arg), which is predicted to be functionally deleterious; therefore, this variant is likely responsible for EA2 in this pedigree. Subject terms

Hereditary episodic ataxia (EA) is a heterogeneous group of movement disorders characterized by recurrent spells of truncal ataxia and incoordination 1 . EA type 2 (EA2, MIM: 108500) is an autosomal dominant hereditary EA caused by heterozygous variants in the calcium voltage-gated channel subunit alpha-1A gene ( CACNA1A , MIM: 601011). Here, we studied a patient with hereditary EA originating from Shikoku Island, Japan (Fig. 1A ). The proband (Patient III-1) was a 20-year-old male who had experienced brief episodes (<30 min) of ataxia since childhood that were precipitated by actions such as running or riding a bicycle. The frequency of the episodes was once a week at most. Witnesses said that his eyes were bloodshot while he was experiencing EA. Interictal neurological examination showed no abnormalities. Brain magnetic resonance imaging (MRI) did not show obvious cerebellar atrophy (Fig. 1B ). Interictal brain single photon emission computed tomography (SPECT) using N-isopropyl-p-(iodine-123)-iodoamphetamine ( 123 I-IMP) with three-dimensional stereotactic surface projections showed hypoperfusion in the cerebellum, brainstem, and lateral occipital lobe (Fig. 1C ). His symptoms were initially improved with acetazolamide (125 mg/day), but the effect did not last. According to the proband, his mother (II-3) and maternal grandfather (I-1) experienced the same ataxic symptoms until they were 20 years old. His 16-year-old younger brother (III-2) and 9-year-old sister (III-3) also exhibit the same ataxic symptoms. We first confirmed the absence of repeat expansion in genes known to be responsible for spinocerebellar ataxia (SCA) 1–3, 6–8, 10, 12, 17, and 36 and dentatorubral-pallidoluysian atrophy. We also confirmed the absence of pathogenic variants in SCA31. We then performed exome sequencing of the proband using SureSelect Human All Exon v6 (Agilent Technologies, Santa Clara, CA, USA) on the Illumina NovaSeq 6000 platform (Illumina, Inc., San Diego, CA, USA). We achieved a sequencing depth of 168× and identified 24,655 variants in the proband. Since the proband was diagnosed with EA, we selected 21 variants located in eight genes known to be associated with EA (Supplementary Table 1 ). By excluding variants already registered in public databases [the 1000 Genomes Project ( http://www.1000genomes.org ), ExAC ( http://exac.broadinstitute.org/ ), and gnomAD ( https://gnomad.broadinstitute.org/ )], we identified a novel nonsynonymous variant located in exon 14 of the calcium voltage-gated channel subunit alpha-1A gene ( CACNA1A ), NM_001127222.2:c.1805T>G (p.Leu602Arg) (Fig. 2 ). The variant was predicted to be “probably damaging” by PolyPhen-2 ( http://genetics.bwh.harvard.edu/pph2/ ), “deleterious” by SIFT ( https://sift.bii.a-star.edu.sg/ ) and “disease-causing” by Mutation Taster ( https://www.mutationtaster.org ) with a CADD score of 29.4. Given the patient’s symptoms, genes known to be associated with dystonia, episodic kinesigenic dyskinesia and SCAs were also examined (Supplementary Table 1 ). However, no pathological variants were detected in any genes associated with these three conditions. We validated the CACNA1A variant in the patient by Sanger sequencing (forward primer, 5′-GGGAAAGTGAGCCTCGTGT-3′ and reverse primer 5′-GGAGTTGGAATTCCTGTGAAG-3′). We confirmed that the patient was heterozygous for the variant, consistent with the autosomal dominant mode of disease inheritance (Fig. 2A ). We also examined the CAG repeat length in CACNA1A and confirmed that the proband was homozygous for an 11-repeat allele that is nonpathogenic. According to the ACMG/AMP/CAP guidelines, the p.Leu602Arg variant is classified as “likely pathogenic”, meeting the PM1, PM2, PP3 and PP4 criteria 2 . The CACNA1A variant data have been deposited in ClinVar ( https://www.ncbi.nlm.nih.gov/clinvar/variation/1809801/ ). We described here a Japanese EA2 patient carrying a novel nonsynonymous heterozygous variant [ CACNA1A , NM_001127222.2:c.1805T > G (p.Leu602Arg)]. The patient exhibited a typical EA2 phenotype 1 , 3 . The neuroradiological features of our patient included hypoperfusion of the cerebellum on brain SPECT despite no marked cerebellar atrophy on brain MRI. Brain SPECT is a functional neuroimaging technique for evaluating cerebrovascular disorders, neurodegenerative diseases, and epilepsy and may be able to detect lesions that do not produce abnormal findings on MRI 4 , 5 . As in the present case, brain MRI revealed no cerebellar atrophy in other EA2 cases 6 , 7 . To our knowledge, this is the first report of brain SPECT in a patient with EA2 caused by a heterozygous point mutation in the CACNA1A gene, although a brain SPECT study has been reported for a patient with familial hemiplegic migraine 1 (FHM1, MIM: 141500) carrying a heterozygous point mutation in the CACNA1A gene 8 . Additionally, several brain SPECT studies in patients with SCA6 (MIM: 183086) caused by a CAG trinucleotide repeat expansion in the CACNA1A gene have been reported 9 – 11 . All reported patients with CACNA1A variations showed atrophy and hypoperfusion localized in the cerebellum 8 – 11 . There are no previous reports of brain SPECT in patients with EA2; therefore, it is currently unclear whether reduced perfusion of the brainstem is common. It is important to collect brain SPECT data from more EA2 patients. The nonsynonymous variant is located within one of the ion-transport domains that is essential for the channel function of the CACNA1A protein (Fig. 2B ). Although the nonsynonymous variant is predicted to be highly pathogenic by multiple prediction tools, including “probably damaging” by PolyPhen-2 and “disease-causing” by Mutation Taster, we were unable to perform segregation analysis of the variant because additional family members were not available. We conclude that the novel CACNA1A variant NM_001127222.2:c.1805T>G (p.Leu602Arg) is highly likely to be responsible for EA2 in the current pedigree. Clinically, when patients complain of recurrent ataxic episodes, it is important to look for cerebellar hypoperfusion by brain SPECT even when brain MRI does not show cerebellar atrophy. Supplementary information

Supplementary information The online version contains supplementary material available at 10.1038/s41439-023-00261-w. Acknowledgements This work was supported by Grants-in-Aid from MEXT, Japan (#23K06853) and the Cooperative Research Project Program of the Medical Institute of Bioregulation, Kyushu University. We thank Jeremy Allen, PhD, from Edanz ( https://jp.edanz.com/ac ) for English language correction of this manuscript. HGV Database The relevant data from this Data Report are hosted at the Human Genome Variation Database at 10.6084/m9.figshare.hgv.3357. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:59

Hum Genome Var. 2024 Jan 15; 11:3

oa_package/09/3a/PMC10788331.tar.gz

PMC10788332

38221521

Introduction Holographic display technology can completely record and reconstruct the wavefront information of 3D objects, and it is one of the most promising naked-eye 3D display technologies 1 , 2 . Color holographic 3D display with large viewing angle has always been pursued, and it has important application values in medical treatment, industrial inspection, education and entertainment 3 – 5 . However, the color and viewing angle of holographic 3D display mainly depend on the wavelength of laser and the pixel size of current spatial light modulator (SLM). Inevitable color differences and narrow viewing angle seriously affect the holographic display effect and hinder the application of holographic 3D display in many fields. In order to enlarge the viewing angle of holographic 3D display, many methods have been proposed 6 – 9 . For example, the method based on time or space multiplexing can enlarge the holographic viewing angle or realize full color by increasing the space bandwidth product of the system 10 – 15 . However, under the limited space bandwidth product, there lies a trade-off between color and large viewing angle. In addition, holographic optical element is also used in holographic systems to expand the viewing angle 16 – 19 . However, due to the wavelength selectivity of the holographic optical element, its application in color holographic display is very difficult. Besides, new optical modulation elements are used in holographic systems 20 – 23 . In 2017, researchers used scattering film and wavefront shaper to break through the limitation of SLM and achieved a viewing angle of 36°, but only monochrome holographic display was achieved 24 . In 2020, researchers used a coherent backlight unit to achieve a 30° color holographic display effect 25 . Liquid crystal is a structured dynamic soft material that can be controlled by various external stimuli and presents several different textures based on its geometric constraints and applied external stimuli 26 – 35 . In order to eliminate the dispersion of liquid crystal diffraction devices, the Pancharatnam-Berry optical elements with specifically designed spectral response are proposed 1 , 36 . In 2022, we proposed a monochromatic holographic 3D display system based on liquid crystal grating, and a viewing angle of 57.4° was realized 37 . The basic optical mechanism of liquid crystal grating is to generate periodic electric field distribution in liquid crystal 38 . However, the diffraction angle of liquid crystal grating depends on the wavelength, which leads to the inevitable chromatic aberration problem and the application of liquid crystal grating in color display has also been seriously hindered. In recent years, although there have been many methods to expand the viewing angle of monochromatic holographic 3D display, color holographic 3D display technology with large viewing angle has not achieved a breakthrough, which limits the application of holographic display. Here, a color liquid crystal grating based holographic 3D display system with large viewing angle is proposed, as shown in Fig. 1a . Different from the traditional liquid crystal grating that has inevitable chromatic aberration, the proposed color liquid crystal grating can perform secondary diffraction modulation on red, green and blue reconstructed images with the same diffraction angle when the voltage is applied, thus avoiding chromatic aberration. In addition, a chromatic aberration-free hologram generation mechanism is proposed to cooperate with color liquid crystal grating to achieve large viewing angle color display. The proposed system shows a color viewing angle of ~50.12°, without any chromatic aberration. When the proposed system displays a monochrome holographic 3D image, the viewing angle can reach ~91.5°. The proposed system solves the problems of small viewing angle and serious chromatic aberration in the traditional holographic 3D display system, which has a decent display effect and broad application prospect.

Materials and methods Sample fabrication In order to fabricate the color liquid crystal grating, firstly, a planar indium tin oxide electrode is deposited on the inner surface of the bottom substrate, and the planar indium tin oxide electrode of the bottom substrate is etched into three different periodic strip indium tin oxide electrodes. Secondly, the polyimide layer is coated on the inner surface of the bottom substrate and rubbed in the direction which is perpendicular to the periodic strip indium tin oxide electrodes. Thirdly, a top glass substrate without indium tin oxide electrode is coated with polyimide layer and rubbed in the same direction as the rubbing direction of the bottom substrate. Fourthly, the top glass substrate and the bottom substrate are encapsulated into a cell. Finally, the liquid crystal material is poured into the cell, and then the liquid crystal director distribution is homogeneously aligned perpendicular to the periodic strip indium tin oxide electrodes direction.

Results Structure of the proposed system As shown in Fig. 1b , the proposed system consists of a red laser, a green laser, a blue laser, three shutters, three beam splitters (BSs), a mirror, a signal controller, a beam expander (BE), three lenses, an SLM, a filter and a color liquid crystal grating. Among them, the red laser, green laser, blue laser and three shutters are used to generate red, green and blue light emitted in time sequence. The mirror and two BSs are used to coincide the optical axes of red, green and blue light. The red, green and blue light with overlapping optical axes are reflected to the SLM by BS after passing through the BE and lens I. The corresponding blazed gratings are superimposed on the holograms. The loading order of red, green and blue holograms is consistent with the emergent order of red, green and blue light. The signal controller is a synchronization control system based on a programmable microcontroller. It is used to control the switching time of three shutters and three color holograms, and to control the voltage applied to the color liquid crystal grating. The color liquid crystal grating is located at the back focal plane of lens II and at the front focal plane of lens III. The holographic diffraction images of red, green and blue channels are incident on different liquid crystal layer regions of the color liquid crystal grating. The holographic diffraction images of three colors are modulated by secondary diffraction in different liquid crystal layers of the color liquid crystal grating, and the color diffraction images with identical intervals are reconstructed. Finally, the color holographic reconstructed image with large viewing angle can be received by the camera. Design of the color liquid crystal grating In order to make the three colors of light passing through the color liquid crystal grating have the same diffraction angle, we design and fabricate a liquid crystal grating with three different-pitch regions in one liquid crystal cell for different incident wavelengths, namely region I, region II and region III, as shown in Fig. 2a . The three regions of the color liquid crystal grating are controlled by the same voltage. The color liquid crystal grating consists of a top substrate, liquid crystal layer, pixel electrodes, common electrodes and a bottom substrate. The pixel and common electrodes are etched by planar indium tin oxide electrode. The common electrode widths of the three regions are w r , w g and w b , respectively. In each region, the width of the pixel electrode is the same as that of the common electrode. The electrode arrangement is optimized by using a separate periodic electrode arrangement to eliminate the fringe electric field and diffraction crosstalk among the three regions. The pitches of the color liquid crystal grating in region I, region II and region III are d r , d g and d b , respectively. The gaps between the common electrode and pixel electrode in region I, region II and region III are l r , l g and l b , respectively. The same voltage is applied to the pixel electrodes in three regions, and the common electrodes in all regions are grounded. With this design, the arrangement periods of liquid crystal molecules in regions I, II and III of the color liquid crystal grating are different from each other when the voltage is applied, so the pitches of different regions are different accordingly. The pitch of region I is the largest, followed by region II and region III is the smallest. In addition, the adjacent electrodes between different regions are all set as common electrodes, thus the fringe electric field of adjacent regions can be eliminated and the angle disturbance of the liquid crystal molecules in adjacent regions decreased by 80%. Therefore, the control accuracy of different regions of the color liquid crystal grating can be improved. Chromatic aberration-free hologram generation mechanism for color holographic 3D display with large viewing angle The generation mechanism of a hologram without chromatic aberration is as follows. Firstly, the information of red, green and blue color channels of a 3D object is extracted. Then, the signal controller is used to generate the hologram of each color channel, and the corresponding blazed grating is superimposed on the hologram. The resolution of hologram and blazed grating is the same as that of SLM, which is a × b . Where a is the lateral resolution and b is the vertical resolution. The parameters of blazed grating satisfy the following equations: where d is the pitch of the blazed grating, γ is the blazed angle of the blazed grating, λ is the wavelength, p is the pixel interval of the SLM, and n is the number of pixels in a single period of the blazed grating. γ is expressed by Eq. ( 3 ): where φ is the phase change of the blazed grating in each period, and 0 < φ ≤ 2π. When the blazed grating is superimposed on the hologram and the collimated light irradiates the hologram, the diffracted light field of the hologram has a certain deflection angle θ = 2 γ . The red channel hologram and blazed grating I, the green channel hologram and blazed grating II, and the blue channel hologram and blazed grating III are superimposed respectively, and then loaded on the SLM in time sequence. At time T 1 , only shutter I is opened. At this time, the SLM is irradiated by the red light, and the red hologram and blazed grating I are loaded on the SLM. At time T 2 , only shutter II is opened. At this time, the SLM is irradiated by the green light, and the green hologram and blazed grating II are loaded on the SLM. At time T 3 , only shutter III is opened. At this time, the SLM is irradiated by the blue light, and the blue hologram and blazed grating III are loaded on the SLM. According to the visual persistence effect of human eyes, three color reconstructed images can be seen at the same time. The holographic diffraction images are diffracted and modulated after passing through the color liquid crystal grating, and second-order diffraction images will be generated. The wavelengths of the red light, green light and blue light are recorded as λ r , λ g and λ b respectively, and the holographic diffraction light fields of red, green and blue color channels respectively irradiate different liquid crystal layer regions of the color liquid crystal grating after passing through lens II. As shown in Fig. 2b , at time T 1 , the deflection angle of the holographic diffracted light field of the red channel is θ r . The holographic diffracted light field of the red channel passes through lens II with the focal length f , and converges to a point on the focal plane of lens II. Then, the light field is modulated by region I to generate M red second-order diffracted images. At this time, the interval L r between two adjacent red second-order diffraction images and the distance H r from the convergence point on the focal plane of lens II to the optical axis can be expressed as follows: At time T 2 , the deflection angle of the holographic diffracted light field of the green channel is θ g . The holographic diffracted light field of the green channel passes through lens II with the focal length f , and converges to a point on the focal plane of lens II. Then, the light field is modulated by region II to generate M green second-order diffracted images. At this time, the interval L g between two adjacent green second-order diffraction images and the distance H g from the convergence point on the focal plane of lens II to the optical axis can be expressed as follows: At time T 3 , the deflection angle of the holographic diffracted light field of the blue channel is θ b . The holographic diffracted light field of the blue channel passes through lens II with the focal length f , and converges to a point on the focal plane of lens II. Then, the light field is modulated by region III to generate M blue second-order diffracted images. At this time, the interval L b between two adjacent blue second-order diffraction images and the distance H b from the convergence point on the focal plane of lens II to the optical axis can be expressed as follows: We adjust the deflection angles θ r , θ g and θ b of the holographic diffraction light fields for the red, green and blue color channels. When H r - H g = L r , H g - H b = L g , and H r - H b = 2 L b , ( M -2) color second-order diffraction images can be received by the camera. At this time, the viewing angle of the color holographic reconstructed image is ( M -2) β , where β is color holographic 3D display angle when the color liquid crystal grating is in the voltage-off state. In order to obtain crosstalk-free color reconstructed images, the width of region II should be <2 L g . Reconstruction process In the experimental system, the wavelengths of the red, green and blue lasers are 638 nm, 520 nm and 450 nm, respectively. The SLM is a reflective phase-only SLM and its pixel size is 3.6 μm. The resolution and the refresh rate are 3840 × 2160 and 180 Hz, respectively. The phase modulation capability of the SLM can reach to 2π. The focal lengths of lens I, lens II and lens III are all 40 cm. The signal controller synchronously controls the on/off state of the three shutters and the generation and loading of holograms on the SLM. The signal controller is used to generate the holograms of red, green and blue channels and the corresponding blazed gratings (Supplementary Material S1 ). The holograms and blazed gratings are loaded on the SLM in time sequence. The resolutions of the holograms and blazed gratings are both 3840 × 2160. Simulation and experiments of the color liquid crystal grating In the color liquid crystal grating, the electrode widths of regions I, II and III are 13 μm, 10 μm and 9 μm respectively. The pitches of regions I, II and III are 26 μm, 20 μm and 18 μm respectively. Figure 3a–c show the electrode structure and arrangement distribution in different regions of the color liquid crystal grating under the microscope in the voltage-off state. When a voltage is applied to the pixel electrode, a spatially uneven gradient electric field distribution is formed between the pixel electrode and the common electrode, which induces the liquid crystal molecules to form a parabolic phase distribution, resulting in a light splitting effect similar to that of a phase grating. Figure 3d–f show the electrode structure and arrangement distribution in different regions of the color liquid crystal grating in the voltage-on state. As shown in Fig. 3d–f , in the voltage-on state, due to different pitches of each region of the liquid crystal grating, the three regions actually form three liquid crystal gratings with different periods. Besides, with the decrease of the pitch of the liquid crystal grating, the number of bright lines in the viewing area gradually increases accordingly. Moreover, the bright lines are evenly distributed and continuous. Figure 3g shows the simulated effective refractive index distributions of the liquid crystal layer in different regions of the color liquid crystal grating. For ease of understanding, we simulated two periodic distributions for each region. As shown in Fig. 3g , in the voltage-on state, the effective refractive index distributions of the liquid crystal layer in the different regions are all axisymmetric, resulting in good diffraction effect. The periodic peak between adjacent grating regions is caused by the direction change of the liquid crystal molecules above the electrode. This periodical peak is very small and does not affect the holographic reconstructed results. The switching time is also a very important performance parameter of the color liquid crystal grating. The color liquid crystal grating is placed in the cross polarizers, and the switching time is tested by measuring the change of transmittance in the voltage on/off state. The rubbing direction of the liquid crystal molecules is set to 0°, and the directions of the polarizer and the analyzer are −45° and +45° respectively. As shown in Fig. 3h , in the voltage-on state, the time for the liquid crystal molecules to form the color liquid crystal grating is 20.2 ms. Due to the anchoring effect of the alignment layer, the recovery time after removing the voltage is 14.3 ms. The distribution of diffracted light when red, green, and blue lasers irradiate the traditional liquid crystal gratings is shown in Fig. 4a–c . The traditional liquid crystal gratings have the same pitch. It can be seen that at this time, the distance of the adjacent diffraction orders between the red diffracted light is larger than that of the green diffracted light, while the distance between the blue diffracted light is the smallest. That is to say, when the traditional liquid crystal grating is used for secondary diffraction, chromatic aberration exists. Then, the designed color liquid crystal grating is used for diffraction, as shown in Fig. 4d–f . When the red, green and blue lasers respectively irradiate the proposed color liquid crystal grating, the distances between the red, green and blue diffracted lights are the same, and there is no chromatic aberration at this case (Supplementary Material S2 ). Experiment results of the proposed system with large viewing angle The holographic diffraction light fields of three channels generated by the SLM pass through lens II and then enter the corresponding regions of the color liquid crystal grating. The holographic diffraction light fields of the three channels of red, green and blue are diffracted by the second order respectively, so that the interval between the second-order diffraction images of each color channel is 1 cm. In order to ensure that the spectral information of red, green and blue diffraction images can pass through the corresponding regions of color liquid crystal grating, we superimpose different blazed gratings on the holograms. The green reconstructed image (the resolutions of the object is 950 × 1350) is taken as an example, when blazed gratings with different blazed angles are superimposed on the hologram (Supplementary Material S3 ), the reconstructed image moves accordingly, as shown in Fig. 4g–l . By controlling the incident angle of the light beam entering the color liquid crystal grating, the holographic diffracted light fields of the three channels of red, green and blue are modulated, respectively, and nine second-order diffracted images with equal spacing are generated. The diffraction order is influenced by the performance of the color liquid crystal grating itself. This has a high requirement for machining accuracy. In the experiment, we hope that the intensity of each second-order diffraction image is consistent. In the final experiment, nine uniform diffraction orders are realized. The maximum viewing angle is affected by the wavelength. The larger the wavelength is, the larger the viewing angle of the holographic display is. When only the red holographic 3D image is displayed, the viewing angle can reach ~91.5°, which is the largest viewing angle as far as we know. So, seven completely coincident color second-order diffracted images are obtained, and the large viewing angle color holographic 3D display can be realized. In color reconstruction, the viewing angle of the color reconstructed image is influenced by the blue wavelength. The real reference “straightedge” is placed on the same depth plane as the reconstructed image, as shown in Fig. 5a–g , and the resolution of the 2D color object is 1260 × 1150. The camera moves horizontally from the leftmost side of the viewing plane to the right side. When the proposed system is used, nine secondary diffraction images of red, green and blue color channels appear. At this time, the viewing angle of the color holographic reconstructed image is ~50.12°. While the viewing angle of the color holographic reconstructed image without using the liquid crystal grating is only ~7.16°. In order to see the details of the reproduced image, an enlarged image of the reproduced image is placed in the lower right corner of each sub-image. In addition, the color reproduction of 3D object is verified by experiment. The “windmill” and “cartoon bear paw” in different depths are used as the 3D object and their depths are 5 cm and 15 cm respectively. The resolution of the 3D object is 2100 × 1000. The real reference “straightedge” is used as a reference. When the reproduction distance is 5 cm, with the movement of the camera, the color holographic 3D display effect of focused “windmill” can be seen at different positions (Supplementary Material S4 ), and the reproduced image has no color difference, as shown in Fig. 5h–n . When the reproduction distance is 15 cm, with the movement of the camera, we can see the color 3D display effect of large viewing angle holography when “cartoon bear paw” is in focus, as shown in Fig. 5 o-u. In the lower right corner of each sub-image is an enlarged view of the reproduced image. In the experiment, two depths are used for 3D demonstration. Theoretically, holographic 3D display technology can achieve continuous depth holographic reconstruction. However, because the camera has a certain depth of field, it is difficult to distinguish the positional relationship when the distance between objects in two adjacent depth planes is too small, and even there will be serious crosstalk. In addition, the motion video of the 3D object is taken. The video of the “cartoon bear paw” from appearing to disappearing when the “windmill” is rotating is recorded (see Supplementary video 1 and Supplementary video 2 ). At different moments of the moving 3D object, the camera is used to capture the result when “windmill” is focused, as shown in Fig. 6a–c . Figure 6d–f shows the movement results at different moments when “cartoon bear paw” is focused. In the holographic reconstruction, the high coherence of laser will introduce speckle noise. Besides, the optical elements used in the holographic system may introduce additional phase difference, thus introducing noise. In addition, the random phase introduced by the hologram in the encoding calculation process will also cause speckle noise. There are many methods to suppress the speckle noise, including light source optimization method, iterative algorithm, deep learning algorithm and error diffusion method. In this paper, we optimize the phase distribution of the hologram, thus reducing the interference between adjacent pixels and effectively suppressing the noise of the reconstructed image. Speckle noise suppression is also an important direction in holographic 3D display, and we will continue to study it in the future. In the aspect of system integration, we can consider integrating RGB light source and replacing 4 f system with holographic optical elements, thus reducing the volume of the whole system.

Discussion We have first demonstrated an easy-to-implement method to realize color holographic 3D display system with large viewing angle. To this end, a specially structured color liquid crystal grating is designed on demand and a novel chromatic aberration-free hologram generation mechanism is put forward. In this case, chromatic aberration-free regulation of holographic reconstructed images with different wavelengths can be realized, thereby achieving large-viewing-angle and color holographic 3D display. The proposed holographic 3D display system enlarges the viewing angle of color holography to ~50.12°, which solves the problems of small viewing angle and serious chromatic aberration in the traditional holographic 3D display. The proposed mechanism can enrich the holographic 3D display theory. The proposed system has a simple structure and is expected to be applied in medical, industrial and other fields.

Holographic 3D display is highly desirable for numerous applications ranging from medical treatments to military affairs. However, it is challenging to simultaneously achieve large viewing angle and high-fidelity color reconstruction due to the intractable constraints of existing technology. Here, we conceptually propose and experimentally demonstrate a simple and feasible pathway of using a well-designed color liquid crystal grating to overcome the inevitable chromatic aberration and enlarge the holographic viewing angle, thus enabling large-viewing-angle and color holographic 3D display. The use of color liquid crystal grating allows performing secondary diffraction modulation on red, green and blue reproduced images simultaneously and extending the viewing angle in the holographic 3D display system. In principle, a chromatic aberration-free hologram generation mechanism in combination with the color liquid crystal grating is proposed to pave the way for on such a superior holographic 3D display. The proposed system shows a color viewing angle of ~50.12°, which is about 7 times that of the traditional system with a single spatial light modulator. This work presents a paradigm for achieving desirable holographic 3D display, and is expected to provide a new way for the wide application of holographic display. A color liquid crystal grating based color holographic 3D display system is proposed with large viewing angle. Subject terms

Supplementary information

Supplementary information The online version contains supplementary material available at 10.1038/s41377-023-01375-0. Acknowledgements This work is supported by the National Key Research and Development Program of China (2021YFB2802100) and the National Natural Science Foundation of China (62020106010, 62275009, and U22A2079). We would like to thank Center for Micro-Nano Innovation (Beihang Nano) for technique consultation. Author contributions D.W., S.D.L., and Q.H.W. conceived the project. Y.L.L., F.C., and D.W. designed the system, performed the simulations and conducted the experiments; N.N.L., Y.L.L., D.W., C.L., Z.S.L. and Z.Q.N. designed the color liquid crystal grating and analyzed the data; All authors discussed the results and commented on the paper. Conflict of interest The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:59

Light Sci Appl. 2024 Jan 15; 13:16

oa_package/1a/51/PMC10788332.tar.gz

PMC10788333

38221523

Subject terms

Relapse remains common in adults with acute myeloid leukemia (AML), with long-term disease-free survival estimated at 30–40% for patients 60 years or younger treated with chemotherapy [ 1 ]. Pretreatment factors, including age, antecedent hematologic disorder (AHD), prior exposure to chemotherapy, karyotype, and molecular mutations, have been used for the prediction of post-treatment outcomes [ 2 , 3 ]. It has been previously shown that AML patients who achieve initial complete remission (CR) are unlikely to relapse after three years of ongoing CR [ 4 ]. In an analysis of over 1000 patients in the first CR, treatment failure was found to be closely related to adverse cytogenetics and older age. Since this prior work, the assessment of treatment response now includes additional factors, including the type of remission (CR vs. CR with incomplete count recovery [CRi]), time to count recovery (if achieved), and presence of measurable residual disease (MRD) [ 5 – 7 ]. We set out to examine the impact of quality of remission, time to count recovery, and presence of MRD in addition to previously identified factors on relapse or death among patients with AML or other high-grade myeloid neoplasms up to 3 years after initial CR. Using our institutional database, we retrospectively identified 972 adults who had a confirmed diagnosis of AML or other high-grade myeloid neoplasm (≥10% blasts in blood or marrow) and underwent initial induction treatment at the University of Washington (UW)/Fred Hutchinson Cancer Center between November 2008 and December 2018. The data cutoff was April 5, 2022. This study was approved by the UW Institutional Review Board. Disease status was defined as de novo vs. secondary, with secondary including AHD (antecedent myelodysplastic syndrome or myeloproliferative neoplasm) or therapy-related. Induction chemotherapy was divided into three groups: high intensity included multiagent chemotherapy with cytarabine at ≥1 g/m 2 /dose/day (such as CLAG-M [ 8 ], FLAG-Ida, or similar); intermediate intensity included 7 + 3 or similar; and low intensity was defined as a hypomethylating agent with or without venetoclax. MRD was quantified using multiparameter flow cytometry and detected once a patient achieved morphologic CR or CRi with <5% blasts in the marrow; MRD was measured at a single time point, usually around day 30 after induction, and was not reassessed later. MRD assessment was performed on marrow using multiparameter flow cytometry with a minimum sensitivity of 0.1% [ 9 ]. Relapse was considered to have occurred if the bone marrow contained ≥5% blasts or if circulating peripheral blasts were identified. The primary outcomes were binary endpoints of survival without relapse (relapse-free survival, RFS) at 1, 2, or 3 years after the achievement of morphologic CR following initial therapy. Patients without relapse and censored before the landmark dates were excluded from the respective analyses. Multivariable logistic regression models were used to evaluate covariate associations with these outcomes; covariates evaluated were: age (assessed as a continuous variable), ELN 2017 risk classification, presence of MRD, gender, WBC count at diagnosis (assessed as a continuous variable), secondary disease, ECOG performance status (0–1 vs. 2–4), treatment-related mortality (TRM) score [ 10 ], induction intensity, CR vs. CRi, early blood count recovery (defined as absolute neutrophil count (ANC) ≥ 1000/μL and platelet count ≥ 100,000/μL within 30 days), and transplantation status. The Kaplan-Meier method was used to estimate RFS and overall survival. All analyses were performed using R version 4.2.2. In all, 656 patients achieved a morphologic remission (CR or CRi) defined by European LeukemiaNet (ELN) 2017 criteria [ 11 ] and were included in our analysis. The characteristics of the 656-patient cohort are summarized in Table 1 . The median age was 60, with a range of 18–91 years. Three hundred seventy-three patients (57%) had de novo AML, and 283 (43%) were defined as secondary due to prior chemotherapy and/or AHD. Pretreatment molecular and cytogenetic information was available for all but 6 patients, and 72% were classified as intermediate or adverse by ELN 2017 criteria. For patients that did not have all data points to classify risk according to ELN, clinical assessment at the time of diagnosis, as well as available data (e.g., molecular markers, cytogenetics/FISH), were used. The majority of patients received high- or intermediate-intensity induction (88%), and most (82%) patients had an ECOG performance status of 0–1 at diagnosis. Five hundred forty (82%) patients achieved a CR (<5% marrow blasts with peripheral count recovery). MRD was identified in 173 patients (26%). Two hundred and seventy patients (41%) received subsequent allogeneic hematopoietic cell transplant (HCT). Median follow-up among all patients was 5.2 years (range 1 month to 13.2 years). Older age was significantly associated with decreased RFS and was significant at years 2 and 3 of the landmark analyses [year 2 OR 1.19, 95% CI 1.03–1.37; year 3 OR 1.17, 95% CI 1.01–1.36; Table 2 ). ELN 2017 intermediate (year 1 OR 3.77, 95% CI 2.22–6.4; year 2 OR 3.49, 95% CI 2.05–5.94; year 3 OR 3.43, 95% CI 1.95–6.04) and adverse (year 1 OR 5, 95% CI 2.76–9.03; year 2 OR 3.63, 95% CI 2–6.58; year 3 OR 3.23, 95% CI 1.73–6.06) risk groups were significantly associated with increased risk of relapse or death and remained significant at each of the three landmark analyses. Certain patient characteristics were associated with significantly higher odds of relapse or death by years 1 and 2, including elevated WBC (year 1 OR 2.84, 95% CI 1.41–5.74, year 2 OR 2.65, 95% CI 1.27–5.54) and those with higher PS (year 1 OR 2.99, 95% CI 1.59–5.63; year 2 OR 1.95, 95% CI 1.02–3.76). However, these factors lost significance in the year 3 analysis. Of all patients that relapsed or died in the first year, 83% (238) were intermediate or adverse risk using ELN 2017 criteria. In the models summarized in Table 2 , treatment response (CR with MRD vs. CR without MRD) had an association for relapse or death at each time point tested (year 1 OR 6.99, 95% CI 4.04–12.1; year 2 OR 7.87, 95% CI 4.34–14.28; year 3 OR 7.91, 95% CI 4.14–15.13). MRD status was also significantly associated with RFS and OS (RFS: MRDneg 47%, 95% CI 43–52%, MRDpos 19%, 95% CI 14–26%; OS: MRDneg 56%, 95% CI 52–61%, MRDpos 28%, 95% CI 22–36%). Other covariates had an initial significant OR for relapse or death at years 1 and 2 but this finding was lost over time. For example, elevated WBC carried a significant risk for relapse or death through years 1 and 2 but lost significance at year 3. Additionally, higher performance status (ECOG 2–4) conferred an increased risk of relapse or death at years 1 and 2 but was not significant at year 3. The evidence of an association between time to count recovery and outcome was less strong (year 1 OR 1.49, 95% CI 1–2.23; year 2 OR 1.32, 95% CI 0.89–1.97; year 3 OR 1.35 95% CI 0.89–2.03). Treatment intensity was not a significant predictor of RFS in multivariable models at any timepoint. Receipt of HCT was significantly associated with a decreased risk of RFS in all models. In our analysis of 656 patients, the strongest and most consistent predictors of RFS were treatment response (MRD vs. no MRD for patients in CR) and ELN 2017 risk since they remained significant at all three time points examined (year 1, year 2, and year 3). Increased age and incomplete count recovery were also associated with relapse or death, though not significant at every time point. Similarly, other factors, including ECOG performance status and high WBC count at diagnosis, were associated with relapse or death in years 1 and 2 but not year 3. These results stress the importance of post-therapy data (e.g., MRD status) in prognostication of outcomes after completing initial therapy. Therapy intensity was not associated with RFS in multivariable models, though notably, the majority of patients received high (64%) or intermediate (24%) intensity induction. Recent data published by Bazinet and co-authors suggest that the goal of any AML-directed treatment should be MRD-negative CR [ 12 ]. The findings in our study that the presence of MRD was significantly associated with RFS—but therapy intensity was not—would also support this conclusion. Our strengths include the large number of patients included in our study with substantial follow-up, which allowed us to evaluate the impact of a number of covariates in multivariable analyses. Furthermore, our population was heterogeneous, with various treatment regimens reflecting the current treatment paradigm of the disease. Our study was limited in that it is retrospective in nature and performed at a single center. The major goal of our study was to evaluate the prognostic importance of several covariates for the outcomes of relapse and non-relapse death through years 1–3 following initial CR for patients with AML and other high-grade myeloid neoplasms. Based on our findings, patients who are older, MRD positive, intermediate or adverse cytogenetic risk, or demonstrate incomplete recovery of their peripheral counts have an ongoing increased risk of relapse or death in the first 3 years. Future studies should evaluate strategies to treat patients with MRD since prognosis is poor and effective treatment options are limited. Other unanswered questions include evaluating the ideal number of intensive consolidation cycles, the role of maintenance chemotherapy, and the benefit of allogeneic HCT, particularly in older patients.

Acknowledgements We would like to acknowledge the contributions of the late Dr. Eli Estey. It was through his guidance that this project was made possible. His kindness, curiosity, and generosity, along with his time and knowledge, left a lasting impact on the authors of this paper and will be greatly missed. This research was supported in part by Cancer Center Support Grant P30 CA015704 through the National Cancer Institute/National Institute of Health (NCI/NIH), Bethesda, MD, USA. Author contributions JJL, EHE, and MEMP designed the project and wrote the report. JJL and CMS performed data abstraction from the institutional database and from medical records as needed. MO performed all statistical analyses. KR, ABH, JSA, PH, and RBW provided critical input on the report. Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:59

Blood Cancer J. 2024 Jan 15; 14(1):5

oa_package/ad/3d/PMC10788333.tar.gz

PMC10788334

38221538

Introduction Lung function gradually decreases over time and varies greatly among individuals 1 . Individuals with impairment of lung function are more likely to suffer from chronic respiratory diseases and have an increased risk of all-cause death 2 . Therefore, it is of particular interest to identify biomarkers associated with this impairment 3 . Uric acid (UA) is the final decomposition product of purine degradation and is present in the plasma and epithelial lining fluid of the respiratory tract 4 , 5 . Previous studies have indicated that serum UA is associated with cardiovascular diseases, including hypertension, stroke, coronary heart disease, and congestive heart failure 6 – 9 . Similarly, there is also a variety of research investigating the relationship between serum UA and respiratory disease, including chronic obstructive pulmonary disease (COPD), pulmonary hypertension and obstructive sleep apnea. Significant correlation has been found 10 – 12 . Serum UA has been reported to be linked to lung function, but controversial results have been observed in an epidemiological setting: a few studies have shown a negative association between serum UA and lung function in the general healthy population 13 – 15 . Meanwhile, only in the female population, the negative association between serum UA and lung function was found in the study by Jeong et al. 16 . Nevertheless, a positive relationship between serum UA and lung function was observed in a healthy Korean population 17 . Consequently, this study aimed to investigate the association between serum UA and lung function in the US population using data from the National Health and Nutrition Examination Survey (NHANES 2007-2012).

Methods Population research Data was obtained from the National Health and Nutrition Examination Survey (NHANES) (2007–2012). NHANES is a complex, stratified, multistage probability sample survey of the non-institutionalized US population. These cross-sectional surveys are used to assess the severity and prevalence of various diseases, as well as to explore potential new directions for medical research and public health policies. Inclusion and exclusion criteria The population included participants with complete data on serum UA and lung function (specifically forced expiratory volume in one second (FEV1) and forced vital capacity (FVC)). The total number of participants from NHANES (2007–2012) was 30,442. After excluding subjects under 19 years old (n = 1960), those with missing FEV1 (n = 10,392), baseline FEV1 Quality Attribute not A or B (n = 1841), missing serum UA (n = 3776), gout (n = 152), asthma (n = 1190), FEV1/FVC < 0.7 (n = 1813), pregnant women (n = 82), coronary heart disease (n = 185), liver disease (n = 286), tumor (n = 497), kidney disease (n = 119), and diabetes (n = 635), a total of 7514 subjects remained for the final analysis (Fig. 1 ). Ethical approval Participants aged ≥ 18 years furnished informed consent on their own. The NCHS Ethics Review Board approved the conduct of NHANES, and written informed consent was obtained from all participants. Study variables The principal variables of this study were lung function (dependent variable) and serum UA (independent variable). Serum UA was measured using a Beckman Synchron LX20 (Beckman Coulter, Inc., Brea, CA). Lung function was measured using Ohio 822/827 dry-rolling seal volume spirometers. The following covariates were included: age, gender, race, income-poverty ratio, body mass index, alcohol drinking, smoke, systolic blood pressure, diastolic blood pressure, blood urea nitrogen, creatinine, calcium, total cholesterol, total protein, fractional exhaled nitric oxide (FeNO) and total bilirubin. Details regarding the measurement process of serum UA and lung function, as well as the acquisition process for other covariates, are available at www.cdc.gov/nchs/nhanes . Statistical analyses NHANES sample weights were taken into account when calculating all estimates. Continuous variables are presented as mean ± standard deviation (SD) or medians (25th percentile–75th percentile), while categorical variables are presented as frequency (%). This research utilized chi-square tests for categorical variables and linear regression models for continuous variables to assess significant differences in these variables. Dummy variables were used to indicate missing covariate values. After adjusting for confounders, multivariable linear regression models were constructed to determine the independent relationship between serum UA and lung function. Generalized additive models and smooth curve fittings were employed to evaluate any non-linear relationship between serum UA and lung function. Stratified and interaction analyses were performed to assess whether covariates influenced the association between serum UA and lung function, ensuring the robustness of data analysis. A 2-tailed P < 0.05 was considered statistically significant for all analyses. Data analysis was conducted using the statistical software packages R ( http://www.R-project.org ) and EmpowerStats ( http://www.empowerstats.com ).

Results Baseline characteristics of selected participants A total of 7514 participants were included in the final analysis. The mean age was 40.85 ± 14.28 years old. The mean serum UA levels were 5.34 ± 1.34 mg/dl. Table 1 presents the weighted baseline characteristics according to the serum UA tertiles. All variables showed statistically significant differences among the different serum uric acid groups. Except for the income-poverty ratio, race, and female gender, all other variables exhibited an increasing trend according to the tertiles of serum uric acid (Table 1 ). Association between serum UA and lung function (FEV1 and FVC) in the total population The relationship between serum UA and lung function was explored using a generalized additive model. The adjusted smoothed plots indicated a linear relationship between serum UA and lung function both in the total population and male or female populations (Fig. 2 ). Three models were constructed to analyze the independent role of serum UA in FEV1 and FVC in different populations: the general population, the male population, and the female population. Model 1 involved no modification variables, Model 2 included adjustments for gender or age and race, and Model 3 incorporated adjustments for the covariates presented in Table 1 (Tables 2 and 3 ). In the general population, Model 1 revealed a positive association between serum UA and FEV1 and FVC, whereas Models 2 and 3 identified a negative association between serum UA and FEV1 and FVC. Model 3 showed a significant negative association between serum UA and FEV1 (β = − 24.77; 95% CI − 36.11 to − 13.43) and FVC (β = − 32.93; 95% CI − 47.42 to − 18.45) (Table 2 ). Each 1 mg/dl increase in serum UA was associated with a 24.77 ml decline in FEV1 and a 32.93 ml decline in FVC. Sensitivity analysis using tertiles of serum UA as a categorical variable yielded similar results, with negative effect sizes for tertiles 2 and 3 in Models 2 and 3. The p -values for trend in all models were significant (all p < 0.05) (Table 2 ). In both the male and female populations, except for a small magnitude positive β value in Model 1 for FVC in males (with a p -value > 0.05), all other models showed negative effect sizes. Furthermore, all p-values in Model 3 were significant, indicating a negative relationship between serum UA and lung function in both male and female populations. When serum UA was categorized into tertiles as a categorical variable, the results remained consistent with the findings obtained when serum UA was treated as a continuous variable (Table 3 ). Subgroup analyses The role of other covariates on the association between serum UA and FEV1 and FVC was further examined in the male and female populations. The association between serum UA and FEV1 and FVC remained consistent and robust in both male and female populations across various subgroups, including age, race, BMI, alcohol drinking, smoke, and income-poverty ratio. Furthermore, all effect sizes of these subgroups were negative, and there was no significant p-interaction (Figs. S1 , S2 in the Supplementary Appendix).

Discussion The primary objective of this study was to investigate the independent association between serum UA and lung function in healthy US adults. Our findings revealed a negative association between serum UA and lung function (FEV1 and FVC) after adjusting for potential confounding factors. This relationship was consistent both in males and females. Further research is required to explore whether there exists a causal, pathophysiological mechanism linking serum UA and lung function, as well as to determine the potential utility of serum UA as a biomarker for identifying impaired lung function. The positive relationship between serum UA and lung function was observed when no covariates were adjusted. However, after only adjusting for gender, the association changed to a negative one. Moreover, even after adjusting for other covariates, the negative association between serum UA and lung function remained. The observational analysis found a correlation between high plasma urate levels and low lung function. Furthermore, our study found a robust negative relationship between serum UA and lung function across various subgroups, including age, race, BMI, smoke, and alcohol consumption, in the overall study population. This negative relationship was also observed within the male and female subgroups mentioned above. Possible explanations for the negative association between serum UA and lung function can be identified as follows. Firstly, the relationship might be attributed to reverse causation. Previous research has indicated that serum UA levels increase in hypoxic conditions, such as chronic heart failure and COPD 18 . It has also been suggested that pulmonary hypoxia triggers the breakdown of purines, resulting in the elevated production of serum UA 19 . However, it remains uncertain whether the mild hypoxia observed in the general healthy population could impact serum UA levels, as our study excluded individuals with evident clinical conditions such as asthma, airflow limitation, and coronary heart disease. Secondly, the formation of uric acid necessitates the catalytic activity of xanthine oxidase, a process that occurs in the epithelial lining fluid of the respiratory tract and is accompanied by the generation of superoxide 20 . As superoxide is a free radical, it has the potential to induce oxidative damage to biological molecules. Therefore, it can be speculated that lung tissue damage may occur in the presence of high xanthine oxidase activity due to oxidative stress. Our study makes a valuable contribution to the existing literature by demonstrating a consistent negative association between serum uric acid (UA) and lung function in both male and female individuals in the US general healthy population. Previous studies have attempted to assess the link between serum UA and lung function in the general population, but the results have been inconsistent. For instance, Hong et al. found a negative relationship between hyperuricemia and FEV1% or FVC% in a sample of 2901 participants from the Korean general population 15 . Similar findings were reported by Aida et al. and Kobylecki et al. 13 , 14 , which align with our own findings. However, Song et al. reported a potential positive effect of serum UA on lung function in middle-aged, healthy populations 17 . We speculate that the discrepancy in results could be attributed to several factors: First, participant exclusion criteria differed between the studies. Song et al. excluded individuals with chronic lung disease or abnormal chest radiograph findings, but the definition of these diseases was vague, and details were not provided. As such, it is challenging to determine the influence of these excluded participants on the overall conclusion. In contrast, our study excluded participants with asthma and FEV1/FVC < 0.7, and in these individuals, a negative relationship between serum UA and lung function was observed 21 , 22 . Thus, our conclusion is more robust, given the exclusion of these participants. Second, Song et al. did not assess the adjusted non-linear relationship between serum UA and lung function, nor did they conduct subgroup analyses as a sensitivity analysis for their conclusion. In our study, all effect sizes of subgroups were found to be negative, indicating the robustness of our findings. Third, the study populations differed, with our study focusing on US individuals while Song et al. targeted individuals from South Korea. Fourth, compared to our research, Song et al. did not consider the effects of FeNO, total bilirubin, and income-poverty ratio when adjusting for covariates in the relationship between serum UA and lung function. However, previous studies have confirmed that these variables are associated with lung function 23 – 25 . Kobylecki et al. 14 conducted a Mendelian randomisation study to investigate the causal relationship between serum uric acid and lung function. Their observational analysis found a correlation between high plasma urate levels and worse lung function. However, genetically high plasma urate levels did not show a direct causal association with the outcomes. Likely potential explanations include reverse causation and unmeasured confounding factors. It is important to note that the study population mainly consisted of individuals of Danish descent and had specific inclusion and exclusion criteria that significantly differed from our study, as the United States is a multicultural country with diverse ethnicities. Further research is still required to assess the causal relationship between serum UA and lung function. The study possesses several strengths. Firstly, a nonlinearity assessment was incorporated, providing a more comprehensive understanding of the relationship between serum uric acid (UA) and lung function. Secondly, additional confounding variables such as Fractional exhaled nitric oxide (FeNO), total bilirubin, and income-poverty ratio were adjusted for. Thirdly, various sensitivity and subgroup analyses were performed, and the findings consistently upheld their robustness. Lastly, the study had a relatively large sample size in comparison to previous studies with similar objectives. The study has several limitations. Firstly, due to its cross-sectional design, a causal relationship between serum uric acid (UA) and lung function cannot be established. Secondly, the use of uric acid-lowering medications was not accounted for, although individuals with gout were excluded. Lastly, the study population was derived from NHANES (2007-2012), and certain exclusion criteria were applied, potentially limiting the generalizability and extrapolation of the findings. In conclusion, serum uric acid is negatively associated with FEV1 and FVC in the US general healthy population. This negative relationship is significant in both the male and female populations. These outcomes emphasize the significance of serum uric acid as a potential mechanism underlying FEV1 and FVC decline. Further epidemiologic studies will still be required to confirm this reverse association.

The relationship between serum uric acid and lung function has been controversial. This study aims to determine whether there is an independent relationship between serum uric acid and lung function in the National Health and Nutrition Examination Survey (NHANES) from 2007 to 2012. Serum uric acid was considered the exposure variable, and lung function (FEV1 and FVC) was the outcome variable. Multivariable linear regression was conducted with adjustments for potential confounders. The total number of participants from NHANES (2007–2012) was 30,442, of which 7514 were included in our analysis after applying exclusion criteria. We observed that serum uric acid was negatively associated with FEV1 and FVC after adjusting for confounders (β for FEV1 [− 24.77 (− 36.11, − 13.43)] and FVC [− 32.93 (− 47.42, − 18.45)]). Similarly, serum uric acid showed a negative correlation with FEV1 and FVC after adjusting for confounding variables both in male and female populations. The relationship between serum uric acid and FEV1 and FVC remained consistent and robust in various subgroups within both male and female populations, including age, race, BMI, alcohol consumption, smoking status, and income-poverty ratio. Serum uric acid is negatively associated with FEV1 and FVC in the US general healthy population. This negative relationship is significant in both the male and female populations. Subject terms

Supplementary Information

Supplementary Information The online version contains supplementary material available at 10.1038/s41598-024-51808-y. Author contributions The authors’ responsibilities were as follows—W.L., C.W., W.Y.W., Y.H.L., D.H.W.: contributed to data collection, analysis, and manuscript writing; Y.H.L., W.L., X.Y.Y., F.L.: devoted to research design and implementation. All authors declare that they have read and approved the manuscript and final version. Funding This study was supported by the Xiamen Health Care Guideline Project (3502Z20214ZD1005). The funder had no role in the design of the study, the collection and analysis of the data, or the preparation of the manuscript. Data availability Data described in the manuscript will not be made available because the data used in this study were from the NHANES database, which is a free and open database for all researchers around the world. The link to the database is https://wwwn.cdc.gov/nchs/nhanes/Default.aspx . Competing interests The authors declare no competing interests.

CC BY

no

2024-01-16 23:41:59

Sci Rep. 2024 Jan 14; 14:1300

oa_package/9d/88/PMC10788334.tar.gz