Datasets:

suolyer

/

pile_pubmed-central

like 1

Introduction {#sec1} ============ Cutaneous leishmaniasis (CL) is the disease caused by *Leishmania* spp., an intracellular parasite. More than 70% of all CL cases are seen in 10 countries worldwide, which are Afghanistan, Algeria, Brazil, Colombia, Costa Rica, Ethiopia, the Islamic Republic of Iran, Peru, Sudan, and the Syrian Arab Republic \[[@cit0001]\]. It was reported that the causative agents of the disease are *L. tropica* and *L. major* in Syria \[[@cit0002], [@cit0003]\]. Turkey is one of the countries where CL is also prevalent \[[@cit0001]\]. Studies conducted in Turkey have shown that the predominant causative agent of CL is *L. tropica* \[[@cit0004]\]. Cutaneous leishmaniasis is endemic in Turkey, especially in the Southeast and the Mediterranean region \[[@cit0005]\]. However, in the last 5--6 years, many Syrian people have immigrated to Turkey due to the civil war in Syria. As a result, the incidence of the disease has increased considerably, especially in the areas of intense emigration \[[@cit0006], [@cit0007]\]. Hatay province of Turkey, located in the Mediterranean region, is one of the cities most affected by the migration due to its border with Syria. Cutaneous leishmaniasis is more common especially in the childhood and young age group (0--20 years of age) \[[@cit0008], [@cit0009]\]. Although the studies reported from Turkey previously described the clinical and demographic data of the disease, a relatively small number of studies have focused on this age group in detail \[[@cit0010]\]. Aim {#sec2} === The aim of this retrospective study was to evaluate the clinical and demographic data of Syrian immigrant and Turkish children diagnosed with CL, who were consulted in our Dermatology Outpatient Clinic, and to compare Syrian patients (SP) and Turkish patients (TP) with respect to age, gender, disease duration, and localization, number and type of lesions. So far, this is the first study in which pediatric CL patients from the Hatay region have been examined in detail. Material and methods {#sec3} ==================== In the present study, we included CL patients aged 0-18 years and recorded their clinical and demographic data. The patients were admitted to our outpatient clinic in the period 2015--2017 and in the first half of 2018. A total of 121 patients (55 (63.2%) male, 32 (36.8%) female SP; 19 (55.88%) male and 15 (44.1%) female TP) were included in the study. The diagnosis of CL was made by showing parasites in the smears taken from the lesions. Demographic and clinical data of patients were recorded from patient files. Statistical analysis was performed by dividing the patients into 0--6, 7--12, and 13--18 age groups. Statistical analysis {#sec3.1} -------------------- Data were analyzed using SPSS for Windows (version 21; IBM Corp, Armonk, New York, USA), χ^2^, Kruskal-Wallis, Mann-Whitney *U*, and Yates's continuity correction tests were used for statistical analysis. Values were presented as the mean ± standard deviation and *p* \< 0.05 was considered as significant. Results {#sec4} ======= Patients' ages ranged from 1 to 18 years (mean: 9.63 ±5.2). The mean ages of TP and SP were 12.06 ±4.47 (3--18 years) and 8.68 ±5.18 (1--18 years) respectively. Fifty percent of the TP were in the 13-18 age group, and 39% of the SP were in the 7--12 age group ([Table 1](#t0001){ref-type="table"}). There was no relationship between age groups and gender in all patients (*p* = 0.629). The SP and TP also did not show any relationship regarding gender and age group within themselves (SP *p* = 0.644, TP *p* = 0.460). ###### Distribution of TP and SP by age group and gender Age TP SP Total -------- ---- ---- ------ ---- ------- ----- ----------- 0--6 3 2 14.7 19 14 38 38 (31.4) 7--12 5 7 35.3 22 12 39 46 (38) 13--18 11 6 50 14 6 23 37 (30.6) Total 19 15 100 55 32 100 121 (100) The average disease duration was 5.15 ±3.57 months (range: 1 week--12 months) (definition of "disease duration": the time period between the appearance of the first lesion and the arrival of the patient at the hospital for examination). There was no significant difference between the age groups regarding disease duration (*p* = 0.943). The disease durations in SP and TP were 1 week-12 months (mean: 4.73 ±3.39 months) and 1--12 months (mean: 6.25 ±3.86 months) respectively (*p* = 0.049). In the same analysis, we did not include the disease durations of 6 patients because when the outlier analysis was performed, the disease duration was deflecting. The disease durations of these patients were in the range 30--60 months and 4 of them were SP while 2 of them were TP. The total number of lesions was 247 (*n* = 42 in TP, *n* = 205 in SP), and the mean number of lesions per patient was 2.04 ±2.01. There was one lesion in 61.2% of the patients and 2 lesions in 19.8% of the patients. When TP and SP were analyzed separately, 82.4% (*n* = 28) of the TP had one lesion, 11.8% (*n* = 4) had 2 lesions, and 5.9% (*n* = 2) had 3 lesions, while 52.9% (*n* = 46) of SP had one lesion, 23% (*n* = 20) had 2 lesions, and 23.9% (*n* = 21) had ≥ 3 lesions. Although there was no statistically significant difference between the age groups in terms of number of lesions (*p* = 0.207, Kruskal-Wallis test), the mean number of lesions was the highest in the 7--12 age group (2.54 ±2.43). When TP and SP were evaluated within themselves, mean lesion number was 2.35 ±2.28 (1--11 lesions) in SP, and 1.23 ±0.55 (1--3 lesions) in TP. Between two groups, the number of lesions per patient was significantly higher in SP (*p* = 0.002) compared to TP and two and multiple lesions were significantly higher in SP (*p* = 0.005). Regarding age group, when TP and SP were evaluated separately, the number of lesions in SP was the highest in the 7--12 age group (mean: 3.029, *p* = 0.049). This was followed by the 13--18 age group. There was no significant difference with respect to number of lesions in all three age groups of TP (*p* = 0.653; Bonferroni corrected Mann-Whitney *U* test) ([Table 2](#t0002){ref-type="table"}). ###### Distribution of number of lesions according to age groups Group Age groups Lesion number Mean ± SD *p*~1~ *p*~2~ --------------- ------------ --------------- ----------- -------- -------- SP 0.049 0.002 0--6 50 1.51 ±0.795 7--12 103 3.02 ±2.657 13--18 52 2.60 ±2.854 Total 205 2.35 ±2.282 TP 0.653 0--6 7 1.40 ±0.894 7--12 14 1.16 ±0.577 13--18 21 1.23 ±0.437 Total 42 1.23 ±0.553 Overall total 247 2.04 ±2.018 p1 -- number of lesions according to age groups in Syrian and Turkish patients, p2 -- evaluation of lesion numbers of Syrian and Turkish patients. It was found that the lesions were most frequently located in the head/neck (HN) region (*n* = 93, 76.9%) in the patients. 44.1% (*n* = 41) of the patients with HN localization were in the 7--12 age group. This was followed by the age groups 0--6 (32.3%, *n* = 30) and 13--18 (23.7%, *n* = 22). When the age groups of SP and TP were evaluated within themselves, HN localization was most frequently observed in the 7--12 age group in both groups (SP; *n* = 29, 42.6%; TP; *n* = 12, 48%). This was followed by the 0-6 age group in SP, and the 13--18 age group in TP. When the SP and TP were compared in terms of location of the lesions, it was found that the localization of HN was more frequent in TP than in SP (73.5% and 59.8% respectively, *p* = 0.05). We found that 57.1% of the HN lesions were of the papule type. In addition, 63.6% (*n* = 7) of the extremity-localized lesions were of the ulcer type. Clinical samples are shown in [Figure 1](#f0001){ref-type="fig"}. When SP and TP were evaluated separately, again the most common type of lesion was papule (with 52.9% and 55.9% respectively; for both *p* = 0.001). {#f0001} The cheek area lesions in both patient groups have the highest frequency (SP 19.5%, TP 19%). Diffuse infiltration was present in 3 of the patients with lip lesions. The localization and the number of the lesions in TP and SP are shown in [Figure 2](#f0002){ref-type="fig"} and the distribution of lesion types is shown in [Table 3](#t0003){ref-type="table"}. ###### Distribution of lesion types according to age group Group Age group Papule *n* (%) Nodule (%) Plaque (%) Ulcer (%) Other\* (%) -------- ------------ ---------------- ------------ ------------ ----------- ------------- TP 0--6 1 (14.28) 1 (14.29) 4 (57.14) 1 (14.29) -- 7--12 14 (93.33) -- 1 (6.67) -- -- 13--18 8 (40.00) 6 (30.00) 4 (20.00) 2 (10.00) -- SP 0--6 28 (56.00) 6 (12.00) 11 (22.00) 4 (8.00) 1 (2.00) 7--12 85 (76.58) 9 (8.11) 10 (9.01) 5 (4.50) 2 (1.80) 13--18 30 (68.18) 4 (9.09) 5 (11.36) 5 (11.36) -- Total 166 (67.21) 26 (10.53) 35 (14.17) 17 (6.88) 3 (1.21) {#f0002} Discussion {#sec5} ========== Although CL occurs at all ages, it is most common among children and teenagers \[[@cit0006], [@cit0010], [@cit0011]\]. In our study, we found that the mean age of TP was significantly higher than SP. This difference in our study may be due to the difference between the numbers of patients in both groups as well as the higher recognition of the disease among SP. The most frequently affected age group in our study was 7--12 in SP and 13--18 in TP. It has been previously reported that CL disease was more frequently observed between the ages of 11--15 \[[@cit0012]\] and 13--24 \[[@cit0013]\] in Hatay province where all age groups were included in these studies, which is also compatible with our study. Uzun *et al.* reported that the disease was most commonly seen in the 10--19 age group \[[@cit0008]\]. In another study comprising the 0--15 age group, the 6--10 age group was identified as the most infected group \[[@cit0010]\]. If we look at the results of other studies comprising pediatric age groups bearing CL in Iran and Tunisia, we see that 6--9 \[[@cit0014]\] and 6--15 age groups \[[@cit0015]\], respectively, were more frequently affected. Among our patients, gender was homogeneously distributed within all age groups. Some of the reports conducted in these age groups are consistent with our results \[[@cit0011]\]. However, Aksoy *et al.* reported that CL was more frequently present in females \[[@cit0010]\]. We found that the disease duration in our study was significantly shorter in SP than in TP (SP: 4.73 ±3.39 months, TP: 6.25 ±3.86 months). We did not find any difference in the disease duration among the age groups. However, Aksoy *et al.* reported that the 0--5 age group had the longest disease duration \[[@cit0010]\]. In our study, the average number of lesions per patient was 2.04 ±2.01. However, when the groups were compared, the mean number of lesions in SP was 2.35 ±2.28, while it was 1.23 ±0.55 in TP (*p* = 0.002). 82.4% of TP and 52.9% of SP had only one lesion and two and multiple lesions were significantly higher in SP (*p* = 0.005). Our results concerning TP were consistent with the results of Aksoy *et al.* and Gürel's *et al.* studies reported from Turkey \[[@cit0010], [@cit0011]\]. The results of the studies carried out in other countries involving the pediatric age group are as follows: a study in Pakistan reported that 75% of patients had a single lesion and the mean number of lesions per patient was 1.43. 16.6% of patients had two lesions and 8.3% of patients had \> 2 lesions \[[@cit0016]\]. In India, a single lesion (72.2%) was most frequently detected in the 0-5 age group of patients. The authors reported that 19.9% of patients had two lesions and 7.9% of patients had multiple lesions \[[@cit0017]\]. In another study from Iran, only one lesion was detected in 82.3% of patients, two lesions were detected in 12.4% of patients, and ≥ 3 lesions were detected in 5.3% of patients \[[@cit0018]\]. When the results of studies reported from other countries as well as Turkey were evaluated, the number of patients with multiple lesions was higher in SP (23% of patients had two lesions and 23.9% of patients had ≥ 3 lesions). This may be due to inadequate shelter, a crowded environment, and lack of hygiene caused by adverse conditions of migration. Similar to the results of other studies, it was determined that the HN site was the most common localization of CL in our study \[[@cit0010], [@cit0014], [@cit0015]\]. When assessed according to age groups, 44.1% (*n* = 41) of patients with HN localization were in the 7--12 age group. Aksoy *et al.* reported that HN localization was more frequently observed in the 0--5 age group \[[@cit0010]\]. In our study, the cheek area was the part where CL was most frequently localized in both SP and TP. In other studies conducted with the childhood age group, it has been reported that the cheek is the region most frequently affected by CL \[[@cit0014]--[@cit0016]\]. The cheek was proposed to be the area most exposed and the least protected from sandfly bites \[[@cit0016]\]. In our study, papules were the most common type of lesions and most frequently seen in the 7--12 age group. 57.1% of the HN lesions were of the papule type. Another interesting result of our study is that in 63.6% (*n* = 7) of patients with ulcer type the lesions occurred on extremities. When other studies were reviewed, it was observed that papules \[[@cit0014]\], plaques \[[@cit0016]\], plaques or papulonodules \[[@cit0017]\] or ulcers \[[@cit0010]\] were reported as the most common types of lesions. In a study conducted by Turan *et al*., the number of lesions was higher and disease duration was shorter in SP, similar to the results of our study \[[@cit0018]\]. In our study, we also found that the papular type of CL lesion was frequent in both TP and SP. Turan *et al*. reported that nodules were more common in SP whereas ulcers were more frequent in TP \[[@cit0019]\]. Another study from the same region of Turkey reported that ulcers were the more common type of CL lesions \[[@cit0010]\]. This may be because of the difference in the species of microorganism that causes CL. Conclusions {#sec6} =========== In the present study, we obtained results similar to previous studies in terms of age, gender, and the localization and duration of CL lesions in general. When SP and TP were compared, we found that the number of lesions was higher, disease duration was shorter, and the mean age was lower in SP. The illustrations in [Figure 2](#f0002){ref-type="fig"} are taken from the following internet addresses: <https://www.rch.org.au/uploadedImages/Main/Content/clinicalguide/guideline_index/adult-head-front-and-back.gif> (Head/neck)). <https://www.rch.org.au/uploadedImages/Main/Content/clinicalguide/guideline_index/anterior-body-from-the-right.gif> (Body). Conflict of interest ==================== The authors declare no conflict of interest.

Introduction {#s1} ============ With the increasing affordability of personal genome testing (PGT), the incorporation of patient genotype data into the practice of medicine is becoming more pervasive. Multiple medical centers across the country have begun introducing genetic and genome-wide analysis to make pharmacogenetic testing available to patients [@pone.0068853-Johnson1], [@pone.0068853-Pulley1]. PGT companies offering direct-to-consumer (DTC) tests have empowered individuals to independently obtain their personal genomic profiles, which provide them with a view of their genetic risks for hundreds of diseases and atypical drug responses. Further, applications of genetic testing are expanding to pre-conception genetic screening [@pone.0068853-Srinivasan1], selection of embryos for *in vitro* fertilization [@pone.0068853-Johnson2], non-invasive screening for fetal chromosomal abnormalities [@pone.0068853-Fan1], and the diagnosis of complex medical conditions [@pone.0068853-Worthey1]. Despite this expansion, most medical schools have not kept pace in providing state-of-the-art education in genetics and genomics to medical trainees [@pone.0068853-Salari1]. Healthcare authorities and medical educators now agree that there is a strong need to train medical students and physicians to understand basic principles of genomics and to be able to interpret PGT results [@pone.0068853-Guttmacher1], [@pone.0068853-Wiener1]; however, there has been significant debate over the best educational models to deploy [@pone.0068853-Salari2], [@pone.0068853-Walt1]. Several institutions, including ours, have considered offering students the opportunity to undergo PGT themselves as part of an updated medical school genetics curriculum, with some institutions ultimately deciding against it [@pone.0068853-Walt1]. At Stanford School of Medicine, after a school-wide task force rigorously evaluated potential risks and benefits, PGT was offered to students as part of a first-of-its-kind medical school elective course on genomics and personalized medicine, where students learn principles of genetics and genomics through a combination of interactive lectures and hands-on analysis of genomic data, using either their personal genotype data or publicly available datasets [@pone.0068853-Salari2]. Given the novelty of this educational initiative, there was no data on how PGT impacts student learning and whether its use in the classroom enhances education. Therefore, we used a survey instrument administered before and after the course to examine associations between the use of PGT and student knowledge and attitudes about genomics. Based on previous evidence of the benefit of participatory learning in medical education [@pone.0068853-Genzen1], [@pone.0068853-Knoell1], [@pone.0068853-Mazmanian1], we hypothesized that the use of personal genome data in the classroom would improve knowledge and the learning experience for students. Materials and Methods {#s2} ===================== Subjects {#s2a} -------- Subjects were medical and graduate students enrolled in an elective 8-week course on genomics and personalized medicine (Genetics 210; <http://gene210.stanford.edu/>) offered in the Summer 2010 quarter at Stanford School of Medicine. Forty-six students were enrolled in the course, and participation in this study was voluntary and anonymous. Genotyping {#s2b} ---------- The course started with two weeks of instruction and class discussion led by a clinical geneticist (L.H.), a genetic counselor (K.E.O.), and a bioethicist/lawyer about the risks, benefits, uses, and limitations of PGT; these sessions provided the students with necessary background to provide informed consent should they proceed with PGT. At the end of the second week of instruction, students decided whether to personally undergo genotyping using the PGT services of either one of two PGT companies (23andMe or Navigenics). Of note, at the time of the course offering, 23andMe provided customers with their genotypes for all ∼600 K SNPs on their microarray while Navigenics provided genotypes for only the ∼300 SNPs used in their clinical reports. The subsequent six weeks of instruction included lectures and hands-on data analysis exercises on various topics related to human genetics, genomics, and personalized medicine (see course website for more details of the curriculum; <http://gene210.stanford.edu/>). Each week students were led through classroom exercises to analyze various aspects of whole-genome single nucleotide polymorphism (SNP) data. A dataset comprising 12 diverse individuals from the HapMap project genotyped on Illumina HumanHap 650 K SNP microarrays were provided to all students. Students who underwent PGT were able to complete data analysis exercises using their personal genotype data, and students who did not undergo testing used publicly available genotype data from the 12 HapMap patients. A number of safeguards were implemented to ensure student privacy, confidentiality, and safety, including the provisioning of free genetic and medical counseling and mechanisms that students using their own data could ask questions if they had difficulty in resolving the class exercises without disclosing their genotype results [@pone.0068853-Salari2]. Survey Instrument {#s2c} ----------------- At the start and conclusion of the course, we electronically administered a survey that assessed student attitudes and knowledge about genomics and personalized medicine. The survey (extending the questionnaire developed by Ormond *et al*. [@pone.0068853-Ormond1]) included basic demographic information; assessed attitudes and knowledge about PGT; and solicited students\' feedback on the experience of undergoing testing as it related to the class and their learning experience (**[Methods S1](#pone.0068853.s003){ref-type="supplementary-material"}**). Student attitudes were assessed either via yes/no questions or by asking for extent of agreement with statements on a 5-point Likert scale. Knowledge was assessed by both subjective and objective questionnaire items. For objective knowledge assessment, 6 multiple-choice questions and one free response question were asked; responses were scored blinded to subjects\' genotyping status. Separate from the surveys presented in this study, students were also invited to participate in individual interviews discussing their experience (presented separately [@pone.0068853-Vernez1]). The Stanford University Institutional Review Board approved all study methodology. Data Analysis {#s2d} ------------- We analyzed responses from students who completed both pre- and post-course surveys and attended at least 50% of the eight class sessions. Student responses to the pre- and post-course surveys were linked using a randomly assigned numeric code to maintain anonymity. We considered separating the students who underwent PGT before the course from those who underwent it during the course, but since preliminary statistical comparisons were underpowered to show differences, we elected to combine these groups in our study analysis. Student attitudes assessed on a 5-point Likert scale were collapsed and reported as the percentage of students who agree or strongly agree with the stem statement. Paired pre-course and post-course responses were tested for change using paired non-parametric statistics (McNemar\'s test for binary response questions and the Wilcoxon signed-rank test for Likert items). Comparisons between responses of genotyped and non-genotyped students were made using Fisher\'s exact test for binary response questions and the Mann-Whitney *U*-test for Likert items. The change in student knowledge assessed by pre-course and post-course knowledge scores was evaluated by paired *t*-test. The difference in knowledge improvement between genotyped and non-genotyped students was assessed by Student\'s *t*-test. Results {#s3} ======= Forty-three class participants completed the pre-course survey (93% response rate) and 34 class participants completed the post-course survey (74% response rate). We present data from 31 students who completed both pre- and post-course surveys, and attended at least 50% of the class sessions (67% of the course enrollees). Demographics from this study population are presented in [**Table 1**](#pone-0068853-t001){ref-type="table"}; subjects were evenly split between genders, with slightly fewer medical versus non-medical trainees, and most frequently in their first year of training. Thirteen of the 17 students (76%) who indicated on the pre-course survey that they planned to undergo PGT did ultimately undergo testing. Seven students were initially unsure, of which 3 proceeded with testing. Another seven students had already undergone PGT prior to the course (all by 23andMe); these students indicated that they did not plan to undergo testing again and used their previously obtained data in the course. Due in part to a more limited genotype dataset provided by Navigenics, all 16 students who underwent PGT in the course did so via 23andMe. Thus, 23 students formed the genotyped group, and 8 students formed the non-genotyped group. There were no significant differences in demographics between the genotyped and non-genotyped groups ([**Table 1**](#pone-0068853-t001){ref-type="table"}) or between class participants who completed the study and those who were lost to follow-up (**[Table S1](#pone.0068853.s001){ref-type="supplementary-material"}**). 10.1371/journal.pone.0068853.t001 ###### Subject characteristics. {#pone-0068853-t001-1} Genotyped[a](#nt101){ref-type="table-fn"} Non-genotyped[a](#nt101){ref-type="table-fn"} ---------------------------------------- ------------------------------------------- ----------------------------------------------- ------ Gender (female) 13 (56.5) 3 (37.5) 0.43 Program 0.21 Medical (MD, Clinical Resident/Fellow) 7 (30.4) 5 (62.5) Biomedical (PhD, Post-doctoral Fellow) 16 (69.6) 3 (37.5) Year in Program 0.27 1 11 (47.8) 2 (25.0) 2 1 (4.3) 1 (12.5) 3 3 (13.0) 3 (37.5) 4+ 8 (34.8) 2 (25.0) Previous personal genome testing 7 (30.4) N/A The number (and percentage) of subjects is reported. Fisher\'s exact test comparing genotyped and non-genotyped subjects. Pre and post-course attitudes toward personal genome testing {#s3a} ------------------------------------------------------------ We surveyed student attitudes towards personal genome testing on pre- and post-course surveys ([**Table 2**](#pone-0068853-t002){ref-type="table"}). Among genotyped students, 35--39% indicated that they would recommend PGT for a patient, with no significant difference between pre- and post-course responses. In contrast, among students who elected to not undergo testing, 50% indicated they would recommend PGT for patients before the course, but only 12% maintained this position after the course. Among proponents of PGT for patients, the most common reasons for support were to satisfy general curiosity about their genetic make-up (67%) and to see if a specific disease runs in their family or their DNA (56%). Those opposed to PGT for patients felt that such testing has limited clinical utility (82%), limited clinical validity (77%), individuals have a limited ability to understand and interpret their test results (68%), and not enough trained health care providers are available to help them interpret results (55%). 10.1371/journal.pone.0068853.t002 ###### Student attitudes toward personal genome testing. {#pone-0068853-t002-2} Genotyped group *N* = 23 Non-genotyped group *N* = 8 Genotyped *vs*. Non-genotyped -------------------------------------------------------------------------------------------------------- -------------------------- ----------- ----------------------------- ---------- ------------------------------- ----------- ------ If you were to undergo PGT, would you share your results with a physician? 23 (100.0) -- 6 (75.0) 8 (100.0) If you were to undergo PGT, would you ask a health care provider for help in interpreting the results? 12 (52.2) -- 4 (50.0) 4 (50.0) Would you at this time recommend PGT for a patient? 9 (39.1) 8 (34.8) 1 4 (50.0) 1 (12.5) 0.25 0.38 Most people can accurately interpret their PGT results 0 (0.0) 1 (4.3) **0.025** 0 (0.0) 0 (0.0) 1 0.38 PGT companies provide an accurate analysis and interpretation of genotype data 2 (8.7) 10 (43.5) **0.02** 0 (0.0) 0 (0.0) 1 0.14 PGT companies should be regulated by the federal government 15 (65.2) 18 (78.3) 0.36 4 (50.0) 7 (87.5) **0.037** 0.47 For yes/no questions, the number (and percentage) of subjects responding yes is reported. For Likert items, the number (and percentage) of subjects who agreed or strongly agreed with the statement is reported. McNemar\'s test for binary response questions and Wilcoxon signed-rank test for Likert-scale items comparing pre- to post-course responses. Fisher\'s exact test for binary response questions and Mann-Whitney *U*-test for Likert-scale items comparing post-course responses between genotyped and non-genotyped groups. At the start of the course, 100% of students felt that most people cannot accurately interpret their PGT results. Also, very few students felt that PGT companies provide an accurate analysis and interpretation of genotype data. However, after the course, significantly more students who underwent genotyping themselves believed that people could accurately interpret their results (*P* = 0.025, Wilcoxon signed-rank test) and that PGT companies provide an accurate analysis and interpretation (*P* = 0.02, Wilcoxon signed-rank test) ([**Table 2**](#pone-0068853-t002){ref-type="table"}). In contrast, the non-genotyped group continued to feel that both patients and companies cannot accurately analyze or interpret PGT results. More than half of students felt that PGT companies should be regulated by the federal government, with significantly more non-genotyped students holding this opinion by the end of the course than at the beginning (*P* = 0.037, Wilcoxon signed-rank test; [**Table 2**](#pone-0068853-t002){ref-type="table"}). Notably, the majority of students (62%) indicated that they would undergo whole-genome sequencing in the future once it became affordable to them, including 50% of the students who chose not to undergo SNP-based genotyping at this time. Students overall felt PGT is an important educational topic, as 71% of students agreed or strongly agreed that it will likely play an important role in their future career. Knowledge of genetics and personal genome testing {#s3b} ------------------------------------------------- We next examined students\' reflections on their own knowledge of genetics and personal genome testing as well as that of practicing physicians ([**Table 3**](#pone-0068853-t003){ref-type="table"}). Nearly all students felt that most physicians do not have enough knowledge to help individuals interpret PGT results, on both pre- and post-course surveys. Regarding their own knowledge, by the end of the course students in the genotyped group more strongly indicated that they understood the risks and benefits of using PGT services (*P* = 0.008, Mann-Whitney *U*-test) and that they knew enough about genetics to understand PGT results (*P* = 0.012, Mann-Whitney *U*-test) than students in the non-genotyped group ([**Table 3**](#pone-0068853-t003){ref-type="table"}). 10.1371/journal.pone.0068853.t003 ###### Student perceptions of knowledge about genetics and personal genome testing. {#pone-0068853-t003-3} Genotyped group *N* = 23 Non-genotyped group *N* = 8 Genotyped *vs*. Non-genotyped --------------------------------------------------------------------------------------------------------------------------------------- -------------------------- ----------- ----------------------------- ---------- ------------------------------- ---------- ----------- Most physicians have enough knowledge to help individuals interpret results of personal genome tests 0 (0.0) 1 (4.3) 0.41 0 (0.0) 0 (0.0) 0.77 0.96 I understand the risks and benefits of using PGT services 20 (87.0) 23 (100) **0.003** 4 (50.0) 7 (87.5) 0.2 **0.008** I know enough about genetics to understand PGT results 18 (78.3) 23 (100) **0.005** 4 (50.0) 7 (87.5) **0.02** **0.012** I have a better understanding of principles of human genetics on the basis of undergoing personal genotyping -- 16 (69.6) -- -- -- -- -- Undergoing personal genotyping was an important part of my learning in GENE210 -- 15 (65.2) -- -- -- -- -- I would have learned just as much from GENE210 had I not undergone personal genotyping and only used publicly available genotype data -- 7 (30.4) -- -- -- -- -- I would have learned more from GENE210 had I undergone personal genotyping instead of using publicly available geno type data -- -- -- -- 3 (37.5) -- -- The number (and percentage) of subjects who agreed or strongly agreed with each statement is reported. Wilcoxon signed-rank test comparing pre- to post-course responses. Mann-Whitney *U*-test comparing post-course responses between genotyped and non-genotyped groups. Among genotyped subjects, 70% felt that they acquired a better understanding of principles of human genetics on the basis of undergoing PGT, and 65% felt that undergoing PGT was an important part of their learning in the course. Since all students were provided publicly available genotyping data from HapMap subjects to complete the in-class computer exercises, we specifically asked students to reflect on the use of personal versus publicly available genotype data. Only 30% of students who used personal genotype data felt that they would have learned just as much in the course had they not undergone testing and only used publicly available genotyping data. Conversely, a similar proportion of students in the non-genotyped group (37%) felt that they would have learned more in the course had they used personal genotype data instead of publicly available data. To assess student knowledge of genetics and personal genome testing more objectively, we incorporated a short knowledge assessment in the pre- and post-course surveys, covering basic principles of genetics and clinical scenarios requiring the interpretation of PGT results (the same seven knowledge questions were asked on both surveys). At the start of the course, there was no significant difference between knowledge scores of students who did and students who did not undergo genotyping. However, by the end of the course we noted a significant improvement in knowledge scores only among students who underwent PGT (*P* = 3.5×10^−6^, paired *t*-test; [**Figure 1**](#pone-0068853-g001){ref-type="fig"}). Students in the non-genotyped group did not demonstrate significant improvement in their knowledge scores. The extent of improvement among genotyped students was significantly greater than that of non-genotyped students (31% *vs*. 1%, *P* = 0.002, Student\'s *t*-test). {#pone-0068853-g001} Genotyping process and experience {#s3c} --------------------------------- Students in the genotyped group most frequently reported having undergone testing due to general curiosity about their genetic make-up (100%), to help them understand principles of human genetics (57%), to help them understand what patients learn/experience (52%), and to see if a specific disease runs in their family or is in their DNA (52%). Non-genotyped students decided against testing due to concern that a for-profit company would have their DNA or genotype data (50%), concern that their data would not remain private (50%), and feeling that the information from SNP-based genotyping tests would not be useful (50%). Genotyped students were more likely than non-genotyped students to feel the course helped them understand a patient\'s experience in undergoing PGT (*P* = 0.00057, Mann-Whitney *U*-test; [**Table 4**](#pone-0068853-t004){ref-type="table"} **)** and to be pleased with their decision to undergo genotyping (*P* = 0.00058, Mann-Whitney *U*-test; [**Table 4**](#pone-0068853-t004){ref-type="table"}). Most of those who were not genotyped were neutral about their decision (75%). 10.1371/journal.pone.0068853.t004 ###### Student reflection on genotyping offer and experience. {#pone-0068853-t004-4} Question[a](#nt109){ref-type="table-fn"} Genotyped group*N* = 23 Non-genotyped group *N* = 8 ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------- ----------------------------- ------------- This course helped me understand what a patient\'s experience might be like if they chose to undergo personal genotyping 23 (100.0) 4 (50.0) **0.00057** Pleased with decision regarding personal genotyping 19 (82.6) 1 (12.5) **0.00058** Experienced anxiety when deciding whether to undergo personal genotyping 3 (13.0) 4 (50.0) **0.0087** Experienced anxiety when awaiting PGT results 3 (13.0) -- Experienced anxiety after receiving PGT results 2 (8.7) -- The opportunity to ask healthcare professional (e.g. genetic counselor, medical geneticist, or other physicians) for help in interpreting the results is an important component to a personal genotyping offer 21 (91.3) 7 (87.5) 0.23 The number (and percentage) of subjects who agreed or strongly agreed with each statement is reported. Mann-Whitney *U*-test comparing post-course responses between genotyped and non-genotyped groups. When deciding whether to undergo PGT, only 13% of students who ultimately underwent testing reported experiencing anxiety, compared to 50% of students who did not undergo testing (*P* = 0.0087, Mann-Whitney *U*-test; [**Table 4**](#pone-0068853-t004){ref-type="table"}). Few students who elected to undergo testing reported anxiety while awaiting their test results (13%) or after receiving the results (8.7%). Nearly all students in both the genotyped and non-genotyped groups agreed that the opportunity to ask healthcare professionals for help in interpreting test results was an important component to the PGT offer ([**Table 4**](#pone-0068853-t004){ref-type="table"}). Effect of genotyping on behavior {#s3d} -------------------------------- When asked prior to the course, 100% of genotyped students and 75% of non-genotyped students indicated a willingness to share their PGT results with their physician ([**Table 2**](#pone-0068853-t002){ref-type="table"}); most indicated they would do so only if they discovered they were at an elevated risk for a condition. Although students indicated a strong willingness to share, significantly fewer (approximately half in each group) indicated they would ask a healthcare professional for help interpreting the results. After actually undergoing PGT, even fewer students reported that they had already asked (13%) or were planning to ask (13%) a healthcare professional for help interpreting their test results; the remainder indicated that they do not plan to ask a healthcare professional for help. All 23 genotyped students reported taking at least one action specifically on the basis of their PGT results (**[Table S2](#pone.0068853.s002){ref-type="supplementary-material"}**). Most frequently, students held discussions with their family about test results (78%) and to learn more about their family history (52%), and 70% of students performed internet searches to educate themselves on conditions for which they were found to be at risk. Three students reported changing their diet in a positive manner and 4 students reported contemplating positive diet, exercise, or smoking habits, but had not yet made changes. The actions were most commonly things that subjects reported already planning to do or were actively doing (30%), but some actions had been previously attempted and students indicated that their PGT results moved them to try again (22%). In 30% of instances, they reported that PGT results moved them to contemplate and/or attempt various positive behavior changes. Course curriculum reflection {#s3e} ---------------------------- Among numerous safeguards built into the course, students who elected to undergo PGT did so privately with a PGT company, and course instructors never asked students whether they had undergone genotyping or for their raw genotype data. Course instructors also asked that students not disclose whether they had undergone genotyping. We asked students to reflect on the experience of using personal genotype data in the classroom and specifically, how well any concerns about privacy and confidentiality were addressed. Few students (2 genotyped, 1 non-genotyped) felt that the professors of the course knew whether they had undergone PGT; no student reported feeling at a disadvantage in the class as a result of this. Overall, most genotyped students (83%) felt they were easily able to go back and forth between their personal genotype data and the publicly available genotype data provided to them when working on the computer exercises, none felt required to divulge their genotype information in order to ask questions of the course professors, and 43% indicated they would have felt comfortable sharing their genotype data in order to ask questions of the course professors. Overall, all students felt that PGT should be made available to medical and graduate students as part of their genetics curriculum in some manner, but varied in their feelings towards whether it should be incorporated as an option in an elective course (61%) or core course (32%). Discussion {#s4} ========== We report here the first study of educational outcomes in a course where students have the option to undergo personal genome testing. Overall, our results suggest that utilizing personal genotype data can augment the educational value of courses teaching concepts of genomics and personalized medicine. At the end of the course, genotyped and non-genotyped students alike viewed the option to undergo genotyping favorably, and PGT was incorporated into the curriculum in a manner that effectively maintained student safety, privacy, and confidentiality. Most students who participated in this study took the course with the intention of undergoing PGT and adhered to their initial plan. However, the decision of a substantial fraction of students was influenced by the first two weeks of the course, which was spent discussing the risks, benefits, uses, and limitations of PGT services. We found that students were more likely to elect to undergo testing if they felt that they understood the risks and benefits of the test and enough about genetics to understand the results. Our data also suggest that students who experience anxiety during the decision-making process are more likely to decide against testing than students who do not. This is not surprising and reflects the self-selection process that is often seen in predictive genetic testing [@pone.0068853-Sanderson1]. Together, these observations highlight the importance of a rigorous informed consent process prior to offering PGT, whether in a classroom setting or elsewhere. Few genotyped students reported anxiety at any point in the process (deciding to undergo PGT, waiting for results, and after results were received), and none of the genotyped students reported regret with their decision on the post-course survey (4 weeks after receiving test results). These results are consistent with a recent report of subjects who underwent PGT with the Navigenics Health Compass [@pone.0068853-Bloss1], where such testing did not result in any measurable short-term changes in psychological health and over 90% of subjects experienced no test-related distress. Students frequently cited gaining a better understanding of the patient experience as a reason that compelled them to undergo PGT. This parallels findings of a recent study of 137 Cleveland Clinic physicians who were offered PGT as a way to increase their familiarity with clinical genetics and PGT [@pone.0068853-Sharp1]. A majority of respondents in that study (77%) felt their personal experience pursuing PGT would benefit their patients directly by improving their ability to advise patients on the testing process and to relate to patients\' experiences interpreting PGT results. Indeed, 100% of the genotyped students in our study reported that the course helped them understand the patient experience of undergoing PGT, compared to only half of non-genotyped students who indicated such an understanding. Despite this, most students still reported that, at this time, they would not recommend PGT for patients. However, students who underwent genotyping more often recommended it for patients than did students who did not undergo genotyping. These results are congruent with those of a recent study, in which primary care physicians currently offering PGT services as part of their practice were more likely to order the test for their patients if they felt well-informed about PGT and if they had undergone testing themselves [@pone.0068853-Haga1]. The primary hypothesis of this study was that undergoing PGT would enhance the learning of students in the course. Since all students were provided publicly available genotype data from HapMap subjects to complete in-class computer exercises, we were able to specifically evaluate the educational utility of analyzing personal versus publicly available genotype data. Regardless of whether they used personal or public genotype data, by the end of the course nearly all students felt confident that they understood the risks and benefits of PGT and the underlying genetics required to understand PGT results. This stands in contrast to the previous study of students in our core medical school genetics course without PGT, in which only 20% of students felt they knew enough about genetics to understand PGT results by the end of the course [@pone.0068853-Ormond1]. While there are significant differences between the two courses (e.g., PGT is a smaller focus of the core course for medical students, and students in this study likely started with a greater level of understanding and interest based on prior coursework), our finding suggests that using genotype data of any sort (personal or public) to perform exercises on data analysis and interpretation enhances the learning experience of students. We also found evidence specifically suggesting that PGT positively impacts learning for those students who self-select to undergo it. The majority of genotyped students felt they acquired a better understanding of the principles of human genetics on the basis of undergoing PGT and that the genotyping was an important part of their learning in the course. Substantiating these beliefs, genotyped students significantly improved their knowledge scores by an average of 31%, while non-genotyped students showed no significant difference in knowledge scores. The performance of non-genotyped students is similar to that described in the study of students in our core medical school genetics course without PGT, where only a modest improvement was noted between pre-course and post-course knowledge scores [@pone.0068853-Ormond1]. Together, these data suggest that some students derive greater educational benefit by undergoing PGT and using personal genotype data in the classroom than students who strictly use publicly available data or no data at all. As has been suggested in other educational contexts [@pone.0068853-Genzen1], [@pone.0068853-Knoell1], [@pone.0068853-Mazmanian1], analyzing and interpreting data with personal relevance may encourage students to be more engaged with the material, leading to greater understanding and retention of knowledge. For example, a recent report describes a genotyping exercise in a pharmacy class where 10 student volunteers provided DNA samples that were subjected to genotype analysis and presented to the class in the context of a genetic counseling session [@pone.0068853-Knoell1]. Students indicated in a survey that the exercise engaged them with the course content and would positively influence their ability to apply pharmacogenetic principles to patient care. Undergoing PGT and interpreting test results also led some students to make or consider behavioral changes. Almost one-third of genotyped students indicated that due to elevated risks, they had already changed or were contemplating changes to their diet, exercise, or smoking habits. However, a longer-term qualitative study conducted on a small number of our students indicates that 6 months after receiving PGT results, none had taken significant behavioral actions [@pone.0068853-Vernez1], suggesting that early behavioral changes may not be sustained. These results mirror the recent study by Bloss *et al*.; while they found no significant change between baseline and follow-up in dietary fat intake or exercise behavior of subjects who underwent PGT with the Navigenics Health Compass, they also found that a substantial fraction of subjects contemplated behavioral changes or intended to undergo more medical tests [@pone.0068853-Bloss1]. The studies\' differences may reflect differences in study population (mean age was 46.7 compared to our younger population of students in their mid-twenties), or more likely, length of follow-up (mean 5.6 months compared to 4 weeks in our study). As an exploratory study of the first iteration of the course, this study has several limitations. The small sample size, self-selection of enrollees in an elective course, and single-institution setting of our study make broad generalizability of our results difficult. We also do not know the extent to which the educational benefits noted here extend to other types of learners, such as undergraduate students or practicing physicians, since our study was conducted primarily on medical and graduate students who expressed a specific interest in the topic of genotyping and personal genome testing. Finally, the survey instrument used in the study is not validated, and thus we cannot exclude the possibility that unclear wording in the questions may have affected some of our findings. These limitations notwithstanding, our study represents the first line of evidence that the use of personal genome testing can enhance genetics education for at least a subset of learners. As personal genome testing becomes more widely-used in the classroom, future work should focus on conducting a randomized study where students who would like to undergo PGT are randomized to either undergo testing and work with their own genotype data or not undergo testing and work with publicly available genotype data. Such a study design would help control for any bias in educational outcomes resulting from self-selection and the results would be of great interest. We believe it is imperative that medical school educators think creatively about how to incorporate education on this rapidly emerging area of medicine and science into their curricula. Our study finds that the interactive and participatory approach of using PGT in the classroom has the potential to increase students\' knowledge and awareness of genetic testing. Although further study of its pedagogical utility is warranted, we believe when thoughtfully implemented, PGT can be used as a powerful and effective tool in genetics education. Supporting Information {#s5} ====================== ###### **Characteristics of subjects who completed study vs. lost to follow-up.** (DOC) ###### Click here for additional data file. ###### **Actions taken specifically as a result of receiving PGT results.** (DOCX) ###### Click here for additional data file. ###### (PDF) ###### Click here for additional data file. The authors would like to thank the Stanford Genotyping Task Force and the Genetics 210 course instructors (in particular Stuart Kim, PhD) for their important contributions to the course and this research study; Mildred Cho, PhD and Jennifer Deitz, MA for assistance in constructing the survey instrument; and Deans Philip Pizzo, MD and Charles Prober, MD for their leadership in medical education that allowed this innovative course to be implemented. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KS LH KEO. Performed the experiments: KS KJK. Analyzed the data: KS. Wrote the paper: KS KJK LH KEO.

There are several errors in Table 2. Please refer to the corrected version of Table 2 here: {#pone-a8dfd4ee-f4b7-443e-bf78-ebb0dab4e55b-g001} **Competing Interests:**No competing interests declared.

1. Introduction {#sec1} =============== The role of growth hormone (GH) in growth promotion is well known by clinicians, however, less appreciated is its effect on metabolism in the well state. In 1936, Bernardo Houssay, M.D. in the New England Journal of Medicine, proposed that the anterior pituitary gland after the liver and pancreas plays a key role in glucose metabolism. That key role was later shown to be due in part to GH \[[@B1]\]. The effect of GH on glucose metabolism involves two phases: an initial phase which involves a decrease in glucose (an insulin-like effect) and a second phase which includes its effects on gluconeogenesis and fat mobilization \[[@B2]\]. In both states of growth hormone deficiency and excess, these effects on glucose metabolism will be altered. Altered glucose and fat metabolism are important components of fatty liver (FL) and type 2 diabetes (T2DM), both states of hepatic insulin resistance (IR). In fact, a seminal case of fatty liver (FL) resolution with GH administration in a 17-year-old patient with panhypopituitarism treated initially with levothyroxine and hydrocortisone alone was reported by*Takano*et al. in 1997 \[[@B3]\]. This suggests that GH treatment of T2DM which may be associated with GHD has the potential to reduce the IR which may be present. We outline a case of untreated childhood onset growth hormone deficiency (CO-GHD) who presented with type 2 diabetes mellitus (T2DM) and also steatohepatitis. We discuss her management and evidence from the basic sciences and clinical studies which show that her presentation with T2DM and steatohepatitis was likely associated with untreated GHD and that with GH supplementation her condition was ameliorated. Lastly, we discuss the implications of this case. 2. Case Presentation {#sec2} ==================== A 17-year-old nonobese Caucasian female who had a history of a medulloblastoma diagnosed at 7 years of age was treated with radiation therapy. She subsequently developed TSH and GnRH deficiencies. Though GHD was suspected based on height (z-score of -- 3.1; see [Figure 1](#fig1){ref-type="fig"}), treatment had not been initiated based on the initial management focus being to treat her medulloblastoma. At 15 years of age when her bone age showed full skeletal maturity, her parents were informed that GH therapy could not be pursued because her linear growth was complete. On presentation, the patient\'s height was 141.3 cm (z= -3.1) and weight was 53 kgs (36^th^ percentile for age). Body mass index was 25.8 kg/m^−2^ (86^th^ percentile for age). Surveillance labs done at the oncology clinic showed glucosuria. Further testing showed HbA1c of 9.6% and on another day her fasting glucose was 277 mg/dL. Based on these results, diabetes mellitus was diagnosed. When glutamic acid decarboxylase (GAD-65; Esoterix), islet-cell (Esoterix), insulin (Esoterix), and zinc transporter 8 (ARUP Laboratories) antibodies as well as DNA panel for maturity onset diabetes of youth (MODY) genes (HNF4*α*, GCK, IPF1, HNF1*α*, and HNF1*β*, \[Athena Diagnostics\]) returned all negative along with an elevated fasting C-peptide level of 3 ng/mL (normal: 0.4 - 2.1), T2DM was diagnosed. With the initiation of traditional basal/bolus insulin therapy using conventional dosing, a rapid escalation to peak total daily insulin dose of 2.9 units/kg/day (\~ 155 units/day) was required to treat her refractory hyperglycemia. Treatment nonadherence was thought to be the unlikely cause of her increased insulin requirements based on the agreement between her insulin dosing and prescription refill data. A comprehensive evaluation for conditions associated with IR was negative. However, based on Arginine/Clonidine stimulation testing showing peak GH level of 0.8 (normal: ≥ 10 ng/mL), a diagnosis of GHD was made. GH supplementation was initiated at 0.3 mgs daily and titrated based on IGF-1 levels. After GH was started, her systolic and diastolic blood pressures (BP) which were mildly elevated between 124-136 and 77-89, respectively, became more normal. Despite this, lisinopril 5 mgs once daily was added for microalbuminuria. With the diagnosis of T2DM and our patient having a significant family history of adverse cardiovascular risk factors, she was started on atorvastatin 10 mgs once daily. Within 2 months of therapy, her LDL cholesterol (LDL-C) decreased to 74 mg/dL. [Table 1](#tab1){ref-type="table"} shows serial lipid profiles. Though her diabetes was not fully reversed with GH, her HbA1c decreased to 5.9% and 5.8% at 6 and 19 months, respectively. Her insulin therapy requirement decreased to 1.9 units/kg/day (\~ 100 units) at 12 months after the start of GH. Magnetic Resonance Imaging (MRI) of the brain and abdomen indicated a small anterior pituitary gland and liver masses, respectively. Liver biopsy showed steatohepatitis with bridging fibrosis ([Figure 2](#fig2){ref-type="fig"}). With GH therapy, her liver transaminases trended to normalcy ([Table 1](#tab1){ref-type="table"}). Repeat MRI abdomen at 20 months after the start of GH showed stability of the liver lesions when compared to that done at 14 months. These hyperintense lesions like the initial ones were located in the liver\'s parenchyma and the appearance of the liver was otherwise normal. With GH therapy, the patient\'s stamina improved. She was now able to work for 20 hours weekly without becoming fully exhausted and her Quality of Life-Assessment of Growth Hormone Deficiency in Adults (QoL-AGHDA) and Quality of Life Satisfaction (QLS) scores, both questionnaire-based, improved ([Figure 3](#fig3){ref-type="fig"}). 3. Discussion {#sec3} ============= This case demonstrates that GH supplementation in an adolescent with CO-GHD led to improvements in transaminases, insulin requirements, and glucose control. Several mouse models have corroborated the association of GHD with IR. In mice with liver GH receptor (GH-R) knockout, metabolic syndrome (MetS), steatohepatitis, increased inflammation, liver fibrosis, and hepatic tumor develop \[[@B4]\]. Additionally, a similar mouse model resulted in hyperinsulinemia, hyperglycemia, and IR. With the restoration of the liver\'s IGF-1 expression, there was an improvement in both insulin sensitivity and serum lipid profile. This, however, did not protect against hepatic inflammation induced by steatosis. This shows that GH and not IGF-1 directly affects lipid uptake and lipogenesis \[[@B5]\]. Also in a prior study again involving a liver specific GH receptor knockout mice, de novo lipogenesis was increased; however, this increase was not associated with the classic insulin mediated pathway \[[@B6]\]. So, our patient\'s IR and hepatic steatosis can be explained by her GHD based on data from some studies as well as the effect of GH in inhibiting this. Evidence for the association of hepatic steatosis with GHD is also provided by the abnormalities involving the downstream pathway of GH signal transduction. Mice with hepatocyte-specific deletion of Janus kinase 2 (JAK2L), involved in the postreceptor phase of GH signaling, were lean but had FL. They also showed increased levels of GH, triglycerides, and plasma free fatty acids. Since GH in some instances can cause lipolysis, GH-deficient*little*mice which were crossed to JAK2L mice had both a rescuing of the FL and an increased expression of a fatty acid transporter \[[@B7]\]. Though this provides a mechanism for the FL observed with the liver specific disruption of GH signaling, in the same mice, elevated GH levels occurring as a consequence of disrupted signaling can cause an increase in resting energy expenditure \[[@B8]\]. In this situation, steatohepatitis is prevented based on increased fatty acid utilization. So the putative lipolytic action of GH can be offset by hyperinsulinemia; hence, the action of GH is variable and depends on the physiological context. Additionally, mice with signal transducer and activator of transcription (STAT) 5 mutations, another downstream signal from GH, also develop steatohepatitis \[[@B9]\]. Since pathologies involving the downstream pathways in GH signal transduction are associated with steatohepatitis, this supports the role of GH in lipid metabolism and the notion that the FL changes in our patient may be due to GHD. Clinical reports about Laron syndrome, primary GH insensitivity involving a molecular defect in human GH-R, have also documented the development of FL, IR, and T2DM \[[@B10]\]. In addition, men with hypopituitarism have a high prevalence of nonalcoholic FL disease (NAFLD) in the absence of GH therapy. When these men were treated with GH, there was histological improvement in the liver. This demonstrates that NAFLD is predominantly attributable to GH \[[@B11]\]. Though a follow-up liver biopsy was not done, the trend to normal in our patient\'s transaminases ([Table 1](#tab1){ref-type="table"}) supports the effect of GH on the liver and the likelihood that these changes were induced by GHD. Since our patient\'s steatohepatitis improved with GH treatment, she likely had a reduction in her hepatic IR. The evidence for this was seen in her decreased insulin requirements. This reduction in IR also was associated with her decreased blood pressures and these decreased pressures can be explained since IR negatively impacts endothelial cell function \[[@B12]\]. Moreover, the decrease in our patient\'s insulin resistance may also be explained by the positive impact of GH on *β*-cell function which is well described. In adults with lifetime congenital untreated GHD, there is reduced *β*-cell function \[[@B13]\]. In children with GHD who are supplemented with GH, *β*-cell secretory capacity is enhanced \[[@B14]\]. Additionally, studies in mice have shown that, with isolated GHD, *β*-cell function deteriorates. This deterioration is not due to changes in *β*-cell mass \[[@B15]\]. Studies by*Nielsen* et al. have shown that GH stimulates *β*-cell proliferation, glucose induced insulin release, and insulin gene expression*in vitro*\[[@B16]\]. These also provide a possible mechanism for the decrease in our patient\'s IR. Atorvastatin was added to our patient\'s treatment because although an amelioration of her LDL-C was expected with GH therapy, this did not happen. Plausible explanations for this include a genetic basis for her hyperlipidemia. This could have been superimposed on an accumulated adverse CV risk profile which developed based on the time period of her untreated CO-GHD \[[@B17]\]. Atorvastatin was also added to her treatment as well since based on her history of T2DM, her LDL-C levels needed improvement. Based on the points discussed, with evidence from adult studies, clinical reports, and, more recently, mouse models (even with the lack of pediatric studies, especially long-term ones) explaining the important contribution of GH to normal lipid and glucose metabolism, there is enough data to reinforce the benefits of treating GHD even when linear growth is completed. Pediatric clinicians should highlight and reinforce that GH supplementation has the potential to prevent adverse metabolic consequences in untreated states of severe deficiency. Also important is the fact that, even with growth cessation, supplementation with GH may improve QoL-AGHDA and QLS scores \[[@B18]\]. These questionnaire-based scores address the impact of GHD on issues of relevance to patients with GHD and can be useful for tracking the patient\'s response to treatment. The QoL-AGHDA tool addresses the general impact of GHD on each patient whereas the QLS score accounts for the level of importance which each individual may place on the issues affecting his/her life and gives a summarized weighted score based on these. In conclusion, this case illustrates not only that both NAFLD and T2DM are potential associations of untreated GHD but also that they may represent points along the natural history of hepatic IR secondary to untreated GHD. Furthermore, clinicians should ensure that patients with CO-GHD are not only treated in childhood but also appropriately transitioned to adult GH dosing after growth has ceased. This is important as GH can have a crucial role beyond the period of linear growth. With delayed GH initiation, it is possible that, with irreversible hepatic injury (such as bridging fibrosis in this patient), there may not be total amelioration of the metabolic manifestations seen in patients with GHD. The authors would like to thank Robert Hoffman, M.D., for his critical review of this manuscript. Additional Points ================= *Learning Points.* (i) Growth hormone is integral to both glucose and lipid metabolism even after the period of linear growth. (ii) Insulin resistance (IR) can be a feature of untreated growth hormone deficiency (GHD). (iii) Hepatic steatosis and type 2 diabetes mellitus (T2DM) being both hepatic IR states could be a feature of the natural history of untreated GHD. (iv) While growth hormone supplementation may ameliorate IR seen in the setting of T2DM and GHD, it is likely that delayed supplementation may not fully ameliorate IR if simple steatosis progresses to steatohepatitis with fibrosis. Ethical Approval ================ No IRB approval was required by our institutions. Consent ======= Consent was obtained from the patient prior to the submission of this publication. Conflicts of Interest ===================== The authors have no conflicts of interest relevant to this article to disclose. Authors\' Contributions ======================= Rohan K. Henry and Ram K. Menon contributed to drafting the article. Rohan K. Henry contributed to the critical revision of the article. All authors approved the final version of the article to be published. {#fig1} {#fig2} {#fig3} ###### Some of the patient\'s lab tests with reference to growth hormone start. TEST RESULT REFERENCE RANGE -------------------------------- -------- ----------------- ----- --------------- **Aspartate aminotransferase** 92 36 33 \< 40 U/L **Alanine aminotransferase** 87 24 56 15- 50 U//L **Total Cholesterol** 247 99 132 95- 195 mg/dL **HDL Cholesterol** 32 27 43 40- 58 ng/dL **LDL Cholesterol** 171 57 65 73- 117 mg/dL **Triglyceride** 267 70 120 20-200 mg/dL **IGF-1** 74 146 294 121-566 ng/mL [^1]: Academic Editor: Lucy Mastrandrea

Introduction {#sec1-1} ============ Liver cirrhosis has an increasing incidence in the general population. It is estimated that, in Europe, there are 170,000 annual deaths due to this cause with major differences between regions. Thus, in the south-east cirrhosis mortality rate is 10-20 times higher than the rest of the continent \[**[@R1]**\]. Romania is situated on the second place with figures of 64 males and 26,7 females considering the death rate per 100,000 population. The main causes of the disease are represented by chronic alcohol consumption, infections with hepatitis viruses B and C, and obesity related metabolic syndromes \[**[@R2]**\]. In these circumstances, abdominal surgery in patients with liver cirrhosis has an increasing frequency. Liver cirrhosis is a chronic and slowly progressive disease characterized by a replacement of normal liver tissue with scars. Liver function is progressively altered leading to hepatic failure. The effects of cirrhosis are not limited to liver function, but also affect other organs and systems. The renal, cardiac and respiratory functions are also adversely affected. These, along with coagulation disorders and immune deficiency create premises of a high-risk surgical ground \[**[@R3]**\]. Surgery for "acute abdomen" in such patients has a high mortality rate and requires special measures of resuscitation \[**[@R4]**\]. Copeland and al. proposed, in 1991, The Physiologic and Operative Score for enUmeration of Mortality and Morbidity (POSSUM) as a scoring system for the surgical audit \[**[@R5]**\]. Considering both pre- and intra-operative commonly measured parameters, this score is easy to use and has a wide application for general surgery in emergency and elective procedures \[**[@R6]**,**[@R7]**\]. The aim of the present study is to evaluate the accuracy of POSSUM score in predicting outcomes after emergency abdominal surgery in patients with liver cirrhosis. Material and method {#sec1-2} =================== Our study is based on a prospective analysis between January 2008 and December 2012 on 115 consecutive patients with liver cirrhosis hospitalized and operated for acute surgical abdomen in Surgery Clinic of "Sf. Pantelimon" Hospital in Bucharest. The criteria for the patients' inclusion in the group were the following: \- Liver cirrhosis as a background disease \- The presence of an abdominal surgical problem whose solution was imposed in the first 24 from admission as an emergency solution In order to assess the determinant factors of postoperative morbidity and mortality in these patients we considered the following: \- Evolutive stage of the liver disease \- Physiological status of the patient at the moment of operation \- Type and magnitude of the surgical procedure For the staging of the liver cirrhosis we choose Child -- Pugh classification. This classification is an useful instrument in emergency conditions because of its small number of parameters which can be rapidly evaluated -- serum albumin level, total bilirubin, prothrombin time INR, the presence and grade of ascites and grade of encephalopathy \[**[@R8]**\]. This score has proven its usefulness in many studies involving patients with liver cirrhosis, whether it was the rupture of esophageal varices, hepatocellular carcinoma, Budd-Chiari syndrome or portal subclinical encephalopathy \[**[@R9]**-**[@R12]**\]. In a study conducted over a period of 12 years Mansour used Child -- Pugh score as a prognosis factor for non hepatic abdominal surgery \[**[@R13]**\]. Despite the criticism that brings into question that the score is based upon components empirically chosen or that its variables are not independent predictors, studies show that the data obtained with this score have statistical significance \[**[@R14]**\]. In light of the above, Child-Pugh score remains an instrument which is easy to use in the evaluation of the prognosis of the patient with liver cirrhosis and its predictive value is at least comparable to some elaborated scores such as MELD \[**[@R15]**\], elements which have imposed their use in our study. The physiological status of patients and magnitude of surgery procedure were evaluated based on the POSSUM score. This score quantifies twelve significant and independent physiological variables which evaluate the physiological status of the patient at the time of surgery. The patient\'s respiratory and cardiac function are assessed by using information that can be obtained quickly and easily by clinical and laboratory common methods - the presence and degree of dyspnea and its degree of severity, chest X-ray appearance, presence of arrhythmias on the electrocardiogram, blood pressure, pulse, the presence of angina or an underlying heart disease requiring inotropic therapy or anticoagulant. In addition to these are the laboratory tests such as hemoglobin, WBC count, blood urea nitrogen, sodium and potassium. Moreover, the patient\'s age and Glasgow score are also considered. The values obtained for the physiological score calculated at the time of intervention can have values between 12 and 88 (**[Table 1](#F1){ref-type="fig"}**). {#F1} Surgical trauma is evaluated based on six factors of the severity of procedure. This is classified into four categories taking into account the extent of surgery, emergency or elective nature of the operation, amount of blood lost during surgery, duration, peritoneal contamination and the presence / extension of a neoplasm (Table 2). Once the score is known, it is possible to estimate the risk of morbidity (R2) and mortality (R1) by using two equations. For morbidity the equation is the following: Log (R2 / (1-R2) = -5.91 + (O.16 X physiological score) + (0.19 X severity score of the surgical procedure) For mortality the equation is the following: Log (R1 / 1 R1) = -7.04 + (0.13 X physiological score) + (0.16 X severity score of the surgical procedure). The physiological score was assessed at the time of surgery and the operative score at the moment of the patient's discharge. Comparing the obtained results with those estimated by using POSSUM score, we tried an appreciation of the influence upon postoperative outcome of the physiological status and evolution stage of liver cirrhosis in these patients. Statistical data analysis was based on X²(chi-square) test and the p value of less than 0,05 was considered statistically significant. {#F2} Results {#sec1-3} ======= Our group was composed of 115 patients, 45 females (39%) and 70 males (61%) and their ages ranged from 29 to 85 years, with an average of 60.27. The etiology of liver cirrhosis in patients represented in our group was especially chronic consumption of ethanol and hepatitis virus infections, groups including 102 cases - 60 and 42 patients, respectively. To these 8 cases of cardiac cirrhosis and other 5 in which the etiology could not be documented, being labeled as cryptogenic, were added. The classification of patients into Child groups depending on the evolutionary stage of liver disease was the following: 33 of them (28.7%) were classified in group Child A, 54 (47%) in Child B stage and 28 (24.3%) in group Child C(**[Fig. 1](#F3){ref-type="fig"}**). {#F3} The preoperative assessment based on the physiological POSSUM score resulted in dividing the patients into the following groups score: below 15,5 cases; 23 cases with a score between 16 and 18, 28 cases with 19 to 21 points, 32 cases with 22 to 24 points, 18 cases with 25 to 27 points, 7 cases with 28 to 30 points, 2 cases over 30 points (**[Fig. 2](#F4){ref-type="fig"}**). {#F4} The pathological conditions in our group of 115 patients who required emergent surgery were the following: 22 cases of complicated umbilical hernias, 12 complicated postoperative incision hernias, 8 complicated inguinal hernia, 35 acute cholecystitis, 14 peptic ulcer bleeding, 6 perforated ulcer, 3 penetrating abdominal wounds, 4 acute appendicitis, 2 traumatic ruptures of spleen, 2 choledocholithiasis with cholangitis, 7 bowel obstructions. Surgery procedures practiced in emergency conditions -- in the first 24 hours from admission -- on our group of patients were the following: splenectomies -- 2 cases, umbilical hernia repair -- 22 cases, inguinal hernia repair -- 8 cases, incision hernia repair -- 12 cases, suture of ulcer perforation -- 3 cases, excision of ulcer -- 2 cases, gastric resection -- 5 cases, gastrotomy with hemostasis \"in situ\" for bleeding ulcer -- 10 cases, resection of small intestine -- 3 cases, appendectomy -- 4 cases, right hemicolectomy -- one case, transverse colon segmental resection -- one case, Hartmann operations -- 2 cases, laparoscopic cholecystectomy -- 21 cases, open cholecystectomy -- 14 cases, laparotomy -- 3 cases, choledochotomy with Kehr drainage -- 2 cases. Calculating POSSUM operator score for our patients resulted in a classification into the following groups: 31 patients with a POSSUM operator score below 10, 42 patients with a score between 11 and 13; 14 patients had a score of 14 to 16; 26 cases with 17 to 19 points; 2 patients with 20 -- 22 points (**[Fig. 3](#F5){ref-type="fig"}**). {#F5} The estimated mortality risk in our group based on POSSUM score was the following: less than 5% - 22 patients (19%); risk between 6% and 10% - 37 patients (32%); risk 11% to 15% - 23 patients (20%); 16% to 20% mortality risk -- 9 patients (7,8%); 21% to 25% - 10 patients (8,7%); over 25% risk -- 14 patients (12%) (Fig. 4) {#F6} The average estimated risk of mortality for the entire group was 12.73%. The number of deaths observed in our group of patients was 52, representing 45.21%. This results in a report of cases observed / expected at 3.55 / 1 for the whole group. The bundling patients by Child evolutionary stages led to the observation that the reporting observed to the estimated results based on POSSUM score figure varies from one group to another (Table 3). {#F7} The group of patients in stage Child A of liver cirrhosis had a morbidity of 48.5% (16 cases) and a mortality rate of 9.7% (4 cases). Based on POSSUM score estimation, morbidity in these patients would have a percentage of 42.54 and a mortality rate of 9.1%, values which are quite similar to those registered by us. The ratio O / E for patients with liver cirrhosis in stage Child A was 1.1 for morbidity and 1,2 for mortality. The null hypothesis was that the evolutionary stages of liver cirrhosis had no influence on POSSUM score prediction ability. The statistical analysis based on X² test (\"goodness of fit\") for mortality occurred in patients belonging to Child class A, which resulted in the following value: x² =(1, n=33) =0.36, p\<0,05 The critical value for X2 is 3.84, placing the result in the acceptance area. Regarding the capacity of POSSUM score to estimate morbidity for our patients in stage Child A of liver disease, the statistical analysis based on X² test resulted in the following value: x²=(1 , n=33)=0.35, p\<0.05 Consequently, the predictive value of POSSUM scores in patients with liver cirrhosis in stage Child A was satisfactory both in terms of morbidity and mortality. We found an exponential increase in the mortality rate -- 29 deaths from 54 -- for patients in Child B stage of liver cirrhosis representing 53,7%. The estimated rate of mortality based on POSSUM score for this group of patients was of 14% - 8 deaths. Thus, the observed/estimated ratio was 3,8/1 and the value for x²=(1, n=54)=64,7, p\<0,05, placing the result in the rejection area. Of the 25 patients of Child B group who were discharged, 22 had postoperative complications. Patients in stage Child C of liver disease had the highest death rates -- 19 from a total of 28, representing 67,85%. Compared to a total of 4 deaths expected in this group based on POSSUM score prediction the ratio O/E is 4,75 and the value for X² in rejection area. All 9 patients belonging to Child C stage who survived surgery had postoperative complications requiring prolonged hospitalization. Discussion {#sec1-4} ========== Surgery for acute abdomen in a patient with liver cirrhosis is an unavoidable risk, but its quantification is particularly significant given that modern medicine is based on analysis, objectivity and rigorous control of medical care quality. The outcome of surgery depends not only on the skills of the operator, but is the result of several factors such as the physiological status of the patient, the condition requiring surgery, type of operation, but also the resources for preoperative and postoperative resuscitation which the hospital unit can provide. Operative risk in a patient with cirrhosis can be approximated in a wide range on the evolutionary stage of the liver disease \[**[@R13]**,**[@R16]**\]. For the final result, the magnitude of the surgery and the patient\'s condition at the time of intervention are important factors to consider. POSSUM score has a satisfying predictive value quantifying and integrating in its equations both the physiological parameters and the magnitude of the procedure. Cirrhotic patients undergoing abdominal surgery are at a high risk of hepatic decompensation due to reduced hepatic blood flow during such interventions \[**[@R17]**,**[@R18]**\]. Considering all these elements in a 5-year study on a group of 115 patients hospitalized and operated in emergency for various conditions that can be defined as \"acute abdomen\" we assessed the grade of influence of underlying liver disease, physiological status and amplitude of surgical procedure on the postoperative outcome. The data obtained in our study showed that when there is a functional satisfactory hepatic reserve, as in stage Child A of liver cirrhosis, the patient can tolerate the surgical act satisfactorily. At this stage of liver disease, patients can fit in the same category with the other patients in terms of predictive ability of POSSUM score. For the advanced stages of cirrhosis the high rates of mortality brought out the results from the prediction area provided by the POSSUM score. It suggests that patients in stages Child B and C of liver cirrhosis, emergency abdominal surgery has a major impact on the homeostasis and the resources for compensation are overloaded. Thus, in advanced stages, the liver disease has a prominent value for prognosis as against the POSSUM score. Conclusions {#sec1-5} =========== Surgery for acute abdomen in patients with liver cirrhosis is a therapeutic challenge and for the surgeon it means taking an inevitably risk. The outcome of the surgical procedure is a resultant of multiple factors and some of them can be quantified and integrated in valuable predictive scores such as POSSUM. According to our study, POSSUM score offers a satisfactory prediction for morbidity and mortality in emergency abdominal surgery for patients in compensated stages of liver cirrhosis -- Child A. For the advanced cirrhosis stages, Child B and C, high levels of mortality cannot be predicted by using POSSUM score and liver disease has a prominent value in prognosis of this category of patients.

1. Introduction {#s0005} =============== Severe myoclonic epilepsy of infancy (SMEI), also known as Dravet syndrome, is an epileptic encephalopathy that presents during the first year of life and is one of the most severe types of infant epilepsy that is resistant to drugs [@bb0005; @bb0010]. Patients with Dravet syndrome display multiple seizure types including tonic--clonic, myoclonic, absence, and focal seizures. In addition to epilepsy, SMEI is associated with cognitive delays and behavioral disorders [@bb0010]. De novo SCN1A mutations are a major cause of SMEI [@bb0015]. In addition, mutations in *SCN1B*, *SCN2A*, and *GABRG2* also cause SMEI, and recently, a Dravet-like phenotype in which *PCDH19* and *CHD2* genes are involved was described [@bb0020; @bb0025]. Pharmacoresistance is one of the major problems in SMEI, and many antiepileptic drugs do not seem to be able to control all the different kinds of seizures, including generalized tonic--clonic seizures (GTCS) and myoclonic, absence, and focal seizures, which characterize this syndrome. Eslicarbazepine acetate (ESL) is a relatively new antiepileptic drug which has been approved in 2009 by the European Medicines Agency and in 2013 by the US Food and Drugs Administration as adjunctive therapy in adults with partial onset seizures with and without secondary generalization. Eslicarbazepine acetate shows an important action on the blockade of voltage-gated sodium channel (VGSC). It has been recently reported that there are some differences between AEDs acting on VGSC that are also related to their different affinities for the channel based on the functional state of VGSC. In fact, three distinctive states have been shown for VGSC which include the resting state, the open state, and the inactivated state. The affinity of ESL for the inactivated state of these channels is similar to other AEDs (carbamazepine), while its affinity for the resting state is three times lower than the other drugs, thus preventing sustained repetitive neural firing but disturbing physiological mechanisms to a lesser extent. Another peculiarity of ESL has been reported by an experimental study which showed that this drug could reduce VGSC availability through enhancement of slow inactivation, as reported for lacosamide, while it seems not able to alter fast inactivation of VGSC, unlike carbamazepine (CBZ) and oxcarbazepine (OXC) [@bb0030]. In addition, the eslicarbazepine metabolite effectively inhibited high- and low-affinity CaV3.2 inward currents with greater affinity than CBZ [@bb0035]. We report a short-term open study about two patients affected by SMEI who both showed favorable response with regard to seizure frequency reduction when treated with ESL and good tolerability. 2. Methods {#s0010} ========== The first patient is a 17-year-old Italian boy with SMEI and a severe clinical outcome, in whom we identified a punctiform mutation (5053del AA fs1685X1691) in the α subunit of the neuronal sodium channel SCN1A gene. He started to have febrile and afebrile seizures at the age of 6 months. No drug treatment, including valproate, lamotrigine, carbamazepine, phenobarbital, vigabatrin, topiramate, clonazepam, ethosuximide, phenytoin, felbamate, nitrazepam, rufinamide, zonisamide, or lacosamide had proven effective ([Table 1](#t0005){ref-type="table"}). The best results in the reduction of seizure frequency was obtained with a combination of phenobarbital, clonazepam, and zonisamide, showing a reduction of about 50% of partial complex seizures during sleep and 70% of daytime GTCS. During this time, the EEGs and clinical and epileptic characteristics of this patient showed typical features of SMEI such as slowing of background EEG activity with multifocal and diffuse discharges prevalent in the frontal regions, myoclonic seizures at the age of 2 years and 6 months, ataxia, and progressive intellectual disability. At the age of 16 years and 6 months, we added ESL in combination with phenobarbital, clonazepam, and zonisamide at the initial dosage of 400 mg once daily and after two weeks at 800 mg/day. After just a few weeks from the start of therapy, we observed a complete response with disappearance of monthly GTCS and sleep frontal-partial complex seizures (with 3--4 occurrences per night previously observed). We did not observe adverse events during the treatment period of six months with a significant improvement in behavior, particularly oppositional defiant disorder and the maintenance of attention. The second patient is a 21-year-old Italian girl. At the age of 11, we had identified a punctiform mutation (931G \> T E311X) in the α subunit of the neuronal sodium channel SCN1A gene. The febrile and afebrile complex-partial seizures started at the age of 3 months. Similar to the first patient, the drug treatment for this patient included acetazolamide and stiripentol that had not been proven effective ([Table 1](#t0005){ref-type="table"}). In particular, at the age of 8, a combination of phenobarbital, clonazepam, and acetazolamide (250 mg/day) caused a decrease in seizure frequency of about 40%. The number of seizures was previously much higher and associated with several eyelid myoclonic seizures at a frequency of hundreds of occurrences daily and several massive myoclonias from the age of 3 years. Magnetic resonance imaging showed moderate brain atrophy, which did not progress during maturation, and brain interictal SPECT at the age of 2 showed hypoperfusion of the left frontal cortex. The EEG showed diffuse and multifocal paroxysmal activity that was more evident in the bioccipital regions. In the first year of life, the child showed unimpaired neuromotor development and hypotonia. She presented progressive intellectual disability, which became profound, as well as severe ataxia, making it impossible for her to walk. In addition, her speech was absent. At the age of 10, a VNS device was implanted without any results. At the age of 20, we added eslicarbazepine acetate to phenytoin, clonazepam, and valproate at the initial dosage of 400 mg once daily, which was increased after two weeks to 800 mg/day. We have also observed in this case good response with regard to seizure frequency reduction in just a few weeks after starting therapy. The girl does not present any daily seizures, just one seizure every four days currently. 3. Results {#s0015} ========== We observed a significant and fast effect using ESL as add-on therapy to other AEDs in both patients similar to what was reported in literature for adult patients with partial-onset seizures [@bb0040]. The peculiarity in our cases was the age when ESL therapy was started, 16 years and 6 months for the boy and 20 years for the girl, respectively, underlying the possibility of using ELS with good efficacy as an add-on therapy in young patients [@bb0045]. Of particular interest is ESL\'s efficacy in both patients with SMEI carrying SCN1A mutation because it is well known that seizure control is very difficult to achieve in this syndrome. The reason why ESL was effective in patients whose seizures are typically drug-resistant is not yet known; we could theorize that it could be due to the peculiar action of ESL on VGSC. Indeed, carbamazepine, which alters fast inactivation of VGSC, can worsen seizures and should not be used in patients with SMEI [@bb0045]. It is interesting to note that in our cases, ESL has been used in combination with different AEDs, although benzodiazepine was the only AED common in both cases. Thus, it does not seem plausible that the effectiveness of ESL in patients with SMEI is attributable to the association between different drugs, but by analyzing the drugs in particular, we can observe their similar mechanisms, especially those on GABAergic system and voltage-gated sodium channel. Another peculiarity of our report is the effective dose of ESL, which was a low recommended dosage. Moreover, we observed a fast response to the initial dose of ESL [@bb0050]. 4. Conclusions {#s0020} ============== Adjunctive eslicarbazepine led to seizure reduction in two patients with severe myoclonic epilepsy of infancy. Although this is a short-term open study, the results are promising but need confirmation. Conflict of interest {#s0025} ==================== The authors report no conflicts of interest. This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial-No Derivative Works License, which permits non-commercial use, distribution, and reproduction in any medium, provided the original author and source are credited ###### Response and time of administration of AEDs in both patients. AEDs Time of administration Response[a](#tf0005){ref-type="table-fn"} ----------------- ------------------------ ------------------------------------------- ---- --- Valproate 2 years 18 years 1 2 Lamotrigine 6 months 5 months 1 1 Clonazepam 10 years 12 years 2 3 Carbamazepine 9 months 6 months 1 1 Phenobarbital 12 years 1 month 3 0 Vigabatrin 3 months 4 months 1 1 Topiramate 1 month 6 months 0 1 Ethosuximide 1 year 8 months 1 1 Phenytoin 2 months 10 years 1 3 Felbamate 1 month 3 months 1 1 Nitrazepam 3 months 2 months 1 1 Rufinamide 2 months 3 months 1 1 Zonisamide 4 years 4 months 2 1 Lacosamide 3 months 1 month 1 0 Acetazolamide 4 months 6 months 1 1 Stiripentol 4 months 4 months 1 1 VNS -- -- 1 Eslicarbazepine 6 months 6 months 4 4 Number of daily seizures: 0 = increased daily seizures; 1 = unchanged seizures; 2 = reduction of seizures (less than 50% of frequency); 3 = reduction of seizures (less than 50--70% of frequency); 4 = reduction of seizures (more than 70% of frequency).

One and a half billion people live in conflict-affected and fragile states. At the last estimate in 2012, 172 million people were directly affected by war, including refugees, internally displaced people and those who were affected but did not flee.[@R1]^,^[@R2] Children are twice as likely to be malnourished and twice as likely to die by the age of five years in low-income countries affected by conflict compared with similar but stable countries. Their families are twice as likely to live without clean water.[@R3] Conflict does more than short-term damage; it decimates a country's infrastructure and impairs the social contract between the state and citizens. Food supplies are disrupted; health services collapse. Pregnant women and people who are ill do not receive the life-saving services they need.[@R4] Less often measured are the long-term consequences of conflict on people's mental health and social functioning. People made vulnerable by conflict are being bypassed by global progress. The World Bank warned in 2011 that no low-income, conflict-affected country was on course to achieve any of the millennium development goals (MDGs).[@R3] Indeed, four years later, of 55 conflict-affected and fragile states, 37 (67%) had met only two or fewer of the 15 MDG targets.[@R5] The inequity is not simply about the differences between stable and unstable countries. Even within countries, conflict-affected areas fare worse than areas with less or no conflict. In the Democratic Republic of Congo, for example, under-five mortality in the conflict-affected South Kivu province is nearly double that of Kinshasa province.[@R6] Despite this experience, the sustainable development goals (SDGs) for the year 2030 include barely more guidance on conflict than did the MDGs, which did not specifically mention conflict. SDG 16 explicitly recognizes the need to resolve conflict and mitigate its circumstances, but this intention does not translate into specific action points for other SDGs, such as SDG 3, which focuses on health. Unless we learn how to achieve the targets in conflict settings, the benefits of the SDGs will not reach many of the people who need them most. The first question is whether the list of 17 SDGs and 169 targets should be adapted by each country. There is experience of simpler, more modest goals being associated with greater success. In Afghanistan, the government and its partners worked together to adapt the MDGs to meet their needs, setting more realistic interim targets and agreeing on objectives and approaches adapted to the country's unique realities.[@R7] Methods of measurement need to be realistic too. Widely used mechanisms to monitor progress in health at a national scale, such as the demographic and health surveys and the multiple indicator cluster surveys, sometimes leave out whole conflict-affected areas and routinely exclude internally displaced persons and refugees.[@R8] There are alternatives, such as data collected by nongovernmental organizations that work in conflict zones. These data, which are collected to identify needs and monitor the progress of humanitarian interventions, are arguably underused for monitoring development goals.[@R8] Population data are also scarce among displaced and disrupted communities, although humanitarian organizations commonly conduct and update small-scale censuses of difficult-to-access areas. Using these sources, while not a solution to the problem, could be a first step towards better monitoring. We also need to be more realistic about the level of investment needed to effect even modest changes. Conflict-affected countries often have greater needs both in terms of capital costs to rebuild destroyed infrastructure and of recurring costs to operate in environments with transport and security challenges. Yet conflict-affected countries have received less investment than others. In 2012, for example, the Central African Republic received one-fifth the per-capita direct assistance for health that Malawi did.[@R9] With less than 200 km of paved roads in South Sudan, a country larger than France, air transport is often the only option. Overall, levels of international humanitarian funding routinely face a large shortfall: in 2014, 7.2 out of 18 billion United States dollars of the United Nations (UN) coordinated appeal for humanitarian action worldwide went unfunded.[@R9] Realistic goals, better measurement of progress and increased investment would help, but these will not be enough to make a difference without a fourth adaptation: the most difficult one. We need to change what we do and how we do it. Investment in scalable, cost-effective interventions, such as family planning, insecticide-impregnated bed nets and integrated community case management, have helped stable countries to make dramatic reductions in maternal and child mortality. Countries in crisis could and should benefit from similar public health interventions, but have often instead been served by short-term actions, such as provision of field hospitals and mobile clinics, which have higher costs, smaller scale and less potential for sustained impact. People already burdened by conflict receive aid that reaches fewer people, is more expensive and has a shorter impact than aid in non-conflict settings. This does not mean implementing the same large-scale, long-term health programmes in fragile states as in stable low-income countries. Interventions should be effective and cost-effective and have wide coverage but they need to be adapted to the context of conflict. For example, integrated community case management for childhood diseases, traditionally considered a development intervention, has been adapted and scaled up in conflict-affected areas to increase access to care. In South Sudan, for example, the International Rescue Committee (IRC) supports over 2600 community health workers, many of them in the most intense areas of conflict such as Unity state. These community workers have continued to work even when conflict has shut down formal health structures. Tools were developed for illiterate community health workers, in a setting in which literacy rates are very low. Supply-chain adaptations, such as the use of boats and a network of transit villages to move materials, were developed to address problems of flooding and the lack of infrastructure. Similarly, although family planning has nominally been included in the United Nations Population Fund (UNFPA) minimum initial service package for reproductive health, its family planning component has seldom been implemented in acute or chronic emergencies. In the last 10 years, a coalition of organizations with experience in reproductive health in conflict settings has demonstrated that modern and long-acting contraceptive methods can be provided at low cost and with high quality, and that there is demand for these, even in the most precarious and transient conditions.[@R10]^,^[@R11] As a result, the full range of contraception methods are being provided by Save the Children, CARE, IRC and other agencies as a standard in places like the eastern Democratic Republic of Congo and northern Uganda, even in the midst of security upheavals. In other areas, such as Zaatari camp in Jordan, the United Nations High Commissioner for Refugees and its partner agencies are meeting the demand for modern contraception which existed before the Syrian refugee crisis and which continues.[@R12] These approaches not only deliver proven, cost-effective interventions at much larger scale than most classic emergency interventions, but also help to mitigate the risks to humanitarian workers and health workers. Hospital workers in the Syrian Arab Republic and vaccine workers in Pakistan appear to have been deliberately targeted during recent conflicts. Attacks of this type are not new, but they are being increasingly documented. Community health workers are less visible than facility workers, and can provide vital services with less danger to themselves in settings where formal health workers are deliberately targeted. They consequently have more options for continuing services in conflict. Such new approaches to aid in conflicts will not happen without a new understanding of coordination that lasts beyond the acute phase of conflict, and that includes both acute and long-term actors. The cluster system, a UN-led, country-specific coordination system for acute emergencies, has helped to bring better geographical distribution of aid and better communication between agencies. What is needed, however, is coordination not just among acute responders but also among health actors with complementary mandates and spheres of action. Agencies with expertise in specific interventions, such as UNFPA or Planned Parenthood affiliates in family planning, need to support acute responders, such as Médecins Sans Frontières. Agencies just arriving into a conflict area need to collaborate with agencies within a country who can contribute greater knowledge of the local context and assets such as networks of community health workers. Establishing more effective public health responses will also require better coordination with governments ‒ a challenging task in many conflict settings, but an important one. Last but not least, an approach combining large-scale public health interventions with responses in conflict settings will require coordination across different kinds of donors and sometimes different teams within the same agency who may not be used to coordinating among themselves. As the world embarks on another 15-year enterprise in global aid planning, implementation and tracking, we owe it to populations affected by conflict -- who are some of the most vulnerable people on earth -- to apply public health principles to deliver better aid. Changing the politics that drive conflict is beyond the sphere of the global health community. However, it is well within our power to maintain our high standards of effectiveness and common sense regarding cost‒effectiveness, and to coordinate better between global health actors. With a smarter, more adapted, more ambitious approach to assistance in conflict settings, we have an opportunity to make the SDGs more effective and more equitable than previous development goals. None declared.

Background {#Sec1} ========== Turner's syndrome (TS) is depicted as a total or partial absence of one X chromosome, and occurs in approximately 1/2200 of live born females \[[@CR1]\]. Nearly 43-49 % of the patients are cases with classical TS who are monosomic for X chromosome (45,X). The remaining patients are mosaic cases carrying normal and abnormal cell lines together (most of them had 45,X/46,XX karyotype) (15-23 %), those with isochromosome in long arm of X chromosome (i(Xq)) (14 %), those with ring X chromosome (r(X)) (3-11 %) and those with 46,XX karyotype but having partial losses in one X chromosome (9 %). Y chromosome fragments are detected in 10-11 % of the cases \[[@CR2]--[@CR6]\]. Patients with classical TS demonstrate characteristic clinical features such as short stature, web neck, cardiovascular and renal abnormalities and hearing loss; and besides, gonadal dysgenesis in TS results in pubertal delay or failure, infertility and premature ovarian failure (POF) in most patients \[[@CR7], [@CR8]\]. Although correlation between genotype and phenotype is not well understood, mosaic cases present milder phenotypic abnormalities compared to those with 45,X karyotype \[[@CR9]\]. Mosaic TS patients are more likely to experience normal pubertal development, regular menstrual cycles and to conceive spontaneously compared to those with 45,X monosomy \[[@CR2]\]. Since mosaic patients are diagnosed following karyotype analysis due to recurrent pregnancy loss, repeated in vitro fertilization (IVF) failure and history of an abnormal offspring, our knowledge concerning reproductive and obstetric outcomes relies on case reports and case series \[[@CR4], [@CR10]--[@CR12]\] and more comprehensive studies investigating the fertility outcomes of these patients. The purpose of this study was to evaluate reproductive and obstetric outcomes of natural conception and IVF procedures in mosaic TS cases. Materials and methods {#Sec2} ===================== This retrospective study evaluated 706 female patients who underwent karyotyping between 2009--2013 in the laboratory of Medical Genetics Department of our tertiary health institution. The approval from the local ethics committee had been obtained prior to the initiation of the study. Informed consent from all individual participants included was obtained. Chromosomal analysis was performed by G-banding technique at high resolution. A hundred metaphases were counted for each patient and International System for Human Cytogenetic Nomenclature (ISCN, 2009) guidelines were used when performing karyotype analysis \[[@CR13]\]. Mosaicism ratios of the cases were calculated by proportioning total number of abnormal cell lines. In karyotype analysis, mosaic cell line ratio of ≤ %10 was defined as low-grade mosaicism, and \> %10 as high-grade mosaicism \[[@CR14]\]. Medical history regarding prior hormone therapy, prior assisted reproductive techniques attempts in infertility cases, obstetric outcomes in patients who had conceived through spontaneous or IVF cycles and perinatal outcomes were obtained by face-to-face interview or assessment of hospital medical records. All patients were assessed with physical examination and underwent a set of diagnostic tests including thyroid function tests, abdominal ultrasonography and echocardiography. Statistical analysis {#Sec3} -------------------- SPSS version 15.0 (Statistical Package for Social Science, Spss Inc. Chicago, IL, USA) was used for statistical analysis. Parametric variables were given as median (range), mean ± standard deviation or ± standard error. Chi-square test was used for the comparison of parametric variables. Non-parametric tests such as Mann--Whitney U test was used to compare non-parametric variables between cases with high-grade mosaicisim and low-grade mosaicism. Possibility of a total of 2-year fecundability was calculated at 6-month intervals by time-table analysis. The effect of menarche age, marriage age and mosaicism ratio on the time until spontaneous or IVF pregnancy was assessed by Cox-regression analysis. p \< 0.05 was considered as statistically significant. Results {#Sec4} ======= A total of 22 mosaic TS patients were extracted from 706 patients who underwent genetic karyotyping for varying indications including, recurrent implantation failure (5.1 %), recurrent pregnancy loss (defined as three or more consecutive miscarriage) (2.2 %), POF (2.2 %) and history of an offspring with any chromosomal or structural abnormality (4 %). Clinical characteristics of our study population were presented in Table [1](#Tab1){ref-type="table"}. Menarche was achieved with hormone replacement therapy in three cases at the ages of 16,17 and 18 years and continued regularly. Menstruation was regular in 18 cases at the time of study enrollment whereas it was irregular in the two and the remaining two, who were diagnosed as POF at the ages of 28 and 38, were on hormone replacement therapy. There was one case with a short stature (\<150 cm) phenotype and no cases with cardiovascular or renal abnormality, hearing loss and mental retardation among the included patients. The following systemic disorders were diagnosed in the patients; hypothyroidism in three, type 2 diabetes in three, asthma in one and generalized anxiety disorder in one patients. Uterine hypoplasia was observed in one case. In addition, one patient underwent hysteroscopic septum resection for uterine septum, another case underwent Strasmann metroplasty for bicornuat uterus and one underwent hysteroscopy and cavity expansion with fundal and lateral incisions for T-shaped uterus.Table 1Clinical characteristics of the study groupPatients age at the time of enrollment (years)^a^37 (26--47)Patients age at the time of diagnosis (years)^a^34.5 (18--46)Age of menarche (years)^a^13 (11--18)Age at marriage (years)^a^25 (15--40)Patients age at first pregnancy (years)^a,b^23 (18--32)Time from the marriage to the first conception (months)^a,b^12 (6--49)Height of the patients (cm)^a^163 (132--174)Body mass index at the enrollment (kg/m^2^)^c^28.43 ± 1.21^a^Data presented as median and range, ^b^In the cases who conceived spontaneously, ^c^ data presented as mean ± standart error A total of 22 female patients, who were diagnosed as mosaic TS after karyotyping and who attended our IVF clinic with the diagnoses of recurrent implantation failure (*n* = 10), recurrent pregnancy loss (*n* = 9), infertility due to POF (*n* = 1), and history of a prior offspring with a chromosomal abnormality (*n* = 2), were included in the study. Out of 22 patients, five despite IVF treatment and one who never sought treatment, could not ever conceive. Out of remaining 16 cases, 11 conceived spontaneously and five conceived following IVF cycles; resulting in a total of 52 pregnancies of which 17 (32.7 %) resulted in live birth and 35 (67.3 %) resulted in abortion. Of 22 mosaic TS patients' karyotypes, 17 were 45,X/46,XX and five were 45,X/46,XX/47,XXX. There were no cases including 45,X/46,Xr(X); 45,X/46,X(i(Xq)); Y chromosome fragment or 46,XX karyotype with structural abnormalities in X chromosome. One patient was determined to have 45,X/46,XX inv(9)p11q13 karyotype. The comparison regarding the number and the percentages of pregnancies, miscarriages and live births between the different karyotypes of TS were presented in Table [2](#Tab2){ref-type="table"}.Table 2Comparison of the ratios of live birth and miscarriage or terminated fetus between groupsMaternal karyotype45,X/46,XX (*n* = 17)45,X/46,XX/47,XXX (*n* = 5)All cases (*n* = 22)pNo of pregnancies (n)351752.678Live birth (n (%))10 (28.6)7 (41.2)17 (32.7).611Miscarriage (n (%))25 (71.4)10 (58.8)35 (67.3).712*p* \< .05 Statistically significant Mosaic cell line ratio was below 10 % in 17 and above 10 % in five cases. The comparison regarding the number and the percentages of pregnancies, miscarriages and live births between high-grade and low-grade mosaicism cases were presented in Table [3](#Tab3){ref-type="table"}. Miscarriages included 29 spontaneous abortions, four biochemical abortions and two induced abortions (due to anencephaly and trisomy 21). Mean abortion week was found to be 8.16 ± 2.98 (±SD) in cases who experienced spontaneous abortion (Biochemical abortions were excluded from the calculation).Table 3Comparison of the ratios of live birth and miscarriage or terminated fetus between low and high grade mosaic cell line groupsMosaic cell line ratioCases with low grade mosaic cell line (*n* = 17)Cases with high grade mosaic cell line (*n* = 5)All cases (*n* = 22)pNo of Pregnancies (n)46652.062Miscarriage (n (%))30 (65.2)5 (83.3)35 (67.3).468Live Birth (n (%))16 (34.8)1 (16.7)17 (32.7).127*p* \< .05 Statistically significant A total of 82 embryo transfers were performed on 10 patients during 35 intracytoplasmic sperm injection (ICSI) treatment cycles. Five patients could not conceive despite ICSI and five cases could conceive following ICSI treatment. Out of these pregnancies, one clinical pregnancy resulted in spontaneous abortion, two pregnancies resulted in biochemical abortion and two live babies were taken home. Implantation rate per cycle was 3.7 % (3/82), clinical pregnancy rate was 8.6 % (3/35) and take home baby rate was 5.7 % (2/35). 90.4 % of the pregnancies (*n* = 47) (88.2 % of live born babies (*n* = 15)) have occurred in spontaneous cycles. Perinatal outcomes of the 17 pregnancies that resulted in live birth are presented in Table [4](#Tab4){ref-type="table"}. Mean birth weight of these newborns was found as 3355 ± 140 gr (±SE).Table 4Perinatal outcomes of the pregnancies that resulted in live birthNumberPercentageRoute of deliveryVaginal952.9 %Abdominal (C/S)847.1 %Fetal genderFemale1270.6 %Male529.4 %Adverse perinatal outcomesIUGR15.9 %P.previa15.9 %GDM211.8 %C/S, cesarean section; IUGR, intrauterine growth restriction; P.previa, placenta previa; GDM, gestational diabetes mellitus Prenatal or postnatal cytogenetic examination was performed in three pregnancies, two as prenatal and one as postnatal investigations. Out of two prenatal cytogenetic examinations, one patient was diagnosed as Trisomy 21 and subsequently underwent a pregnancy termination; whereas the other revealed a normal karyotype. The postnatal investigation was of a patient who underwent karyotype analysis at the age of 15 due to mental retardation and a deletion was detected between the regions of 18q21.3 and q23. None of the miscarriages were evaluated genetically from the abortus materials. Time table analysis revealed that pregnancy hazard rate within the first 2 years at 6-month intervals was found to be 0.01 in the first 6 months, 0.04 in the second 6 months, 0.02 in the third 6 months and 0.01 in the last 6 months. Neither spontaneous nor IVF pregnancy was detected beyond 60th month of marriage. COX regression analysis revealed that marriage age, menarche age and mosaicism ratio did not have an effect on the time until first pregnancy (*p* = .685; *p* = .350 and *p* = .149, respectively). Discussion {#Sec5} ========== Gonadal dysgenesis in women with Turner Syndrome might depend on chromosome pairing failure during meiotic prophase, causing failure in synaptic formation at the zygotene and oocyte loss \[[@CR15]\]. Majority of germ cells, which trigger spontaneous puberty in 10-30 % and provide pubertal development, start to diminish in the third month of intrauterine life, resulting in only 5-10 % of affected patients could menstruate regularly \[[@CR16]--[@CR19]\]. POF is another frequent clinical feature of TS and mean age of menopause was reported to be 29.3 years \[[@CR18]\]. However; the degree of gonadal dysgenesis depends on the size of impaired regions of homologous chromosomes. Severe pairing failures induce the degeneration of all oocytes prior to puberty and are associated with rimary amenorrhea and poor sexual development, whereas mild pairing failures contribute to the survival of a considerable number of oocytes until puberty, leading to secondary amenorrhea and secondary impaired sexual development \[[@CR15]\]. Thus, it is possible that puberty and reproductive capacities are less affected or even preserved in TS cases with mosaic karyotype \[[@CR19]\]. When compared to the classical form, mosaic karyotype TS cases are more likely to present spontaneous puberty, normal levels of serum sex steroids and gonadotropins and follicles in ovarian biopsies \[[@CR20]\]. In addition, the chance of spontaneous conceiving in women with TS was reported as 2-10 %, most of which are the cases of mosaic pattern and those with 45,X monosomy are candidates for oocyte donation \[[@CR16], [@CR21], [@CR22]\]. X chromosome monosomy and mosaicism are encountered in 1.5 % of all amenorrhea cases although the incidence of X chromosome mosaicism in the general population still remains challenging \[[@CR23]\]. In our study, we detected mosaic Turner karyotype in 2.2 % of the cases with a diagnosis of recurrent pregnancy loss and 5.1 % of the cases with recurrent implantation failure. Our study results are compatible with the previously published reports which indicated the rate of mosaic Turner karyotype in the recurrent pregnancy loss groups as 2.6 % \[[@CR24]\]. X chromosome mosaicism ratio in female partner of ICSI candidate couples was reported to be 2.77-4.12 % whereas 45X/46XX mosaicism in couples with IVF failures due to severe male infertility and fertilization failures were reported to be 3.5-9.6 % \[[@CR25]--[@CR28]\]. Simpson et al. suggested that mosaicism has been underestimated as a cause of repeated failure in assisted reproduction \[[@CR29]\]. The median age of marriage was 25 (15--40) and median age of first pregnancy in spontaneously pregnant women was 23 years (18--32) and these results are in agreement with the previous studies reporting median age of first pregnancy in TS syndrome cases as 23.5-27.2 years \[[@CR4], [@CR10]\]. ROC analysis revealed that marriage age had no impact on the chance of conceiving. Moreover, Cox-regression analysis demonstrated that the age of menarche, age of marriage and mosaicism ratio did not affect the time-to-first pregnancy. There were no significant difference between high-grade and low-grade mosaicism cases in terms of pregnancy, time-to-first spontaneous pregnancy, abortion and take home baby rates in our study. Several previously published reports indicated a correlation between mosaicism ratio and phenotypic abnormalities and reproductive capacity of the patients; on the contrary, some other publications did not find a consistent relationship \[[@CR2], [@CR30]\]. Scholtes et al. showed a correlation between mosaicism and a low implantation rate \[[@CR31]\]. In contrast, Sonntag et al. could not find any significant impact of low-grade mosaicism on the course or outcome of ICSI in 20 couples \[[@CR14]\]. No required minimum percentage of abnormal cells has been established to define true versus "low-grade" mosaicism. Thus, some studies disregard the importance of low-grade mosaicism \[[@CR26]\]. In our study, first two-year fecundability analyses revealed that 5 % of the cases experienced their first pregnancy within 6 months and 8 % within first 2 years. No pregnancy was detected after 60 months of marriage. With the expectation of reduction in ovarian reserve with advanced age in TS, our results revealed that chance of fertility in cases with mosaic karyotype was high at younger ages but their chance will decrease if they do not conveive within the first 2 years. Majority of TS cases who conceive spontaneously are known to have mosaic cell lines. In the study by Birkebaek et al., evaluating 410 Danish women with TS, 27 out of 31 women who could spontaneously conceive, at least once, had a TS diagnosis \[[@CR4]\]. Similarly in the study by Bryman et al. there were 45,X/46,XX mosaicism in 25 out of 27 Swedish women with TS. While 23 cases had a spontaneous pregnancy, three cases conceived by assisted reproductive technologies \[[@CR32]\]. Clinical pregnancy rate and implantation rate were reported to be 46 % and 30 %, respectively, by fresh embryo transfer in oocyte donation cycles, and 28 % and 19 %, respectively, in frozen embryo transfers in TS cases \[[@CR33]\]. Pregnancy rates of these cases were found to be comparable with other women in donation programs or probably lower due to diminished endometrial receptivity \[[@CR34], [@CR35]\]. Many studies have shown that TS women who are able to conceive are at increased risk for miscarriage, stillbirths and malformed babies \[[@CR7], [@CR10], [@CR36], [@CR37]\]. Tarani et al. analyzed 160 spontaneous pregnancies in 74 women with TS and reported that 67.3 % of the pregnancies with 45,X/46,XX karyotype and 70.8 % of the pregnancies with 45,X/46,XX/47,XXX karyotype resulted in miscarriage or malformed fetus \[[@CR10]\]. Similarly, Bryman et al. have reported that 45 % of cases with mostly 45,X/46,XX mosaicism who conceived with their own oocytes ended with miscarriage and 10 % of them with induced abortion \[[@CR32]\]. Moreover, Kuo et al. have reported miscarriage rate as 68.6 % in patients with diminished ovarian reserve and 44.1 % in patients without diminished ovarian reserve among cases who had a history of prior recurrent spontaneous abortions with X chromosome mosaicism \[[@CR38]\]. Among our patients, 67.3 % of the pregnancies were resulted in abortion or termination. However karyotyping was performed for nine cases in our study group due to recurrent pregnancy loss. This condition was a reason of how we detected a high abortion rate and besides, this rate is quite higher than the abortion rate of 10-15 % in the population \[[@CR39]\]. Miscarriages that are frequently seen in TS cases are explained by chromosomal abnormalities in fetus, autoimmune disorders, ovarian and uterine factors \[[@CR10], [@CR11], [@CR38], [@CR40]\]. Aborted fetuses of TS women or their live born children are more susceptible to trisomy 21 (4 % vs 0.4 %, respectively) and TS (15 % vs 0.5 %, respectively) risks compared to general population \[[@CR10]\]. Hereditary in nature was reported in mosaic TS cases and especially cases with ring chromosome \[[@CR10], [@CR41]\]. Singh et al. revealed in their study investigating 97 conceptions in 31 pregnant women with sex chromosome mosaicism that 75 % of the fetuses were abnormal and 50 % of these pregnancies were resulted in spontaneous abortion and 25 % had genetic or congenital abnormalities \[[@CR42]\]. Besides, it was reported that abnormal karyotype ratio was increased from 42.9 % up to 73.7 % in abortion samples in the presence of diminished ovarian reserve \[[@CR38]\]. Birkebaek et al. detected chromosomal aberrations in 6 of 25 pregnancies who underwent prenatal or postnatal cytogenetic analysis among TS cases with classical and mosaic forms \[[@CR4]\]. In our study fetal karyotype analysis was not performed in all pregnancies terminated by abortion, hence the exact abnormality rates cannot be predicted however it seems likely to be higher. Cerebral paresis, neuropsychological disorder, aortic coarctation, cleft lip and palate and congenital tumor were detected in 5 (7 %) of 68 children born to women with TS \[[@CR32]\]. Even mosaic, preimplantation diagnosis, chorionic villus sampling or amniocentesis should be recommended for the patients undergoing infertility treatment if pregnancy is planned with their own oocytes since biological children of TS women are under risk for chromosomal abnormalities; and their children should be investigated for birth defects after delivery. Uterine hypoplasia and related reduced uterine perfusion secondary to significant changes in utero-ovarian vascular anatomy, and subclinical uterine abnormalities in TS cases have been implicated in the etiology of miscarriages \[[@CR33], [@CR38], [@CR43]\]. Although abortion rate is higher in pregnant women conceived with their own oocytes, use of donor oocytes does not reduce pregnancy loss rate in mosaic TS cases who underwent IVF (45 % with their own oocytes vs 26-30 % with donor oocytes) \[[@CR32], [@CR44]\]. This condition can be suggested as the evidence of the effect of diminished endometrial receptivity as well as oocyte-associated factors in TS cases. In our study group, uterus hypoplasia was present in one case and surgery-corrected uterine abnormality was present in three cases. It has been reported that uterine size were often normal in cases with mosaic karyotype and that they experience spontaneous puberty \[[@CR45], [@CR46]\]. Khastgir et al. have reported bicornuat uterus in four (13.8 %) of 29 TS cases, 10 of whom had mosaic karyotypes and miscarriage rate associated with uterus hypoplasia, bicornuat uterus and low fertilization rate was 50 % \[[@CR34]\]. Kuo et al. have reported that they detected uterine abnormality in 16.7 % of mosaic X-chromosome aneuploidies with a history of recurrent spontaneous abortions and 5.2 % in the control group \[[@CR38]\]. In our study, we found the ratio of Mullerian abnormality in all mosaic TS cases as 13.6 % and all pregnancies had terminated with abortions before the surgical corrections of the malformations. However, after the surgical corrections, four pregnancies were achieved and three terminated with delivery. These results suggest that the rate of uterine abnormalities in TS cases is high and that they may benefit from the surgery. TS patients may experience complications during pregnancy due to congenital malformations and endocrine diseases, and should be evaluated for the presence of these pathologies before IVF or pregnancy \[[@CR7]\]. Gestational diabetes mellitus was present in two (11.8 %) of 17 pregnancies that achieved live birth. These rates are close to those reported by Bryman et al. who affirmed a pregnancy induced hypertension and gestational diabetes rate of 5 % in TS cases, similar to the rate in general population \[[@CR32], [@CR47], [@CR48]\]. In our study population, there was one pregnancy with a history of SGA fetus delivery (5.9 %) and none of the cases had preterm delivery. In TS, preterm delivery rate was reported between 8--37.1 % and low birth weight was between 8.8-27.5 % \[[@CR49], [@CR50]\]. Given that the chromosomal aberrations arise possibly from the transmission of the imbalances in the genetic arrangements in mother to the offsprings (10), genetic counseling should be offered to all who conceived with autologous oocytes (11). Preimplantation genetic diagnosis (PGD) may improve the chance of conceiving in patients with recurrent ART failure and TS (12). Limitations of our study include its retrospective design, presence of some data that are based on patients' memory, absence of a control group and inability to perform a karyotype analysis in all newborns. In the context of lack of a control group, we should emphasize that it was not possible to constitute a control group since the present was not a population-based study. In addition, it is obvious that it is difficult to conduct such a study in this issue in a prospective fashion. Besides, maternal age significantly affects gains and losses in sex chromosomes \[[@CR51]--[@CR53]\]. A significant correlation was reported between maternal age and incidence of 45,X cell after 51 years old, and incidence of 45,X cell was reported to be 3.2-5.1 % among the women older than this age \[[@CR52], [@CR54]\]. Although cases in our study were younger than 46 years old, mosaicism ratio in some cases, whose pregnancy age was young but diagnosis age was advanced, may not reflect exact karyotype profile. Moreover, in the study by Hanson et al. percentage of cases with mosaic karyotype was found to be 45 % only with karyotyping but it reached to 70 % when FISH was used \[[@CR3]\]. Mosaicism ratio was given by the assessment of 100 metaphase plaques in all cases in our study, and it should be considered that FISH method was not used while evaluating our results. Last, the results obtained in the present study might have been affected by the non-mosaicism-related individual factors given that some patients included had more than one pregnancy as we stressed that 16 cases experienced 52 pregnancies. Conclusion {#Sec6} ========== Only a small proportion of the mosaic TS patients conceive in the first 2 years of marriage, hence any possible interventions should be considered within this period and at as much younger ages as possible. In TS patients who conceived, only 5.7 % take home baby whereas 67.3 % abort. Age of menarche and age of marriage appear not to have any impact on the chance of conceiving. Mosaicism ratio does not affect the time to the first pregnancy. Patients should be informed about high abortion rates after pregnancy. TS : Turner's Syndrome IVF : In Vitro Fertilization ICSI : Intra Cytoplasmic Sperm Injection FISH : Fluorescent In Situ Hybridization **Competing interests** This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors. The authors stated that there are no conflicts of interest regarding the publication of this manuscript. The authors also state that they have had full control of all primary data and that they agree to allow the Journal to review their data if requested. **Authors' contribution** ED conceived of the study, participated in the design, obtained the data and drafted the manuscript. YÇ participated in the design, obtained the data. YC participated in the design, obtained the data, performed the statistical analysis. EU carried out the genetic analysis. ÖÖ participated in the design and drafted the manuscript. EÇ performed the statistical analysis and edited the manuscript. All authors read and approved the final manuscript.

Measurement of cerebral blood flow (CBF) is of interest, both in the clinical setting and for research. Absolute quantification of global CBF is of key importance when investigating factors affecting the entire brain, eg, altered physiological states or aging.[1](#jmri25442-bib-0001){ref-type="ref"}, [2](#jmri25442-bib-0002){ref-type="ref"}, [3](#jmri25442-bib-0003){ref-type="ref"} Different invasive and noninvasive methods have been used to measure global CBF.[4](#jmri25442-bib-0004){ref-type="ref"}, [5](#jmri25442-bib-0005){ref-type="ref"} Phase‐contrast mapping (PCM) magnetic resonance imaging (MRI) allows measurement of total flow in the cerebral arteries, and calculation of global CBF by subsequently normalizing to brain volume.[6](#jmri25442-bib-0006){ref-type="ref"} Measurement of global CBF by PCM MRI has a number of advantages compared to other techniques. It is completely noninvasive, fast, and can be combined with other noninvasive MRI techniques for absolute quantification of global cerebral metabolic rate of oxygen[7](#jmri25442-bib-0007){ref-type="ref"}, [8](#jmri25442-bib-0008){ref-type="ref"} or be used for scaling of regional CBF measured by arterial spin labeling.[9](#jmri25442-bib-0009){ref-type="ref"}, [10](#jmri25442-bib-0010){ref-type="ref"} The technique has been used in a number of studies on variation in CBF in healthy individuals[11](#jmri25442-bib-0011){ref-type="ref"}, [12](#jmri25442-bib-0012){ref-type="ref"} and in larger population‐based studies of aging showing that global CBF decreases with age and that decreased global CBF is associated with accelerated signs of brain aging and increased all‐cause mortality.[2](#jmri25442-bib-0002){ref-type="ref"}, [3](#jmri25442-bib-0003){ref-type="ref"}, [6](#jmri25442-bib-0006){ref-type="ref"}, [13](#jmri25442-bib-0013){ref-type="ref"}, [14](#jmri25442-bib-0014){ref-type="ref"}, [15](#jmri25442-bib-0015){ref-type="ref"} Values of global CBF obtained by PCM MRI in healthy subjects[6](#jmri25442-bib-0006){ref-type="ref"}, [16](#jmri25442-bib-0016){ref-type="ref"} are generally in good agreement with accepted textbook CBF values of ∼50 mL/100 g/min.[17](#jmri25442-bib-0017){ref-type="ref"}, [18](#jmri25442-bib-0018){ref-type="ref"} Previous studies have also shown excellent reproducibility for CBF measurements in vivo[12](#jmri25442-bib-0012){ref-type="ref"}, [16](#jmri25442-bib-0016){ref-type="ref"}, and shown PCM MRI to be accurate for measuring flow in phantoms[19](#jmri25442-bib-0019){ref-type="ref"} and in large vessels.[20](#jmri25442-bib-0020){ref-type="ref"} In order to confidently interpret the physiological significance of these measurements, in vivo validation of the accuracy of PCM MRI for CBF measurements is required. However, comparative studies with accepted reference methods are generally lacking. For human studies, ^15^O‐H~2~O positron emission tomography (PET) is generally considered the best available method for absolute CBF measurements.[9](#jmri25442-bib-0009){ref-type="ref"}, [21](#jmri25442-bib-0021){ref-type="ref"}, [22](#jmri25442-bib-0022){ref-type="ref"}, [23](#jmri25442-bib-0023){ref-type="ref"}, [24](#jmri25442-bib-0024){ref-type="ref"} One previous study failed to show a correlation of CBF measured by PCM MRI and ^15^O‐H~2~O PET, but these measurements were obtained months apart and did not account for a number of important physiological covariates.[16](#jmri25442-bib-0016){ref-type="ref"} The aim of the present study was to validate PCM MRI for measurement of global CBF by same‐day measurements of global CBF by PCM MRI and ^15^O‐H~2~O PET. Materials and Methods {#jmri25442-sec-0006} ===================== Twenty‐two healthy males (mean age: 27.4 years, range: 18--40 years) participated in the study. The measurements were obtained as a part of a placebo‐controlled, crossover study investigating the effect of erythropoietin on cerebral metabolism. Measurements from PCM MRI have previously been published.[8](#jmri25442-bib-0008){ref-type="ref"} In the present analysis only data during placebo treatment are included. The study was approved by the Danish National Committee on Health Research Ethics (H‐4‐2012‐167) and was conducted according to the Declaration of Helsinki. Written informed consent was obtained from all participants. General Experimental Setup {#jmri25442-sec-0007} -------------------------- All experimental measurements in each subject were performed on the same day. Height and weight were measured at the time of inclusion in the study. After an overnight fast, a short catheter was inserted in the radial artery of the nondominant hand for blood sampling. After the ^15^O‐H~2~O PET scans, the subjects had a small meal before being transported to the MRI facility. The PET and MRI scans were performed 2--6 hours apart. Before the PET scans a venous blood sample was obtained and analyzed for hemoglobin. Arterial blood samples were obtained between the two PET scans and again after the MRI CBF measurement, and analyzed for oxygen saturation (SaO~2~), and partial pressures of O~2~ (PaO~2~) and of CO~2~ (PaCO~2~) using a Radiometer ABL800 Flex system (Radiometer, Copenhagen, Denmark). The arterial blood sample obtained at the MRI session was also analyzed for hemoglobin. PET {#jmri25442-sec-0008} --- PET scans were performed on a Siemens High Resolution Research Tomograph (HRRT) brain PET scanner (Siemens, Knoxville, TN).[25](#jmri25442-bib-0025){ref-type="ref"}, [26](#jmri25442-bib-0026){ref-type="ref"} The scanner had an axial field of view of 25 cm and a near isotropic resolution of 2 mm. During PET scanning the subject\'s head rested in a foam‐cushioned headrest, and a head strap was used to minimize head movement. Initially, a 6‐minute transmission scan with a rotating ^137^Cs single‐photon point source was performed for attenuation correction. Approximately 800 MBq of ^15^O radiolabeled water (half‐life: 123 sec) produced online was injected as a bolus using an Automatic Water Injection System (Scansys Laboratorieteknik, Værløse, Denmark). Arterial blood sampling was initiated 15 seconds before isotope injection. Emission scans were acquired after injection of intravenous bolus of tracer for 7 minutes in dynamic frames of 1 × 30 seconds, 18 × 5 seconds, 9 × 10 seconds, 10 × 15 seconds, and 2 × 30 seconds. Radioactivity concentration in the arterial blood was continuously measured by an automatic blood sampling system and drawn at 8 mL/min (Allogg ABSS, Mariefred, Sweden). The detectors in the system were cross‐calibrated against the PET scanner and the sampling frequency was 2 Hz. The inner diameter of the tube connected to the arterial catheter was 1.2 mm. The clocks of the scanner and sampling system were synchronized. Two consecutive scans were acquired for each subject with an interval of at least 10 minutes between the two injections to allow for washout and isotope decay. Scans were reconstructed using 3D‐ordered subset expectation maximum and point spread function (3D OSEM‐PSF).[27](#jmri25442-bib-0027){ref-type="ref"} Each map consisted of 207 image planes in a 256 × 256 matrix with an isotropic voxel size of 1.22 × 1.22 × 1.22 mm^3^. All images were corrected for randoms, scatter, attenuation (TXTV method),[28](#jmri25442-bib-0028){ref-type="ref"} decay and dead time, and filtered with a 3D Gaussian 5 mm filter. Postprocessing {#jmri25442-sec-0009} -------------- Quantitative regional CBF‐maps were calculated using a 1‐tissue 2‐compartment model: $$\frac{dC_{t}\left( t \right)}{dt} = K_{1}C_{a}\left( t \right) - k_{2}C_{t}\left( t \right)$$where *C* ~*t*~ is the tissue compartment concentration, *C* ~*a*~ is the arterial concentration, *K* ~*1*~ is the influx rate constant, which describes the unidirectional clearance of water from the blood to the tissue in mL/100 g/min, scaled to perfusion by a factor of one when assuming full extraction of water, and *k* ~*2*~ is the efflux rate constant (in min^−1^). The model also accounts for the contribution to the measured voxel concentration *C* ~*tot*~ from the vascular volume component *vB*: $$C_{tot}\left( t \right) = \left( {1 - vB} \right)C_{t}\left( t \right) + vBC_{a}\left( t \right)$$where *C* ~*tot*~ is the measured tissue time activity and *vB* is the blood volume fraction. The parametric maps were calculated using linear ridge regression with a spatial constraint parameter to increase signal‐to‐noise ratio (SNR), as described by Zhou et al.[29](#jmri25442-bib-0029){ref-type="ref"} The arterial input curves used in modeling were corrected for dead‐time, decay, delay, and dispersion. The calculation was done using the software PMOD 3.0 (PMOD Technologies, Zürich, Switzerland). Using the software PVElab,[30](#jmri25442-bib-0030){ref-type="ref"} each regional CBF‐map was coregistered to the volume segmented brain tissue mask from the 3D *T* ~1~‐weighted anatomical MR‐scan (Fig. [1](#jmri25442-fig-0001){ref-type="fig"}), and global CBF was calculated as the mean of all brain voxels. {#jmri25442-fig-0001} MRI {#jmri25442-sec-0010} --- ### Flow Measurement {#jmri25442-sec-0011} MR scans were performed on a 3T Philips Achieva MRI scanner (Philips Medical Systems, Best, The Netherlands) using a 32‐channel phase array head coil. The blood velocity in the carotid and vertebral arteries were measured using a through‐plane phase‐encoding technique. The sequence acquires a reference phase image and a velocity sensitive phase image by using a bipolar gradient in the slice‐selection direction. By subtracting the velocity sensitive phase image with the reference phase image, a phase‐contrast map is calculated. The phase‐contrast maps can be scaled to velocity according to the velocity‐encoding factor (V~enc~). Based on an initial 2D inflow angiogram, the measurement slice was positioned orthogonal to the carotid and vertebral arteries (Fig. [2](#jmri25442-fig-0002){ref-type="fig"}a). The imaging parameters for the sequence were: field of view (FOV) = 240 × 240 mm^2^, voxel size = 0.75 × 0.75 × 8 mm^3^, 1 slice, TE = 7.33 msec, TR = 27.63 msec, flip angle = 10°, 10 repeated measures, and total scan time = 1 minute 42 seconds. In order to reduce scan time, cardiac triggering was not applied. When acquiring PCM MRI without cardiac triggering, the *k*‐space is random‐sampled in an interval including multiple cardiac cycles causing time averaging of the velocity.[19](#jmri25442-bib-0019){ref-type="ref"}, [31](#jmri25442-bib-0031){ref-type="ref"} A V~enc~ of 100 cm/s was applied in order to avoid underestimation of flow velocities due to aliasing of the phase at high velocities. When performing measurements without cardiac triggering, potential aliasing from high velocities is not clearly visible because the average velocity is measured. The V~enc~ was therefore chosen around the upper normal peak velocity values in the internal carotid and vertebral arteries in young subjects while maintaining a reasonable dynamic range.[32](#jmri25442-bib-0032){ref-type="ref"} {#jmri25442-fig-0002} ### Postprocessing {#jmri25442-sec-0012} The total flow was calculated by measuring the mean velocity and the cross‐sectional area of the cerebral arteries by drawing regions of interests (ROI) corresponding to each cerebral artery (Fig. [2](#jmri25442-fig-0002){ref-type="fig"}b). The ROIs were initially drawn manually on the magnitude image of the first measurement and then copied to the corresponding velocity map. Only voxels with a positive mean velocity were included. ROIs were subsequently copied to the following measurements and inspected for misalignment from possible motion and corrected by moving the ROI if necessary. Cross‐sectional area and diameter of the cerebral arteries were calculated from the ROIs. Flow was calculated for each measurement by multiplying mean velocity with the cross‐sectional area and integrating over time. The total flow of the four arteries was normalized to whole‐brain tissue weight to attain quantitative physiological global CBF values in mL/100 g/min. For comparison with PET, the average of all 10 measurements was used. ### Anatomical Scan {#jmri25442-sec-0013} An anatomical scan for segmenting brain tissue was obtained with a 3D *T* ~1~‐weighted turbo field echo sequence (FOV = 241 × 180 × 165 mm^3^, voxel size = 1.09 × 0.81 × 1.1 mm^3^, TE = 2.78 msec, TR = 6.9 msec, flip angle = 9°). The FSL BET and FAST tools (FMRIB Software Library, Oxford University, Oxford, UK)[33](#jmri25442-bib-0033){ref-type="ref"} were used to produce a whole‐brain tissue mask including cerebral hemispheres (excluding the ventricles), cerebellum, and the brainstem. The mask was inspected (by coauthor M.B.V.) and manually edited if necessary. The same brain volume mask was used for segmentation of the PET scan and for calculation of brain weight assuming a brain tissue density of 1.05 g/mL.[34](#jmri25442-bib-0034){ref-type="ref"} Statistics {#jmri25442-sec-0014} ---------- For comparison of CBF values, the mean of the 10 repeated PCM MRI measurements and of the two PET scans were used. A paired *t*‐test was used to compare mean values of measurements from PET and MRI. Agreement was assessed by linear regression and calculation of Pearson\'s correlation coefficient (R^2^), and by Bland--Altman analysis. Both absolute (CBF~PCM~ -- CBF~PET~) and relative (\[CBF~PCM~ -- CBF~PET~\]/ CBF~PET~) differences were calculated. As spontaneous fluctuations in PaCO~2~ have been shown to introduce within‐subject variability of CBF measurements,[35](#jmri25442-bib-0035){ref-type="ref"}, [36](#jmri25442-bib-0036){ref-type="ref"} the possible influence of PaCO~2~ fluctuations were investigated by including the difference in PaCO~2~ (PaCO2~PCM~ -- PaCO2~PET~) as a covariate in a multiple regression model with CBF~PCM~ as the dependent variable and CBF~PET~ as the independent variable. Method precision was assessed as the intrasubject variability derived from a mixed linear model including all repeated measurements of global CBF by each method. Subject was entered as random effect and measurement number as fixed effect in the model. From the mixed linear model within‐subject and between‐subject variance can be separated. The corresponding within‐subject and between‐subject coefficients of variation were calculated as the respective standard deviations divided by the mean value of the measurements. Except where indicated otherwise, all results are reported as mean ± standard deviation. Results {#jmri25442-sec-0015} ======= Results of mean velocity, vessel cross‐sectional area, and diameter from PCM measurements are presented in Table [1](#jmri25442-tbl-0001){ref-type="table-wrap"}. Physiological measurements at the two sessions, and brain and body size of participants, are shown in Table [2](#jmri25442-tbl-0002){ref-type="table-wrap"}. ###### Results of Phase‐Contrast Mapping Measurements Flow \[mL/min\] Flow fraction \[%\][a](#jmri25442-note-0001){ref-type="fn"} Velocity \[cm/s\] Area \[mm^2^\] Diameter \[voxels\][^b^](#jmri25442-note-0001){ref-type="fn"} ----------- ----------------- ------------------------------------------------------------- ------------------- ---------------- --------------------------------------------------------------- Right ICA 293.1 ± 53.7 35.4 ± 2.4 21.0 ± 3.1 23.6 ± 4.3 7.3 ± 0.8 Left ICA 292.2 ± 42.0 35.5 ± 2.1 21.0 ± 2.5 23.4 ± 4.0 7.1 ± 0.9 Right VA 102.5 ± 45.9 12.4 ± 4.8 12.4 ± 2.2 13.5 ± 4.7 5.3 ± 1.0 Left VA 135.5 ± 40.5 16.7 ± 4.8 13.8 ± 2.4 16.5 ± 5.0 5.9 ± 0.8 Total 823.3 ± 112.4 Percentage of total flow, ^b^narrowest diameter. ICA, internal carotid artery; VA, vertebral artery. ###### Physiological Measurements and Brain and Body Size ^15^O‐H~2~O PET PCM MRI *P*‐value for difference ------------------------------------------------------ -------------------------------------- ------------ -------------------------- SaO~2~ \[%\] 98.3 ± 0.3 98.1 ± 0.5 0.04 PaO~2~ \[kPa\] 15.6 ± 1.0 14.2 ± 1.3 \<0.01 PaCO~2~ \[kPa\] 5.6 ± 0.3 5.5 ± 0.5 0.17 Hemoglobin \[mmol/L\] 8.7 ± 0.6 8.8 ± 0.6 0.71 Brain weight \[kg\] Body weight \[kg\] Height \[cm\] 1.439 ± 0.114 81.2 ± 8.9 187.5 ± 7.0 Mean global CBF was on average 34.9 ± 3.4 mL/100 g/min using ^15^O‐H~2~O PET and 57.0 ± 6.8 mL/100 g/min using PCM MRI. Absolute and relative differences between ^15^O‐H~2~O PET and PCM MRI were 22.0 ± 5.2 mL/100 g/min and 63.4 ± 14.8%, respectively (*P* \< 0.0001 for difference). Linear regression (Fig. [3](#jmri25442-fig-0003){ref-type="fig"}a) showed a highly significant positive correlation of PCM MRI and ^15^O‐H~2~O PET (*P* = 0.0008, R^2^ = 0.44). The slope of the regression was higher than one, indicating that the difference between the methods was perfusion‐dependent and increased with higher perfusion values. The positive slope of the regression line in the Bland--Altman plot (Fig. [3](#jmri25442-fig-0003){ref-type="fig"}b) was significantly different from zero (*P* = 0.0014), confirming a systematic perfusion‐dependent difference between CBF values obtained by the two methods. {#jmri25442-fig-0003} When including PaCO~2~ difference in the linear model, the effect of PaCO~2~ difference was near‐significantly correlated with PCM CBF (95% confidence interval = \[--0.58;10.4\], *P* = 0.077) and R^2^ increased to 0.53. When comparing the models with and without CO~2~‐difference by F‐test the improvement in fit was shown to be nonsignificant (*P* = 0.091). Intrasubject variability of global CBF was 6.5% using PCM MRI and 5.7% using PET. The corresponding values of intersubject variability were 11.6% and 8.6%, respectively. Discussion {#jmri25442-sec-0016} ========== The present study compared global CBF values obtained by MRI using PCM and by PET using ^15^O‐H~2~O in healthy volunteers. The main finding is that global CBF obtained by PCM MRI is highly correlated with values obtained by ^15^O‐ $H_{2}^{15}O$ PET, thereby confirming the ability of PCM MRI to obtain quantitative measures of global CBF. However, the analysis also demonstrated a systematic relative perfusion‐dependent difference between CBF values obtained by the two methods. The average mean CBF values obtained by the two methods differed in opposite directions from the generally accepted textbook normal global CBF values of 46--50 mL/100 g/min.[17](#jmri25442-bib-0017){ref-type="ref"}, [18](#jmri25442-bib-0018){ref-type="ref"} Nevertheless, the values obtained by each method were very similar to those previously reported in healthy subjects,[12](#jmri25442-bib-0012){ref-type="ref"}, [16](#jmri25442-bib-0016){ref-type="ref"}, [37](#jmri25442-bib-0037){ref-type="ref"} indicating that the differences more likely reflect general methodological biases rather than the current implementation of the methods or data processing. The PCM MRI technique has some well‐known errors and limitations, primarily related to the limited resolution and the suboptimal measurement geometry of the PCM MRI measurements. First of all, voxels in the periphery of the vessel will contain signal from moving and stationary tissue, causing an underestimation of velocity and overestimation of the cross‐sectional area of the vessel. The result of these two opposite effects on flow quantitation has in simulation studies been found to be very small if the artery diameter is larger than 5--6 voxels, but will cause overestimation of the flow at smaller relative diameters.[38](#jmri25442-bib-0038){ref-type="ref"} In the present study the average diameter of the internal carotid arteries assessed from PCM MRI measurements was 7.2 voxels and of the vertebral arteries 5.6 voxels, corresponding to actual diameters of 5.4 mm and 4.2 mm, respectively. These values are somewhat higher than the corresponding values of 4.8 and 3.3 mm previously reported in young subjects using ultrasound,[39](#jmri25442-bib-0039){ref-type="ref"} indicating a possible overestimation of the luminal diameter due to partial volume effect (PVE). Improving in‐plane resolution may reduce the partial volume error. Indeed, a recent in vivo study investigating the effects of varying resolution found that PCM flow values acquired at a resolution of 0.7 mm were 13% higher in the vertebral arteries and 6% higher in internal carotid arteries compared to high‐resolution imaging at 0.4 mm.[40](#jmri25442-bib-0040){ref-type="ref"} However, higher‐resolution imaging is associated with prolonged acquisition time and poorer SNR, and the authors of the study concluded that a resolution of 0.5 mm might offer a reasonable trade‐off.[40](#jmri25442-bib-0040){ref-type="ref"} Second, a single imaging slice was used to measure velocity in all of the feeding arteries. If the slice is not perpendicular to the vessels, linear velocities will be underestimated and the cross‐sectional area will be overestimated. Again, the two effects will tend to balance out, and both simulation and in vivo studies have shown that the effect on flow is negligible if the imaging planes deviate less than 10°, but will cause overestimation at higher degrees of deviation.[38](#jmri25442-bib-0038){ref-type="ref"}, [40](#jmri25442-bib-0040){ref-type="ref"} The angle of misalignment on each artery was measured from the angiography images and found to be on average less than 5° on the carotid arteries and less than 6° on the vertebral arteries. The errors related to misalignment are therefore probably very small. PCM MRI measurements were performed without cardiac triggering in order to reduce scan time. A similar approach has been used in several other studies.[6](#jmri25442-bib-0006){ref-type="ref"}, [15](#jmri25442-bib-0015){ref-type="ref"}, [41](#jmri25442-bib-0041){ref-type="ref"} One previous study has shown that nontriggered measurements produced 6% higher flow compared to triggered measurements and was also associated with slightly poorer reproducibility.[12](#jmri25442-bib-0012){ref-type="ref"} Other studies, however, have not demonstrated any systematic differences.[19](#jmri25442-bib-0019){ref-type="ref"}, [20](#jmri25442-bib-0020){ref-type="ref"}, [42](#jmri25442-bib-0042){ref-type="ref"}, [43](#jmri25442-bib-0043){ref-type="ref"} Nontriggered measurements are less susceptible to irregular or varying heart rate and applying cardiac triggering may prolong acquisition time and potentially cause underestimation of flow, as the entire cardiac cycle is not sampled. Although often considered a reference standard, the absolute quantification of CBF using ^15^O‐H~2~O PET is restricted at high CBF values by the limited water diffusion across the blood--brain barrier. The correct extraction fraction of water depends on the exact tissue measured, the perfusion itself, and interindividual variation, and has been suggested to be in the range of 0.84--0.90 for average global extraction.[44](#jmri25442-bib-0044){ref-type="ref"}, [45](#jmri25442-bib-0045){ref-type="ref"} Consequently, when reporting global CBF as the *K* ~*1*~ of the one‐tissue model, CBF is underestimated on average by 10--16%. A further cause of global CBF PET underestimation may be related to a partial volume effect when the low‐resolution CBF maps are masked with a high‐resolution brain mask, which leads to loss of signal in high‐CBF cortical voxels. White matter and cerebrospinal fluid may further contribute to dilution of the cortical signal. A previous ^15^O‐H~2~O PET study has shown that a kinetic model incorporating a physiological correction for the segments of the ROI that are not perfused (the nonperfusable tissue fraction) will increase the global CBF value ∼17%.[46](#jmri25442-bib-0046){ref-type="ref"} The underestimations of CBF using ^15^O‐H~2~O PET due to limited water diffusion and partial volume effects are both perfusion‐dependent, causing larger underestimation at high perfusion values. We do not expect the overestimation of flow by PCM MRI to be flow‐dependent within the normal perfusion range. The slope of the regression line being larger than one demonstrates this perfusion‐dependent underestimation of ^15^O‐H~2~O PET. Method precision of PCM MRI as assessed by the intrasubject coefficient was similar to a previous study that found a corresponding value of 7.4% using a cardiac triggered PCM MRI with similar resolution, also at 3T MRI.[16](#jmri25442-bib-0016){ref-type="ref"} This finding thus confirms the very high short‐term reproducibility of PCM MRI for measurement of CBF, and does not support the previous report of poorer reproducibility of nontriggered measurements.[12](#jmri25442-bib-0012){ref-type="ref"} The intrasubject variability of ^15^O‐H~2~O PET CBF measurements was very similar to that of PCM MRI, whereas the intersubject variability was slightly higher using PCM MRI compared to PET. The latter observation may be a consequence of the also slightly higher variability of PaCO~2~ at the MRI session. A limitation to the present study is that ^15^O‐H~2~O PET and PCM MRI CBF measurements were not performed simultaneously, but separated by a 2--6 hours interval. Spontaneous random variation in CBF could thus contribute to method disagreement. Studies on within‐subject variability in CBF are limited and do not separate true physiological fluctuations from methodological imprecision. Including also day‐to‐day variability, such studies have reported overall within‐subject coefficients of variation of between 8.3% and 12.9%, but do not indicate large low‐frequency variation in CBF.[18](#jmri25442-bib-0018){ref-type="ref"}, [35](#jmri25442-bib-0035){ref-type="ref"}, [47](#jmri25442-bib-0047){ref-type="ref"} As documented by the arterial blood gas and hemoglobin values, the participants were studied in stable, resting conditions at both sessions. We did observe slightly higher PaO~2~ and oxygen saturation values during the PET study compared to MRI, but these subtle differences are not likely to influence the CBF measurements and most likely reflect differences in blood sample handling at the two sessions and in calibration of the two blood gas analyzers used. Previous studies have shown that spontaneous fluctuations in PaCO~2~ is a major source of within‐subject variability[35](#jmri25442-bib-0035){ref-type="ref"}, [36](#jmri25442-bib-0036){ref-type="ref"} and in the present study residual variability tended to decrease when accounting also for changes in PaCO~2~. Finally, changes in CBF and cerebral metabolism from circadian cycle variation may also have contributed to method disagreement and residual variation.[41](#jmri25442-bib-0041){ref-type="ref"}, [48](#jmri25442-bib-0048){ref-type="ref"} Such effects cannot be assessed from the present study, but the awake‐state circadian changes are relatively small and cannot account for the large offset between ^15^O‐H~2~O PET and PCM MRI CBF measurements. The overall highly significant positive correlation of PCM MRI with ^15^O‐H~2~O PET confirms the use of PCM MRI for absolute quantification of global CBF, and thus lends further support for the use of PCM MRI in quantitative studies of cerebral physiology and in population‐based studies of cerebrovascular function and brain aging.[6](#jmri25442-bib-0006){ref-type="ref"}, [15](#jmri25442-bib-0015){ref-type="ref"}, [40](#jmri25442-bib-0040){ref-type="ref"} In particular, it allows us to more confidently interpret other quantitative physiological MRI techniques relying on accurate measures of global CBF obtained by PCM MRI.[7](#jmri25442-bib-0007){ref-type="ref"}, [10](#jmri25442-bib-0010){ref-type="ref"}, [43](#jmri25442-bib-0043){ref-type="ref"} However, it should be stressed that PET and PCM MRI measurements cannot be used interchangeably, as the very large difference will influence all CBF‐derived physiological measures directly. In conclusion, the present study demonstrates that measurement of mean global CBF with PCM MRI and ^15^O‐H~2~O PET are highly correlated, thereby validating the use of PCM MRI for quantification of global CBF. The study also showed considerable differences between the two methods, most likely resulting from methodological biases prohibiting interchangeable use of the methods. The authors thank laboratory technician Helle Juhl Simonsen for assistance with blood sampling analysis. The authors also thank nuclear medicine technologist Bente Dall at the Department of Clinical Physiology, Nuclear Medicine and PET for assistance in the ^15^O‐H~2~O PET studies, the Cyclotron Unit at the Department of Clinical Physiology, Nuclear Medicine and PET for reliable delivery of ^15^O‐H~2~O, and The John and Birthe Meyer Foundation, who donated the HRRT PET scanner to Copenhagen University Hospital Rigshospitalet.

This article is part of the thematic issue \"Frontiers in pharmaceutical nanotechnology\". Introduction ============ Nanocrystals for pharmaceutical use were invented in the early 1990s \[[@R1]--[@R4]\]. They are composed of 100% substance, are stabilized with only small amounts of surfactants, and possess particle sizes below 1 µm. According to the Ostwald--Freundlich equation nanocrystals possess a higher curvature leading to an enhanced dissolution pressure and thus to an enhanced kinetic saturation solubility \[[@R5]\]. Due to their small size they possess an increased surface area, resulting in an increased dissolution rate expressed by the Noyes--Whitney equation. In addition, they also possess an increased adhesiveness and thus, represent a universal, powerful and well-known formulation principle to overcome poor aqueous solubility and poor bioavailability of class-II and class-VI active ingredients of the biopharmaceutics classification system (BCS) \[[@R6]--[@R7]\]. Nanocrystals are already used in various pharmaceutical drug products for oral use. Examples are Rapamune^®^ (Sirolimus, Wyeth), Emend^®^ (Aprepitant, Merck), Tricor^®^ (Fenofibrate, Abbott), Megace ES^®^ (Megestrol, Par Pharm) or Triglide^®^ (First Horizon Pharmaceuticals). In 2009 the first parenteral drug product, Invega Sustenna^®^ (Paliperidone palmitate, Johnson & Johnson), was approved by the FDA. However, besides oral or parenteral administration, nanocrystals can also be used to improve the bioactivity of poorly soluble active ingredients via other routes of administration. Examples include pulmonal, ocular or dermal application \[[@R8]--[@R12]\]. Nanocrystals can be produced by different methods. Examples are precipitation, wet milling, high-pressure homogenization or combinations of these methods \[[@R1]--[@R4]\]. Regardless of the process used, all these methods will yield nanosuspensions, i.e., nanocrystals dispersed in a liquid. As liquid formulations are not always a convenient dosage form for the final drug product, in most cases nanosuspensions need to be formulated into other, more convenient, dosage forms. Depending on the route of administration this could be tablets, pellets, powders, gels or creams. However, prior to the formulation into final drug products, the aqueous nanosuspensions need to be stored, which certainly requires a sufficient stability of the nanosuspension. For this, besides chemical and physical stability, also the microbial stability needs to be considered. One method to avoid microbial contamination of aqueous formulations during storage is the use of preservatives. In previous studies it was already found that preservatives can strongly impair the physical stability of the nanosuspensions. Reasons for this are changes of pH value or of the conductivity of the dispersion medium, or the adsorption of the preservatives onto the surface of the particles, which changes the charge of the particles (zeta potential) and forces agglomeration of the nanocrystals. To avoid instabilities of nanosuspensions only very hydrophilic and non-charged preservatives, which will not interact with the nanocrystals, should be used. Due to the above-mentioned reasons, only a few preservatives are available for the preservation of nanocrystals. Suitable preservatives for the preservation of nanocrystals include different alcohols, i.e., pentylene glycol or mixtures of phenoxyethanol and ethyl hexyl glycerol \[[@R13]--[@R15]\]. The limited number of preservatives and sometimes the incompatibility of these preservatives with other excipients in the final formulation are inconvenient for a successful formulation of nanosuspensions. Therefore, to enable a more convenient formulation of nanocrystals in the future, this work aimed at investigating an alternative method to maintain the microbial stability of nanosuspensions during storage. Considering that bacterial growth strongly depends on the temperature, it was hypothesized that freezing of non-preserved nanosuspensions might prevent bacterial growth of the nanosuspensions during storage. However, the harsh conditions during freezing and thawing might also impair the physical stability of the nanocrystals and might cause agglomeration of the particles, which would then lead to a loss of the "nano properties". Hence, in this case the method could not be exploited to preserve nanosuspensions without preservative. To investigate whether the freeze--thaw method is suitable for the production of long-term stable non-preserved nanosuspensions with high microbial quality, previously developed nanosuspensions containing the flavonoid rutin as model substance and either Plantacare 2000 or Poloxamer 188 (PLX 188) as stabilizers were produced by high-pressure homogenization as described previously \[[@R16]--[@R19]\]. Each of the nanosuspensions obtained was allocated into two parts. One part was preserved, and the other part remained non-preserved. All formulations were stored at different temperatures for a period of three months and size, zeta potential, antioxidant capacity and the microbial quality were determined and monitored over this period of time ([Fig. 1](#F1){ref-type="fig"}). {#F1} Results and Discussion ====================== Production and characterization of nanosuspensions -------------------------------------------------- High-pressure homogenization yielded rutin nanosuspensions with a relatively broad size distribution, i.e., polydispersity indices (PdI values) above 0.3 and some larger particles with sizes above 4 µm ([Table 1](#T1){ref-type="table"}). Because of this, the nanosuspensions were expected to be prone to Ostwald ripening, i.e., particle growth during storage was expected. Limited physical stability of suspensions is advantageous if a study aims at investigating different stabilizing and destabilizing effects, because in comparison to highly physically stable formulations, destabilizing effects can be detected earlier during storage, making a discrimination between stabilizing and destabilizing effects clearer. ###### Overview of results obtained from the characterization of the nanocrystals at the day of production. ------------------- ----------------- ------------- ----------------------- ------------------- -------------- ------------- ------------- stabilizer preservative DLS data zeta potential \[mV\] LD data \[µm\]^a^ z-ave \[nm\] PdI in water in medium *d*(*v*)0.50 *d*(*v*)0.95 PLX 188 no preservative 408 ± 45 0.31 ± 0.08 −29.4 ± 2.9 −24.8 ± 2.6 1.3 ± 0.17 4.09 ± 0.15 with preservative 412 ± 29 0.30 ± 0.05 −27.0 ± 7.8 −24.6 ± 3.0 1.3 ± 0.12 4.13 ± 0.15 Plantacare 2000 no preservative 436 ± 30 0.32 ± 0.06 −30.6 ± 4.0 −39.5 ± 2.7 1.2 ± 0.11 4.02 ± 0.11 with preservative 447 ± 15 0.33 ± 0.06 −29.3 ± 2.7 −37.6 ± 3.3 1.2 ± 0.12 3.99 ± 0.11 ------------------- ----------------- ------------- ----------------------- ------------------- -------------- ------------- ------------- ^a^ *d*(*v*): volumetric median diameter. The suspension stabilized with PLX 188 yielded sizes of about 410 nm. Slightly larger nanocrystals with a slight agglomeration were obtained when Plantacare was used as stabilizer ([Table 1](#T1){ref-type="table"} and [Table 2](#T2){ref-type="table"}). From this it was expected that non-preserved nanosuspensions stabilized with PLX 188 might possess a slightly better physical stability than the Plantacare-stabilized formulations. Upon the addition of the preservatives only very minor changes in size were observed for both formulations ([Table 1](#T1){ref-type="table"}) and for the Plantacare-stabilized formulation even a slight deagglomeration was determined ([Table 2](#T2){ref-type="table"}). Also the zeta potential values did not change significantly in both water and original dispersion medium ([Table 1](#T1){ref-type="table"}), again indicating no, or only a very limited, impairment of the stabilization mechanisms of the nanocrystals by the hydrophilic preservative \[[@R20]\]. ###### Microscopic images of the nanosuspensions stabilized with Poloxamer 188 (left) and Plantacare 2000 (right) at the day of production. Magnification: 400-fold. --------------- ----------------------------------------------- ----------------------------------------------- stabilized with Poloxamer 188 stabilized with Plantacare 2000 non-preserved   preserved   --------------- ----------------------------------------------- ----------------------------------------------- Physical stability ------------------ The physical stability was assessed by size measurements over a period of three months of samples under all storage conditions. Increases in *d*(*v*) values, *z*-average, and polydispersity index (PdI) over time indicated instability. The laser diffractometry (LD) data and the dynamic light scattering (DLS) data obtained from this part of the study are shown in Figures 2--7. {#F2} {#F3} {#F4} {#F5} {#F6} {#F7} For the non-preserved nanosuspensions, increasing storage temperatures reduced the physical stability of the non-preserved suspensions ([Fig. 2](#F2){ref-type="fig"}, [Fig. 3](#F3){ref-type="fig"}, [Fig. 6](#F6){ref-type="fig"}, [Fig. 7](#F7){ref-type="fig"}). Consequently, the least stable formulations were obtained when the formulations were stored at 30 °C. Reasons for this are the more pronounced particle growth due to Ostwald ripening at elevated temperatures and/or destabilization due to accelerated bacterial growth, which might excrete compounds that contribute to changes in pH value or conductivity in the medium or interact with the particles. Such changes should become visible in the zeta potential values. However, this was not the case in this study ([Fig. 8](#F8){ref-type="fig"}). Therefore, the observed destabilization at elevated temperatures of the non-preserved nanosuspensions might be more related to Ostwald ripening. The assumption is also underlined by the fact that the Poloxamer-stabilized formulations, which possessed a narrower size distribution, i.e., no agglomerates at the day of production (c.f. [Table 2](#T2){ref-type="table"}), were found to be more stable than the Plantacare-stabilized formulations with a broader size distribution due to a slight agglomeration of the nanocrystals at the day of production. {#F8} The trend for physical stability was different for the preserved nanosuspensions. The most stable formulations were obtained when the samples were stored at room temperature. Lower (except freezing) and higher temperatures reduced the stability (Figures 4--7). Reasons for this cannot be explained completely but might be due to the presence of the preservatives that were added to the samples at room temperature. Changes in temperature might change the interaction between preservative and particles and thus the stability. More research is needed to understand these phenomena in detail. Most interesting results were obtained for the samples that were stored in frozen state. Physically stable formulations, i.e., without pronounced changes in *z*-average, PdI and LD values, were obtained for the non-preserved suspensions when stored at −20 °C. These results were not expected, because in general it is assumed that freezing of colloidal formulations leads to changes in the stabilizing layer and the particle interactions, being the cause for agglomeration of the nanoparticles. Especially freezing of nanosuspensions that contain dissolved active ingredients in "supersaturation" is hazardous, because due to the reduction in temperature, re-crystallization of dissolved active ingredients can easily occur. Nevertheless, the data obtained from this study indicate that it is possible to freeze--thaw nanosuspensions with good steric and/or electrostatic stabilization, without impairing their physical stability. The later fact seems to be highly important, especially when looking at the data obtained for the frozen and preserved nanosuspensions. Here, it was found that preserved nanosuspensions that were stabilized with PLX 188 became unstable during the freeze--thaw process, whereas Plantacare-stabilized formulations remained stable. Reasons for the differences might be the different stabilizers that interact differently with the preservative. Poloxamer 188 is a non-ionic surfactant, providing steric stabilization for the nanocrystals. In general, a thick surfactant layer that is indicated by a zeta potential (ZP) near zero should be obtained for good steric stabilization \[[@R21]\]. However, in our study ZP analysis revealed that the thickness of the layer is relatively low (ZP \> 20 mV, cf. [Table 1](#T1){ref-type="table"}). Hence, steric stabilization of the formulation stabilized with PLX 188 is relatively poor. Upon the addition of the preservative to the formulations stabilized with PLX 188 a very limited decrease in ZP was detected when the suspensions were analysed in original dispersion medium (c.f. [Table 1](#T1){ref-type="table"}). This might indicate that small amounts of the non-charged preservative are adsorbed onto the surface of the nanocrystals where it might interact with the polymer layer. This interaction might cause a re-arrangement of the already thin stabilizing layer around the nanocrystals and might therefore explain the decreased stabilization efficacy of the poloxamer in the preserved formulations. The destabilizing effect of PLX 188 in combination with other excipients in nanocrystal formulations was also shown by a study of Beirowski and co-workers, who showed that some combinations of poloxamer and cryoprotectant were unsuitable for stabilizing nanocrystals during a freezing process \[[@R22]\]. In contrast to PLX 188, Plantacare, which is an alkyl polyglucoside, is mostly providing electrostatic stabilization. This can be seen by the differences in ZP analysed in water and original dispersion medium, respectively. The zeta potential is about −40 mV in original dispersion medium and is reduced to about −30 mV when analysed in water, because upon dilution with water surfactant is "washed off" from the particle surface, which results in less electrostatic stabilization of the particles. However, due to its chemical structure, Plantacare is also able to provide steric stabilization. This leads to an excellent stabilization capacity combining steric and electrostatic stabilization \[[@R23]--[@R24]\]. Upon addition of the preservative only very small differences in ZP values were detected for both, water and original dispersion medium (c.f. [Table 1](#T1){ref-type="table"}), indicating only very minor impairment of the preservative. In fact, Plantacare provides a very efficient stabilization mechanism, which is not significantly impaired by the addition of the preservative. This explains the slightly better physical stability than that of the preserved PLX-stabilized nanosuspensions. Antioxidant capacity -------------------- The antioxidant capacity (AOC) was measured with the DPPH assay in which the IC~50~ value is determined. The IC~50~ value determines the amount of antioxidant needed to scavenge 50% of the free radical. Consequently, the smaller the IC~50~ value the higher is the antioxidant capacity. In this study, the IC~50~ values for the different formulations did not change during storage, independent on preservative, storage time and storage temperature ([Fig. 9](#F9){ref-type="fig"}). Hence, all these parameters did not affect the AOC of the formulations. As the AOC is an indirect measure for the chemical stability, data indicate excellent chemical stability of all aliquots during storage. The data are in good agreement with a recent study by Müller et al. in which the authors could prove chemical stability of a rutin nanosuspension for more than nine years \[[@R25]\]. {#F9} Microbial quality ----------------- The growth of bacteria and fungi was determined for all formulations after one, two and three months of storage. All preserved formulations showed excellent microbial quality. No fungi or bacteria were detected during the three months of storage ([Fig. 10](#F10){ref-type="fig"}). For the non-preserved nanosuspensions, data indicated that for all formulations the number of bacteria was fairly low upon the production with high-pressure homogenization, which is a well-described technique to reduce the number of bacteria in liquids \[[@R26]\]. The growth of microorganisms during storage was temperature-dependent and was also found to be slightly influenced by the type of stabilizer, i.e., a slightly lower and slower increase in microbial growth was found for the Plantacare-stabilized formulations ([Fig. 10](#F10){ref-type="fig"} and [Table 3](#T3){ref-type="table"}). A possible reason for this observation could be the antimicrobial activity of the stabilizer Plantacare, which was already described in previous works by Jurado and co-workers \[[@R27]--[@R28]\]. {#F10} ###### Determination of microbial quality during three months of storage. P188: rutin nanosuspension stabilized with Poloxamer 188, PLC: rutin nanosuspension stabilized with Plantacare 2000, (+) = preserved with Euxyl 9010, (−) non-preserved. -------------- ----------------------------------------------- ----------------------------------------------- ----------------------------------------------- ----------------------------------------------- storage temperature −20 °C 5 °C 21 °C 30 °C d28 bacteria     d28 fungi     d60 bacteria     d60 fungi     d90 bacteria     d90 fungi     -------------- ----------------------------------------------- ----------------------------------------------- ----------------------------------------------- ----------------------------------------------- Finally, it was found that for all formulations that were stored in frozen state at −20 °C no bacterial growth occurred. Hence, the hypothesis, that storing nanosuspensions after production in frozen state might prevent bacterial growth during storage without the use of preservatives, could be confirmed by this set of data. The same trend was also observed for the growth of fungi. However, there was one exception, which was observed for the Plantacare-stabilized formulation at day 60 of storage. At this time point a slight contamination with fungi was observed for one aliquot ([Fig. 10](#F10){ref-type="fig"} and [Table 3](#T3){ref-type="table"}). However, no contamination with fungi was observed at the next time point, i.e., after 90 days of storage. In fact, after three months of storage at −20 °C and upon thawing all non-preserved nanosuspensions were found to possess an excellent microbial quality, as no fungal and no bacterial growth was detected. These findings so far are very promising and could enable a new concept to produce preservative-free nanosuspensions that can be stored over a longer period until further use or processing into final dosage forms. Preservative-free aqueous nanosuspensions would be a convenient formulation principle, because there will be no need to take possible interactions with preservatives and/or other excipients into consideration. Allergies of consumers and/or regulatory hurdles can also be circumvented with this concept. The freeze--thaw concept is simple and can be exploited not only in industry but also in early drug development, where nanosuspensions are often used for early formulation of poorly soluble drug candidates. In this environment, the freeze--thaw concept could improve the predictability of screenings. At present, due to the lack of microbial stability, nanosuspensions need to be prepared shortly before the experiments, i.e., assays, cell culture or in vivo studies, are performed. Any repeating of the tests or continued tests will require the production of new suspensions, which might possess slightly different properties, which in turn might then cause differences in the (in vivo) data. By using thawed nanosuspensions from only one batch, these variations could be circumvented. In conclusion, the freeze--thaw concept was shown to be a simple method to prevent microbial contamination during storage of aqueous rutin nanosuspensions. The new method is believed to enable new possibilities for the use of nanosuspensions and thus can be seen as a highly promising concept, not only in pharma, but also in food and cosmetics. Next steps should now investigate if the concept can also be exploited for other active ingredients and other stabilizers. Conclusion ========== Preserved and non-preserved rutin nanosuspensions stabilized with different stabilizers were produced in this study and were stored for three months at different storage temperatures. During this time physical stability and microbial quality were monitored. In addition, the antioxidant capacity, as an indicator for the chemical stability, was assessed. All formulations were chemically stable over the whole time of observation. Physical stability was influenced by the type of surfactant, the preservative and the storage temperature. Preserved samples were only stable when stored at room temperature. Storage at higher or lower temperatures strongly impaired their physical stability. However, the microbial quality was excellent for all preserved nanosuspensions. Non-preserved samples possessed a better physical stability than the preserved nanosuspension. Proving again that preservatives impair the physical stability of nanosuspensions. Most interestingly, it was found that freezing did not alter the physical stability of the non-preserved suspensions. Hence, nanosuspensions could be frozen, stored up to three months at −20 °C and possessed unchanged particle sizes upon thawing. Storage at −20 °C also prevented bacterial growth of the non-preserved nanosuspensions, whereas storage at higher temperatures caused microbial contamination of the suspensions. The freeze--thaw concept was therefore found to be a suitable method to produce not only physically and chemically but also microbially stable rutin nanosuspensions. More research is now needed to investigate if the method can also be transferred to other nanosuspensions or nanosized formulations. All in all, the method seems to be a promising method to enable long time storage of aqueous nanosuspensions with excellent stability and without the use of preservatives. It can be used for improved formulation development of poorly soluble active ingredients in both lab scale and industrial scale. Experimental ============ Materials --------- Rutin was purchased from Denk Ingredients GmbH (Germany). The stabilizers Poloxamer 188 (PLX 188, Kolliphor^®^ P 188) and alkyl polyglucoside C8-C16 (Plantacare^®^ 2000 UP) were kindly provided from BASF AG (Germany). The preservative was composed of 90% (w/w) 2-phenoxyethanol and 10% (w/w) of 1,2-propanediol as ready to use mixture (Euxyl^®^ 9010) and was obtained from Schülke & Mayr GmbH (Germany). Purified water was obtained from a PURELAB Flex 2 (ELGA LabWater & Veolia, Germany). All other analytical chemicals were of analytical grade and used as received. Methods ------- ### Production of nanosuspensions Rutin nanosuspensions \[[@R16]--[@R19]\] were produced by high-pressure homogenization (HPH) using an EmulsiFlex-C50 (Avestin, Germany). For this, bulk suspensions containing 5% (w/w) rutin and 1% (w/w) surfactant were prepared. The pre-dispersions were homogenized with a high-speed stirrer (D-27, Miccra GmbH, Germany) at 24,000 rpm for 5 min in continuous mode and were subsequently subjected to HPH (20 cycles at 1500 bar). During homogenization and between each cycle the suspensions were cooled to below 10 °C by using a cooling bath to avoid heating of the suspensions and subsequent agglomeration of the crystals \[[@R29]\]. ### Characterization of nanosuspensions **Determination of particle size and physical stability:** Nanosuspensions were characterized regarding size by three different and independent methods. The hydrodynamic diameter and the polydispersity (*z*-average (*z*-ave) and PdI) were analysed by dynamic light scattering (DLS) with a Zetasizer Nano ZS (Malvern Panalytical GmbH, Germany). As DLS measurements, when used as stand-alone method for the characterization of submicron-sized particles, can be misleading because larger sized particles are not detected \[[@R30]--[@R33]\], light microscopy (Olympus BX53, equipped with an Olympus SC50 CMOS color camera, Japan) and laser diffraction (LD) were used as additional techniques to securely detect possible larger particles and agglomerates within the suspensions. By LD analysis the volumetric median diameters *d*(*v*)0.1, *d*(*v*)0.5, *d*(*v*)0.9, *d*(*v*)0.95 and *d*(*v*)0.99 were analysed with a Mastersizer 3000 (Malvern Panalytical GmbH, Germany). Particle diameters were calculated with Mie-theory by using 1.57 as real refractive index and 0.01 as imaginary refractive index. The zeta potential (ZP) of the nanocrystals was determined in water (adjusted to a constant conductivity of 50 µS/cm) and in original dispersion media, i.e., surfactant solution, containing either 1% (w/w) PLX 188 or Plantacare 2000, respectively. Measurements were performed via laser-Doppler-anemometry (LDA) by using a Zetasizer Nano ZS (Malvern Panalytical GmbH, Germany), which determines the electrophoretic mobility (EM), which was then converted into the ZP by using the Helmholtz--Smoluchowski equation \[[@R21]\]. **Determination of antioxidant capacity:** The antioxidant capacity was assessed by calculating the IC~50~ value, which was determined by using the DPPH assay \[[@R34]\]. DPPH (1,1-diphenyl-2-picryl-hydrazyl, Sigma-Aldrich, Germany) is a free radical that can be reduced by antioxidants. Upon reduction the colour of the free radical changes and thus the amount of reduced DPPH can be accessed via UV--vis spectroscopy. For the determination of the IC~50~ values, 100 µL of the samples containing different concentrations of the nanocrystals (200, 100, 50, 25, 12.5, 6.25, 3.125 µg/mL) were added to 100 µL of a 0.3 mM methanolic solution of DPPH. After 30 min incubation time in the dark, the absorbance was measured by a UV--vis plate reader (Multiskan GO, Thermo scientific, Germany) at 517 nm. The inhibition activity (inhibition \[%\]) was calculated as  where *A* ~sample~ is the absorbance of the sample and *A* ~0~ is the absorbance of the control (DPPH solution). The resulting linear function of inhibition against concentration was used to calculate the IC~50~ value (µg/mL). The IC~50~ value represents the concentration needed to scavenge 50% of the free radical. Rutin, which was used in this study as model drug, is a well-known antioxidant. Hence, if chemical degradation of the active ingredient occurs, changes in the IC~50~ value during storage can be observed \[[@R35]\]. As methanol is a good solvent for rutin, the addition of the nanocrystals to the methanolic DPPH solution led to a complete dissolution of the rutin nanocrystals. Hence, all rutin remaining in the formulations was dissolved during the test and thus the DPPH assay was used as a surrogate for the determination of the chemical stability. **Determination of microbial quality:** To verify the biological stability, a simple agar plate test after Ph. Eur. 8 was used. The agar (Müller--Hinton agar, Sigma) was dispersed in water, autoclaved and directly poured into sterile petri dishes (60 mm in diameter) at 28, 60 and 90 days after preparation of the nanosuspensions and used immediately after cooling. Subsequently, 10 μL of the nanosuspension or a 1:100 dilution in water were distributed evenly over the entire surface by means of a cell spreader. For detection of existing bacteria, the agar plate containing the suspension or dilution was incubated for 24 h at 36 °C and 90% humidity. Fungal contamination was detected after 7 days at 25 °C storage by using undiluted nanosuspensions. For evaluation, the visible bacterial colonies were counted, or the presence of fungal growth was noted. With dense colonization, the number of colony forming units (CFU) was set to 10000 CFU/μL. [^1]: Phone: +49(0)6421 2825885

1. Introduction {#sec1-ijms-19-03456} =============== Potassium (K^+^) is an essential macronutrient and is essential for plant growth and development \[[@B1-ijms-19-03456]\]. K^+^ is associated with or involved in several physiological processes that support plant growth and development, such as photosynthesis, enzyme activation, osmoregulation, electrical neutralization, pH and ion homeostasis, anion-cation balance, membrane electrical potential, protein and starch synthesis, sugar and nutrient transport, and stomatal movement \[[@B2-ijms-19-03456]\]. K^+^ also plays a major role in enhancing the tolerance of plants to various stresses \[[@B3-ijms-19-03456],[@B4-ijms-19-03456]\]. The concentrations of K^+^ in the soil solution range from only 0.1--1 mM, and can be much lower at the root surface due to local depletion \[[@B5-ijms-19-03456]\]. K^+^ deficiency in most arable fields is limiting for optimal plant growth \[[@B6-ijms-19-03456],[@B7-ijms-19-03456]\]. K^+^ deprivation leads to a strong increase in chlorophyll degradation; K^+^ deficiency-related symptoms include brown scorching and curling of leaf tips, as well as interveinal chlorosis \[[@B8-ijms-19-03456]\]. Reduced leaf area under K^+^ deficiency has also been reported \[[@B9-ijms-19-03456],[@B10-ijms-19-03456]\]. In addition, K^+^ deficiency affects root development, as primary root growth is negatively affected \[[@B11-ijms-19-03456],[@B12-ijms-19-03456]\]. Various K^+^ shortage-activated signaling cascades exist; these cascades involve reactive oxygen species (ROS) \[[@B13-ijms-19-03456]\], phytohormones (ethylene, auxin, and jasmonic acid) \[[@B14-ijms-19-03456],[@B15-ijms-19-03456]\], calcium \[[@B16-ijms-19-03456]\], and phosphatidic acid \[[@B17-ijms-19-03456]\]. Among these signaling cascades, calcium signaling is the most important signaling system within plant cells. In this review, the possible roles of calcium signaling in plant responses to low-K^+^ stress are discussed ([Figure 1](#ijms-19-03456-f001){ref-type="fig"}). 2. Molecular Mechanisms of Calcium Signaling Involved in Plant Responses to K^+^ Deficiency {#sec2-ijms-19-03456} =========================================================================================== 2.1. Generation of Calcium in Response to K^+^ Deficiency {#sec2dot1-ijms-19-03456} --------------------------------------------------------- The concentration and distribution of cytosolic free calcium form the basis of calcium signaling. Under normal conditions, levels of cytosolic free calcium are low, but some organelles, including the vacuole, endoplasmic reticulum, mitochondria and so on, contain high concentrations of calcium, henceforth referred to as the calcium pool. Elevations in intracellular calcium \[Ca^2+^\]~i~ have been recorded in the responses of both lower and higher plants to a wide variety of both biotic and abiotic stimuli \[[@B18-ijms-19-03456],[@B19-ijms-19-03456]\]. In plants, \[Ca^2+^\]~i~ levels that are altered in response to multiple abiotic stresses result in calcium signatures that exhibit temporal and spatial features \[[@B20-ijms-19-03456],[@B21-ijms-19-03456],[@B22-ijms-19-03456]\]. These calcium signatures can take the form of single calcium transients \[[@B23-ijms-19-03456],[@B24-ijms-19-03456]\], oscillations \[[@B25-ijms-19-03456],[@B26-ijms-19-03456],[@B27-ijms-19-03456]\], or repeated spikes \[[@B28-ijms-19-03456],[@B29-ijms-19-03456]\]. Alterations to cytosolic calcium signals can be perceived by calcium sensors, which can result in a series of downstream responses, such as protein modification and transcriptional regulation \[[@B30-ijms-19-03456],[@B31-ijms-19-03456],[@B32-ijms-19-03456],[@B33-ijms-19-03456],[@B34-ijms-19-03456],[@B35-ijms-19-03456],[@B36-ijms-19-03456],[@B37-ijms-19-03456],[@B38-ijms-19-03456]\]. Calcium sensors in *Arabidopsis* root are involved in both K^+^ uptake and responses to K^+^ deficiency. Low K^+^ induces \[Ca^2+^\] to increase in *Arabidopsis* guard cells \[[@B39-ijms-19-03456]\] and in the pollen tubes \[[@B40-ijms-19-03456]\]. The results of a recent study revealed that K^+^ deficiency triggers two successive but distinct calcium signals in roots, and that those two signals exhibit spatial and temporal specificity \[[@B16-ijms-19-03456]\]. Calcium flows into or out of the cytoplasm via calcium channels located within the plasma membrane and endomembrane system \[[@B18-ijms-19-03456],[@B41-ijms-19-03456],[@B42-ijms-19-03456]\]. Most calcium channels are nonselective for ions \[[@B41-ijms-19-03456],[@B43-ijms-19-03456]\]. In plants, these calcium channels mainly include nonspecific cation channels located within the cell membrane \[[@B43-ijms-19-03456],[@B44-ijms-19-03456]\], including members of the cyclic nucleotide-gated channel (CNGC) family \[[@B26-ijms-19-03456],[@B45-ijms-19-03456],[@B46-ijms-19-03456],[@B47-ijms-19-03456]\] and the glutamate receptor channel (GLR) family \[[@B48-ijms-19-03456],[@B49-ijms-19-03456],[@B50-ijms-19-03456]\], hyperosmolality-gated calcium-permeable channels \[[@B51-ijms-19-03456],[@B52-ijms-19-03456]\], and annexins proteins \[[@B53-ijms-19-03456]\], and mechanosensitive channels (MCAs) \[[@B54-ijms-19-03456]\], as well as two-pore calcium channels (TPCs) \[[@B55-ijms-19-03456],[@B56-ijms-19-03456]\] located in the vacuolar membrane. However, the mechanisms of their action remain unclear. 2.2. Initial Sensing of K^+^ Deficiency by Calcium Channels {#sec2dot2-ijms-19-03456} ----------------------------------------------------------- In plants, although some calcium channel activation occurs via depolarization \[[@B57-ijms-19-03456],[@B58-ijms-19-03456],[@B59-ijms-19-03456]\], most channels operate at highly-negative membrane voltages, and are often described as hyperpolarization-activated calcium channels (HACCs) \[[@B41-ijms-19-03456],[@B53-ijms-19-03456],[@B60-ijms-19-03456],[@B61-ijms-19-03456],[@B62-ijms-19-03456],[@B63-ijms-19-03456]\]. After low-K^+^ stress, calcium channels located within the root epidermis and root hair zone can be activated by hyperpolarization of the PM \[[@B60-ijms-19-03456],[@B64-ijms-19-03456]\]. Calcium increases in the cytosol can activate additional calcium channels located within the inner membrane, thereby causing the calcium pool to release calcium. For instance, two-pore channel 1 (TPC1), which is a voltage-gated channel and is located within the vacuolar membrane, is involved in the influx of calcium to the cytoplasm from the vacuole \[[@B55-ijms-19-03456]\]. In addition, the low-K^+^-inducing \[Ca^2+^\]~i~ increase is also mediated by ROS-activated calcium channels in the PM \[[@B61-ijms-19-03456]\]. ROS are important signaling molecules that mediate many physiological stimuli and lead to the generation of \[Ca^2+^\]~i~ signals under stress \[[@B65-ijms-19-03456],[@B66-ijms-19-03456]\]; increases in ROS are induced by K^+^ deficiency \[[@B67-ijms-19-03456],[@B68-ijms-19-03456]\]. Elevated \[Ca^2+^\]~i~ can induce NADPH oxidase-mediated production of ROS, which in turn activates calcium-permeable ion channels, thereby resulting in further calcium influx \[[@B62-ijms-19-03456],[@B69-ijms-19-03456]\]. Additionally, ROS have been also suggested to participate in long distance signaling with calcium, and are likely involved in generating calcium waves \[[@B70-ijms-19-03456],[@B71-ijms-19-03456],[@B72-ijms-19-03456],[@B73-ijms-19-03456]\]. 2.3. Calcium Sensors Involved in the Sensing of K^+^ Deprivation {#sec2dot3-ijms-19-03456} ---------------------------------------------------------------- How plant cells sense transient increases in \[Ca^2+^\]~i~ in response to low-K^+^ stress is ambiguous. Calcium signals are likely perceived by calcium sensors, decoded, and further transduced. Different calcium sensors must exist in the cytoplasm of plant cells so that different calcium signals can be recognized. In higher plants, putative calcium sensors include CaM, CML, CBL, and CDPK, which are derived mainly from four gene families \[[@B18-ijms-19-03456],[@B74-ijms-19-03456],[@B75-ijms-19-03456],[@B76-ijms-19-03456],[@B77-ijms-19-03456]\] and which are collectively encoded by approximately 100 genes in the *Arabidopsis* genome \[[@B30-ijms-19-03456]\]. Calcium sensors can be divided into two types: sensor responders and sensor relays. Sensor responders, such as CDPK, usually undergo a calcium-induced conformational change that alters the protein's own structure and activity. On the other hand, sensor relays, such as CaM, CML, and CBL, lack responder domains; these sensors first combine with calcium and subsequently undergo a conformational change that is relayed to an interacting partner. The interacting partner then responds with some change in its enzyme activity or structure. The two types of calcium sensors differ by their action: sensor responders function via intramolecular interactions (e.g., CDPK), whereas sensor relays function via biomolecular interactions (e.g., CBL interact with CIPK) \[[@B78-ijms-19-03456],[@B79-ijms-19-03456]\]. Calcium sensor responders mainly consist of CDPKs. CDPKs, which are a class of calcium-dependent protein kinases first discovered in plants, are serine/threonine protein kinases, and are probably the best-studied protein kinases involved in signal transduction in plants \[[@B76-ijms-19-03456],[@B77-ijms-19-03456],[@B80-ijms-19-03456]\]. CDPKs are monomeric proteins with a molecular mass of 40 to 90 kDa, and consist of five domains: an N-terminal variable domain, a protein kinase catalytic domain, an autoinhibitory domain, a regulatory domain, and a C-terminal domain of variable length \[[@B75-ijms-19-03456],[@B76-ijms-19-03456],[@B77-ijms-19-03456]\]. The autoinhibitory domain contains a pseudo-substrate sequence that can interact with the active site and inhibit its activity. Once a calcium signal is generated, the autoinhibition of CDPKs is relieved; therefore, the CDPKs become activated, and the activated CDPKs can subsequently phosphorylate target enzymes or molecules, leading to physiological responses \[[@B81-ijms-19-03456],[@B82-ijms-19-03456],[@B83-ijms-19-03456]\]. In addition to CDPKs, Ca^2+^/CaM-regulated kinases and chimeric Ca^2+^ and Ca^2+^/CaM regulated kinases (CCaMKs) are also members of this family of Ca^2+^-regulated protein kinases \[[@B30-ijms-19-03456]\]. CPK10 is involved in the Ca^2+^-dependent inhibition of K^+^~in~ channels in guard cells \[[@B84-ijms-19-03456]\]. Together, CPK11 and CPK24 mediate Ca^2+^-dependent inhibition of the activity of shaker pollen inward K^+^ channels (SPIK/AKT6) in pollen tubes, further increasing our understanding of the CDPK-mediated regulatory mechanisms of K^+^ channels \[[@B40-ijms-19-03456]\]. CPK13 specifically inhibits guard cell-expressed KAT2 and KAT1 shaker K^+^ channels \[[@B85-ijms-19-03456]\]. The results of a recent study \[[@B86-ijms-19-03456]\] showed that two *CDPKs* are up-regulated in tobacco seedlings under low-K^+^ stress, whereas *OsCPK9* and *OsCCaMK1* are down-regulated in rice root responses to K^+^ deficiency \[[@B87-ijms-19-03456]\]. Calcium sensor relays mainly consist of CaMs and CML \[[@B88-ijms-19-03456]\]; CaM is an important member of this class of calcium sensor proteins. Plant CaMs are small, acidic proteins with a molecular weight ranging from 16.7--16.8 kDa; CaM contains 4 EF-hand repeated domains that bind 4 individual Ca^2+^ ions. In addition, the four binding sites display cooperativity in Ca^2+^ binding; as such, unbound CaM is fully opened \[[@B89-ijms-19-03456],[@B90-ijms-19-03456]\]. CaM binding with Ca^2+^ leads to a conformational change. This change exposes the hydrophobic surface of CaM, which helps to interact with target proteins in a Ca^2+^-dependent manner, altering their activities. CaM binding with Ca^2+^ forms an activated Ca^2+^-CaM complex, and the binding of this complex to target enzymes leads to their activation \[[@B91-ijms-19-03456],[@B92-ijms-19-03456]\]. CaM has no catalytic activity of its own, but upon binding with Ca^2+^, CAM modulates the activities of several enzymes and non-enzymic proteins involved in a variety of cellular processes \[[@B93-ijms-19-03456],[@B94-ijms-19-03456],[@B95-ijms-19-03456]\]. CML also acts in signaling 'cross-talk' and is involved in the co-ordination of plant responses to biotic and abiotic stresses \[[@B21-ijms-19-03456],[@B96-ijms-19-03456],[@B97-ijms-19-03456]\]. CML25 is an important transducer involved in the Ca^2+^ ions-mediated regulation of K^+^ influx \[[@B98-ijms-19-03456]\]. Ma et al. (2012) \[[@B87-ijms-19-03456]\] reported that the expression levels of the calcium sensor protein genes *OsCML1*, *OsCML18*, *OsCML20*, and *OsCML31* are up-regulated in rice under low-K^+^ stress. Several genes that code for members of CML families are differentially expressed in tobacco seedlings under low-K^+^ stress \[[@B99-ijms-19-03456]\]. In addition, a Raf-like MAPKK kinase (AtILK1) directly interacts with AtHAK5 in conjunction with the AtCML9, promoting AtHAK5 accumulation on the membrane \[[@B100-ijms-19-03456]\]. These results indicate that calcium sensor proteins may play vital roles in connecting calcium signaling and downstream target proteins during plant responses to K^+^ deficiency. Another important member of the calcium sensor relays is CBL, which, like CaM and CML, has been shown to lack responder domains \[[@B101-ijms-19-03456]\]. CBLs and CBL-interacting protein kinases (CIPKs) often form the CBL/CIPK complexes, perceiving calcium signals and relaying the signals to downstream responses in plants under low-K^+^ stress \[[@B31-ijms-19-03456],[@B32-ijms-19-03456],[@B33-ijms-19-03456],[@B34-ijms-19-03456],[@B35-ijms-19-03456],[@B37-ijms-19-03456],[@B102-ijms-19-03456],[@B103-ijms-19-03456],[@B104-ijms-19-03456],[@B105-ijms-19-03456]\]. 3. Role of the CBL/CIPK Complex in Response to K^+^ Deficiency {#sec3-ijms-19-03456} ============================================================== 3.1. CBL Proteins {#sec3dot1-ijms-19-03456} ----------------- CBL proteins, which are involved in the salt-overly-sensitive (SOS) pathway of salt stress signal transduction, were first identified in *Arabidopsis* \[[@B101-ijms-19-03456]\]. CBL proteins in plants share significant sequence similarities with the calcineurin B subunit in yeast and the neuronal calcium sensors in animals \[[@B74-ijms-19-03456],[@B101-ijms-19-03456],[@B106-ijms-19-03456]\]. In *Arabidopsis*, CBLs are encoded by at least 10 members of a multigene family \[[@B106-ijms-19-03456],[@B107-ijms-19-03456],[@B108-ijms-19-03456]\]. Each CBL protein harbors four EF-hand motifs that facilitate Ca^2+^ binding; in all CBL proteins, these motifs are arranged in fixed spacing. Therefore, the weights of CBLs are nearly equal (23--26 kD); the N-terminal and C-terminal domains account for the differences in CBL weights. Unlike in other species, in *Arabidopsis*, the first CBL EF-hand domain consists of an unconventional 14 amino acids, not 12 amino acids; this *Arabidopsis* CBL EF-hand domain likely lacks the normal amino acid numbers required for Ca^2+^ binding \[[@B107-ijms-19-03456],[@B109-ijms-19-03456]\]. The crystal structures of AtCBL2 and AtCBL4 indicate that two calcium ions are coordinated in the first and fourth EF-hand motifs despite the presence of two additional amino acids \[[@B110-ijms-19-03456],[@B111-ijms-19-03456]\]. Sequence variations in EF-hand motifs most likely control the overall Ca^2+^-binding affinity of individual CBL proteins. This phenomenon may explain why plants can decode different calcium signals \[[@B109-ijms-19-03456],[@B112-ijms-19-03456],[@B113-ijms-19-03456],[@B114-ijms-19-03456],[@B115-ijms-19-03456],[@B116-ijms-19-03456]\]. With the exception of CBL10, whose hydrophobic N-terminal region is a special transmembrane domain, other CBLs can be classified into one of two categories based on variations in their N-terminal domain. The first category represents CBL proteins with a short N-terminal region, which consist of 27--32 amino acids. Examples in this category include CBL1, CBL4, CBL5, CBL8, and CBL9; other CBL proteins, with exception of CBL8, all have a MGXXX(S/T) consensus sequence for N-myristoylation \[[@B117-ijms-19-03456]\]. The second category represents CBL proteins with an extended N-terminal region, which consist of 41--43 amino acids. Examples in this category include CBL2, CBL3, and CBL6, which lack recognizable lipid modifications \[[@B117-ijms-19-03456]\]. Sequence comparisons place CBL7 within this category; therefore, the CBL7 protein appears to have lost its N-terminal extension during evolution \[[@B106-ijms-19-03456],[@B112-ijms-19-03456]\]. CBL proteins were first identified in model plant *Arabidopsis* \[[@B74-ijms-19-03456],[@B101-ijms-19-03456]\]; subsequent bioinformatics analysis has revealed that these kinds of proteins also exist in other species. The *Oryza sativa* (rice) genome contains 10 genes that encode CBL proteins \[[@B108-ijms-19-03456]\], and the *Populus trichocarpa* (poplar) genome also contains 10 genes that encode CBL proteins \[[@B118-ijms-19-03456]\]. The fully-sequenced genomes of the dicotyledonous plant *Vitis vinifera* (grape) and the monocotyledonous plant *Sorghum bicolor* (sorghum) have been analyzed in attempt to detect the presence of CBLs; these analyses revealed 8 CBLs in grape and 6 CBLs in sorghum \[[@B112-ijms-19-03456]\]. *Gossipium raimondii* (cotton) was found to contain the highest number of CBL genes (13) among the 38 plant species analyzed \[[@B119-ijms-19-03456]\]. And 19 members of the *BrrCBL* genes were identified in *Brassica rapa* var. *rapa* (turnip) \[[@B120-ijms-19-03456]\]. A comparative analysis of CBLs from all these species further supports the classification of these proteins according to their N-terminal domain \[[@B112-ijms-19-03456]\]. An analysis of the genomic sequences of algae and nonvascular plants revealed that the genome of the moss *Physcomitrella patens* encodes 4 CBL proteins, and that the genome of the fern *Selaginella moellendorffii* also encodes 4 CBL proteins. In addition, one CBL protein was identified in the green alga *Chlorella* sp., as well as in the genome of the smallest known free-living eukaryotic alga, *Ostreococcus tauri* \[[@B121-ijms-19-03456]\]. In general, these studies of lower plants have enabled us to address the general evolutionary aspects of this signaling network \[[@B112-ijms-19-03456]\]. 3.2. CBL-Interacting Protein Kinases (CIPKs) {#sec3dot2-ijms-19-03456} -------------------------------------------- Protein kinases that specifically interact with CBL proteins have been identified; these kinases are referred to as CIPKs, which were mentioned earlier with salt-overly-sensitive 2 (SOS2), SOS3-interacting proteins (SIPs), and protein kinase S (PKS) \[[@B122-ijms-19-03456],[@B123-ijms-19-03456],[@B124-ijms-19-03456]\]. All of these CIPKs share a typical two-domain structure comprising an N-terminal kinase domain and a C-terminal catalytic domain, which are separated by a junction domain \[[@B109-ijms-19-03456],[@B112-ijms-19-03456]\]. Sucrose non-fermenting 1 (SNF1)-related protein kinases (SnRKs) are important kinases in plants, and exhibit high sequence homology to metabolic regulators found in mammals (5′-AMP-activated protein kinases \[AMPKs\]) and in yeast (SNF1); the three combined kinases form the protein kinase superfamily. Amino acid sequence identification and expression pattern analyses have revealed that the SnRK family of protein kinases can be categorized into three classes: SnRK1, SnRK2, and SnRK3 \[[@B125-ijms-19-03456],[@B126-ijms-19-03456],[@B127-ijms-19-03456]\]. *Arabidopsis* CIPKs constitute a kind of serine-threonine kinase, and have a highly-conserved N-terminal catalytic domain; these proteins have been classified as SnRK3s \[[@B125-ijms-19-03456],[@B128-ijms-19-03456]\]. The N-terminal domain of a protein kinase contains a conserved activation loop. Assays of mutants revealed that the Thr168, Ser156, or Tyr175 to Asp change in the activation loop of CIPK24/SOS2 protein kinase \[[@B129-ijms-19-03456],[@B130-ijms-19-03456],[@B131-ijms-19-03456]\], the Thr178 to Asp change in the activation loop of CIPK9/PKS6 \[[@B132-ijms-19-03456]\], the Thr161 to Asp change in the activation loop of CIPK8/PKS11 \[[@B133-ijms-19-03456]\], and the Thr183 to Asp change in the activation loop of CIPK3 \[[@B134-ijms-19-03456]\] can cause strong activation of CIPK protein kinases, even in the absence of Ca^2+^ or CBLs. Thus, these several conserved amino acid residues may be phosphorylation sites of CIPK protein kinases, whose phosphorylation results in the activation of CIPK \[[@B112-ijms-19-03456]\]. Furthermore, an additional phosphorylation site (Ser 228) has been identified as a target of autophosphorylation activity in the C-terminal region of the CIPK24/SOS2 kinase domain \[[@B135-ijms-19-03456]\]. Within the otherwise divergent C-terminal regulatory domain, CIPKs have a conserved domain, a NAF motif, and a 24-aminoacid domain with the conserved amino acids N, A, and F that are required for the CBL-CIPK interactions; the NAF motif is also referred to as a FISL motif because of the complete conservation of the six amino acid residues A, F, I, S, L, and F. The NAF/FISL motif of CIPKs is sufficient for mediating protein interactions with all CBL proteins \[[@B128-ijms-19-03456],[@B129-ijms-19-03456],[@B136-ijms-19-03456],[@B137-ijms-19-03456]\]. The NAF/FISL motif is also necessary and sufficient for keeping CIPKs inactive, and serves as an autoinhibitory domain; removal of the NAF/FISL domain can increase the activity of CIPKs \[[@B129-ijms-19-03456],[@B132-ijms-19-03456]\]. Another important functional domain is the protein phosphatase interaction (PPI) motif in the C-terminal region of CIPKs; this motif is adjacent to the NAF/FISL motif. The PPI motif consists of 37 amino acid residues, and was first identified in CIPK24/SOS2, which is necessary and sufficient for interaction with ABA-INSENSITIVE 2 (ABI2) \[[@B138-ijms-19-03456]\]. The PPI motif is conserved in CIPK protein kinases; all of these motifs interact with 2C-type protein phosphatase (PP2C) \[[@B34-ijms-19-03456],[@B128-ijms-19-03456],[@B138-ijms-19-03456],[@B139-ijms-19-03456]\]. In plants, PP2C is a strong negative regulator of the stress-activated, mitogen-activated protein kinase (MAPK) pathway, which is involved in plant responses to abiotic stresses and growth regulation \[[@B140-ijms-19-03456],[@B141-ijms-19-03456],[@B142-ijms-19-03456]\]. PP2C interaction with CIPK results in the complete replacement of the combination between CBL proteins and the NAF domain or a portion of the PPI domain of CIPKs \[[@B112-ijms-19-03456]\]. The dissociation of CBL proteins from the NAF domain of CIPK prevents the autophosphorylation of CIPK, thereby transforming the kinase into an inactive state \[[@B129-ijms-19-03456],[@B132-ijms-19-03456]\]. 3.3. Involvement of the CBL/CIPK Complex in the Sensing of K^+^ Deficiency {#sec3dot3-ijms-19-03456} -------------------------------------------------------------------------- Some CBL protein family members that interact with CIPKs function in plant responses to K^+^ deficiency ([Figure 2](#ijms-19-03456-f002){ref-type="fig"}) \[[@B31-ijms-19-03456],[@B32-ijms-19-03456],[@B33-ijms-19-03456],[@B34-ijms-19-03456],[@B35-ijms-19-03456],[@B37-ijms-19-03456],[@B102-ijms-19-03456],[@B103-ijms-19-03456],[@B104-ijms-19-03456],[@B105-ijms-19-03456]\]. The first CBL proteins identified to be involved in the K^+^ deficiency response were the PM-localized calcium sensors CBL1 and CBL9; these sensors interact with the cytoplasm-localized Ser/Thr kinase CIPK23, and recruit it to the root cell PM, where the complex subsequently phosphorylates AKT1 \[[@B31-ijms-19-03456],[@B102-ijms-19-03456],[@B143-ijms-19-03456]\]. AKT1, a shaker inward K^+^ channel \[[@B144-ijms-19-03456],[@B145-ijms-19-03456],[@B146-ijms-19-03456]\], is considered a major component involved in K^+^ uptake in *Arabidopsis* root cells under low-K^+^ conditions \[[@B147-ijms-19-03456],[@B148-ijms-19-03456],[@B149-ijms-19-03456]\]. In addition, AtKC1, a K^+^ channel regulatory subunit that negatively modulates many inward K^+^ channels, interacts with AtAKT1, forming an AtAKT1-AtKC1 heteromeric channel, and modulates AtAKT1 activity together with AtCIPK23, to synergistically regulate AtAKT1-mediated low-K^+^ stress responses \[[@B150-ijms-19-03456],[@B151-ijms-19-03456]\]. Another CBL protein, CBL10, was recently shown to be a negative regulator of the AKT1 channel. CBL10 may compete with CIPK23 for binding to AKT1, and CBL10 interacts directly with the AKT1 channel and inhibits AKT1-mediated K^+^ flux into the cytoplasm. In *Arabidopsis*, this inhibition ultimately maintains K^+^ homeostasis under ion stress conditions in a CIPK-independent manner \[[@B35-ijms-19-03456]\]. The calcium sensor CBL4, together with the interacting protein kinase CIPK6, modulates the activity and PM targeting of the K^+^ channel AKT2 in *Arabidopsis*; CBL4 in conjunction with CIPK6 mediates the translocation of AKT2 from the endoplasmic reticulum membrane to the PM in a kinase-interaction-dependent but phosphorylation-independent manner in plant cells, and enhances AKT2 activity in oocytes \[[@B33-ijms-19-03456]\]. AKT2 is unique among the nine shaker-type K^+^ channel subunits expressed in *Arabidopsis*, because AKT2 exhibits weak inward-rectifying activity in oocytes \[[@B152-ijms-19-03456],[@B153-ijms-19-03456]\]; several studies have suggested that this channel is regulated by unknown protein kinases and by the protein phosphatase PP2CA \[[@B154-ijms-19-03456],[@B155-ijms-19-03456],[@B156-ijms-19-03456]\]. A recent study demonstrated that CBL3 and CIPK9 work together and function in K^+^ homeostasis under low-K^+^ stress; this complex mediates the regulation of putative tonoplast-localized outward K^+^ channels \[[@B104-ijms-19-03456]\]. CIPK9 is the CIPK family member which is most similar to CIPK23, and CIPK9 loss of function results in a phenotype that is tolerant to K^+^ deficiency conditions \[[@B31-ijms-19-03456]\]; in contrast, CIPK9 overexpression lines are sensitive to K^+^ deficiency stress. Furthermore, because K^+^ deficiency symptoms first appear in relatively old leaves, CIPK9 may be involved in K^+^ reallocation from older leaves to the younger leaves during K^+^ deficiency \[[@B32-ijms-19-03456],[@B157-ijms-19-03456]\]. In addition, the AtCBL1/AtCIPK23 complex can phosphorylate AtHAK5, which is a KT/KUP/HAK-type transporter whose expression occurs mainly in the roots \[[@B37-ijms-19-03456]\]. The transcription of *HAK5* is induced by K^+^ deficiency via the transcription factor RAP2.11 \[[@B147-ijms-19-03456],[@B158-ijms-19-03456]\], which is considered to function predominantly in the uptake of K^+^ from the soil \[[@B37-ijms-19-03456],[@B145-ijms-19-03456],[@B147-ijms-19-03456],[@B159-ijms-19-03456],[@B160-ijms-19-03456],[@B161-ijms-19-03456]\]. These CBLs-CIPKs-AKT1/AKT2/HAK5 pathways are important mechanisms in the response to low-K^+^ stress in *Arabidopsis*. Similar mechanisms have also been identified in other plant species. The OsCBL1-OsCIPK23-OsAKT1 pathway was identified in rice \[[@B36-ijms-19-03456]\]. Rice OsCIPKs show high amino acid sequence similarity to *Arabidopsis* CIPKs; eight OsCIPK genes *(OsCIPK2*, *6*, *9, 10*, *14*, *15*, *23*, and *26*) are upregulated under low-K^+^ stress, whereas two OsCIPK genes (*OsCIPK29* and *31*) are down-regulated \[[@B108-ijms-19-03456]\]. Another study demonstrated that the AtCBL9/AtCIPK23 kinase complex activates DmKT1, which has been identified as a K^+^-selective channel of voltage-dependent high capacity and low affinity; the first proton-driven high-affinity K^+^ transporter with weak selectivity (DmHAK5) is also activated by the same kinase complex \[[@B38-ijms-19-03456]\]. Furthermore, CBL/CIPK (CIPK6, CIPK16, and CIPK23) complexes can interact with the AKT1-interacting PP2CA (AIP1) and AIP1 homologue (AIP1H), both of which are protein phosphatase PP2Cs, via the direct binding of the kinase domain of CIPKs to indirectly deactivate AKT1 by inhibiting phosphorylation. Several CBLs have been reported to interact with and inhibit the activity of PP2Cs, thereby enhancing CIPK-induced AKT1 activation; this phenomenon forms a kinase/phosphatase partnership that enables AKT1 activity to be switched on and off \[[@B34-ijms-19-03456],[@B103-ijms-19-03456]\]. In addition, AtPP2CA can interact physically and functionally with AKT2, and can inhibit AKT2-mediated K^+^ currents via the direct phosphorylation of AKT2 \[[@B154-ijms-19-03456]\]. 4. Conclusions and Perspectives {#sec4-ijms-19-03456} =============================== K^+^ is an essential macronutrient and is associated with or involved in several physiological processes supporting plant growth and development, such as photosynthesis, enzyme activation, osmoregulation, electrical neutralization, pH and ion homeostasis, anion-cation balance, membrane electrical potential, protein and starch synthesis, sugar and nutrient transport, and stomatal movements. K^+^ deprivation leads to a strong increase in chlorophyll degradation. Various K^+^ shortage-activated signaling cascades exist; these cascades involve ROS, phytohormones, calcium, and phosphatidic acid. Among these signaling cascades, calcium signaling is the most important signaling system in plant cells. Thus far, our knowledge on the molecular mechanisms of calcium signaling in plant responses to K^+^ deficiency is still limited. Calcium plays a critical role in plant responses to low-K^+^ stress. Although tremendous progress has been made in understanding plant responses to low-K^+^ stress, one important question that remains unanswered is how calcium as a messenger can relay information that distinguishes different extracellular signals, triggering different processes and specific responses in cells. The results of the previous studies suggest that K^+^ deficiency can induce changes in intracellular calcium levels that exhibit temporal and spatial features. The decoding of calcium signatures in plant cells might depend largely on the presence of various calcium sensors, including CaM, CML, CDPK, and CBL, as well as their targets. In particular, the CBL-CIPK signaling system is a central and critical signaling system for decoding calcium signatures and for translating those calcium signatures into downstream responses to K^+^ deficiency. Over the last two decades, the extensive genomic, genetic, and molecular physiological studies have begun to shed light on the transport regulation and signaling mechanisms of plant responses to K^+^ deficiency. Further identification of important calcium signaling components involved in plant responses to K^+^ deficiency is important. Detailed functional characterization of these calcium-signaling components is also needed to elucidate the complex network of plant signaling in response to K^+^ deficiency. Even though many K^+^ transporters and channels in higher plants have been functionally characterized, specific K^+^ sensors remain unknown. Thus, future investigations should give attention to further functional characterization of K^+^ sensors and to the regulatory mechanisms of these sensors. Given that K^+^ deficiency could induce \[Ca^2+^\]~i~ to increase in *Arabidopsis* guard cells and in the pollen tubes or triggers calcium signals in *Arabidopsis* roots, the perception-defective of K^+^ deficiency mutants are likely to be screened via detection of the plant calcium signal. We thank Zhen-Ming Pei, Zhangli Hu, two anonymous reviewers and the Editor for their valuable comments on the manuscript. This work was funded by China Postdoctoral Science Foundation, grant number (2018M633109). The authors declare no conflict of interest. {#ijms-19-03456-f001} {#ijms-19-03456-f002}

All relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ In arid and semi-arid environments, environmental gradients differ at local scales because shrubs generally improve soil fertility and produce microclimates under their canopies that favor plant-plant interactions \[[@pone.0133559.ref001]\]. According to the stress-gradient hypothesis (SGH) \[[@pone.0133559.ref002]\], interactions among plants vary with environmental conditions, shifting from competition to facilitation as environmental stress increases. Such environmental stress comprises stresses from abiotic and biotic (i.e., consumers) sources. It has been found that plant seedlings growing beneath shrubs have different probabilities of survival than do conspecific seedlings growing in open interspaces between shrubs \[[@pone.0133559.ref003]\]. The occurrence of facilitation interactions under high abiotic stress conditions has been verified by numerous studies in many ecosystems and has provided a crucial basis for empirical studies and theory governing plant-plant interactions in stress-prone Mediterranean environments (see \[[@pone.0133559.ref001]\] and references therein, \[[@pone.0133559.ref003]\]) where tree seedlings benefit from habitat amelioration, especially during summer drought \[[@pone.0133559.ref003]--[@pone.0133559.ref009]\]. However, information on biotic impacts caused by consumers in these environments is still scarce (see \[[@pone.0133559.ref010]\] and references therein). Positive plant-plant interactions should be especially important in plant communities subjected to both climatic stress and biotic stress \[[@pone.0133559.ref002]\], as is the case in Mediterranean-type ecosystems. Surprisingly, the role of herbivory, a common and widespread interaction in arid and semiarid environment, has attracted little attention in this context \[[@pone.0133559.ref011]--[@pone.0133559.ref014]\]. In Mediterranean central Chile, the exotic mammalian herbivores have been altering the structure and functioning of ecosystems \[[@pone.0133559.ref015]\] by, promoting homogenization of species like *Acacia caven* and preventing the recolonization of clearings by woody matorral species on the slopes \[[@pone.0133559.ref008], [@pone.0133559.ref015]\]. It is expected that the impacts of introduced herbivores on native vegetation will be different from the native fauna \[[@pone.0133559.ref016]\], but how plant-plant facilitation relationship may be modulated by animal-plant interactions, specifically in this case, by herbivory, has been frequently neglected. Valiente-Banuet, Rumebe, \[[@pone.0133559.ref017]\] have indicated that facilitation (or the nurse effect) prevented Tertiary species extinction when climatic conditions became more xeric during the Quaternary, the period during which the current Mediterranean climate emerged. This effect should be of particular importance for the persistence of fleshy-fruited Tertiary species with recalcitrant seeds \[[@pone.0133559.ref017]\] as palms (Arecaceae). Originating in the Cretaceous, palms experienced a spectacular radiation in the Tertiary Early Cenozoic and severe extinction rates due to rainforest decline during the drier Quaternary \[[@pone.0133559.ref018]\]. Thus, relict palms in arid ecosystems, such as the endemic wine palm (*Jubaea chilensis* (Molina) Baill.) in Mediterranean central Chile, constitute a proper candidate for evaluating the relative importance of abiotic and biotic (herbivory) stresses for seedling survival and growth, and for assessing the relationship between animal-plant and plant-plant interactions in arid environments. Accordingly, we formulated four hypotheses (i) Shrub canopies, by creating a more humid, mesic, and shaded environment relative to open interspaces, differentially affect palm seedling performance. (ii) Shrubs interact positively with wine palm seedlings to facilitate recruitment, buffering the seedlings performance from potentially limiting stresses (SGH) or (iii) seedling performance is facilitated because shrubs protect the seedlings against browsers. In order to study these hypotheses, we monitored growth and herbivore defoliation of 300 wine palm seedlings over a period of 18 months. Seedlings were transplanted outside and under the canopy of the seven most representative shrub species and were grown with and without exclosure protection from browsing. This study can improve our understanding of the relative importance of biotic and abiotic mechanisms for shrub facilitation in the Mediterranean hotspot ecosystem region, which has experienced an increasing frequency, duration, and severity of drought in recent decades \[[@pone.0133559.ref019]\] and has also faced risks associated with the presence of a high proportion of invasive non-native vertebrate herbivores \[[@pone.0133559.ref008], [@pone.0133559.ref015], [@pone.0133559.ref020]\]. Also, this study will help us to examine to fate of wine palm in a context of increasing of aridity and herbivory pressure. Materials and Methods {#sec002} ===================== Study species and site {#sec003} ---------------------- This study was conducted in the Oasis La Campana private reserve in central Chile (hereafter La Campana; 71°04\'W; 32°57'S). The CONAF/Parque Nacional la Campana and Reserva Oasis de La Campana issued all of the required permits for the work conducted in those sites. Located in the core zone of the Campana-Peñuelas UNESCO Biosphere Reserve, La Campana covers 17,095 ha and has irregular topography varying from 300 to 1,800 m altitude. The area has a Mediterranean climate, closely equivalent to that of southern California but with a six-month offset. The mean annual rainfall and temperature are 109 mm and 20μC, respectively \[[@pone.0133559.ref021]\]. Rainfall is concentrated in spring and autumn, alternating with hot and dry summers and cold winters. Mean temperatures for December-February are 18°C, and a mean of only 30 mm of precipitation falls from November to April. Winters are mild and moist, and the mean winter temperature, from June to August, is 11°C \[[@pone.0133559.ref022]\]. Mediterranean-type ecosystems have been substantially altered by human activities worldwide, but it is probable that such ecosystems persist to a greater extent in central Chile than in other regions of the world \[[@pone.0133559.ref023]\]. The natural recovery of sclerophyllous vegetation in many open and abandoned areas of central Chile is a slow process \[[@pone.0133559.ref024], [@pone.0133559.ref025]\] due to unsuitable conditions for seedling recruitment and a lack of seed sources. The limiting factors for recruitment include seed deposition in unsafe sites for seed germination and for seedling establishment \[[@pone.0133559.ref024], [@pone.0133559.ref025]\]. The most common evergreen trees and shrubs in La Campana are *Cryptocarya alba* (Lauraceae), *Lithraea caustica* (Anacardiaceae), *Quillaja saponaria* (Quillajaceae), *Maytenus boaria* (Celastraceae), *Kageneckia oblonga* (Rosaceae), and *Peumus boldus* (Monimiaceae) \[[@pone.0133559.ref026]\]. Sclerophyllous forests commonly exhibit a patchy spatial structure, but they may present a continuous canopy in ravines, deep creeks, along permanent water courses and on south-facing slopes \[[@pone.0133559.ref027]\]. On dry north-facing slopes or in frequently disturbed sites \[[@pone.0133559.ref028]\], the sclerophyllous vegetation is often replaced by a xerophytic thorn scrub, with a combination of deciduous shrubs, such as *Retanilla trinervia* (Rhamnaceae), *Colliguaja odorifera* (Euphorbiaceae), *Baccharis linearis* (Compositae), *Acacia caven* (Leguminosae) and *Puya berteroniana* (Bromeliaceae) \[[@pone.0133559.ref026]\]. The palm forest, where this study was conducted, is narrowly distributed within native vegetation and frequently neglected component with numerically abundant populations of the endemic wine palm *Jubaea chilensis* (Molina) Baill ([Fig 1](#pone.0133559.g001){ref-type="fig"}). The wine palm population has been severely fragmented and currently occurs mainly on slopes and in deep canyons of the central Chile Coastal Range \[[@pone.0133559.ref029]\]. It is probable that these sites, with soils of granitic origin, are favorable because of their maritime, moisture-laden air and limited temperature oscillations \[[@pone.0133559.ref023]\]. {#pone.0133559.g001} La Campana, the heart of the wine palm\'s range, contains the world's largest remnant population of the endemic wine palm, harboring ≈ 60,000 wine palms that represent more than 58% of the total number of native stands \[[@pone.0133559.ref029]\]. The wine palm has a limited distribution but is numerically abundant where it is found (22--113 ind. ha^-1^) \[[@pone.0133559.ref030]\]. It is usually dominant in its favored habitat of slopes and deep canyons with moisture-laden marine air and limited temperature oscillations \[[@pone.0133559.ref027]\]. In central Chile, the biota has been widely and severely impacted by agriculture, human-caused fires, and livestock grazing \[[@pone.0133559.ref022]\], activities that still persist in areas surrounding the study site \[[@pone.0133559.ref031]\]. Introduced species of seedling predators are commonly found within the area. These species, including the invasive European rabbit (*Oryctolagus cuniculus*) \[[@pone.0133559.ref032]\] and livestock \[[@pone.0133559.ref033]\], have potential negative effects on natural regeneration. The study area also holds most of the Mediterranean-type vegetation of central Chile, including hygrophyllous and sclerophyllous forests, *Nothofagus* trees and palm forests \[[@pone.0133559.ref031]\]. The wine palm can attain a height of 25 to 30 m and a trunk diameter of up to 180 cm. The distinctive trunk is usually bulb shaped, with the thickest portion of the trunk above the base of the plant. *J*. *chilensis* produces prolific ovoid fruits in the form of yellow-orange drupes; within each drupe there is a single spherical fruit that is approximately 2--3 cm (0.79--1.18 in) in diameter. The fruit has a very hard endocarp (shell) and a whitish endosperm. The fresh nuts are commonly sold in the areas where the palms grow during their fruiting season (L.A. González, pers. comm.). Experimental design {#sec004} ------------------- During the fruiting season of 2004, wine palm seeds were collected from several palm trees. For more than 1,000 individual seeds, seed mass was recorded and the seeds sown in an adjacent nursery cubicle (Reserva Oasis de La Campana). Seedlings were grown from early to late January 2005. Five months later, more than 300 wine palm seedlings were randomly selected, with initial seedling aerial biomass weighing 0.94 ± 0.02 g (mean ± SE, n = 300). Because the distribution of palms is strongly influenced by the microhabitats within their range \[[@pone.0133559.ref030], [@pone.0133559.ref034]\], which are distributed non-randomly \[[@pone.0133559.ref027]\], three areas with suitable wine palm populations were selected. The conditions of the physical environment under shrub canopies and interspaces between shrubs differ due to variations in vegetation cover (open interspaces between shrubs 7.67 ± 1.36%; under shrub 63.33 ± 1.27%, n = 150) and moisture soils (open interspaces between shrubs 3.23 ± 0.22%; under shrub 5.25 ± 0.22% moisture content), especially during hot, dry summers. These differences can be assumed to result in distinct microhabitats \[[@pone.0133559.ref035]\]. The 75 seedling-performance experimental stations were simultaneously established at minimum distance of 50 m intervals. Seedling-performance experimental stations were spatially independent from another such that we did not detect spatial autocorrelation in seedling survival or growth among stations (average Moran's I-values were not significant for all distance classes). Each station consisted of four wine palm seedlings and a native shrub of over 1 m of tall, that completely shading the ground surface bellows its crown. Although using different nurse shrub species could cause potential confounding factors, five shrub species have been used due to the species presence and abundance: *A*. *caven*, *R*. *trinervia*, *L*. *caustica*, *B*. *linearis*, *Q*. *saponaria*, *Azara* sp. and *P*. *boldus*, thus ranging a wide spectrum of nurse life forms and architecture. While we cannot altogether discount the possibility of shrub species interacting on herbivory patterns or soil type patterns, our preliminary analysis found no difference between shrub species and wine palm seedling survival (General Linear Model, F = 0.62, df = 6, P = 0.71). In addition, previous studies have shown that the phylogenetic relationship of plants is a weak predictor of the similarity of associated herbivore assemblages, and it may not influence herbivore behavior \[[@pone.0133559.ref036]\]. A two-way fully factorial design was used for the study. The seedlings experienced one of four possible selective regimes generated by selectively weakening the effects of browsers and nurse shrubs independently ([Fig 2](#pone.0133559.g002){ref-type="fig"}). The study design involved the following four combinations of factors: (i) browsers excluded but seedlings under shrub cover; plants were surrounded by a cylinder of wire netting with 1/2-inch mesh openings with a height of 70 cm and buried at least 20 cm deep to exclude browsers, planted under a shrub canopy directly beneath the zone of maximum canopy cover of the most common shrub species ([Fig 3A](#pone.0133559.g003){ref-type="fig"}); (ii) exposed to browsers but under shrub cover, seedlings unfenced to allow herbivore access and planted under the canopy of the most common shrubs ([Fig 3B](#pone.0133559.g003){ref-type="fig"}); (iii) browsers excluded but seedlings without the protection of a shrub canopy, plants surrounded with a 70-cm-high cylinder of wire netting that hampered access by browsers and lacked the potential facilitative effect of nurse plants ([Fig 3C](#pone.0133559.g003){ref-type="fig"}); and (iv) fully exposed plants, unfenced seedlings planted interspaces between shrubs ([Fig 3D](#pone.0133559.g003){ref-type="fig"}). {#pone.0133559.g002} {#pone.0133559.g003} The seedlings were surveyed every four weeks during the first third of the year-long experiment (July 2005 to March 2006) and once in summer (January 2007); hence, 11 sets of measurements were obtained. At each census, all seedlings were identified as alive or dead, distinguishing between abiotic and biotic stress (i.e., dried by stressful environmental conditions or consumed by herbivores); the number of shoots and shoot length individually measured; and examined for herbivore damage, expressed as the difference between the length of an affected shoot and the length previously measured. Two herbivore species, the exotic European rabbit (*Oryctolagus cuniculus*) and the endemic rodent degu (*Octodon degus*) were assessed, as both species usually defecate very close to the areas where they eat \[[@pone.0133559.ref032]\]. Domestic livestock were excluded from the experimental areas. Data analysis {#sec005} ------------- The temporal patterns of seedling death were addressed to compare seedling survival among the browser and microhabitat treatments. Curves of seedling survival over time were generated using Kaplan-Meier estimates \[[@pone.0133559.ref037], [@pone.0133559.ref038]\]. The survival functions describe the probability that an individual survives longer than a specified period, considering individuals at risk at the beginning of each interval, i.e., the lapse between two mortality events, and excluding censored values. Browser exclusion, microhabitat, and shrub type were introduced as factors \[[@pone.0133559.ref038]\]. A log rank (Mantel-Cox) test \[[@pone.0133559.ref039]\] was used to compare pairs of survival functions. The consequences of microhabitat, browsers, and their interaction for *J*. *chilensis* seedlings were assessed by fitting General Linear Mixed Models (GLMMs) separately to the seedling survival and seedling growth data for 2005--2007. The dependent variable in the models was the number of seedlings surviving in each plot or the difference in seedling biomass from the beginning on the experiment to the end. Exclusion from browsers and microhabitat and two-way interactions were inserted in the models as fixed effects. Time was included in the 'repeated' statement, where the scores were nested within individuals. Computations were performed with the SAS MIXED procedure and restricted maximum likelihood estimation \[[@pone.0133559.ref040]\]. A Pearson correlation coefficient was used to quantify the relationship between seed size and log-transformed seedling growth. We tested the data for conformity with the assumption of homogeneous variances. All comparisons reported used two-tailed significance tests at the 0.05 level and were performed with R \[[@pone.0133559.ref041]\] except for GLMMs fitted with the PROC MIXED package of SAS version 9.1 (SAS Institute Inc., Cary, NC). All relevant data are presented in [S1 Table](#pone.0133559.s001){ref-type="supplementary-material"}. Results {#sec006} ======= Our results show that for *Jubaea chilensis*, seedling survival and growth were not affected by seed size (Pearson correlation, P \> 0.05 for both comparisons). In fact, no differences in seedling growth were observed among treatment groups. The mean seedling growth was 0.94 ± 0.02 g and did not vary according to the microhabitat, shrub type or herbivore access (General Linear Mixed Model, P \> 0.05 for all comparisons, [Table 1](#pone.0133559.t001){ref-type="table"}). However, survival depended heavily on protection from vertebrate herbivores. By the end of the 18-mo period, 42 (14%) of the 300 seedlings remained alive, with 109 affected by abiotic unsuitable conditions (dried seedlings) and 149 lethally attacked by vertebrates. 10.1371/journal.pone.0133559.t001 ###### General linear mixed model effects. {#pone.0133559.t001g} Atribute Treatment *F* *df* *P* ----------------- ---------------------------------- ------- ------ ------ Survival Brownser access × nursing effect 21.43 1 0.00 Nursing effect 3.86 1 0.05 Brownser access 488.19 1 0.00 Growth Brownser access × nursing effect 1.65 1 0.20 Nursing effect 0.38 1 0.54 Brownser access 1.65 1 0.20 The seedlings that were accessible to browsers suffered higher mortality rates than those that were enclosed and inaccessible to vertebrates (F = 488.19, df = 1, P \< 0.0001, [Table 1](#pone.0133559.t001){ref-type="table"}). Regardless of microhabitat (outside and under shrub canopy), unfenced seedlings had negligible survival (n = 1, Figs [4](#pone.0133559.g004){ref-type="fig"} and [5](#pone.0133559.g005){ref-type="fig"}), with more than 95% of deaths evidently produced by biotic stress and a minor proportion triggered by abiotic stress (1.3% under canopy shrubs; 3.5% open canopy), showing that shrubs do not provide protection against herbivores as seedlings showed the same survival probability when exposed to browsers in the open interspaces between shrubs and under shrubs. However, an interaction between microhabitat and seedling protection was observed for seedling survival (F = 21.43, df = 1, P \< 0.0001, [Table 1](#pone.0133559.t001){ref-type="table"}), with better performance of fenced seedlings under shrub canopies ([Fig 4](#pone.0133559.g004){ref-type="fig"}). {#pone.0133559.g004} {#pone.0133559.g005} The probability of seedlings remaining alive differed significantly between treatments ([Fig 5](#pone.0133559.g005){ref-type="fig"}), with earlier deaths in wine palm seedlings that were exposed to herbivores. Fenced seedlings had an estimated mean time (± SE) to failure of 331 ± 19 and 433 ± 18 days for open interspaces between shrubs and growth under shrub canopy, respectively, much longer than the mean time predicted for seedlings accessible to browsers (under canopy = 85 ± 9 days; fully exposed seedlings = 114 ± 11 days). The proportion of caged seedlings died under the shade of shrubs canopies was lower than those exposed to browsers attack but under shrub cover (log-rank, χ^2^ = 128.34, P \< 0.0001), or than those inaccessible to browsers but seedlings without the protection of a shrub canopy (χ^2^ = 10.20, P = 0.001) or than those fully exposed seedlings, i.e., unfenced seedlings planted in open interspaces between shrubs (χ^2^ = 118.99, P \< 0.0001). In fact, caged and unshaded wine palm seedlings survived longer than those under shrub canopies exposed to browsers (χ^2^ = 102.41, P \< 0.0001) and or than those fully exposed seedlings (χ^2^ = 116.24, P \< 0.0001). However, the survival of wine palm seedlings was not influenced by nurse shrub protection on those seedlings open to browser (χ^2^ = 3.5, P \> 0.05, [Fig 4](#pone.0133559.g004){ref-type="fig"}). Discussion {#sec007} ========== Our findings from experimentally decoupling the herbivory and abiotic effects on wine palm seedling performance support the scarce empirical evidence indicating that semi-arid ecosystems have a strong impact on plant-plant interactions \[[@pone.0133559.ref011], [@pone.0133559.ref013], [@pone.0133559.ref014]\]. The intensity of shade-induced facilitation of growth did not differ during the course of the experiment, whereas its effect on survival did vary. The effects of shrub interactions on seedling survival differ from their effects on growth have been previously observed \[[@pone.0133559.ref042], [@pone.0133559.ref043]\], and that the intensity of facilitation of growth by shade did not vary with adverse environmental conditions may be consistent with other studies \[[@pone.0133559.ref042], [@pone.0133559.ref044], [@pone.0133559.ref045]\]. Once wine palm seedlings are able to escape from herbivores, the facilitative effect on modifying environmental conditions seem to null or at least reduce as seedlings showed similar growth in the ameliorated conditions under the shrubs and in the harsh open canopy. The effects of facilitation on survival and growth may vary simultaneously, probably as a consequence of differences in environmental requirements according to the plant life-stage. Desiccation during the summer and consumption by vertebrate herbivores, most likely due to the introduced and invasive European rabbit (*Oryctolagus cuniculus*), are important sources of wine palm seedling mortality. Contrary to our hypothesis, herbivory, rather than harsh environmental conditions, is the strongest inhibiting effect on *Jubaea* seedling establishment. Other studies have also shown that biotic filters mediated by animal-plant interactions might be more important than abiotic filters in the early life stages of plants \[[@pone.0133559.ref046]--[@pone.0133559.ref048]\]. Nevertheless, in the absence of consumer pressure, the abiotic mechanism of facilitation by nurse shrubs was observed for survival but not for seedling growth. Our findings provide only partial support for the SGH \[[@pone.0133559.ref002]\], as facilitation did not increase seedling survival or growth under high herbivore pressure (biotic stress). However, seedling establishment and growth depend presumably on the combined effect of biotic and abiotic filters and shrubs cannot provide the facilitative effects of abiotic stress amelioration and protection against herbivores. Thus, under extreme conditions, survival may not be possible even in the presence of positive interactions because shrubs do not facilitate seedling performance by providing protection against browsers, the primary and pervasive constraint to *Jubaea* seedling establishment. In contrast, at intermediate or low levels of herbivory, the wine palm seedlings are able to overcome mortality due to herbivore consumption and higher survival of shaded plants is a result of facilitation provided by shrubs due to the creation of a more humid, mesic and shaded environment, especially during summer, thus supporting the SGH. However, at high stress levels, with elevated consumer pressure, the net positive effect disappears. These results concur with the predictions Smit, Rietkerk \[[@pone.0133559.ref010]\] corroborated by Saiz and Alados \[[@pone.0133559.ref013]\], Graff and Aguiar \[[@pone.0133559.ref049]\], and Graff, Aguiar \[[@pone.0133559.ref050]\]. In low-productivity sites such as semi-arid Mediterranean habitats, stress-adapted herbivore species appear to compensate for nutritionally poor food by increasing the content of their diet, changing their foraging behavior by tending to be less selective, and increasing their feeding area \[[@pone.0133559.ref049], [@pone.0133559.ref051], [@pone.0133559.ref052]\]. Fuentes, Jaksic \[[@pone.0133559.ref032]\] demonstrated that in the central Chile, the European rabbit (*Oryctolagus cuniculus*) and the degu (*Octodon degus*), an endemic rodent, show differential killing effects on native vegetation, with the rabbits foraging over wide areas while the degu restricting its search activity to an area a few meters from their refuges. Jaksic, Fuentes, \[[@pone.0133559.ref053]\] add that kitten and juvenile rabbits are ecologically comparable to degu, limiting their foraging to areas under the canopy \[[@pone.0133559.ref016]\] possibility for explaining the observed widespread high mortality of unfenced wine palm seedlings and weakening the facilitative effect of shrub canopies. From a conservation and management perspective, small invasive consumers such as the alien rabbit play a major role in shaping vegetation dynamics, with consequences for biodiversity worldwide (\[[@pone.0133559.ref054]\], see \[[@pone.0133559.ref055]\]). This form of herbivory is one of the main causes of the lack of seedlings in stands of the endangered *J*. *chilensis*, a relict coming from the Tertiary Early Cenozoic that resisted even severe extinction rate due to rainforest decline in the Quaternary. Thus, the sensitivity of how plant-plant facilitation might be affected to perturbation could cause important disruptions in ecosystems modifying the species recruitment patterns. In extreme cases, herbivore overbrowsing can have profound effects on ecosystem processes through overconsumption, effectively reducing the diversity and carrying capacity for consumers and potentially leading to extinction or even to ecological meltdown \[[@pone.0133559.ref056], [@pone.0133559.ref057]\]. Our results show that despite the facilitation effect that nursing shrubs have on wine palm seedling establishment, recolonization of open spaces is possible without the facilitation of a nurse shrub, but protection of young wine palm seedlings is obligatory to enable them to attain a height at which shoots are no longer vulnerable to browsing. Our findings also stress the importance of assessing the influence of sources of biotic stress, specifically herbivory, on plant interactions as part of the effective planning for the preservation, restoration and management of arid and semi-arid ecosystems. Invasive vertebrate herbivores in natural ecosystems has being recorded all over the world (e.g. \[[@pone.0133559.ref057]\], \[[@pone.0133559.ref058]\], \[[@pone.0133559.ref059]--[@pone.0133559.ref063]\]) once both fragmented and continuous landscapes are suffering from the human-induced disturbances, with strong implications for the persistence of key ecosystem processes. We highlight the importance of assessing the impact of invasive mammals on ecosystems, as the invaders might limit seedling establishment not only for the studied species but other plant species, as well as affecting all steps of the recruitment process, with ensuing impacts on the population dynamics of the affected plants. Supporting Information {#sec008} ====================== ###### Wine palm seedling survival and growth among treatments. (XLSX) ###### Click here for additional data file. The authors are thankful for the assistance of Ronny Zúñiga, and for the helpful input and encouragement from Juan Armesto and Javier Simonetti. We thank the anonymous reviewers for their valuable suggestions and constructive comments. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: WM LAG RAV ROB. Performed the experiments: WM. Analyzed the data: MF. Contributed reagents/materials/analysis tools: WM. Wrote the paper: MF. Provided editorial advice: WM LAG RAV ROB. [^3]: Current address: Instituto de Conservación, Biodiversidad y Territorio, Facultad de Ciencias Forestales y Recursos Naturales, Universidad Austral de Chile, Valdivia, Chile

INTRODUCTION ============ Stress myocardial perfusion imaging using single-photon emission computed tomography (SPECT) is commonly used for risk stratification and therapeutic decision making in patients with coronary artery disease (CAD).[@B1][@B2][@B3] Since ischemia is a strong predictor of adverse outcomes, such as death or myocardial infarction (MI),[@B4][@B5] detecting ischemia is an important part of the diagnostic strategy for patients with stable CAD. In addition to the diagnostic and prognostic utility of myocardial perfusion SPECT, the extent of ischemia is one of the primary measures that drives decisions regarding revascularization.[@B6][@B7][@B8] Patients with moderately or severely abnormal myocardial perfusion SPECT have significantly higher mortality rates if treated with medical therapy alone.[@B5][@B7] Previous studies on myocardial perfusion SPECT have focused on the prognostic utility of stress imaging as the initial test for patients with CAD.[@B8][@B9] However, there are only a few studies that show a relationship between the presence and severity of ischemia and prognosis during CAD treatment.[@B10][@B11][@B12] Therefore, according to the current guidelines and appropriate use criteria for follow-up of CAD, routine stress imaging is not recommended, except for special high-risk groups after coronary revascularization.[@B13][@B14] Hence, in the current study, we aimed to assess the clinical implications of serial myocardial perfusion SPECT in patients with CAD who were receiving either medication or revascularization therapy. MATERIALS AND METHODS ===================== Study population ---------------- We identified consecutive patients who underwent serial myocardial perfusion SPECT, had abnormal results on a first study \[which was defined as summed stress score (SSS) ≥3\],[@B15] and had follow-up adenosine stress SPECT at an interval ≥6 months between the two studies that were performed between January 1, 2000 and June 31, 2014. Patients were also excluded if they had MI \<3 months before initial SPECT, previous revascularization therapy, serious non-coronary heart disease, including cancer with a life expectancy less than one year, incomplete nuclear data, multiple coronary revascularizations between the two SPECT procedures, and no clinical follow-up information. As a result, a total of 1153 patients with serial SPECT studies were included ([Fig. 1](#F1){ref-type="fig"}). Because of the retrospective nature of the study, a waiver for individual informed consent was granted by the Institutional Review Board. Myocardial perfusion imaging ---------------------------- Thallium-201 (Tl-201) SPECT was the default stress myocardial perfusion imaging used during the study period. Images were acquired with a standardized protocol.[@B16] Adenosine was intravenously administered at a rate of 140 mcg/kg/minute for 6 min. Three minutes after the initiation of the adenosine infusion, a dose of Tl-201 (range=92.5--148 MBq, as determined by the patient\'s body weight) was intravenously injected. Six minutes after adenosine infusion, post-stress myocardial perfusion images were acquired using two-head gamma cameras equipped with low-energy, all-purpose collimators. The specific acquisition parameters were dependent on the camera. Image interpretation -------------------- Semi-quantitative visual interpretation was performed by independent expert interpreters, using 17 segments for the severity and extent of abnormalities on stress imaging.[@B17] Each segment was scored using a 5-point scoring system (0=normal; 1=equiv-ocal; 2=moderate; 3=severe reduction in radioisotope uptake; 4=absence of detectable tracer uptake in a segment), as previously described.[@B18] The score that was summed from the stress scan was defined as the SSS: SSS was determined by adding the scores of the 17 segments on the stress images. The SPECT study was considered to be abnormal if the SSS was 3 or greater. According to the result of follow-up myocardial perfusion SPECT, patients were categorized into normal and abnormal groups. Procedure and follow-up ----------------------- Coronary angiography was recommended for patients on the basis of their clinical presentation and the results of the noninvasive stress test. Significant stenosis on coronary angiography was defined as \>50% stenosis in an epicardial coronary artery. In patients with significant stenosis, the decision to perform revascularization or medical therapy was at the discretion of the individual cardiologist. Percutaneous coronary intervention (PCI)[@B19] or coronary artery bypass graft (CABG) surgery was performed using standard techniques.[@B20] Medical treatment was performed with a medical regimen that consisted of at least antiplatelet, antianginal, and lipid-lowering therapies.[@B3] After index myocardial perfusion SPECT, patients received either medical treatment or revascularization treatment. Definitions ----------- The primary outcome of interest was the occurrence of major adverse cardiac events (MACE), which was the composite of all-cause death, nonfatal MI, or unplanned revascularization after follow-up myocardial perfusion SPECT. When patients received multiple serial myocardial perfusion SPECTs, the first follow-up SPECT with an interval of ≥6 months after the index myocardial perfusion SPECT was selected for analysis. An MI was defined as elevated cardiac enzymes (troponin I or myocardial band fraction of creatine kinase) more than the upper limit of the normal value with ischemic symptoms or electrocardiography findings that were indicative of ischemia. After follow-up myocardial perfusion SPECT, any further PCI or CABG (excluding planned staged PCI) was considered an unplanned revascularization. Death, non-fatal MI, and unplanned revascularizations were verified by reviewing medical records. Statistical analysis -------------------- The continuous and categorical covariates are summarized as a mean±standard deviation or count (%). According to the follow-up myocardial perfusion SPECT results, all patients were divided into normal and abnormal groups. The baseline patient characteristics were compared between the two groups using the t test or Fisher exact test for continuous and categorical variables, respectively. The cumulative incidence of MACE for the normal and the abnormal groups was obtained using the Kaplan-Meier method and compared between the two groups using the log-rank test. To examine the effect of abnormal results on MACE and its individual events, the unadjusted and adjusted Cox proportional hazards regression models were fitted. Covariates that were statistically significant in univariate analysis and/or those that were clinically relevant were considered candidate variables for multivariate models. In the Cox model, the proportionality assumptions were assessed using the Scho-enfeld residual test, and no relevant violations were detected. To reduce treatment selection bias and potential confounding, propensity score (PS)-matching analysis was performed. The PS of obtaining the abnormal myocardial perfusion SPECT results was estimated using the nearest-neighbor matching method with a caliper width of 0.2. The considered variables for the PS were age, sex, body mass index, hypertension, diabetes mellitus, current smoking, hyperlipidemia, prior revascularization treatment, chronic pulmonary disease, chronic renal failure, creatinine, total cholesterol, left ventricular (LV) ejection fraction \<50%, and the use of beta blockers, calcium channel blockers (CCB), angiotensin converting enzyme inhibitor (ACEi), angiotensin receptor blocker (ARB) agents, or statins, which are listed in [Table 1](#T1){ref-type="table"}. In general, covariate balancing was considered to be achieved as long as the absolute standardized difference of the means or proportions was \<0.25. In the PS analyses, no violations of covariate balancing were detected ([Supplementary Fig. 1](#S2){ref-type="supplementary-material"}, only online). For the PS-matched cohorts, continuous variables were compared using the paired t test or the Wilcoxon signed-rank test, as appropriate, and the categorical variables were compared using the McNemar\'s or Bowker\'s test of symmetry, as appropriate. A subgroup analysis was performed according to the treatment groups. Furthermore, univariate and multivariate logistic regression analyses were performed to identify independent clinical predictors of having abnormal results on follow-up myocardial perfusion SPECT. In the multivariable logistic regression, we employed a backward variable selection approach based on the *p* values. The significance level for staying in the model was set to 0.05. All statistical analyses were performed using SPSS (version 19.0 software; IBM Corp., Armonk, NY, USA) and R software (version 2.13; R Foundation for Statistical Computing, Vienna, Austria; <http://www.r-project.org>). Additionally, the R package MatchIt was used to conduct the PS analysis.[@B21] All tests were two-tailed, and *p*\<0.05 was considered statistically significant. RESULTS ======= Overall population ------------------ ### Baseline characteristics The median follow-up interval between the index and follow-up myocardial perfusion SPECT procedures was 474 days \[interquartile range (IQR)=243--1107 days\]. Abnormal results on follow-up myocardial perfusion SPECT were noted in 591 patients (51.3%). The baseline clinical characteristics and myocardial perfusion SPECT results, according to the results of the follow-up myocardial perfusion SPECT procedures and treatment strategy, are presented in [Table 1](#T1){ref-type="table"}. The patients in the abnormal group were more likely to be male and have a higher incidence of hypertension, diabetes mellitus, and a low ejection fraction. On myocardial perfusion SPECT, the abnormal group had greater baseline perfusion defects than the normal group. The revascularization therapy group was older and more likely to have diabetes mellitus, greater extent of CAD, and high perfusion defect on the initial myocardial perfusion SPECT ([Supplementary Table 1](#S1){ref-type="supplementary-material"}, only online). ### Clinical outcomes Patients were followed for a median of 6.3 years (IQR=3.7--9.1 years) after the follow-up myocardial perfusion SPECT. During the follow-up period, 188 patients died, 70 patients developed nonfatal MI, and 93 patients underwent unplanned revascularization ([Table 2](#T2){ref-type="table"}). The incidence of MACE was significantly higher in the abnormal group in the overall population and both treatment groups ([Fig. 2](#F2){ref-type="fig"}), and there was also a significant difference in the clinical outcomes of patients according to the result of serial SPECT ([Fig. 3](#F3){ref-type="fig"}). There was also a tendency for a higher risk of events in patients with abnormal follow-up SPECT in comparison with normal SPECT in all subgroups after adjustment using the multivariate Cox model, as indicated in [Table 2](#T2){ref-type="table"}. Propensity-matched population ----------------------------- ### Baseline characteristics After performing PS matching, 346, 112, and 212 matched pairs of patients in the overall population, medication alone, and revascularization groups were created, respectively ([Table 3](#T3){ref-type="table"}). There were no significant differences in the baseline clinical characteristics of the PS-matched patients, except the extent of CAD and the results of SPECT. ### Clinical outcomes In the matched patients, there were 177 MACE events within the median follow-up period of 6.6 years. With respect to the MACE in the matched cohorts, the abnormal group was associated with worse clinical outcomes in the overall population and the revascularization groups ([Fig. 4](#F4){ref-type="fig"}). We noted a consistent pattern in that the abnormal follow-up SPECT group was at a higher risk of MACE and its individual events than the normal group in PS-matched overall patients, medication alone and revascularization groups ([Table 4](#T4){ref-type="table"}). Predictors of abnormal follow-up SPECT -------------------------------------- [Table 5](#T5){ref-type="table"} lists the independent predictors of abnormal results on follow-up SPECT. The variables considered for inclusion in the multivariate models were sex, hypertension, diabetes mellitus, body mass index ≥25 kg/m^2^, current smoking, chronic kidney disease, history of previous revascularization therapy, LV ejection fraction \<50%, treatment with CCB, ACEi, or ARB, and the treatment method. Low ejection fraction was a predictor of abnormal follow-up SPECT, the overall population, medication alone and revascularization groups. Revascularization therapy was a protective factor of abnormal SPECT. DISCUSSION ========== In the present study, we evaluated the long-term clinical implications of follow-up SPECT during CAD treatment with comparisons of the clinical outcomes between patients with normal and abnormal follow-up SPECT results. The major findings of this study included the following: 1) abnormal results on follow-up SPECT were associated with worse clinical outcomes, as indicated by an increased risk of death, MI, or unplanned revascularization in the overall study population, and 2) low ejection fraction and medical treatment were independent clinical predictors of having an abnormal result on follow-up SPECT. Previous randomized studies of serial myocardial perfusion SPECT demonstrated significant reductions in myocardial ischemia after diverse interventions, including the administration of medical therapy or coronary revascularization.[@B22][@B23] In the COURAGE sub-study, residual ischemia was an unadjusted predictor of events, although the noted association was not significant when adjusted for the treatment arm.[@B10] Recent studies have further suggested that a reduction in ischemia on myocardial perfusion SPECT with either medical or coronary revascularization therapy may identify a patient at lower risk for a subsequent cardiac event.[@B10][@B24] Despite the potential advantages of follow-up myocardial perfusion SPECT, clinical practice guidelines and appropriate use criteria do not recommend routine serial testing due to the lack of evidence that supports the benefit of follow-up myocardial perfusion SPECT.[@B13][@B25] A report suggested that \>5% worsening ischemia is an independent predictor of death or MI, irrespective of the treatment arm, such as medical therapy or revascularization.[@B26] However, these previous studies were limited by small study populations or case-control designs. Our study, despite its retrospective, single-center, observational study design, is relatively stronger, because it includes a larger study population that was treated at a real-world practice with a longer follow-up period. The present study demonstrates that follow-up ischemia is associated with adverse outcomes, such as death, MI, and unplanned revascularization. The association between mortality and follow-up ischemia is in line with a previous study that reported that revascularization reduces mortality in patients with an ischemic burden ≥10% on baseline myocardial perfusion SPECT.[@B7][@B27] In our study, among the overall population and the revascularization group, abnormal results on follow-up SPECT demonstrated an increased risk of MACE, death, and MI. In the medical group, abnormal SPECT was also associated with a higher risk of MACE and unplanned revascularization. Due to the observational study design, unfavorable clinical factors, such as male sex and lower LV ejection fraction, may contribute to worse clinical outcomes in patients with abnormal follow-up SPECT. However, consistent findings after rigorous adjustment with PS matching support our hypothesis that the prognostic benefit of baseline myocardial perfusion SPECT could be applied to follow-up perfusion SPECT in order to predict long-term clinical prognosis. These results indicate that serial myocardial SPECT after receiving either medication alone or revascularization therapy may be helpful for predicting prognosis and subsequently determining the need for more aggressive treatment. According to current guidelines, myocardial perfusion SPECT is considered appropriate when symptom recurrence, suspected incomplete revascularization, or ≥5 years after CABG.[@B13][@B28] Therefore, it is of interest to select appropriate patients who can receive myocardial SPECT as a risk assessment modality when they demonstrate the ambiguous presentation of symptoms. Our present findings indicate that abnormal myocardial SPECT results are common in patients with low LV ejection fraction and treatment with medication only. These factors may be indirectly associated with abnormal follow-up SPECT due to the risk of revascularization or decompensated symptoms of CAD.[@B29][@B30] Therefore, our analysis implies that patients with the risk factors for abnormal SPECT are potential candidates who could benefit from receiving follow-up SPECT in order to determine the appropriate treatment strategy. Our study had some limitations. This was a single-center, observational, retrospective study with the biases that are inherent to this type of analysis. There were also significant baseline differences between patients with normal and abnormal follow-up myocardial SPECT. In addition, patients did not routinely receive serial myocardial perfusion SPECT; rather SPECT was performed at the discretion of the individual physician. Even after statistical adjustment therefore, unobserved confounders might have influenced the results. However, the patients in this study may be representative of a real-world population in daily clinical practice. Finally, due to the small study population, an ischemic threshold on follow-up SPECT for determining clinical prognosis was not evaluated. In conclusion, abnormal follow-up SPECT results appear to have prognostic implications during CAD treatment in patients receiving either medication alone or revascularization therapy. This study was partly supported by a grant from the Korea Healthcare Technology R&D Project, Ministry of Health and Welfare, Republic of Korea (HI14C0517 and HI15C1790). The authors have no financial conflicts of interest. SUPPLEMENTARY MATERIALS ======================= ###### Supplementary Table 1 Baseline Characteristics According the Treatment Groups ###### Supplementary Fig. 1 Distribution of propensity scores and standardized differences in the mean or proportion of variables before and after matching. (A, B, and C). Distribution of propensity scores before and after propensity score matching. (D, E, and F). Standardized differences in the mean or proportion of variables before and after propensity score matching. Each circle indicates the individual covariates in the propensity score model, and each line indicates how they changed after propensity score matching. {#F1} {#F2} {#F3} {#F4} ###### Baseline Characteristics of the Study Patients with Normal and Abnormal Follow-Up SPECT Results  Variables Total population Medication group Revascularization group --------------------------------- ------------------ ------------------ ------------------------- -------------- -------------- --------- ------------- ------------- --------- Age, yr 61.4±9.6 61.7±10.0 0.564 61.6±10.4 61.1±10.3 0.611 61.3±9.3 62.2±9.8 0.191 Male 380 (67.6) 435 (73.6) 0.030 110 (65.9) 199 (74.0) 0.089 270 (68.4) 236 (73.3) 0.174 BMI, kg/m^2^ 25.1±3.0 25.2±3.5 0.447 25.3±2.9 25.0±3.8 0.421 25.0±3.0 25.4±3.3 0.109 Hypertension 340 (60.5) 395 (66.8) 0.030 98 (58.7) 191 (71.0) 0.011 242 (61.3) 204 (63.4) 0.620 Diabetes mellitus 179 (31.9) 233 (39.4) 0.009 44 (26.3) 94 (34.9) 0.077 135 (34.2) 139 (43.2) 0.017 Hyperlipidemia 201 (35.8) 207 (35.0) 0.841 59 (35.3) 110 (40.9) 0.290 142 (35.9) 97 (30.1) 0.117 Current smoker 162 (28.8) 184 (31.1) 0.149 54 (32.3) 83 (30.9) 0.770 108 (27.3) 101 (31.4) 0.482 Chronic renal failure 23 (4.1) 33 (5.6) 0.298 13 (7.8) 25 (9.3) 0.713 10 (2.5) 8 (2.5) 1.000 Chronic lung disease 15 (2.7) 10 (1.7) 0.349 9 (5.4) 7 (2.6) 0.214 6 (1.5) 3 (0.9) 0.715 Laboratory data  Total cholesterol, mg/dL 180.6±42.5 181.2±41.2 0.801 182.1±40.2 181.2±38.9 0.825 178.0±43.4 181.2±43.0 0.703  Serum creatinine, mg/dL 1.1±1.0 1.3±1.7 0.009 1.2±1.2 1.5±2.2 0.071 1.1±0.9 1.2±1.2 0.217  LV EF, % 59.0±8.1 53.1±11.6 \<0.001 57.5±8.6 53.0±11.6 \<0.001 59.7±7.8 53.2±11.7 \<0.001   LV EF \<50% 36 (6.4) 175 (29.6) \<0.001 13 (7.8) 78 (29.0) \<0.001 23 (5.8) 97 (30.1) \<0.001 Medications  Beta-blocker 424 (75.4) 464 (78.5) 0.243 116 (69.5) 208 (77.3) 0.087 308 (78.0) 256 (79.5) 0.685  CCB 512 (91.1) 489 (82.7) \<0.001 126 (75.4) 192 (71.4) 0.412 386 (97.7) 297 (92.2) 0.001  ACEi or ARB 164 (29.2) 280 (47.4) \<0.001 53 (31.7) 138 (51.3) \<0.001 111 (28.1) 142 (44.1) \<0.001  Statin 340 (60.5) 339 (57.4) 0.307 55 (32.9) 116 (43.1) 0.044 285 (72.2) 223 (69.3) 0.443 Coronary angiography 463 (82.4) 442 (74.8) 0.002 91 (54.5) 144 (53.5) 0.923 398 (100.0) 324 (100.0) 1.000  Disease extent \<0.001 \<0.001 \<0.001   1 vessel 129 (23.0) 105 (17.8) 24 (14.4) 49 (18.2) 85 (21.4) 56 (17.4)   2 vessel 111 (19.8) 110 (18.6) 16 (9.6) 39 (14.5) 95 (23.9) 71 (22.0)   3 vessel 207 (36.8) 229 (38.7) 11 (6.6) 35 (13.0) 196 (49.2) 194 (60.2)  Left main disease 83 (14.8) 72 (12.2) 0.048 3 (1.8) 4 (1.5) 0.365 80 (20.3) 68 (21.1) 0.520 SPECT results  Interval between 2 tests, days 734.2±723.9 817.6±705.7 0.048 1237.0±802.2 1082.8±724.8 0.039 521.6±567.9 596.1±607.1 0.091  SSS at baseline 8.3±4.3 9.9±5.2 \<0.001 7.2±4.0 8.2±4.2 0.011 8.8±4.4 11.3±5.4 \<0.001  SSS at follow-up 0.8±1.2 7.8±3.9 \<0.001 0.9±1.2 7.6±3.7 \<0.001 0.8±1.1 8.0±4.0 \<0.001  Angina at follow-up 114 (20.3) 124 (21.0) 0.770 41 (24.6) 61 (22.7) 0.653 73 (18.5) 63 (19.0) 0.713 ACEi, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; BMI, body-mass index; CCB, calcium channel blocker; LV EF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography; SSS, summed stress score. Values are presented as a n (%) or mean±SD. ###### Clinical Outcomes in Patients with Abnormal Follow-Up SPECT in Comparison with Normal SPECT in the Overall Population  Variables Normal Abnormal Unadjusted HR (95% CI) *p* value Adjusted HR (95% CI)^\*^ *p* value ------------------------------ ------------ ------------ ------------------------ ----------- -------------------------- ----------- Overall population^†^  MACE 111 (19.8) 191 (32.3) 1.735 (1.373−2.193) \<0.001 1.595 (1.249−2.038) \<0.001  Death 74 (13.2) 114 (19.3) 1.472 (1.098−1.972) 0.010 1.235 (0.907−1.681) 0.180  Myocardial infarction 22 (3.9) 48 (8.1) 2.114 (1.276−3.502) 0.004 1.697 (0.997−2.888) 0.051  Unplanned revascularization 37 (6.6) 56 (9.5) 1.477 (0.975−2.237) 0.066 1.634 (1.060−2.517) 0.026 Medication group^‡^  MACE 36 (21.6) 87 (32.3) 1.492 (1.012−2.201) 0.044 1.472 (0.979−2.211) 0.063  Death 32 (19.2) 54 (20.1) 0.974 (0.629−1.509) 0.907 0.978 (0.616−1.550) 0.923  Myocardial infarction 5 (3.0) 23 (8.6) 2.667 (1.013−7.021) 0.047 2.134 (0.770−5.915) 0.145  Unplanned revascularization 5 (3.0) 20 (7.4) 2.309 (0.864−6.168) 0.095 2.118 (0.764−5.872) 0.149 Revascularization group^§^  MACE 75 (19.0) 104 (32.3) 1.772 (1.316−2.384) \<0.001 1.662 (1.218−2.269) 0.001  Death 42 (10.6) 60 (18.6) 1.720 (1.159−2.552) 0.007 1.409 (0.930−2.133) 0.106  Myocardial infarction 17 (4.3) 25 (7.8) 1.791 (0.967−3.317) 0.064 1.613 (0.847−3.072) 0.146  Unplanned revascularization 32 (8.1) 36 (11.2) 1.414 (0.878−2.277) 0.154 1.602 (0.980−2.620) 0.060 ACEi, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; CI, confidence interval; HR, hazard ratio; MACE, major adverse cardiac events; SPECT, single-photon emission computed tomography. ^\*^The adjusted covariates included age, diabetes mellitus, chronic renal failure, chronic lung disease, use of ACEi or ARB, use of statins, and left ventricular ejection \<50%, ^†^Nomal: n=562, abnormal: n=591, ^^‡^^Nomal: n=167, abnormal: n=269, ^§^Nomal: n=395, abnormal: n=322. ###### Baseline Characteristics between the Patients with Normal and Abnormal Follow-Up SPECT Results in the Propensity Score-Matched Population  Variables Total population Medication group Revascularization group ---------------------------------- ------------------ ------------------ ------------------------- -------------- -------------- --------- ------------- ------------- ------- Age, yr 61.8±9.2 61.8±9.6 0.952 60.5±10.6 61.5±9.4 0.480 61.0±9.5 62.4±10.0 0.140 Male 389 (70.1) 404 (72.8) 0.340 74 (66.1) 74 (66.1) 1.000 148 (69.8) 144 (67.9) 0.753 BMI, kg/m^2^ 25.1±2.8 25.2±3.0 0.880 25.3±2.5 25.2±3.1 0.698 24.9±3.2 25.1±2.9 0.477 Hypertension 217 (62.7) 216 (62.4) 1.000 68 (60.7) 78 (69.6) 0.207 132 (62.3) 138 (65.1) 0.614 Diabetes mellitus 112 (32.4) 119 (34.4) 0.629 28 (25.0) 29 (25.9) 1.000 77 (36.3) 82 (38.7) 0.688 Hyperlipidemia 119 (34.4) 113 (32.7) 0.687 39 (34.8) 42 (37.5) 0.781 66 (31.1) 66 (31.1) 1.000 Current smoker 93 (27.5) 97 (29.0) 0.723 38 (33.9) 33 (29.5) 0.712 67 (31.8) 59 (27.8) 0.438 Chronic renal failure 14 (4.0) 12 (3.5) 0.842 9 (8.0) 7 (6.2) 0.795 3 (1.4) 4 (1.9) 1.000 Chronic lung disease 8 (2.3) 7 (2.0) 1.000 8 (7.1) 3 (2.7) 0.216 2 (0.9) 3 (1.4) 1.000 Laboratory data  Total cholesterol, mg/dL 183.0±42.1 182.6±42.7 0.902 185.2±40.4 181.6±34.9 0.480 179.5±44.0 180.9±43.8 0.739  Serum creatinine, mg/dL 1.1±0.0 1.1±1.1 0.876 1.2±1.3 1.1±1.2 0.663 1.1±1.1 1.1±1.1 0.872  LV EF, % 59.2±8.4 58.3±7.9 0.170 58.5±8.0 59.2±6.7 0.496 58.9±7.9 58.2±7.7 0.297   LV EF \<50% 24 (6.9) 23 (6.6) 1.000 7 (6.2) 5 (4.5) 0.767 17 (8.0) 18 (8.5) 1.000 Medications  Beta-blocker 271 (78.3) 272 (78.6) 1.000 80 (71.4) 84 (75.0) 0.651 168 (79.2) 172 (81.1) 0.715  CCB 313 (90.5) 301 (87.0) 0.186 85 (75.9) 84 (75.0) 1.000 206 (97.2) 202 (95.3) 0.445  ACEi or ARB 108 (31.2) 98 (28.3) 0.454 29 (25.9) 30 (26.8) 1.000 74 (34.9) 64 (30.2) 0.351  Statin 200 (57.8) 199 (57.5) 1.000 34 (30.4) 29 (25.9) 0.552 144 (67.9) 146 (68.9) 0.917 Coronary angiography 292 (84.4) 288 (83.2) 0.757 67 (59.8) 69 (61.6) 0.891 196 (92.5) 196 (92.5) 1.000  Disease extent 0.016 0.008 0.025   1 vessel 63 (18.2) 43 (12.4) 19 (17.0) 21 (18.8) 42 (19.8) 22 (10.4)   2 vessel 67 (19.4) 75 (21.7) 13 (11.6) 19 (17.0) 52 (24.5) 46 (21.7)   3 vessel 136 (39.3) 152 (43.9) 9 (8.0) 19 (17.0) 102 (48.1) 127 (59.9)  Left main disease 51 (14.7) 47 (13.6) 0.893 3 (2.7) 2 (1.8) 0.327 37 (17.5) 41 (19.3) 0.854 SPECT results  Intervals between 2 tests, days 728.6±699.0 752.7±657.5 0.640 1230.1±794.9 1056.6±713.0 0.087 507.8±612.9 617.7±577.9 0.058  SSS at baseline 8.1±4.0 9.3±4.9 \<0.001 7.0±3.6 7.4±3.6 0.367 8.8±4.5 10.7±5.0 0  SSS at follow-up 0.8±1.1 7.3±3.7 \<0.001 0.7±1.1 6.5±2.8 \<0.001 0.8±1.2 7.4±3.7 0  Angina at follow-up 71 (20.5) 77 (22.3) 0.643 30 (26.8) 26 (23.2) 0.643 42 (19.8) 48 (22.6) 0.553 ACEi, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; BMI, body-mass index; CCB, calcium channel blocker; LV EF, left ventricular ejection fraction; SPECT, single-photon emission computed tomography; SSS, summed stress score. Values are presented as a n (%) or mean±SD. ###### Clinical Outcomes in Patients with Abnormal Follow-Up SPECT Compared with Normal SPECT in a Propensity-Matched Population  Variables Normal group Abnormal group Adjusted HR^\*^ (95% CI) *p* value ------------------------------------- -------------- ---------------- -------------------------- ----------- Overall population (346 pairs)  MACE 66 (19.1) 111 (32.1) 1.727 (1.273−2.344) \<0.001  Death 44 (12.7) 59 (17.1) 1.230 (0.830−1.821) 0.302  Myocardial infarction 15 (4.3) 26 (7.5) 1.694 (0.896−3.202) 0.105  Unplanned revascularization 25 (7.2) 38 (11.0) 1.566 (0.945−2.595) 0.082 Medication group (112 pairs)  MACE 25 (22.3) 34 (30.4) 1.374 (0.801−2.358) 0.248  Death 22 (19.6) 17 (15.2) 0.701 (0.359−1.367) 0.297  Myocardial infarction 3 (2.7) 12 (10.7) 5.370 (1.401−20.583) 0.014  Unplanned revascularization 5 (4.5) 8 (7.1) 1.488 (0.457−4.842) 0.509 Revascularization group (212 pairs)  MACE 38 (17.9) 64 (30.2) 1.730 (1.154−2.594) 0.008  Death 22 (10.4) 32 (15.1) 1.274 (0.734−2.208) 0.389  Myocardial infarction 7 (3.3) 15 (7.1) 2.094 (0.850−5.160) 0.108  Unplanned revascularization 18 (8.5) 27 (12.7) 1.571 (0.863−2.861) 0.140 ACEi, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; CI, confidence interval; HR, hazard ratio; MACE, major adverse cardiac events; SPECT, single-photon emission computed tomography. ^\*^Adjusted covariates included age, diabetes mellitus, chronic renal failure, chronic lung disease, use of ACEi or ARB, use of statin, and left ventricular ejection \<50%; adjusted HRs represent the risk of each clinical outcome comparing abnormal follow-up SPECT with normal follow-up SPECT. ###### Independent Predictors of an Abnormal Follow-Up SPECT^\*^  Variables Revascularization group Revascularization group Revascularization group ----------------- ------------------------- ------------------------- ------------------------- ------- ------------- ------- ------- -------------- --------- Male 1.342 0.968−1.860 0.077 2.147 1.133−4.069 0.019 1.161 0.788−1.710 0.449 BMI ≥25 kg/m^2^ 1.124 0.851−1.486 0.410 0.854 0.499−1.464 0.567 1.262 0.905−1.759 0.170 LV EF \<50% 5.330 3.393−8.374 \<0.001 3.499 1.462−8.373 0.005 6.041 3.558−10.256 \<0.001 ACEi or ARB 1.240 0.905−1.699 0.181 1.352 0.735−2.485 0.332 1.234 0.847−1.796 0.273 Medical therapy 2.683 1.934−3.721 \<0.001 \- \- \- \- \- \- ACEi, angiotensin converting enzyme inhibitor; ARB, angiotensin receptor blocker; BMI, body-mass index; CI, confidence interval; LV EF, left ventricular ejection fraction; OR, odds ratio; SPECT, single-photon emission computed tomography. ^\*^Adjusted covariates included male sex, hypertension, diabetes mellitus, current smoking, chronic renal failure, BMI ≥25 kg/m^2^, LV EF \<50%, treatment with ACEi or ARB, and medication therapy. [^1]: ^\*^Se Hun Kang and Hyo In Choi contributed equally to this work.

Zhang J, Gao X, Han Y, *et al*. Treatment effects of systematic two-stent and provisional stenting techniques in patients with complex coronary bifurcation lesions: rationale and design of a prospective, randomised and multicentre DEFINITION II trial. *BMJ Open* 2018;8:e020019. doi: 10.1136/bmjopen-2017-020019. The sample size estimation in Statistical Analysis Part should be: From previous studies, we hypothesized that the rate of a 1-year TLF would be 7% in the systematic two-stent technique group and 14% in the provisional stenting group. Accordingly, a total sample size of 600 is needed to detect a power of 0.8 (Type Ⅱ error = 0.2, = 0.05, 2-tailed). Because of the considerable uncertainty, the enrolment is extended to 660 patients (10% increment). Instead of: From previous studies, we hypothesized that the rate of a 1-year TLF would be 15% in the systematic two-stent technique group and 25% in the provisional stenting group. Accordingly, a total sample size of 600 is needed to detect a power of 0.8 (Type Ⅱ error = 0.2, = 0.05, 2-tailed). Because of the considerable uncertainty, the enrolment is extended to 660 patients (10% increment).

{#sp1 .92}

1. Introduction =============== Fluid flowmeter calibration methods use a variety of techniques; they include wide ranges of operational parameters; they cover wide ranges of precision and accuracy, \[[@b1-jresv97n5p509_a1b]--[@b3-jresv97n5p509_a1b]\]. Increasingly, the diverse improvements being sought for flowmeters are producing corresponding improvements in the characteristics of flowmeter calibration systems. Of these systems, the piston displacement-type calibrator offers advantages such as compactness, mobility, efficient change of fluid, and prospects for state-of-the-art performance characteristics. To characterize the performance of a piston displacement calibrator, which produces a pulsed output signal that is proportional to the volumetric flowrate, the objective is, generally, to determine the "pulses per volume displacement" ratio (or its reciprocal) where the pulse output is assumed to come from a source such as a linear encoder. This pulse output is also assumed to be proportional to the piston displacement. High accuracy calibrator performance requires examination of these assumptions. A number of techniques can be used to determine volumetric displacement. It is assumed in what follows that the displaced fluid is a liquid, but the principles apply to gases or mixtures of gases and liquids as well. It is also assumed that both temperature and pressure effects on all the components of the piston displacement system, the cylinder, encoder, and fluid should be considered in order that high accuracy performance can be achieved. Depending upon the desired uncertainty level for the performance of the calibrator, one or more of the pressure and temperature effects on the components of the system may be negligible. When this is so, it may be permissible to disregard such effects to simplify data processing or to reduce the size of the controlling software for the system. Alternatively, and more preferably, all effects can be included in computer software. In this way, the terms which are negligible will not influence the results when higher levels of uncertainty, i.e., less precise performance can be tolerated or is desirable from benefits vs. costs perspectives. More importantly, where high accuracy is required, more of the figures available via the software capabilities can be accepted as significant. The volumetric-type calibrator system, using encoded piston displacement as both the flow source and as the flow determination scheme, is sketched in [Fig. 1](#f1-jresv97n5p509_a1b){ref-type="fig"}. The piston motion produces and measures a fluid volumetric flowrate that is proportional to the encoder frequency. The piston in the cylinder has a seal that is assumed to seal perfectly for all piston velocities. The corresponding fluid flowrate through the meter and the meter output frequency enable a calibration of the flowmeter. Using these elements, three steps are considered. The first step is determining the calibrator factor which is an "encoder pulses per volume displaced" ratio (or its reciprocal) at the defined set of reference conditions. These results can be obtained experimentally in a number of ways: (1) by physical measurement, (2) by the so-called "draw" technique, or (3) by using a transfer standard such as a single or, preferably, a tandem arrangement of calibrated turbine flowmeters. The second step is the use of the characterized calibrator to calibrate a pulsed-output flowmeter such as a turbine meter or other device where the meter factor is a "pulses per volume" quantity or its reciprocal. This process is done with one or more selected fluids and at specified conditions. In what follows turbine flowmeters will be considered. Conventionally, turbine meter results are produced in the form of a meter factor which has units of pulses per volume or volume per pulses (referenced to specified conditions). This meter factor is determined over the desired ranges of fluid conditions and flowrate expressed in terms of a ratio of inertial-to-viscous effects such as a Reynolds number, or equivalent parameter. The third step is the use of the characterized turbine flowmeter to calculate a fluid flowrate under actual conditions of use. The results can be produced with respect to specified reference conditions or to the actual conditions, depending upon the needs of the meter operator. The purpose of this paper is to describe these three steps and give the pertinent relationships that pertain to each procedure. The resulting equations are intended to be used in the software packages used with these types of calibrators and metering units. In this way, it is expected that the measurement performance of both the calibrators and the metering units can be maximized. 2. Calibrator Characterization ============================== 2.1 Geometrical Determination at Reference Conditions ----------------------------------------------------- To perform the required measurements at reference conditions and then calculate the calibrator factor in units of pulses per fluid volume displaced, the system analyzed is that sketched in [Fig. 2](#f2-jresv97n5p509_a1b){ref-type="fig"}. The assumption of reference conditions is a conceptual situation that is impractical to achieve precisely but is done solely for convenience, as will be clear in what follows. For the conditions selected to be the reference conditions which are denoted by the "0" subscript, a specific piston stroke produces a displacement volume, *V*~C0~, and the encoder produces the corresponding number of pulses, *N*~E0~. A list of symbols is given in the Glossary. The reference conditions of calibrator temperature and the pressure in the calibrator are *T*~C0~ and P~C0~, respectively. These properties are assumed to be constant and steady in the calibrator volume. These conditions should be monitored, quantified, and assessed with respect to the performance level of the calibrator. Furthermore, the fluid inside and outside the calibrator is assumed to have the same temperature as the cylinder of the calibrator and no heat is being transferred to or from the calibrator. In all that follows, the reference conditions of *T*~0~ and *P*~0~ will be assumed to be the same for all components. The specified piston stroke, *L*~E0~, produces the pulse total, *N*~E0~, where $$N_{E0} = L_{E0}K_{E0},$$and *K*~E0~ is the encoder constant in pulses per length at the reference temperature condition, *T~0~.* The calibration constant can be written $$K_{C0} = \frac{N_{E0}}{V_{C0}} = \frac{L_{E0}K_{E0}}{{\overline{A}}_{C0}L_{E0}} = \frac{K_{E0}}{{\overline{A}}_{C0}},$$in units of pulses per volume, where, at the reference conditions, *Ā*~C0~ is the averaged cylinder cross-sectional area over the piston stroke, *L*~E0~. This calibrator constant, *K*~C0~ is assumed to be constant over the operating range of the calibrator. For the calibrator configuration shown in [Fig. 1](#f1-jresv97n5p509_a1b){ref-type="fig"}, the area *Ā*~C0~ is an annular one between the cylinder of the calibrator and the rod or tube attached to the piston. The precision with which *K*~C0~ is determined can be written using root-sum-square combinations of component precisions: $$\begin{array}{l} {\frac{\Delta K_{C0}}{K_{C0}} < \left\lbrack {\left( \frac{\Delta N_{E0}}{N_{E0}} \right)^{2} + \left( \frac{\Delta V_{C0}}{V_{C0}} \right)^{2}} \right\rbrack^{1/2}} \\ {= \left\lbrack {\left( \frac{\Delta K_{E0}}{K_{E0}} \right)^{2} + \left( \frac{\Delta{\overline{A}}_{C0}}{{\overline{A}}_{C0}} \right)^{2}} \right\rbrack^{1/2},} \\ \end{array}$$where the numerators of the respective terms refer to the maximum errors of each of the component measurements. [Equation (3)](#fd3-jresv97n5p509_A1b){ref-type="disp-formula"} indicates that high levels of precision in *K*~C0~ can be attained when large pulse sums, *N*~E0~, and large displaced volumes, *V*~C0~, are used. Correspondingly, these precision levels can be achieved with accurate and sensitive linear encoders and accurately measured and large cross-sectional areas. 2.2 Geometrical Determination at Non-Reference Conditions --------------------------------------------------------- To perform the required measurements at non-reference temperature and pressure conditions, it is assumed that these conditions are constant and steady as shown in [Fig. 3](#f3-jresv97n5p509_a1b){ref-type="fig"}. For a specified piston stroke, *L*~E~, the corresponding encoder pulse total, *N*~E~, is $$N_{E} = N_{E}K_{E},$$where the encoder constant, *K*~E~, is assumed to depend only on temperature according to $$K_{E} = K_{E0}\left\lbrack {1 - \alpha_{E}\left( {T_{E} - T_{E0}} \right)} \right.,$$where *α*~E~ is the pertinent linear expansion coefficient for the encoder and *T*~E~ and *T*~E0~ are, respectively, the encoder temperatures at non-reference and reference conditions. The cross-sectional area change of the calibrator cylinder that is produced by temperature, *T*~C~, and internal fluid pressure, *P*~C~, in excess of the reference values is analyzed in [Appendix A](#app1-jresv97n5p509_A1b){ref-type="app"}. In what follows, it is assumed that the annular area contained between the calibrator cylinder and the rod or tube connected to the piston is changed with temperature in the usual way and the pressure effect is considered to enlarge only the cylinder of the calibrator. The pressure effect on the tube or rod is taken to be negligible. The cross-sectional area of the cylinder of the calibrator averaged over the stroke length is given, to first order in temperature and pressure separately, by $${\overline{A}}_{C} = {\overline{A}}_{C0}\left\lbrack {1 + 2\alpha_{C}\left( {T_{C} - T_{0}} \right)} \right\rbrack\left\lbrack {1 + \frac{\left( {P_{C} - P_{0}} \right)D_{C0}}{t_{C0}E_{C}}} \right\rbrack,$$where *α*~C~ is the linear expansion coefficient for the material of the calibrator cylinder, *D*~C0~ and *t*~C0~ are, respectively, the inside diameter and wall thickness of the calibrator cylinder at reference conditions and *E*~C~ is the modulus of elasticity of the material of the calibrator cylinder. In [Eq. (6)](#fd6-jresv97n5p509_A1b){ref-type="disp-formula"} the compressive effects of the pressure in the cylinder on the rod attached to the piston are assumed negligible. If it is assumed that the linear expansion coefficient of the cylinder material is 2 × 10^−5^ °C^−1^, a systematic temperature difference of only 2.5 °C would produce a corresponding systematic error of±0.01% in the calibrator area. Similarly, if it is assumed that the modulus of elasticity for the calibrator material is 2×10^11^ Pa (2×l0^6^atm) and if the calibrator has a diameter to thickness ratio of 20, a systematic pressure error of only 1 MPa (10 atm) will produce a systematic error of±0.01% in the calibrator area. Such calibrator area errors will propagate through all of the successive relationships for calibrating and using meters, and these relationships will be further altered by additional temperature and pressure effects. The calibrator constant, *K*~C~, at non-reference conditions in units of pulses per volume can be written $$K_{C}\frac{N_{E}}{V_{C}},$$since *V*~C~ is the displacement at non-reference conditions. Therefore $$K_{C} = \frac{K_{E}}{{\overline{A}}_{C}}.$$Combining [Eq. (8)](#fd8-jresv97n5p509_A1b){ref-type="disp-formula"} with [Eqs. (2)](#fd2-jresv97n5p509_A1b){ref-type="disp-formula"}, [(5)](#fd5-jresv97n5p509_A1b){ref-type="disp-formula"}, and [(6)](#fd6-jresv97n5p509_A1b){ref-type="disp-formula"} produces $$K_{C} = \frac{K_{C0}\left\lbrack {1 - \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack}{\left\lbrack {1 + 2\alpha_{C}\left( {T_{C} - T_{0}} \right.} \right\rbrack\left\lbrack {1 + \frac{\left( {P_{C} - P_{0}} \right)D_{C0}}{t_{C0}E_{C}}} \right\rbrack},$$or, to first order approximation: $$K_{C0} = K_{C}\frac{\left\lbrack {1 + \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack}{\left\lbrack {1 - 2\alpha_{C}\left( {T_{C} - T_{0}} \right.} \right\rbrack\left\lbrack {1 - \frac{\left( {P_{C} - P_{0}} \right)D_{C0}}{t_{C0}E_{C}}} \right\rbrack}.$$For temperatures and pressures higher than reference conditions the calibrator constant, *K*~C~, is less than the value at reference conditions. In both of the above geometrical methods for obtaining calibrator constants, fluid properties are not involved. The value for *K*~C0~ given by [Eqs. (2)](#fd2-jresv97n5p509_A1b){ref-type="disp-formula"} or [(10)](#fd10-jresv97n5p509_A1b){ref-type="disp-formula"} should be put into the software that operates the displacement calibrator, together with pertinent material constants and component dimensions. Then the computation for the calibrator constant *K*~C~ at non-reference conditions can be done using appropriate measurements for encoder temperature, in addition to the pressure and temperature in the calibrator, as shown in [Eq. (9)](#fd9-jresv97n5p509_A1b){ref-type="disp-formula"}. These values of *K*~C0~ or *K*~C~ will be used for the accurate calibrations of flowmeters. 2.3 Determination of Calibrator Constant by the Draw Technique at Reference Conditions -------------------------------------------------------------------------------------- The configuration for the draw technique is shown in [Fig. 4](#f4-jresv97n5p509_a1b){ref-type="fig"}, where the piston stroke displaces an amount of fluid through a valve into the collection tank; the corresponding encoder pulses total N~EO~. All temperatures and pressures are the reference values. In all that follows, it is assumed that the seal of the piston in the calibrator seals perfectly and is sufficiently pliant to continue to seal perfectly when changes in temperature and internal pressure occur and change the diameter of the cylinder of the calibrator. Applying conservation of mass principles to the constant volume shown in [figure 4](#f4-jresv97n5p509_a1b){ref-type="fig"}, we obtain: $$\frac{\partial}{\partial t}{\int_{V}{\rho dV - {\int_{s}{\rho\mathbf{\upsilon}}}}} \cdot \mathbf{n}\ dS = 0,$$where *ρ* is the fluid density, *V* is the control volume surrounded by the control surface, *S*,***υ*** is the fluid velocity vector, and ***n*** is the unit vector normal to *S* with positive direction pointing to the interior of *V.* When $$\frac{\partial}{\partial t}{\int_{V}{\rho dV = 0,}}$$there is no change of the mass within the control volume in time. This means the effects of fluid friction or of heat transfer are negligible or, taken in total, do not alter in time the mass contained within the control volume. Under such conditions, the inlet and outlet mass fluxes through the control surface, *S* are the same, that is $$\rho_{C0}\ {\overset{˙}{V}}_{C0} = \rho_{{COLL}^{’}T0\ }{\overset{˙}{V}}_{{COLL}^{’}T0},$$where *ρ*~C0~ and *ρ*~COLL'T0~ are the fluid densities, respectively, in the calibrator and collection tank at the same reference conditions and therefore are equal. The quantities ${\overset{˙}{V}}_{C0}$ and ${\overset{˙}{V}}_{{COLL}^{’}T0}$ are the volumetric flowrates out of the calibrator and into the collection tank, respectively. Therefore, since the times of displacement and collection are assumed to be the same, $$V_{C0} = V_{{COLL}^{’}T0},$$and the calibrator constant, in units of pulses per displacement volume can be written $$K_{C0} = N_{E0}/V_{{COLL}^{’}T0},$$where $$V_{{COLL}^{’}T0} = {\overline{A}}_{C0}L_{E0}.$$Therefore, using [Eq. (13)](#fd13-jresv97n5p509_A1b){ref-type="disp-formula"}--[(16)](#fd16-jresv97n5p509_A1b){ref-type="disp-formula"}, we obtain $$K_{C0} = K_{E0}/{\overline{A}}_{C0},$$in agreement with [Eq. (2)](#fd2-jresv97n5p509_A1b){ref-type="disp-formula"}. Accordingly, results obtained via this procedure should duplicate those obtained via the geometrical measurement techniques. It is generally assumed that the calibrator constant *K*~C0~ is independent of the piston speed or the fluid properties. If this assumption is not valid then it is necessary to characterize the dependence. This should be done in a manner analogous to the procedures used to characterize turbine meters as given below. It should be noted that, for the control volume shown in [Fig. 4](#f4-jresv97n5p509_a1b){ref-type="fig"}, if $$\frac{\partial}{\partial t}{\int_{V}{\rho dV \neq 0,}}$$then the mass flux displaced in the cylinder of the calibrator is not equal to that delivered to the collection tank. Furthermore, if the fluid density in the control volume decreases because of a temperature rise then the mass flux to the collection tank would be more than that displaced in the calibrator cylinder. As will be seen below, this produces a turbine-type meter calibration result in pulses per volume units that is larger than it should be. Conversely, if the fluid density in the control volume increases, then the mass flux delivered to the collection vessel is less than that displaced in the calibrator cylinder. This produces a low calibration result for a turbine meter constant in pulses per volume units. 2.4 Determination of Calibrator Constant by the Draw Technique at Non-Reference Conditions ------------------------------------------------------------------------------------------ The configuration for the draw technique in non-reference conditions is shown in [Fig. 5](#f5-jresv97n5p509_a1b){ref-type="fig"}. As before, the piston stroke displaces a volume of fluid in the cylinder and the corresponding encoder pulses total *N*~E~. The temperature and pressure of the fluid in the cylinder are *T*~C~ and *P*~C~, respectively. Applying conservation of mass principles to the control volume in [Fig. 5](#f5-jresv97n5p509_a1b){ref-type="fig"}, for the condition where there is no change with time of the mass within the control volume, [Eq. (11)](#fd11-jresv97n5p509_A1b){ref-type="disp-formula"} and [(12)](#fd12-jresv97n5p509_A1b){ref-type="disp-formula"} indicate that $$\rho_{C}{\overset{˙}{V}}_{C} = \rho_{{COLL}^{’}T}{\overset{˙}{V}}_{{COLL}^{’}T},$$or when the displacement and collected volumes are simultaneous $$\rho_{C}V_{C} = \rho_{{COLL}^{’}T}V_{{COLL}^{’}T},$$where *ρ*~C~ and *ρ*~COLL'T~ are, respectively, the density of the fluid displaced within the cylinder of the calibrator and flowing from the valve and into the collection tank. The volume *V*~COLL'T~ can be determined using volumetric or gravimetric techniques, or both. It is noted that constant conditions of temperature and pressure need to be maintained in the calibrator and in the collection vessel for accurate results. With the volume *V*~COLL'T~ determined, *V*~C~ can be computed via $$V_{C} = \left( {\rho_{{COLL}^{’}T}/\rho_{C}} \right)V_{{COLL}^{’}T},$$where, according to the definitions of fluid thermal expansion effects and compressibility, to first order in temperature and pressure, separately: $$\begin{array}{l} {\frac{\rho_{{COLL}^{’}T}}{\rho_{C}} =} \\ {\left\lbrack {1 - 3\alpha_{F}\left( {T_{{COLL}^{’}T} - T_{C}} \right)} \right\rbrack{\left\lbrack {1 + \frac{\left( {P_{{COLL}^{’}T} - P_{C}} \right)}{E_{F}}} \right\rbrack,}} \\ \end{array}$$where *α*~F~ is the linear expansion coefficient for the fluid, the pressures *P*~COLL'T~ and *P*~C~ are, respectively, those for the fluid passing from the valve into the collection tank and in the cylinder of the calibrator, and *E*~F~ is the modulus of elasticity of the fluid. It is noted that both α~F~ and *E*~F~ are dependent upon temperature and pressure for the specific fluid, but the values used here and in what follows are assumed to be averages taken over the appropriate ranges of temperature and pressure. It is also noted here that the reciprocal of the modulus of elasticity of the fluid is also the compressibility of the fluid. In [Eq. (22)](#fd22-jresv97n5p509_A1b){ref-type="disp-formula"}, it is noted that temperature and pressure effects have opposite signs in producing fluid density changes. However, when the temperature differences between the collection vessel and the calibrator are large and where the pressure in the calibrator is much larger than that in the collection vessel, the effects on the fluid can be significant. If it is assumed that the linear expansion coefficient of a hydrocarbon liquid is 3 × 10^−4^ °C^−1^, a systematic temperature error of only 1 °C will produce a systematic error of 0.1% due to temperature in the determination of the calibrator volume by this draw method. Similarly, if it is assumed that the fluid's modulus of elasticity is 2 × 10^9^ Pa (2 × 10^4^ atm), a systematic pressure error of only 2 MPa (20 atm) will produce a systematic error of ±0.1% in the calibrator volume. As mentioned above, such errors will propagate through all of the relationships for calibrating and using meters and these errors will be further altered by additional temperature and pressure corrections. Furthermore, using [Eq. (7)](#fd7-jresv97n5p509_A1b){ref-type="disp-formula"} we can write $$K_{C} = \frac{N_{E}}{\left( \frac{\rho_{{COLL}^{’}T}}{\rho_{C}} \right)V_{{COLL}^{’}T}},$$then, from [Eq. (22)](#fd22-jresv97n5p509_A1b){ref-type="disp-formula"} $$\begin{array}{l} {K_{C} = \frac{N_{E}}{V_{{COLL}^{’}T}}\left\lbrack {1 + 3\alpha_{F}\left( {T_{{COLL}^{’}T} - T_{C}} \right)} \right\rbrack} \\ {\times \left\lbrack {1 - \frac{\left( {P_{{COLL}^{’}T} - P_{C}} \right)}{E_{F}}} \right\rbrack.} \\ \end{array}$$This result is then related to reference conditions via [Eq. (10)](#fd10-jresv97n5p509_A1b){ref-type="disp-formula"} to give $$\begin{array}{l} {K_{C0} =} \\ \frac{N_{E}\left\lbrack {1 + 3\alpha_{F}\left( {T_{{COLL}^{’}T} - T_{C}} \right)} \right\rbrack\left\lbrack {1 - \frac{\left( {P_{{COLL}^{’}T} - P_{C}} \right)}{E_{F}}} \right\rbrack\left\lbrack {1 + \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack}{V_{{COLL}^{’}T}\left\lbrack {1 - 2\alpha_{C}\left( {T_{C} - T_{0}} \right)} \right\rbrack\left\lbrack {1 - \frac{\left( {P_{C} - P_{0}} \right)D_{C0}}{T_{C0}E_{C}}} \right\rbrack.} \\ \end{array}$$It is noted that the cross-sectional area, *Ā*~C~ can be determined via [Eqs. (8)](#fd8-jresv97n5p509_A1b){ref-type="disp-formula"} and [(23)](#fd23-jresv97n5p509_A1b){ref-type="disp-formula"} to be $${\overline{A}}_{C} = \frac{K_{E}}{N_{E}}\left( \frac{\rho_{{COLL}^{’}T}}{\rho_{C}} \right)V_{{COLL}^{’}T}.$$Using [Eqs. (5)](#fd5-jresv97n5p509_A1b){ref-type="disp-formula"}, [(6)](#fd6-jresv97n5p509_A1b){ref-type="disp-formula"} and [(22)](#fd22-jresv97n5p509_A1b){ref-type="disp-formula"}, this can be written in terms of directly measured quantities and related to reference conditions via $$\begin{array}{l} {{\overline{A}}_{C0} =} \\ \frac{K_{E0}V_{{COLL}^{’}T}\left\lbrack {1 - \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack\left\lbrack {1 - 2\alpha_{C}\left( {T_{C} - T_{0}} \right)} \right\rbrack\left\lbrack {1 - \frac{\left( {P_{C} - P_{0}} \right)D_{C0}}{t_{C0}E_{C}}} \right\rbrack}{N_{E}\left\lbrack {1 + 3\alpha_{F}\left( {T_{{COLL}^{’}T} - T_{C}} \right)} \right\rbrack\left\lbrack {1 - \frac{\left( {P_{{COLL}^{’}T} - P_{C}} \right)}{E_{F}}} \right\rbrack.} \\ \end{array}$$If *K*~Ć~ is defined as $$K_{Ć} = N_{E}/V_{{COLL}^{’}T},$$then $$\begin{array}{l} {K_{C0} = K_{Ć} \times} \\ \frac{\left\lbrack {1 + 2\alpha_{C}\left( {T_{C} - T_{0}} \right)} \right\rbrack\left\lbrack {1 + \frac{\left( {P_{C} - P_{0}} \right)D_{C0}}{t_{C0}E_{C}}} \right\rbrack}{\left\lbrack {1 - 3\alpha_{F}\left( {T_{{COLL}^{’}T} - T_{C}} \right)} \right\rbrack\left\lbrack {1 + \frac{\left( {P_{{COLL}^{’}T} - P_{C}} \right)}{E_{F}}} \right\rbrack\left\lbrack {1 - \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack} \\ \end{array}$$or, to first order $$\begin{array}{l} {K_{Ć} = K_{C0} \times} \\ {\frac{\left\lbrack {1 - 2\alpha_{C}\left( {T_{C} - T_{0}} \right)} \right\rbrack\left\lbrack {1 - \frac{\left( {P_{C} - P_{0}} \right)D_{C0}}{t_{C0}E_{C}}} \right\rbrack}{\left\lbrack {1 + 3\alpha_{F}\left( {T_{{COLL}^{’}T} - T_{C}} \right)} \right\rbrack\left\lbrack {1 - \frac{\left( {P_{{COLL}^{’}T} - P_{C}} \right)}{E_{F}}} \right\rbrack\left\lbrack {1 + \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack}.} \\ \end{array}$$When temperatures exceed those of the reference conditions and pressures are those for the reference conditions: $$K_{Ć} < K_{C0}.$$Thus, the observation made above is repeated here, namely, that when pressure effects can be neglected and when temperatures are above the reference conditions, the calibrator delivers, for the same encoder output pulses, more fluid volume than would occur under reference conditions. It is also noted that [Eqs. (23)](#fd23-jresv97n5p509_A1b){ref-type="disp-formula"}--[(27)](#fd27-jresv97n5p509_A1b){ref-type="disp-formula"} can be combined to give: $$K_{C} = K_{C0}\frac{\left\lbrack {1 - \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack}{\left\lbrack {1 + 2\alpha_{C}\left( {T_{C} - T_{0}} \right)} \right\rbrack\left\lbrack {1 + \frac{\left( {P_{C} - P_{0}} \right)D_{C0}}{t_{C0}E_{C}}} \right\rbrack},$$which is the same as [Eq. (9)](#fd9-jresv97n5p509_A1b){ref-type="disp-formula"}. It is noted that, in these two draw procedures, the decision to collect a sufficiently large number of encoder pulses should precede the operation. This number should be selected according to the desired precision for the calibrator constant, see [Eq. (3)](#fd3-jresv97n5p509_A1b){ref-type="disp-formula"}. In the reference conditions $$N_{E0} = K_{E0}L_{E0},$$and in the non-reference conditions $$N_{E} = K_{E}L_{E},$$but, to first order in temperature, $$K_{E} = K_{E0}\left\lbrack {1 - \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack,$$and $$L_{E} = L_{E0}\left\lbrack {1 + \alpha_{E}\left( {T_{E} - T_{0}} \right)} \right\rbrack.$$Combining [Eqs. (33)](#fd33-jresv97n5p509_A1b){ref-type="disp-formula"}--[(36)](#fd36-jresv97n5p509_A1b){ref-type="disp-formula"} therefore yields: $$N_{E} = K_{E}L_{E} = K_{E0}L_{E0} = N_{E0}.$$ This indicates that the precision criterion specified for the calibrator should be achieved via the number of pulses selected and the choice is not dependent upon whether reference or non-reference conditions prevail. As stated above following [equation (17)](#fd17-jresv97n5p509_A1b){ref-type="disp-formula"}, the calibrator constants determined via the draw procedure are generally assumed to be independent of piston speed and fluid properties. Where this is not valid, efforts should be made to achieve this assumption, i.e., improving the piston seals or the calibrator should be characterized using techniques analogous to those for turbine meters as will be described below. It should also be noted that in the above described draw procedures, the valve and pulse counting techniques must not introduce spurious effects. Spurious counting effects may result from the fluid dynamics in the valve as the flow is started and stopped in conjunction with starting and stopping the pulse count. If such an effect is present then it should be eliminated or proper account made for it so that the appropriate fluid volume is associated with the pulse total. This volume can be determined using the valve compensation techniques that are conventionally applied to diverter systems, see \[[@b4-jresv97n5p509_a1b]\]. To use the now-characterized calibrator, it is required that appropriate instrumentation be properly installed both to assure that [Eq. (12)](#fd12-jresv97n5p509_A1b){ref-type="disp-formula"} is satisfied and to measure the quantities involved in [Eqs. (9)](#fd9-jresv97n5p509_A1b){ref-type="disp-formula"} or [(10)](#fd10-jresv97n5p509_A1b){ref-type="disp-formula"} or [(32)](#fd32-jresv97n5p509_A1b){ref-type="disp-formula"} for the calibrator and [equations (28)](#fd28-jresv97n5p509_A1b){ref-type="disp-formula"} or [(29)](#fd29-jresv97n5p509_A1b){ref-type="disp-formula"} or [(30)](#fd30-jresv97n5p509_A1b){ref-type="disp-formula"} for dispensing precise volumes of fluid at specified conditions of temperature and pressure. Appropriate values are needed for the material constants---the thermal expansion coefficients *α*~E~, *α*~F~ and *α*~C~, the modulus of elasticity of the material of the calibrator cylinder, *E*~C~, and that of the fluid, *E*~F~, and the pertinent dimensions of the cylinder. Once the appropriate value for the constant *K*~C0~ has been installed in the calibrator software or the working procedures for the calibrator, the next step is to use the calibrator to calibrate a flowmeter. 3. Calibrator Use in Calibrating a Turbine Type Flowmeter ========================================================= 3.1 Reference Conditions ------------------------ To calibrate a turbine-type flowmeter at reference temperature and pressure conditions using the calibrator characterized as described above, the arrangement is sketched in [Fig. 6](#f6-jresv97n5p509_a1b){ref-type="fig"}. The temperature and pressure are the reference conditions denoted by *T*~0~ and *P*~0~. As stated above, *T*~0~=*T*~M0~=*T*~C0~ and *P*~0~*=P*~M0~*=P*~C0~. Applying again the conservation of mass principles,. and assuming that there is no change of mass within the control volume with time, [Eqs. (11)](#fd11-jresv97n5p509_A1b){ref-type="disp-formula"} and [(12)](#fd12-jresv97n5p509_A1b){ref-type="disp-formula"} indicate that $$\rho_{C0}{\overset{˙}{V}}_{C0} = \rho_{M0}{\overset{˙}{V}}_{M0},$$where *ρ*~C0~ and *ρ*~M0~ are the fluid densities in the calibrator and meter, respectively, at the reference conditions. The quantities ${\overset{˙}{V}}_{C0}$ and ${\overset{˙}{V}}_{M0}$ are the volumetric flowrates, respectively, in the calibrator and through the meter at the reference conditions. Since reference conditions in cylinder and meter are assumed the same, the densities *ρ*~C0~ and *ρ*~M0~ are equal and therefore: $${\overset{˙}{V}}_{C0} = {\overset{˙}{V}}_{M0}.$$Since the pertinent time intervals are assumed the same: $$K_{C0} = \frac{K_{E0}}{{\overline{A}}_{C0}} = \frac{N_{E0}}{{\overline{A}}_{C0}L_{E0}} = \frac{f_{E0}}{{\overset{˙}{V}}_{C0}},$$where *f*~E0~ is the encoder frequency, and $$K_{M0} = \frac{N_{M0}}{V_{M0}} = \frac{f_{M0}}{{\overset{˙}{V}}_{M0}},$$where *f*~M0~ is the meter frequency. In [Eqs. (40)](#fd40-jresv97n5p509_A1b){ref-type="disp-formula"} and [(41)](#fd41-jresv97n5p509_A1b){ref-type="disp-formula"}, the pulse counts *N*~E0~ and *N*~M0~ are to correspond to the respective volumes *V*~C0~ and *V*~M0~ which are assumed to be the same, or appropriate corrections are made to compensate for any differences between these volumes. Combining [Eqs. (40)](#fd40-jresv97n5p509_A1b){ref-type="disp-formula"} and [(41)](#fd41-jresv97n5p509_A1b){ref-type="disp-formula"} gives $$K_{M0} = \frac{f_{M0}}{f_{E0}}\frac{K_{E0}}{{\overline{A}}_{C0}} = \frac{N_{M0}}{N_{E0}}\frac{K_{E0}}{{\overline{A}}_{C0}} = \frac{N_{M0}}{N_{E0}} \bullet K_{C0} = \frac{f_{M0}}{f_{E0}} \bullet K_{C0},$$where *K*~M0~has units of pulses per volume at the reference conditions. The quantity *K*~E0~ is obtained from manufacturer's specifications, testing or [Eq. (1)](#fd1-jresv97n5p509_A1b){ref-type="disp-formula"}. Conventional non-dimensionalization procedures can be applied to the flowmeter characteristics to produce a set of parameters which interrelate the significant inertial, viscous, and oscillatory effects that constitute the performance of the meter in the calibration conditions, see \[[@b6-jresv97n5p509_a1b],[@b7-jresv97n5p509_a1b]\]. In this way, the performance of the device can be predicted for other fluid and flow conditions where this set of parameters are the pertinent ones to describe the meter's performance. Of course, when other, different effects such as fluid compressibility or gravitational influences become significant, it should be expected that the initial parameterization needs to be modified to include such effects to obtain satisfactory description of meter performance, see \[[@b8-jresv97n5p509_a1b],[@b9-jresv97n5p509_a1b]\]. By normalizing the meter factor, *K*~M0~ and the fluid flowrate using the meter diameter, *D*~M0~ and the mean flow velocity ${\overset{˙}{V}}_{M0}/A_{M0} = U_{M0}$ and the fluid kinematic viscosity *υ*~0~, we can obtain, for example, the Strouhal number $$St = \frac{f_{M0}D_{M0}}{U_{M0}} = C_{1}K_{M0}D_{M0}^{3} \propto K_{M0}D_{M0}^{3},$$where *C*~1~=π/4. The Strouhal number is the ratio of characteristic meter frequency effects such as propeller rotation rate to fluid momentum effects. By effects here is meant either forces or energies. As such, the Strouhal number is a dimensionless meter factor. Conventionally, it is fluid mechanical practice to formulate the Reynolds number as the ratio of fluid inertial to viscous effects, $$Re = \frac{D_{M0}U_{M0}}{\upsilon_{0}}.$$This is generally used to describe the domain of the meter calibration for which the corresponding Strouhal numbers specify the range of meter response. A Strouhal-Reynolds characterization of a pulse-producing flowmeter conforms to conventional fluid mechanical procedures and it is analogous to orifice metering practice where discharge coefficient (a ratio of fluid inertia to differential pressure effects) is described functionally or graphically versus Reynolds number. However, as noted in [Eq. (44)](#fd44-jresv97n5p509_A1b){ref-type="disp-formula"} the Reynolds number requires that the fluid velocity be known. Since this is the purpose for using the meter, an iteration technique is required to calculate the flowrate. To avoid such an iteration, turbine meter manufacturers have designed their products to have high levels of linearity over wide flowrate ranges. As a result of this, it has been conventional turbine meter practice to characterize performance via *K*-factor vs. frequency to kinematic viscosity ratio---the so-called Universal Viscosity Curve (UVC). In accord with the principles of dimensional similitude, the meter frequency to fluid viscous effects can be formulated using, as characteristic length scale, the meter diameter, *D*~M0~ as in [Eq. (44)](#fd44-jresv97n5p509_A1b){ref-type="disp-formula"}. This formulation can also be achieved via the product of Strouhal and Reynolds numbers; this product has been recently referred to, see \[[@b9-jresv97n5p509_a1b]\], as the Roshko number, $$R_{0} = \frac{f_{M0}D_{{M0}^{2}}}{\upsilon_{0}}.$$This Roshko number is the dimensionless version of the frequency-to-kinematic viscosity ratio used for the UVC. It is expected that the dimensionless version of the UVC---the Strouhal-Roshko characterization---should produce superior prediction of meter performance compared to the UVC by virtue of its more complete, i.e., dimensionless description of the inertial, viscous, and oscillatory effects that occur in the meter operation. The question of which set of parameters best describes a turbine meter's performance should be determined using appropriate data sets. The specific design considerations of blade size and shape, internal meter geometry, bearing features, etc., can be expected to play significant roles in the selection of non-dimensional parameters. Where it may happen that the Reynolds number produces a better description of turbine meter performance, i.e., the data collapses better onto a single curve than achieved using the Roshko number, an iterative sequence of computations may be necessary to produce an accurate determination of flowrate. If the meter is very linear, such an iteration may not be required. Typical flowmeter characterization results are sketched in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"}. Because it is not known whether Reynolds or Roshko number is the better parameter with which to characterize the meter performance, both shall be included in what follows. In those instances where meter performance might deviate from the curve shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"}, the interpretation would be that additional factors in the deviant conditions have become significant whereas these factors were insignificant in the calibration conditions. Examples might be fluid frictional effects in the turbine bearings produced by extreme viscosity variations from those prevailing in the calibration conditions, or liquid cavitation effects, etc. The curve shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"} is interpreted as the functional relationship between the Strouhal and Reynolds or Roshko numbers which are assumed to be the salient parameters describing the performance of this flowmeter over these calibration conditions. Inherent in this interpretation is the assumption that in any subsequent use of this functional relationship, the geometries of this meter, for example, the bearings or the propeller diameter are not changed relative to the selected characteristic length of the meter, i.e., the internal diameter, *D.* If such changes do occur it can be expected that the curve shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"} can change. For example, if a smaller propeller were installed, this curve may retain its shape but lie below that shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"}. A non-dimensional parameter which could take into account different propeller diameters is *ß=d/D*, where *d* is the propeller diameter. Accordingly, the curve shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"} would pertain to the specific *β* for which the calibration was done. If other, smaller propellers were also calibrated, these results could be plotted in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"} and parameterized with the smaller value of *β.* When the propeller and meter-body materials are the same and where pressure effects can be neglected, the *β* ratio will remain constant when the temperature changes. When the propeller and the meter-body materials are different, temperature changes can produce different *β* ratios. These different ratios can be computed using pertinent relationships. The computed results should then be used with calibration data taken for different *β* ratios to predict the meter performance at the different temperature conditions, see [Appendix B](#app2-jresv97n5p509_A1b){ref-type="app"}. [Figure 7](#f7-jresv97n5p509_a1b){ref-type="fig"} is different from conventional turbine meter performance plots in which the meter's *K*-factor is plotted versus the ratio of frequency-to-kinematic viscosity. These conventional quantities are different from the Strouhal and Reynolds or Roshko number parameters by constant factors and by factors of the meter diameter raised to different exponents. These powers of the diameter should change only slightly with small changes in temperature and internal pressure. However, when conditions vary widely, the dimensionless formulations should be used and are expected to produce improved meter performance. Plots in the format of [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"} should then, with the exceptions of deviant phenomena becoming influential, apply to a wide range of specific, dimensional fluid property and flow conditions and produce accurate predictions of turbine meter performance. Where conventional turbine meter practice is used and plots are produced for K-factor versus the ratio of frequency-to-kinematic viscosity, improved meter performance can be expected when these calibration results are corrected to specified reference conditions. Accordingly, the corrected results should incorporate temperature and internal pressure corrections for the ratios of meter diameters raised to the relevant exponent to predict meter performance for other fluid and flow conditions. For meters having good linearity characteristics, i.e., constancy of the meter factor over specified flowrate ranges, the more important of these two corrections is that for the *K*-factor, i.e., $$K_{M0} = K_{M}{(D_{M}/D_{M0})}^{3}.$$This stipulates that the meter frequency-to-fluid inertial effects ratio, i.e., Strouhal number be the same in the actual conditions as in the reference conditions. To first order in temperature and pressure, separately, the diametral ratio can be written, $$\begin{array}{l} {\left( \frac{D_{M}}{D_{M0}} \right)^{3} = \lbrack 1 + 3\alpha_{M}(T_{M} - T_{M0})\rbrack} \\ {\times \left\lbrack {1 + \frac{3(P_{M} - P_{M0})D_{M0}}{2t_{M0}E_{M}}} \right\rbrack = \frac{K_{M0}}{K_{M}},} \\ \end{array}$$where *D*~M~ and *D*~M0~ are, respectively, the meter diameters at the non-reference conditions, *T*~M~ and *P*~M~ and reference conditions, *T*~M0~ and *P*~M0~. The quantities α~M~ and *E*~M~ are, respectively, the linear expansion coefficient and the modulus of elasticity of the meter body material for these conditions, and *t*~M0~ is the thickness of the meter body at reference conditions. The more significant of the two correction factors is usually that for temperature deviations from reference conditions. When the pressure correction can be neglected, $$K_{M0} = K_{M}\lbrack 1 + 3\alpha_{M}(T_{M} - T_{M0})\rbrack,$$or, to first order: $$K_{M} = K_{M0}\lbrack 1 - 3\alpha_{M}(T_{M} - T_{M0})\rbrack.$$This relationship duplicates that given in \[[@b10-jresv97n5p509_a1b]\]. 3.2 Non-Reference Conditions ---------------------------- To calibrate a turbine-type flowmeter using the calibrator characterized as described above, the arrangement is sketched in [Fig. 8](#f8-jresv97n5p509_a1b){ref-type="fig"}. The temperature and pressure in the calibrator are the non-reference conditions denoted by *T*~C~ and *P*~C~. Applying again the conservation of mass principles, and assuming that there is no change of mass within the control volume with time, [Eq. (11)](#fd11-jresv97n5p509_A1b){ref-type="disp-formula"} and [(12)](#fd12-jresv97n5p509_A1b){ref-type="disp-formula"} indicate that $$\rho_{C}{\overset{\bullet}{V}}_{C} = \rho_{M}{\overset{\bullet}{V}}_{M} \cdot$$Since the fluid conditions in the cylinder and in the meter can be different, the ratio of the densities is $$\frac{\rho_{C}}{\rho_{M}} = \lbrack 1 - 3\alpha_{F}(T_{C} - T_{M})\rbrack\left\lbrack {1 + \frac{P_{C} - P_{M}}{E_{F}}} \right\rbrack,$$and from [Eq. (32)](#fd32-jresv97n5p509_A1b){ref-type="disp-formula"} and in a manner analogous to [Eq. (40)](#fd40-jresv97n5p509_A1b){ref-type="disp-formula"} $$\begin{array}{l} {K_{C} = \frac{K_{E}}{{\overline{A}}_{C}} = \frac{f_{E}}{{\overset{\bullet}{V}}_{C}} =} \\ {K_{C0}\frac{\lbrack 1 - \alpha_{E}(T_{E} - T_{0})\rbrack}{\lbrack 1 + 2\alpha_{C}(T_{C} - T_{0}\rbrack\left\lbrack {1 + \frac{(P_{C} - P_{0})D_{C0}}{t_{C0}E_{C}}} \right\rbrack}.} \\ \end{array}$$Then, analogously to [Eq. (42)](#fd42-jresv97n5p509_A1b){ref-type="disp-formula"} $$\begin{array}{l} {K_{M} = \frac{f_{M}}{f_{E}}K_{C}\frac{\rho_{M}}{\rho_{C}} = \frac{N_{M}}{N_{E}}K_{C0}} \\ {\times \frac{\lbrack 1 - \alpha_{E}(T_{E} - T_{0})\rbrack\lbrack 1 + 3\alpha_{F}(T_{C} - T_{M})\rbrack}{\lbrack 1 + 2\alpha_{C}(T_{C} - T_{0})\rbrack\left\lbrack {1 + \frac{(P_{C} - P_{0})D_{C0}}{t_{C0}E_{C}}} \right\rbrack\left\lbrack {1 + \frac{(P_{C} - P_{M})}{E_{F}}} \right\rbrack},} \\ \end{array}$$where *K*~M~ has units of pulses per volume at the specific, non-reference meter conditions. As stated above after [Eq. (41)](#fd41-jresv97n5p509_A1b){ref-type="disp-formula"}, the pulse totals from the meter and the encoder have to correspond to the same displaced volume or time interval. If compensations are needed to achieve this correspondence, these should be done and results used in [Eq. (53)](#fd53-jresv97n5p509_A1b){ref-type="disp-formula"}. By normalizing this meter factor, *K*~M~ and the fluid flowrate using the non-reference meter diameter, *D*~M~, and fluid kinematic viscosity, υ, we obtain Strouhal and Reynolds or Roshko numbers $$St = f_{M}D_{M}/U_{M} = C_{1}K_{M}D_{M}^{3},$$where *C*~1~=π/4, and $$Re = D_{M}U_{M}/\upsilon$$or $$Ro = \frac{f_{M}D_{M}^{2}}{\upsilon}.$$Using these dimensionless parameters, the performance for the meter can be plotted; results should be as shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"}. As noted above in [Eqs. (43)](#fd43-jresv97n5p509_A1b){ref-type="disp-formula"}--[(45)](#fd45-jresv97n5p509_A1b){ref-type="disp-formula"}, the characteristics of meter factor and diameter at reference and non-reference conditions are interrelated and the dependence of the fluid's kinematic viscosity can be written functionally as $$\upsilon = \upsilon_{0}\lbrack T,T_{0},P,P_{0}\rbrack.$$With the performance curve given in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"} and the relationships given in [Eqs. (43)](#fd43-jresv97n5p509_A1b){ref-type="disp-formula"}--[(45)](#fd45-jresv97n5p509_A1b){ref-type="disp-formula"}, or [(54)](#fd54-jresv97n5p509_A1b){ref-type="disp-formula"}--[(57)](#fd57-jresv97n5p509_A1b){ref-type="disp-formula"}, one is now ready to use the flowmeter to make a flowrate measurement. 4. Using a Turbine-Type Flowmeter To Make a Measurement ======================================================= 4.1 Reference Conditions ------------------------ Given that the meter performance characteristics are as shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"}, or less preferably but more conventionally as in [Fig. 9](#f9-jresv97n5p509_a1b){ref-type="fig"}, one can quantify the meter linearity over a specified flowrate range. The meter linearity is conventionally the average of the maximum and minimum values of the meter factor (Strouhal number) over this range; normalized by the average meter factor, see \[[@b3-jresv97n5p509_a1b] and [@b10-jresv97n5p509_a1b]\]. For the meter performance shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"}, the mean value of the Strouhal Number, $\overline{St}$ gives the mean meter factor, via $${\overline{K}}_{M0} = \frac{\overline{St}}{C_{1}D_{M0}^{3}},$$where *C*~1~=π/4. The flowrate is determined via $${\overset{\bullet}{V}}_{M0} = \frac{f_{M0}}{{\overline{K}}_{M0}} = \frac{C_{1}D_{M0}^{3}f_{M0}}{\overline{St}}.$$This result would pertain to any flowrate over the range specified for the meter's linearity. If it is desired to improve the accuracy of this flowrate determination, this could be done by using, for example, the curve shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"} or close approximations to it. With advances in today's computer technology, this type of process can be readily installed in the secondary devices used with flowmeters. For the specific frequency from the meter, the Roshko number can be calculated directly and then used to determine the corresponding Strouhal number. For the case where the meter is characterized using Strouhal and Reynolds parameters, the process to determine an accurate flowrate should be iterative. This iteration process should begin using a mean value of meter factor, such as given in [Eq. (58)](#fd58-jresv97n5p509_A1b){ref-type="disp-formula"}, this value of meter factor enables a computation of the flowrate via [Eq. (59)](#fd59-jresv97n5p509_A1b){ref-type="disp-formula"}. Using this flowrate, the Reynolds number can be computed and then used to get the corresponding Strouhal number from the calibration curve and a refined value of flowrate. This process should be repeated until satisfactorily small changes are found in successive results. In this way, the accuracy level for the flowrate determination can be increased over the level associated with the meter's linearity. This can be done to the precision level associated with a specific flowrate as quantified in the calibration process. The resulting enhanced meter performance could amount to significant improvements in measurement accuracy. In the following, a Strouhal-Reynolds characterization of meter performance will be used since it is more conventional in fluid mechanics and since it may require the iteration procedure, described above, to be used. 4.2 Non-Reference Conditions ---------------------------- To describe meter performance in non-reference conditions, the meter characteristics shown in [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"} will be used. The reason for this is that the complete, non-dimensional assessment of meter frequency effects and fluid inertial and viscous effects are not complete in [Fig. 9](#f9-jresv97n5p509_a1b){ref-type="fig"}. For this reason, the non-reference meter diameter, *D*~M~, and frequency,*f*~M~, and the fluid's kinematic viscosity, *v*, should be used, in compatible units, to produce the Reynolds number for the non-reference conditions. This Reynolds number produces, using [Fig. 7](#f7-jresv97n5p509_a1b){ref-type="fig"}, the corresponding Strouhal number which with the meter diameter, *D*~M~, gives the meter factor, *K*~M~-The flowrate measurement is then obtained using $${\overset{\cdot}{V}}_{M} = \frac{f_{M}}{{\overline{K}}_{M}} = \frac{C_{1}f_{M}D_{M}^{3}}{\overline{St}}.$$This flowrate is correctly converted to reference conditions by specifying that the Strouhal number is the same for the non-reference and reference conditions, specified by this Reynolds number, i.e., $$St = C_{1}\frac{f_{M}D_{M}^{3}}{{\overset{\cdot}{V}}_{M}} = C_{1}\frac{f_{M0}D_{M0}^{3}}{{\overset{\cdot}{V}}_{M0}}.$$Therefore $${\overset{\cdot}{V}}_{M0} = {\overset{\cdot}{V}}_{M}\frac{f_{M0}}{f_{M}}\left( \frac{D_{M0}}{D_{M}} \right)^{3}.$$The reference to non-reference frequency ratio is obtained by specifying that Reynolds number similarity exists for these two conditions, i.e., $$Re = \frac{f_{M}D_{M}^{2}}{(St)\upsilon_{M}} = \frac{f_{M0}D_{M0}^{2}}{(St)\upsilon_{M0}}.$$It is noted that this is equivalent to stipulating that Roshko number similarity exists for these two conditions. From [Eq. (63)](#fd63-jresv97n5p509_A1b){ref-type="disp-formula"} $$\frac{f_{M0}}{f_{M}} = \frac{\upsilon_{M0}}{\upsilon_{M}}\left( \frac{D_{M}}{D_{M0}} \right)^{2}$$and $${\overset{\cdot}{V}}_{M0} = \frac{C_{1}f_{M}D_{M0}^{3}}{St}{(\frac{\mu_{M0}}{\mu_{M}})}\frac{\lbrack 1 + 2\alpha_{M}(T_{M} - T_{M0})\rbrack\left\lbrack {1 + \frac{(P_{M} - P_{M0})D_{M0}}{t_{M0}E_{M}}} \right\rbrack}{\lbrack 1 + 3\alpha_{F}(T_{M} - T_{M0})\rbrack\left\lbrack {1 - \frac{(P_{M} - P_{M0})}{E_{F}}} \right\rbrack}.$$ It is apparent that, to obtain high accuracy flowrate measurements using the procedures described above, appropriately high accuracy measurements are required for the component measurement systems and for the pertinent material properties as shown in [Eq. (65)](#fd65-jresv97n5p509_A1b){ref-type="disp-formula"}. In turn, it appears feasible that once systematic uncertainties are satisfactorily removed from calibration facilities, the measurement processes in calibration laboratories, and the measurement systems making on-line measurements will be commensurately improved and the uncertainty levels for these measurements can be predicted using such [Eqs. as (2)](#fd2-jresv97n5p509_A1b){ref-type="disp-formula"}, [(10)](#fd10-jresv97n5p509_A1b){ref-type="disp-formula"}, [(25)](#fd25-jresv97n5p509_A1b){ref-type="disp-formula"}, and [(65)](#fd65-jresv97n5p509_A1b){ref-type="disp-formula"}. 5. Discussion ============= The above-derived results can, for the sake of brevity, be assessed by considering the effects of temperature and pressure on the respective factors--the calibrator and flowmeter constants and the flowrate measurement at reference conditions. To do this [Table 1](#t1-jresv97n5p509_a1b){ref-type="table"} shows, for specific material conditions and geometrical sizes, the variations associated with temperature and pressure effects separately and then summed together. The results shown in the first row of [Table 1](#t1-jresv97n5p509_a1b){ref-type="table"} present the variations in calibrator constant *K*~C0~ determined using geometrical measurement methods for variations of ± 1 °C in temperature and ±l×l0^5^Pa (1 atm) in pressure. The material properties and geometrical assumptions are given under the headings of the respective columns. The worst-case combination of temperature and pressure variations taken separately are given in columns 7 and 12, respectively. These results are obtained by adding the absolute values of the component contributions. The total worst-case combination for temperature and pressure variations taken together is given in the column at the right side of the table. Accordingly, the total temperature effect on *K*~C0~ in the measurement method is five times larger than the pressure effect and the total of these gives an imprecision of ± 0.006%. The results for *K*~C0~ determined using the draw technique are shown in the second row. These give a total temperature variation of ±0.095% owing mainly to fluid expansion effects. The pressure variation is noted to be a factor of about 16 less than this level and the total imprecision totals ±0.101% which is about a factor of 20 larger than the level achieved using the geometrical measurement technique. The results for the meter factor, *K*~M0~, are given in the third row of the table. Here, at reference conditions, no additional uncertainties are shown over those for *K*~C0~ in accord with [Eq. (42)](#fd42-jresv97n5p509_A1b){ref-type="disp-formula"}. The fourth row indicates that the total temperature variation for *St* is increased over that for *K*~M0~ by the amount allocated for meter expansion. This produces the ±0.101% uncertainty total for temperature which when added to the increased pressure total gives ±0.108%. This shows that while the Strouhal number is the preferred non-dimensional parameter to characterize the frequency effects of a meter, it has more factors than the conventional meter factor, *K*~M0~ and, therefore, it can have increased uncertainty. However, because the Strouhal number is a dimensionless ratio of frequency to inertial effects in the meter it should be successful in producing more satisfactory metering results for widely ranging conditions than can be done using a dimensional quantity such as *K*~M0.~ The fifth and sixth rows express the uncertainties for *K*~M~ and the corresponding Strouhal number in a manner analogous to that used to obtain values in rows three and four. However, as shown in [Eq. (53)](#fd53-jresv97n5p509_A1b){ref-type="disp-formula"} the uncertainties attributed to *K*~C0~ are increased appropriately due to encoder, calibrator, and fluid effects; and, in the case for *St*, meter effects. It is noted that the totaled uncertainties have essentially doubled in comparison with those for *K*~M0~. While it is recognized that the conditions of ± 1 °C and ± 1 × 10^5^ Pa (1 atm) pressure variation can be termed large, it should be recognized that no uncertainty has been allocated in [Table 1](#t1-jresv97n5p509_a1b){ref-type="table"} for the uncertainty with which the material properties and geometrical dimensions are known or have been determined. As well, it should be recognized that these assumed conditions of ± 1 °C and ± 1×10^5^ Pa (1 atm) can in reality understate the actual variations that may exist in real situations where flowrate measurements can be attempted in harsh, hostile environments. It should be emphasized that the above discussion deals only with the uncertainties associated with temperature and pressure imprecision and does not take into account contributions from flowmeter imprecision such as the variability of detecting meter pulses. Additionally, systematic errors which can greatly exceed the levels of imprecision are not included in any of the above. A summary of the equations derived for characterizing piston-type, encoded-stroke calibrators, their calibration procedures, and the subsequent use of calibrated turbine meters is given in [Appendix C](#app3-jresv97n5p509_A1b){ref-type="app"}. These equations are found to be quite simple when reference conditions prevail. However, reference conditions are practically fictitious and therefore the performance of all of these devices should be considered as occurring in non-reference conditions. Given the capabilities available in today's computers, it is felt that the complete equations for these types of devices should be installed in the controlling and processing software so that when temperature and or pressure conditions become significant in these processes the results are accurate. Using these equations when temperature and or pressure effects are not significant produces negligible differences from the results at reference conditions. 6. Conclusions ============== High accuracy fluid measurements can be attained when the important factors affecting the performances of both flowmeter calibrators and fluid meters are properly taken into account in the measurement processes of these devices. To do this--to first order temperature and pressure effects on both the calibrator and the meter as well as on the fluid--the equations derived above should be used as the basic system models for calibrator and meter. Specific features of calibrators need to be analyzed and, where pertinent, appropriate modifications to the above-derived equations should be made. Calibrators and metering units need to be properly instrumented and operated according to the required assumptions and techniques for handling the data. Where further increases in flow measurement accuracy are needed, either more accurate descriptions of currently considered factors should be made or additional factors not currently considered should be assessed and included as pertinent, or both. This manuscript has evolved and been assembled as the results of many contributions, direct and indirect, from many sources---NIST staff, DoD metrologists, industrial personnel, and the flow measurement staff members in several national laboratories in Canada, Italy, and the U.K. Special thanks are due to K.R. Benson, NIST, Gaithersburg, for his reactions and contributions all through this effort; ideas and suggestions have also been incorporated from NIST Fluid Flow Group staff members: W. G. Cleveland, D. E. Hess, and N. E. Mease. It is especially pertinent to acknowledge the significant inputs and contributions made to these tests and this paper by V. Grabe, P. Olivier, and D. Ruffner. The secretarial contributions made by Mrs. G. M. Kline, NIST Fluid Flow Group, toward the production of this document are most gratefully acknowledged. In the DoD areas, it is pertinent to cite, alphabetically: Air Force: W. Callis, K. Hartz, B. Keeley, and J. Sharp; Army: J. Miller and D. Sherk; Navy: R. Cheesman, M. Gee, B. McBurney, and D. Todd. *Symbol* *Description* *Dimensions* ------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------- *A*~C0~, *A*~C~ Calibrator cross-sectional areas at reference and non-reference conditions, respectively. Over-bars, where used, denote averages over piston stroke length. L^2^ *A*~t0~, *A*~t~ Cross-sectional areas of tubes connected to calibrator piston at reference and non-reference conditions, respectively. L^2^ *C*~0~, *C* Circumference of cylindrical body in unstressed and stressed conditions, respectively. L *d*~0~, *d* Diameter of rod connected to calibrator piston at reference and non-reference conditions, respectively. L *D*~0~, *D* Inner diameter of calibrator cylinder at reference and non-reference conditions, respectively. L *D*~M0~, *D*~M~ Inside diameter of flowmeter at reference and non-reference conditions, respectively. L *d*~M0~, *d*~M~ Outside diameter of turbine propeller at reference and non-reference conditions, respectively. L *D*~C0~, *D*~C~ Inside diameter of the cylinder of the calibrator at reference and non-reference conditions, respectively. L *D*~T0~, *D*~T~ Outside diameter of the tube connected to the calibrator piston at reference and non-reference conditions, respectively. L *E*~C~ Modulus of elasticity of the material of the calibrator cylinder at pertinent conditions. F/L^2^ *E*~F~ Modulus of elasticity of the fluid at pertinent conditions. F/L^2^ *E*~M~ Modulus of elasticity of the material of the body of the flowmeter at pertinent conditions. F/L^2^ *E*~T~ Modulus of elasticity of the material of the tube connected to the calibrator piston at pertinent conditions. F/L^2^ *f*~E0~, *f*~E~ Frequency of pulses from encoder at reference and non-reference conditions, respectively. pulses/t *f*~M0~, *f*~M~ Frequencies of pulses from the flowmeter at reference and non-reference conditions respectively. pulses/t *K*~E0~, *K*~E~ Encoder constant at reference and non-reference conditions, respectively. pulses/L *K*~C0~, *K*~C~ Calibrator constant at reference and non-reference conditions, respectively. pulses/L^3^ *K*~Ć~ Calibrator constant in pulses per fluid volume collected in collection vessel. pulses/L^3^ *K*~M0~, *K*~M~ Meter factor at reference and non-reference conditions, respectively. Where over-bars are used is meant the average value over a Reynolds number range. pulses/L^3^ *L*~E0~, *L*~E~ Piston stroke length at reference and non-reference conditions, respectively. L *N*~E0~, *N*~E~ Totalized encoder pulses at reference and non-reference conditions, respectively. pulses *N*~M0~, *N*~M~ Totalized meter pulses at reference and non-reference conditions, respectively. pulses ***n*** Unit vector. dimensionless *P*~C0~, *P*~C~ Pressure of the fluid in the cylinder of the calibrator at reference and non-reference conditions, respectively. F/L^2^ *P*~COLL'T~ Pressure of fluid in collection vessel. F/L^2^ *P*~C0~, *P*~C~ Pressure of the fluid in the flowmeter at reference and non-reference conditions, respectively. F/L^2^ *Re* Reynolds number. dimensionless *Ro* Roshko number. dimensionless *St* Strouhal number. Where over-bars are used is meant the average value over a Reynolds number range. dimensionless *S* Surface area of control volume. L^2^ *T*~r~ Temperature of meter rotor or of rod connected to calibrator piston. °C (°F) *T*~C0~, *T*~C~ Temperature of the cylinder of the calibrator at reference and non-reference conditions, respectively. °C (°F) *T*~E0~, *T*~E~ Temperature of the encoder at reference and non-reference conditions, respectively. °C (°F) *t*~C0~ Thickness of the cylinder of the calibrator at reference conditions. L *t*~T0~ Thickness of the tube connected to the to the calibrator piston at reference conditions. L *t* Time. t *T*~COLL'T~ Temperature of fluid in collection vessel. °C (°F) *T*~M0~, *T*~M~ Temperature of the flowmeter at reference and non-reference conditions, respectively. °C(°F) *t*~M0~ Thickness of the body of the flowmeter at reference conditions. L *U*~M0~ Average velocity of fluid through the flowmeter at reference conditions. L/t *V*~C0~, *V*~C~ Calibrator piston displacement at reference and non-reference conditions, respectively. L^3^ ***v*** Fluid velocity vector. L/t *V* Control volume. L^3^ ${\overset{˙}{V}}_{C0}$, *V*~C0~ Fluid volumetric flowrate and volume, respectively, in the calibrator cylinder at reference condition. L^3^/t,L^3^ ${\overset{˙}{V}}_{{COLL}^{’}T0}$, *V*~COLL'TO~ Fluid volumetric flowrate and volume, respectively, into collection vessel at reference conditions. L^3^/t,L^3^ ${\overset{˙}{V}}_{C}$, *V*~C~ Fluid volumetric flowrate and volume, respectively, in the calibrator cylinder at non-reference conditions. L^3^/t,L^3^ ${\overset{˙}{V}}_{M0}$, *V*~M0~ Fluid volumetric flowrate and volume, respectively, in flowmeter at reference conditions. L^3^/t,L^3^ ${\overset{˙}{V}}_{M}$, *V*~M~ Fluid volumetric flowrate and volume, respectively, through the flowmeter at non-reference conditions. L^3^/t,L^3^ *α*~E~ Linear expansion coef-of the encoder at pertinent conditions. °C^−1^(°F^−1^) *α*~C~ Linear expansion coefficient for the material of the calibrator cylinder at pertinent conditions. °C^−1^(°F^−1^) *α*~F~ Linear expansion coefficient of fluid. °C^−1^ (°F^−1^) *α*~M~ Linear expansion coefficient for the material of the body of the flowmeter. °C^−1^ (°F^−1^) *α*~r~ Linear expansion coefficient for the material of rotor or turbine wheel. °C^−1^ (°F^−1^) *β* Ratio of turbine propeller diameter to inside diameter of meter. dimensionless ϵ~θ~ Hoop strain in cylindrical body. dimensionless *ρ* Fluid density. M/L^3^ *ρ*~C0~, *ρ*~C~ Fluid density in calibrator at reference and non-reference conditions, espectively. M/L^3^ *ρ*~COLL'T0,~ *ρ*~COLL'T~ Density of fluid collected in collection vessel at reference and non-reference conditions, espectively. M/L^3^ *ρ*~M0~, *ρ*~M~ Density of the fluid in the flowmeter at reference and non-reference conditions, respectively. M/L^3^ σ~θ~ Hoop stress in cylindrical body. F/L^2^ ν~0~, *v* Fluid kinematic viscosity at reference and non-reference conditions, espectively. L^2^/t *μ*~0~, *μ* Fluid absolute viscosity at reference and non-reference conditions, espectively. Ft/L^2^ The fluid pressure and temperature inside the cylinder of the calibrator can enlarge or reduce the cross-sectional areas of this cylinder, see [Fig. 1](#f1-jresv97n5p509_a1b){ref-type="fig"}. As well, geometrical changes of the rod or tube connected to the piston can also contribute to changes in this cross-sectional area. Of course, knowledge of specific geometries and material properties are crucial to characterizing these changes. In the following it is assumed that a generic cylinder and rod geometry exist as sketched in [Fig. 1](#f1-jresv97n5p509_a1b){ref-type="fig"}. First, thermal expansion will increase the diameters of the cylinder and the connecting tube or rod via the first order approximations $$D = D_{0}\lbrack 1 + \alpha_{C}(T_{C} - T_{0}\rbrack$$and $$d = d_{0}\lbrack 1 + \alpha_{r}(T_{r} - T_{0}\rbrack,$$where, in compatible units, *D* and *d* and *D*~0~ and *d*~0~ are the cylinder and tube diameters at non-reference and reference conditions, respectively. The linear expansion coefficients for cylinder and tube or rod are *α*~C~ and *α*~r~, respectively, and *T*~C~ and *T*~r~ are the corresponding temperatures. These diametral enlargements will produce an increased cross-sectional area via the first order approximation $${\overline{A}}_{C} = {\overline{A}}_{C0}\lbrack 1 + 2\alpha_{C}(T_{C} - T_{0}\rbrack,$$where *Ā*~C~ and *Ā*~C0~ are, respectively, the annular areas between cylinder and rod as averaged over the appropriate piston displacement. The linear expansion coefficient of the rod or tube is assumed to be the same as that for the cylinder. Second, pressure effects can change the cross-sectional area through the enlargement of the calibrator cylinder and the contraction of the rod or tube connected to the piston. The azimuthal or "hoop" stress, *σ~β~*, produced in the calibrator cylinder can be shown to be $$\sigma_{\theta} = (P_{C} - P_{0})D_{C}/2t_{C},$$where *P*~C~--*P*~0~ is the pressure difference between the pressure in the cylinder and the reference pressure outside the cylinder. The quantities *D*~C~ and *t*~C~ are, respectively, the inside diameter and the thickness of the cylinder. The azimuthal strain produced by this stress is assumed to be $$\in_{\theta} = \sigma_{\theta}/E_{C},$$where transverse effects are neglected and where *E*~C~ is the modulus of elasticity of the material of the calibrator. This azimuthal strain increases the circumference of the calibrator cylinder via $$C = C_{0}(1 + \in_{\theta}),$$where *C* and *C*~0~ are, respectively, the inside circumferences of the cylinder in stressed and unstressed conditions. This produces an increase, to first order, in the cylinder's averaged cross-sectional area of $${\overline{A}}_{C} = \frac{{\overline{C}}^{2}}{4\pi} = \frac{{{\overline{C}}_{0}}^{2}}{4\pi}\left\lbrack {1 + \frac{(P_{C} - P_{0})D_{C0}}{t_{C0}E_{C}}} \right\rbrack$$or since ${\overline{A}}_{C0} = {{\overline{C}}_{0}}^{2}/4\pi,$ $${\overline{A}}_{C} = {\overline{A}}_{C0}\left\lbrack {1 + \frac{(P_{C} - P_{0})D_{C0}}{t_{C0}E_{C}}} \right\rbrack,$$where *D*~C0~ and *t*~C0~ are the diameter and thickness, respectively, of the calibrator cylinder at reference conditions. Typical values for the ratio *D*~C0~/*t*~C0~could be 10 to 20. In similar fashion, the cylinder pressure in excess of the reference pressure, *P*~0~ at which condition the reference dimensions are determined can reduce the outside diameter of the tube connected to the piston. Since the deformation of a tube should, for the same material and diameter, exceed that for a solid rod, it shall be considered here. Therefore, the cross-sectional area of the tube connected to the piston would be $${\overline{A}}_{t} = {\overline{A}}_{t0}\left\lbrack {1 - \frac{(P_{C} - P_{0})D_{t0}}{t_{t0}E_{t}}} \right\rbrack,$$where *D*~t0~ and *t*~t0~ are, respectively, the outside diameter and the thickness of the wall of the tube connected to the piston at reference conditions. The quantity *E*~T~ is the modulus of elasticity of the tube material, at reference conditions. It is generally and safely assumed that the tube contraction effect expressed in [Eq. (A.1.9)](#fd74-jresv97n5p509_A1b){ref-type="disp-formula"} is negligible in comparison to the effects of temperature on the inner diameter of the calibrator. Therefore, it is assumed that the annular cross-sectional area between cylinder and tube is that expressed in [Eq. (A.1.3)](#fd68-jresv97n5p509_A1b){ref-type="disp-formula"} and the effect of pressure on this annular area is expressed in [Eq. (A.1.8)](#fd73-jresv97n5p509_A1b){ref-type="disp-formula"}. However, where specifics indicate that other compensations need to be made, other approximations can and should be done. The combined effect of pressure and temperature on the averaged cross-sectional area of the cylinder of the calibrator is, $${\overline{A}}_{C} = {\overline{A}}_{C0}\lbrack 1 + 2\alpha_{C}(T_{C} - T_{C0}\rbrack\left\lbrack {1 + \frac{(P_{C} - P_{0})D_{C0}}{E_{C}t_{C0}}} \right\rbrack.$$ Conventional practice in processing and using turbine-type flowmeter data began with plotting the meter *K* factor in pulses per volume units as a function of meter frequency, *f*; see [Fig. A.2.1](#fA1-jresv97n5p509_a1b){ref-type="fig"} and \[[@b5-jresv97n5p509_a1b]--[@b8-jresv97n5p509_a1b]\]. In this figure, the bracketed dots are intended to represent the mean values and the scatter obtained for the *K* factors at respective frequencies. For fluid and flow conditions which precisely duplicate those for which the meter calibration was performed, this practice produced acceptable results. However, when fluid or flow conditions deviated from those of the calibration, it was found that results could be unacceptable. An example is indicated in [Fig. A.2.1](#fA1-jresv97n5p509_a1b){ref-type="fig"}. The lines for which the liquid viscosities are *μ*~1~ and *μ*~2~, respectively, indicate that different meter performance occurs at low flowrates when the fluid dynamic viscosity, *μ*~2~ is greater than the value *μ*~1~ used for the calibration. {#fA1-jresv97n5p509_a1b} Flowmeter performance which deviates from that of the calibration conditions at low flowrates is explained by citing the deviant phenomena which have, because of the different fluid or flow conditions, become significant in comparison to the phenomena which prevailed during calibration. Shafer and Lee have resolved several situations involving viscous effects in specific flowmeter geometries \[[@b6-jresv97n5p509_a1b],[@b8-jresv97n5p509_a1b]\]. From such efforts have come Universal Viscosity Curves (UVC) i.e., plots of meter factor versus the ratio of meter frequency to fluid kinematic viscosity. A typical UVC is shown in [Fig. A.2.2](#fA2-jresv97n5p509_a1b){ref-type="fig"}. The ordinate, *K*~0~, is the meter factor referenced to a selected reference temperature via, see \[[@b10-jresv97n5p509_a1b]\], $$K_{0} = K\lbrack 1 + 3\alpha_{M}(T - T_{0}\rbrack.$$ {#fA2-jresv97n5p509_a1b} Deviations of flowmeter performance characteristics from UVC's can be explained by citing further deviants, such as lubricity, cavitation, extreme temperature or pressure effects on fluid properties, etc., see \[[@b5-jresv97n5p509_a1b]--[@b8-jresv97n5p509_a1b]\]. When the turbine meter is to be used in temperature conditions different from those of the reference or calibration conditions, temperature effects may be compensated for by producing corrections for the meter factor *K*, as found in some standards, see \[[@b10-jresv97n5p509_a1b]\]. This ISA standard produces an underived correction that is based upon the thermal expansion for the material of the meter body. This correction is effective in predicting trends for the case where the meter body and the turbine wheel material are the same. Where these materials are different, this correction can give erroneous results, for example, in the extreme situation where the turbine wheel has expanded to touch the inner wall of the meter body, thus stopping the wheel's rotation. In such a situation the meter factor should be zero yet the ISA correction would not give this result. Therefore, further evolution is needed for conventional turbine-type flowmeter practice---beyond the UVC methods. Turbine meter performance is most properly characterized by specifying and quantifying, over the pertinent ranges of fluid and flow conditions, the significant dimensionless parameters which influence meter performance. To do this, the dimensional quantities that are involved in turbine-type flowmetering are cited as shown in [Fig. A.2.3](#fA3-jresv97n5p509_a1b){ref-type="fig"}. These five dimensional quantities---fluid density *ρ*, and viscosity *μ*, the flow velocity *U*, the meter frequency *f*, and the meter diameter, *D* are described in terms of three independent units --- length, time, and mass or force. Consequently, there are two dimensionless parameters to characterize the performance of this meter. These dimensionless parameters can take many forms. Conventional fluid mechanical practice would produce parameters expressing ratios of meter frequency effects to fluid inertial effects, i.e., the Strouhal number and the fluid inertial to viscous effects, i.e., the Reynolds number. These are written $$St = fD/U$$and $$Re = \frac{DU_{\rho}}{\mu} = \frac{DU}{\upsilon}.$$ {#fA3-jresv97n5p509_a1b} However, turbine meters are designed to have frequencies proportional to volumetric flowrates. The constant of proportionality is the reciprocal of the meter factor. Since the constancy of this meter factor, i.e., the linearity of the meter over the flowrate (frequency) range is generally taken to be a measure of the quality of the meter, it has become conventional turbine meter practice to characterize meters via their *K* factor versus frequency dependence. In non-dimensional format, this can be done by specifying meter performance using a Strouhal number that is dependent upon the product of the Strouhal and Reynolds numbers, i.e., *fD*^2^/*v.* This parameterization is essentially that known as the UVC except for the factors involving the meter diameter, *D*, raised to an exponent. This product has been referred to as the Roshko number, *Ro*, $$Ro = \frac{fD^{2}}{\upsilon}$$see \[[@b9-jresv97n5p509_a1b]\]. Alternatively, the pair of parameters, *St* and *Re* could also be used to characterize the meter's performance. In either case, only two parameters are needed to characterize the system sketched in [Fig. A.2.3](#fA3-jresv97n5p509_a1b){ref-type="fig"}. In all of the above, the effect of different meter geometries or a different turbine wheel is not considered. This is proper where the meter geometry is not changed and where the same turbine wheel and bearings, etc. are used and where the turbine wheel and the meter body are made of the same material. Under these circumstances, the above parameterization of the five dimensional quantities *ρ,μ,U,D*, and *f* is completed with the formulation of the Strouhal and Reynolds or Roshko numbers. Using this formulation the performance of the calibrated meter should enable the meter to be satisfactorily used over a range of conditions similar to those encountered in the calibration. If the situation is changed as shown in [Fig. A.2.4](#fA4-jresv97n5p509_a1b){ref-type="fig"} so that the turbine wheel diameter, *d*, is included to produce six dimensional quantities, then three dimensionless parameters should be formed. The turbine wheels that are considered here are geometrically scaled versions of each other, i.e., they have the same number of blades, the same blade shape and only differ in the diameter, *d.* The nondimensionalization can be achieved by choosing the third dimensionless parameter to be *β=d/D*, the ratio of turbine wheel to inside pipe diameter. The three parameter performance is sketched in [Fig. A.2.5](#fA5-jresv97n5p509_a1b){ref-type="fig"}. It is noted that when the turbine wheel and meter body are the same material and when only thermal expansion effects are considered, *β* remains constant with changes in temperature, i.e., $$\beta = \frac{d}{D} = \frac{d_{0}\lbrack 1 + \alpha_{r}(T_{r} - T_{0}\rbrack}{D_{0}\lbrack 1 + \alpha_{M}(T_{M} - T_{0}\rbrack},$$when *T*~r~ = *T*~M~ and *α*~r~ *= α*~M~ this diameter ratio is *d*~0~/*D*~0~. When turbine wheel and meter body are different materials and when *α*~r~*\>α*~M~, the beta ratio is not constant and a condition for the linear expansion coefficients and the fluid flowrate can be derived for which *T*~M~ *= T*~r~, and *β* = 1. This would indicate that the turbine blade tips touch the inside of the meter body thus stopping the wheel. This condition is, of course, contrived in order to illustrate the point that *β* variation can produce meter performance variation and in extreme limits can radically alter the graphs of meter performance. {#fA4-jresv97n5p509_a1b} {#fA5-jresv97n5p509_a1b} If pressure effects are not negligible in altering the diameter of the meter body but are negligible in affecting the diameter of the turbine wheel, *β* is altered, to first order in temperature and pressure, separately, via $$\beta = \frac{d_{0}\lbrack 1 + \alpha_{r}(T_{r} - T_{0})\rbrack}{D_{0}\lbrack 1 + \alpha_{M}(T_{M} - T_{0})\left\lbrack {1 + \frac{(P_{M} - P_{0})D_{M0}}{2t_{M0}E_{M}}} \right\rbrack}.$$By so incorporating thermal and pressure effects using *St*, *Re* or *Ro*, and *β* parameters into performance curves as sketched in [Fig. A.2.6](#fA6-jresv97n5p509_a1b){ref-type="fig"}, widely varying metering conditions should be successfully handled. {#fA6-jresv97n5p509_a1b} It is concluded that turbine-type flowmeter performance should be handled non-dimensionally. Pertinent corrections are then most clearly seen and most easily performed using the appropriate, first-order, relationships for the thermal and pressure effects. In this way high levels of flowmeter performance can be expected and achieved. The equations in [Appendix C](#app3-jresv97n5p509_A1b){ref-type="app"} are numbered as in the main text. *Calibrator Characterization:* Geometrical Measurement Method---Reference Conditions (see p. 513) $$K_{C0} = \frac{N_{E0}}{V_{C0}} = \frac{L_{E0}K_{E0}}{{\overline{A}}_{C0}L_{E0}} = \frac{K_{E0}}{{\overline{A}}_{C0}}.$$Geometrical Measurement Method--Non-Reference Conditions (see p. 515) $$K_{C0} = K_{C}\frac{\lbrack 1 + \alpha_{E}(T_{E} - T_{0})\rbrack}{\lbrack 1 - 2\alpha_{C}(T_{C} - T_{0}\rbrack\left\lbrack {1 - \frac{(P_{C} - P_{0})D_{C0}}{t_{C0}E_{C}}} \right\rbrack},$$where $$K_{C} = \frac{N_{E}}{V_{C}},$$Draw Technique--Reference Conditions (see p. 516) $$K_{C0} = N_{E0}/V_{{COLL}^{’}T0}.$$Draw Technique--Non-Reference Conditions (see p. 517) $$\begin{array}{l} {K_{C0} = K_{C}^{\prime}} \\ \frac{\lbrack 1 + 2\alpha_{C}(T_{C} - T_{0})\rbrack\left\lbrack {1 + \frac{(P_{C} - P_{0})D_{C0}}{t_{C0}E_{C}}} \right\rbrack}{\lbrack 1 - 3\alpha_{F}(T_{{COLL}^{’}T} - T_{C})\rbrack\left\lbrack {1 + \frac{P_{{COLL}^{’}T} - P_{C}}{E_{F}}} \right\rbrack\lbrack 1 - \alpha_{E}(T_{E} - T_{0})\rbrack} \\ \end{array}$$where $$K_{C}^{\prime} = N_{E}/V_{{COLL}^{’}T}.$$*Meter Calibration:* *Reference Conditions:* (see p. 519) $$K_{M0} = \frac{f_{M0}}{f_{E0}}\ \frac{K_{E0}}{{\overline{A}}_{C0}} = \frac{N_{M0}}{N_{E0}}\ \frac{K_{E0}}{{\overline{A}}_{C0}} = \frac{N_{M0}}{N_{E0}} \bullet K_{C0} = \frac{f_{M0}}{f_{E0}} \bullet K_{C0}.$$*Data Processing*: (see p. 519) $$St = \frac{f_{M0}D_{M0}}{U_{M0}} = C_{1}K_{M0}D_{M0}^{3} \propto K_{M0}D_{M0}^{3},$$ $$Re = \frac{D_{M0}U_{M0}}{\upsilon_{0}},$$ $$R_{0} = \frac{f_{M0}D_{M0}^{2}}{\upsilon_{0}}.$$*Non-Reference Conditions:* (see p. 521) $$\begin{array}{l} {K_{M} = \frac{f_{M}}{f_{E}}K_{C}\frac{\rho_{M}}{\rho_{C}} = \frac{N_{M}}{N_{E}}K_{C0}} \\ {\times \frac{\lbrack 1 - \alpha_{E}(T_{E} - T_{0})\rbrack\lbrack 1 + 3\alpha_{F}(T_{C} - T_{M})\rbrack}{\lbrack 1 + 2\alpha_{C}(T_{C} - T_{0})\rbrack\left\lbrack {1 + \frac{(P_{C} - P_{0})D_{C0}}{t_{C0}E_{C}}} \right\rbrack\left\lbrack {1 + \frac{(P_{C} - P_{M})}{E_{F}}} \right\rbrack.}} \\ \end{array}$$*Data Processing:* (see p. 522) $$St = f_{M}D_{M}/U_{M} = C_{1}K_{M}D_{M}^{3},$$ $$Re = D_{M}U_{M}/\upsilon,$$ $$Ro = \frac{f_{M}D_{M}^{2}}{\upsilon}.$$*Flow Measurement:* *Reference Conditions:* (see p. 522) $${\overset{˙}{V}}_{M0} = \frac{f_{M0}}{{\overline{K}}_{M0}} = \frac{C_{1}D_{M0}^{3}f_{M0}}{\overline{St}}.$$*Non-Reference Conditions:* (see p. 523) $${\overset{˙}{V}}_{M} = \frac{f_{M}}{{\overline{K}}_{M}} = \frac{C_{1}f_{M}D_{M}^{3}}{\overline{St}},$$ $${\overset{˙}{V}}_{M0} = \frac{C_{1}f_{M}D_{M0}^{3}}{St}{(\frac{\mu_{M0}}{\mu_{M}})}\frac{\lbrack 1 + 2\alpha_{M}(T_{M} - T_{M0})\rbrack\left\lbrack {1 + \frac{(P_{M} - P_{M0})D_{M0}}{t_{M0}E_{M}}} \right\rbrack}{\lbrack 1 + 3\alpha_{F}(T_{M} - T_{M0})\rbrack\left\lbrack {1 - \frac{(P_{M} - P_{M0})}{E_{F}}} \right\rbrack}.$$ **About the author:** Dr. George E. Mattingly is a mechanical engineer who leads the Fluid Flow Group in the Process Measurements Division of the NIST Chemical Science and Technology Laboratory. The National Institute of Standards and Technology is an agency of the Technology Administration, U.S. Department of Commerce. {#f1-jresv97n5p509_a1b} {#f2-jresv97n5p509_a1b} {#f3-jresv97n5p509_a1b} {#f4-jresv97n5p509_a1b} {#f5-jresv97n5p509_a1b} {#f6-jresv97n5p509_a1b} {#f7-jresv97n5p509_a1b} {#f8-jresv97n5p509_a1b} {#f9-jresv97n5p509_a1b} ###### Temperature and pressure effects Factor affected Total diff (°C) Temperature effects Total temperature combined worst case Press diff (atm) Pressure effects Total pressure combined worst case Total worst case -------------------------------------------------------------------------- ----------------- --------------------- --------------------------------------- ------------------ ------------------ ------------------------------------ ------------------ --------- --------- --------- --------- --------- *K*~C0:GEOM~[a](#tfn1-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.001% ±0.004% ±0.005% ±1 ±0.001% ±0.001% ±0.006% *K*~C0:DRAW~[b](#tfn2-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.001% ±0.004% ±0.090% ±0.095% ±1 ±0.001% ±0.005% ±0.006% ±0.101% *K*~M0~[c](#tfn3-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.001% ±0.004% ±0.090% ±0.095% ±1 ±0.001% ±0.005% ±0.006% ±0.101% *St*[d](#tfn4-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.001% ±0.004% ±0.090% ±0.006% ±0.101% ±1 ±0.001% ±0.005% ±0.001% ±0.007% ±0.108% *K*~M~[e](#tfn5-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.002% ±0.008% ±0.180% ±0.190% ±1 ±0.002% ±0.010% ±0.012% ±0.202% *St*[f](#tfn6-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.002% ±0.008% ±0.180% ±0.006% ±0.196% ±1 ±0.002% ±0.010% ±0.001% ±0.013% ±0.209% ${\overset{*}{V}}_{MU}$[g](#tfn7-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.001% ±0.004% ±0.090% ±0.090% ±1 ±0.001% ±0.005% ±0.006% ±0.101% ${\overset{*}{V}}_{M}$[h](#tfn8-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.002% ±0.008% ±0.180% ±0.190% ±1 ±0.002% ±0.010% ±0.012% ±0.202% ${\overset{*}{V}}_{MU}$[i](#tfn9-jresv97n5p509_a1b){ref-type="table-fn"} ±1 ±0.002% ±0.008% ±0.270% ±0.012% ±0.292% ±1 ±0.002% ±0.015% ±0.002% ±0.019% ±0.311% Determined via [Eq. (10)](#fd10-jresv97n5p509_A1b){ref-type="disp-formula"} at non-reference conditions. Determined via [Eq. (29)](#fd29-jresv97n5p509_A1b){ref-type="disp-formula"} at non-reference conditions with no additional uncertainty for the encoder or the collection volume. Determined using *K*~CU:DRAW~ via [Eq. (42)](#fd42-jresv97n5p509_A1b){ref-type="disp-formula"} with no additional uncertainty resulting from encoder or meter outputs. Determined using *K*~CU:DRAW~ via [Eq. (43)](#fd43-jresv97n5p509_A1b){ref-type="disp-formula"} in reference conditions. Determined using *K*~CU:DRAW~ via [Eq. (53)](#fd53-jresv97n5p509_A1b){ref-type="disp-formula"}. Determined using *K*~CU:DRAW~ via [Eq. (54)](#fd54-jresv97n5p509_A1b){ref-type="disp-formula"}. Determined using *K*~M0~ via [Eq. (59)](#fd59-jresv97n5p509_A1b){ref-type="disp-formula"} with no additional uncertainty resulting from the meter output. Determined using *K*~M~ via [Eq. (60)](#fd60-jresv97n5p509_A1b){ref-type="disp-formula"} with no additional uncertainty resulting from meter output. Determined using [Eq. (65)](#fd65-jresv97n5p509_A1b){ref-type="disp-formula"} with not additional uncertainty for meter output, reference meter dimension, and absolute viscosity.

Introduction {#Sec1} ============ The mutation of p53 protein is one of the frequent causes of human cancer since it has a key role in numerous cell stability and proliferation functions^[@CR1]^. On the other hand, MDM2 is an oncogene protein that negatively regulates p53^[@CR2]^. MDMX (MDM4) is also a homolog for MDM2 and acts as p53 inhibitor^[@CR3]^. In vivo and in vitro studies have shown the overexpression of both MDM2 and MDMX proteins in several cancer types^[@CR4]^ (e. g. MDMX overexpression up to 92% of AML cases^[@CR5]^). However, their mechanisms of action are slightly different as MDM2 mainly degrades p53 protein using its E3 ubiquitin ligase activity, while MDMX suppresses p53 by decreasing its transactivation activity and increasing MDM2 function^[@CR6]^. Both MDM2 and MDMX (MDM2/X) are significantly overexpressed in cancer cells harboring wild type p53^[@CR7],[@CR8]^. As a result, the inhibition of MDM2/X-p53 interaction leads to the restoration of p53 activity and subsequent tumor suppression^[@CR9]^. From the structural point of view, MDM2/X share a similar p53-binding domain of the N-terminal hydrophobic pocket. This site engages hydrophobic bonds with p53 residues of Phe19, Trp23, and Leu26, which leads to the p53 transcriptional activity suppression^[@CR10],[@CR11]^. To pharmacologically antagonize the MDM2/X-p53 interaction, several molecules have been developed. In this path, small molecule inhibitors, such as Nutlin-3 were mostly unsuccessful in clinical trials^[@CR12]^. Although its derivative, RG7112, has shown some promising results, its unsuitable inhibition of MDMX was one of the main drawbacks^[@CR13]--[@CR15]^. Alongside small molecules, some lead peptide sequences have also been developed, mostly by the alteration in the p53-MDM2/X binding site (ETFSDLWKLLPE), including 12/1 (MPRFMDYWEGLN)^[@CR16]^, \_ENREF_17PMI (TSFAEYWNLLSP)^[@CR17]^ and pDI (LTFEHYWAQLTS)^[@CR18]^. In what follows, there are several studies to improve the kinetics and dynamics of these peptides such as synthesizing stapled peptides^[@CR19],[@CR20]^. One of the first examples was SAH-8, designed by stapling p53-MDM2/X binding site sequence^[@CR21]^. More recently, ATSP-7041 has been developed as stapled pDI peptide which later led to ALRN-6924 peptide with promising anticancer results^[@CR22]^. Nevertheless, the main limitation in this area is the common low affinity of peptides to MDMX protein which reduces their effectiveness as anticancer agents^[@CR23],[@CR24]^. As a result, the optimization of lead peptides to obtain a "potent" and "dual" inhibitor of MDM2/X is an ongoing challenge. "Potent" is a common term for high affinity and stable therapeutic candidate which targets MDM2/X. On the other hand, any agent introduced to inhibit MDM2 is not robust enough to have an acceptable cytotoxic effect unless it could also inhibit MDMX.^[@CR18]^. Consequently, "dual" refers to the capability of the peptides to target both MDM2 and MDMX proteins. Most of the previous studies focused on the rational design of p53-based peptides are usually lacking the consideration of dual inhibition of MDM2/X^[@CR25],[@CR26]^. In this regard, one of the promising approaches is using in silico and theoretical workflows such as Molecular Dynamic (MD) simulation to optimize and verify drug candidate peptides^[@CR27]--[@CR29]^. Accordingly, in this study, we applied a computational structure-based approach inclusively on experimentally validated p53-based peptides, including p53, pDI, pDIQ, and pMI to; A) study their "potent" and "dual" inhibitor activities against MDM2/X, B) find critical residues playing roles in higher affinity to MDM2/X and helicity conservation, C) follow a computational high-throughput point mutation screening on pDI, (LTFEHYWAQLTS) and pDIQ, (ETFEHWWSQLLS) to analyze and validate structurally and functionally. To these ends, 27 different MD simulations and follow-up analyses were performed to assign the suitable features of native and mutated peptides against MDM2/X. The appropriate correlation between achieved theoretical and confirmed experimental results highlighted the suitability of the suggested methodology in this study, which could lead to the implicit rational peptide design in this area. Materials and methods {#Sec2} ===================== Atomic coordinates of peptide-MDM2/X structures {#Sec3} ----------------------------------------------- The structures of peptides-MDM2/X complexes were extracted from Protein Data Bank (PDB). First, due to the lack of pDI-MDM2 structure with the total number of residues, the two close PDB codes of 3G03 and 3JZR (<https://www.rcsb.org/>) were taken. The final coordinates were achieved by the superimpositions and refinement of the mentioned PDB structures utilizing Chimera software. The coordinates of other structures were directly extracted from PDB codes of 3FDO, 4HFZ, 3DAB, 3EQS, 3EQY, 3JZS, and 3JZQ for pDI-MDMX, p53-MDM2, p53-MDMX, PMI-MDM2, PMI-MDMX, pDIQ-MDM2, and pDIQ-MDMX, respectively. The "repair PDB" command of FoldX v.3 program was employed to optimize the structures and add the missing atoms. High-throughput mutation screening of pDI and pDIQ using FoldX {#Sec4} -------------------------------------------------------------- To screen a large number of possible mutations of pDI and pDIQ peptides, the FoldX program was applied. The FoldX program computes the interaction energy robustly based on the empirical data extracted from protein engineering studies which makes it a notable assistance for peptide designing. First,\_ENREF_29 the "PSSM" interface of FoldX was utilized to mutate all the residues of peptides to 20 native amino acids (mutation to the residue itself was also carried out) in pDI-MDM2/X and pDIQ-MDM2/X complexes. Using BuildModel and AnalyseComplex interfaces upon each mutation, the final optimized structure and the interaction energy of the peptide-MDM2/X complexes were obtained. The FoldX force field was used to calculate the free energy of unfolding (ΔG) as described below.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\Delta G = \left( { 0.33 \times \Delta G_{{{\text{vdw}}}} } \right) + \Delta G_{{{\text{solvH}}}} + \Delta G_{{{\text{solvP}}}} + \Delta G_{{{\text{wb}}}} + \Delta G_{{{\text{hbond}}}} + \Delta G_{{{\text{el}}}} + \Delta G_{{{\text{kon}}}} + T\Delta S_{{{\text{mc}}}} + T\Delta S_{{{\text{sc}}}}$$\end{document}$$ where the terms are free energy difference of total van der Waals energy of all atoms (Δ*G*~vdw~); solvation energy for hydrophobic and polar groups (Δ*G*~solvH~ and Δ*G*~solvP~ , respectively); the water bridges (Δ*G*~wb~); the hydrogen bonds (Δ*G*~hbond~); the electrostatic bonds (Δ*G*~el~ and the electrostatic bonds of subunit molecules (Δ*G*~kon~). Meanwhile, Δ*S*~mc~ is defined as the entropy change when the backbone is changed to a particular conformation, and Δ*S*~sc~ is the entropy change when the side chain is transformed into an appropriate structure. *T* is the temperature^[@CR30]^. The input structures of pDI-MDM2/X and pDIQ-MDM2/X for the FoldX program were described in the "[Atomic coordinates of peptide-MDM2/X structure](#Sec3){ref-type="sec"}s" section. After each mutation on the peptides' residues, the variance of interaction energy (ΔE) was calculated. Subsequently, the extracted structures with the lowest ΔE were selected for further studies. The final atomic coordinates of selected mutant peptides_MDM2/X complexes were used as reference structures for MD simulations. Molecular dynamics simulations {#Sec5} ------------------------------ GROMACS-2018 package^[@CR31]^ was applied to perform 200 ns simulation on 27 different systems (total time of 5.4 μs) using the GROMOS96 54a7 force field. Solvation of each system was conducted via the SPC water model in the a triclinic box that the distances of molecules' centers to the edges were 15 Å. All the simulation systems were neutralized by the suitable number of Na + and Cl- ions with the final ionic concentration of 150 mM. The van der Waals interactions were truncated using energy and pressure correction of long-range dispersion. Particle Mesh Ewald (PME) technique and LINCS algorithm^[@CR32]^ were employed for the calculation of long-range electrostatic interactions and making all bond constraints, respectively. Furthermore, the integration time step was set up as 2 fs and periodic boundary conditions were utilized in all directions. Prior to MD simulation, 200 ps of energy minimization from 0 to 300 K was carried out using a method of steepest descent. Moreover, 200 ps of system equilibration at 300 K was conducted with constant volume (NVT) and constant pressure (NPT) ensembles. UCSF Chimera Viewers was used to visualize the simulation outcomes. In addition, the analysis of Dictionary of Secondary Structure for Proteins (DSSP) of all complexes was conducted during the simulation time. Follow-up "umbrella sampling" and "gRINN" analyses were also conducted as explained in the following sections. Umbrella sampling {#Sec6} ----------------- Umbrella sampling simulation has proven to be a robust method to obtain the ΔG of a protein--protein interaction using the extracted potential of mean force (PMF)^[@CR33]^. In this approach, after the converging of PMF curve at a large center of mass distance, the difference of PMF's highest and lowest values is calculated as ΔG. The center of mass is defined as the representing average point of the mass. Moreover, the "wham utility" of the GROMACS package is used to perform the Weighted Histogram Analysis Method (WHAM) to extract the values of PMF during the simulations. The whole procedure is described as follows. For pulling simulations, the atomic coordinates resulted from the last trajectories of MD simulations, as described in the "[Molecular dynamics simulations](#Sec5){ref-type="sec"}" section, were used as starting structures. Prior to pulling simulation, energy minimization and the only 200 ps of NPT ensemble were conducted. Subsequently, immobile references were adjusted for pulling simulations via the induction of restraints on MDM2/X proteins. The pull rate of 0.001 nm ps^−1^ (0.01 Å ps^−1^) and the spring constant of 1,000 kJ mol^−1^ nm^−2^ were used in the pulling of the peptides outside the structures' centers along the corresponding axis over 2.5 ns. The final COM distance between peptides and MDM2/X proteins was approximately 5 nm. In the next step, an approximate number of 25 snapshots of pulling simulations' trajectories were taken to obtain the primary structures for the windows of umbrella sampling. Moreover, to have an asymmetric distribution of sampling windows, the window spacing was defined as 0.1 nm up to 0.2 nm, beyond 3 nm of COM separation. Afterward, each window was utilized for 5 ns of MD simulation which makes the total umbrella sampling simulation time of approximate 1,125 ns. Ultimately, the PMF curve was produced and the differences of the maximum and minimum values of PMF were used to obtain the ΔG~binding~ for different complexes of peptides_MDM2/X by WHAM utility analyses. Analyses of residue interactions in peptide-MDM2/X complexes using gRINN tool {#Sec7} ----------------------------------------------------------------------------- One of the significant outputs of MD simulations is the functioning behavior of each residue especially in the protein--protein interactions. As a result, to gain information about the peptide_MDM2/X interactions on the residue level, a post-simulation analysis, called gRINN (get Residue Interaction eNergies and Networks) v1.1.0.hf1 tool, was utilized^[@CR34]^. This software uses MD simulation trajectories for building any pairwise amino acids non-bonded interaction energies. Before introducing the files into the software, all water and non-protein molecules were removed. The outputs of MD simulations including "tpr", "top" and "xtc" files were used as input files for gRINN tool. The generated data from 2000 MD simulation trajectories were employed to extract important residues interaction energy based analysis. Results and discussion {#Sec8} ====================== Mutant pDI and pDIQ peptides screening in the complex with MDM2/X {#Sec9} ----------------------------------------------------------------- MDM2 and MDMX are considered as well-known oncoproteins with inhibitory activities on the p53 protein. In addition, they are overexpressed in several types of cancerous cells harboring wild type p53^[@CR4]^. Therefore, MDM2/X suppression strategies have been considered as promising approaches for cancer therapy over the last decade^[@CR7],[@CR9]^. In this path, anticancer peptides are the robust agents to inhibit MDM2/X proteins effectively^[@CR17],[@CR18],[@CR21]--[@CR23]^. Hence, we have studied some p53-based peptides computationally to obtain high affinity, stable, and valid anti-MDM2/X peptides. In addition, we aimed to qualify our theoretical methodology for the development of "dual" anti-MDM/X peptides. To this end, a lead peptide named pDI^[@CR18]^ and its more potent derivative, pDIQ^[@CR17]^ were used as starting peptides for the mutation screening using the FoldX program. Having used the PSSM interface of the FoldX program, the mutations were scored based on their primary interaction energy (∆E) of the mutated pDI and pDIQ peptides in complex with MDM2/X. The corresponding results were displayed in Tables S[1](#MOESM1){ref-type="media"}, S[2](#MOESM1){ref-type="media"}, S[3](#MOESM1){ref-type="media"}, and S[4](#MOESM1){ref-type="media"}. The criterion to select the mutant peptides was the appropriate ∆E for both MDM2/X to propose a dual inhibiting peptide. Excluding the neutral positions showing ineffective ∆E values, some point mutations on pDI with suitable interaction energies with MDM2/X were selected (Tables S[1](#MOESM1){ref-type="media"} and S[2](#MOESM1){ref-type="media"}). The same procedure was performed on pDIQ_MDM2/X complexes (Tables S[3](#MOESM1){ref-type="media"} and S[4](#MOESM1){ref-type="media"}). In summary, Leu1, His5, Ala8, Gln9, Ser12 were considered as neutral positions in pDI_MDM2 complexes as their ∆E showed neither negative nor positive value. Thr2 was also a neutral position to mutate in the pDI_MDMX complex. However, mutations on Phe3, Trp7, Leu10 residues were shown to be non-favorable. This finding is consistent with the key residues of Phe19, Trp23, and Leu26 as p53 binding segment with MDM2/X, mentioned in other studies^[@CR11]^. Conversely, single mutations on Glu4 and Thr11 were identified with favorable ∆Es in both MDM2/X complexes. Moreover, double mutations on the same positions of pDI were carried out and ∆Es were obtained. Accordingly, three pDI mutants, pDI_E4W (pDIm1) (∆E = −1.97 and − 2.89 kcal/mol), pDI_T11M (pDIm2) (∆E = −1.58 and − 2.02 kcal/mol) and the double mutant pDI_E4WT11M (pDIdm) (∆E = −1.40 and − 1.01 kcal/mol) were selected with the most suitable ΔEs for MDM2 and MDMX, respectively (Tables S[1](#MOESM1){ref-type="media"} and S[2](#MOESM1){ref-type="media"}). Regarding pDIQ, single mutations in Thr2 and Glu4 were shown to indicate favorable ∆Es in both MDM2/X complexes. As a result, the best mutant candidates of pDIQ were pDIQ_E4K (pDIQm1) (∆E = −1.26 and − 0.83 kcal/mol) and pDIQ_T2H (pDIQm2) (∆E = −1.28 and − 0.90 kcal/mol) with the best ΔEs for MDM2 and MDMX, respectively (Tables S3[3](#MOESM1){ref-type="media"}and S[4](#MOESM1){ref-type="media"}). The extracted atom coordinates from the mentioned mutants, as well as the p53, pDI, pDIQ, and PMI peptides were collected for further MD simulations in the free and MDM2/X bound states. Secondary structural stability of peptides in the free state and complex with MDM2/X {#Sec10} ------------------------------------------------------------------------------------ The different experimentally validated anti-MDM2/X peptides, including p53, pMI, pDI, pDIQ, and also the selected mutant peptides (see the "[Mutant pDI and pDIQ peptides screening in the complex with MDM2/X](#Sec9){ref-type="sec"}" section) were entered into the MD simulations in the free and complex states. The sequences and available experimental IC~50~ of each peptide are shown in Table [1](#Tab1){ref-type="table"}.Table 1IC~50~ and computational ΔG~binding~ of the peptides_MDM2/X. The experimental IC~50~s recorded in the literature ^17^ for control studied peptides of p53, pDI, pDIQ, and PMI are presented. Also, ΔG~binding~s of all studied peptides of p53, pDI, pDIm1, pDIm2, pDIdm, pDIQ, pDIQm1, pDIQm2 and PMI in complex with MDM2/X using the "umbrella sampling" method are shown.PeptideSequenceIC~50~ (nM)ΔG~binding~^a^ (kCal/mol)MDM2MDMXMDM2MDMXp53ETFSDLWKLLPE2000600021NApDILTFEHYWAQLTS445502220pDIm1LTFWHYWAQLTS----2824pDIm2LTFEHYWAQLMS----2822pDIdmLTFWHYWAQLMS----2528pDIQETFEHWWSQLLS81103116pDIQm1EHFEHWWSQLLS----2519pDIQm2ETFKHWWSQLLS----1813PMITSFAEYWNLLSP20402519^a^The error with the calculated energy is 0.2 kcal/mol for every system.NA is not applicable. Peptide-based agents, especially with the resemblance to the p53-MDM2/X binding site, have been shown promising dual inhibition of MDM2/X. The mentioned binding site consists of 12 residues with an alpha-helix structure. Helical peptides are usually unstable without a complete protein fold which reduces their affinity to the target protein. It is therefore highly significant to preserve the helicity of such peptides. In this regard, secondary structure analysis was performed on the studied peptides both in the free state and in complex with MDM2/X, using DSSP (Dictionary of Secondary Structure of Proteins) program^[@CR35]^. The results are depicted in Fig. [1](#Fig1){ref-type="fig"}.Figure 1Helicity probability of the studied peptides during MD simulation of different studied peptides of p53, pDI, pDIm1, pDIm2, pDIdm, pDIQ, pDIQm1, pDIQm2 and PMI using DSSP program in the free state (**A**), in complex with MDM2 (**B**) and in complex with MDMX (**C**). As shown in Fig. [1](#Fig1){ref-type="fig"}A, the free p53 peptide interestingly showed one of the best conformational stability in the N-terminal residues (Thr2 and Phe3) compared to the same positions in other peptides. Likewise, in the pDIQm2 and PMI peptides, phe3, one of the conserved residues in the p53 binding site, preserved helicity. The pDIQm1 peptide has the lowest helicity scores among all of the free mutant peptides. Regarding C-terminal residues, in the free p53 peptide, Leu9 has a score of 0.48, and Leu10 and Pro11 showed no helicity. In fact, Proline is known for destabilizing alpha-helix structures especially in the C-terminal residues^[@CR36]^ and this is one of the reasons p53 lost its helicity during the 200 ns simulation. However, other free peptides showed diverse range of stability in alpha-helix conformation in residues 8 to 11 (Fig. [1](#Fig1){ref-type="fig"}A). Overall, the mutant pDI and pDIQ peptides showed the best helicity in all Ala8, Gln9, Leu10, and Met11 compared to other studied free peptides. In the complex forms with MDM2/X, as expected, the conformational stability of most residues was increased (Fig. [1](#Fig1){ref-type="fig"}B, C). However, in contrast to the free states, residue 2 of the peptides in complex with MDM2/X showed no or very little score of helicity (Fig. [1](#Fig1){ref-type="fig"}B, C). Moreover, in residue 3, the helicity scores were less and were varied in the peptides_MDM2/X complexes (Fig. [1](#Fig1){ref-type="fig"}B, C). Residue 11 of PMI bound to MDM2 showed no helicity compared to its free state (Fig. [1](#Fig1){ref-type="fig"}B). In the p53_MDMX, peptide residues indicated lower helicity scores in the positions 3 to 5 rather than the scores of the same positions in the p53_MDM2. The conformational stability of pDIQm2 was also increased in complex forms possibly due to the favorable interaction energy between the peptide and MDM2/X. The overall results confirmed the improvement of the conformational stability of the previously designed peptides (pDI, pDIQ, and PMI) both in the free and complex forms compared to p53. Moreover, the newly developed mutant peptides especially pDIm2, pDIdm, and pDIQm1 have shown the best helicity score, particularly in the C-terminal residues. The importance of the C-terminal of p53-based peptides has previously been shown in the binding features to MDM2/X^[@CR37],[@CR38]^. Specifically, Pazgier et al*.* have shown that p53 peptides C-termini mutation from proline to other amino acids lead to an extended alpha-helical conformation in complex with MDM2^[@CR17]^. Overall, data in Fig. [1](#Fig1){ref-type="fig"} clearly displayed that; A) almost all peptides preserved their initial helicity structure during 200 ns simulation in the free and mainly in the complex forms. B) the helicity values of peptides in complex with MDM2 were more dominant than those of MDMX. C) The helicity probability had almost a similar pattern in residues 3 to 10 in the free state. However, the helicity score was higher in N-terminal residues rather than C-terminal residues in the complex states (mainly with MDM2). It should be noted that the first two and the last residues did not participate in helix structure, neither in the free nor the complex states. However, their effect on the helicity extension especially on C-terminal residues was undeniable. The impact of the last residue on the peptides' structures will be discussed in the following sections. Binding free energy analysis of peptides_MDM2/X using the umbrella sampling method {#Sec11} ---------------------------------------------------------------------------------- The binding affinity of the therapeutic agents to the target proteins is an important factor in predicting their effectiveness and value as promising drug candidates. It is therefore essential to precisely calculate the ΔG~binding~ of such interactions. The umbrella sampling method is a trustworthy approach for ΔG~binding~ computation especially for protein--protein interactions^[@CR33],[@CR39]^. Hence, in the current study, the interaction energies of all peptides-MDM2/X complexes were calculated using the umbrella sampling method. The method uses the PMF profile of a pulling simulation to estimate the binding energy between the two molecules. According to Fig. [2](#Fig2){ref-type="fig"} and Table [1](#Tab1){ref-type="table"}, the ΔG~binding~ of pDI mutant peptides to MDM2/X remarkably increased compared to the control peptides mainly p53 and pDI. The ΔG~binding~ results of pDIm1 and pDIm2 were also more suitable compared to the PMI peptide in complex with MDM2/X. The pDIQ peptide as an experimentally valid and potent inhibitory agent with previously measured IC~50~ of 8 and 110 nM^[@CR17]^ showed 31 and 16 kcal/mole ΔG~binding~ for MDM2 and MDMX, respectively (Table [1](#Tab1){ref-type="table"}). These results were highly consistent with the experimental results which were also plausible for p53, pDI, and PMI peptides. As shown in Table [1](#Tab1){ref-type="table"}, the order of peptides affinity to MDM2 is pDIQ \> PMI \> pDI \> p53 based on the experimental results. Interestingly, the computational calculation showed the same order. This was also applicable to MDMX complexes with a slight tolerance. As a result, the suitable binding energies of pDIm1, pDIm2, and pDIdm peptides with MDM2/X could acquire more validity. On the other hand, the pDIQm2 peptide had less binding energy to MDM2 and more affinity to MDMX compared to pDIQ itself. In addition, its binding energy was equal to PMI's. However, the pDIQm1 peptide did not have a suitable affinity to MDM2/X as the pDIQ peptide had. Given the fact that pDIQ is an optimized sequence, it seems more mutations such as E4K could not effectively improve the affinity to MDM2/X proteins.Figure 2Potential of mean force (PMF) curve compared to the mass center between studied peptides and (**A**) MDM2 and (**B**) MDMX using the "umbrella sampling" method. ξ is the center of mass between peptides and MDM2/X in angstrom. Residual interaction analyses of peptides_MDM2/X using gRINN method {#Sec12} ------------------------------------------------------------------- Each residue of a peptide can stimulate a dramatic structural and functional change in the peptide_protein complex. Additionally, every residual contact in protein--protein interactions could be either favorable or unfavorable regarding the binding and affinity points of view^[@CR40],[@CR41]^. Through MD simulations, the dynamicity of residue interactions is usually difficult to be delicately analyzed^[@CR36]^. However, a recently developed tool, named gRINN method, has been introduced to generate and extract residue by residue behavior of the biomolecules during MD simulation. Using gRINN tool, the interaction energies, the significant residues and the correlations of interaction energies are obtainable^[@CR34]^. The sum of total interaction energies in all residue-residue interactions was shown in each peptide_MDM2/X complexes (Fig. [3](#Fig3){ref-type="fig"}). Accordingly, as shown in Fig. [3](#Fig3){ref-type="fig"}A, the total interaction energies in pDIm1 (− 84.15 kcal/mol), pDIm2 (− 51.6 kcal/mol) and pDIdm (− 75.25 kcal/mol) decreased compared to that in pDI (− 95.89 kcal/mol) in complex with MDM2. However, the total interaction energy of pDIdm_MDMX was remarkably higher (− 87.5 kcal/mol) compared to the pDI_MDMX complex (− 32.65 kcal/mol) which made pDIdm peptide a potent dual inhibitor of MDM2/X. On the other hand, although the sum of interaction energies in pDIQ_MDM2 (− 85.7 kcal/mol) was comparable with pDI_MDM2 (− 95.89 kcal/mol), the level of interaction energies of pDIQ_MDMX (− 27.4 kcal/mol) was even lower than that of pDI_MDMX (− 32.65 kcal/mol). The interaction energy of pDIQm2 (− 23.4 kcal/mol) was lower in complex with MDM2 and was not different from that of MDMX (23.4 kcal/mol) compared to the pDIQ_MDM2/X interaction energies. In contrast, pDIQm1_MDM2 with total interaction energies of − 85.35 kcal/mol preserved the level of interaction energies of pDIQ_MDM2. However, its interaction energy quantity with MDMX was unremarkably higher (− 43.2 kcal/mole) compared to the pDIQ_MDMX complex. Overall, the top potent peptides against MDM2 in terms of the sum of residue-residue interaction energies were pDI (− 95.89 kcal/mol), PMI (− 92.6 kcal/mol), pDIQ (− 85.7 kcal/mol), pDIQm1 (− 85.35 kcal/mol) and pDIm1 (− 84.15 kcal/mol). However, regarding dual MDM2/X inhibition, the main potent peptide was pDIdm with − 75.25 kcal/mol and − 87.5 kcal/mol sums of interaction energies to MDM2 and MDMX, respectively.Figure 3Sum of interaction energies between studied peptides of p53, pDI, pDIm1, pDIm2, pDIdm, pDIQ, pDIQm1, pDIQm2, and PMI in complex with MDM2 and MDMX during MD simulation generated by gRINN tool. Black and Blue columns are MDM2 and MDMX complex forms, respectively. (**A**) illustrates the sum of all types of interaction energies, (**B**) indicates the total sum of electrostatic type of interaction energies, and (**C**) shows the total sum of van der Waals type of interaction energies in peptides_MDM2/X. To get a better view of the type of intermolecular forces playing roles in the row of p53-based peptides bound to MDM2/X, the sum of interaction energies were illustrated in the Fig. [3](#Fig3){ref-type="fig"}B, C. As shown in Fig. [3](#Fig3){ref-type="fig"}B, electrostatic forces played an important role in the experimentally modified p53 sequences such as pDI, pDIQ, and PMI. The sum of electrostatic interaction energies of the p53_MDM2 complex increased from − 10.955 kcal/mol to − 64.165 kcal/mol, − 53.56 kcal/mol and − 65.35 kcal/mol in pDI_MDM2, pDIQ_MDM2, and PMI_MDM2 complexes, respectively. Similarly, the mutant pDI and pDIQ peptides showed strong electrostatic interaction energies with MDM2. Exceptions were pDIm2 and pDIQm1 with the lower electrostatic type of interaction energies. On the other hand, as shown in Fig. [3](#Fig3){ref-type="fig"}C, van der Waals forces had a higher level in the modified p53_MDM2 complexes compared to the native p53_MDM2. However, , van der Waals energy values were not much different among the peptides in MDM2 complexes. As shown in Fig. [3](#Fig3){ref-type="fig"}B, C, regarding the type of interaction energies with MDMX, there was a similar level of electrostatic and van der Waals interaction energies (with a slight favor for electrostatic bonds). The only exception was pDIdm which showed the electrostatic energy of − 63.2 kcal/mol for MDMX. This was possibly the reason for the its best binding energy to MDMX (see the previous section) and the highest residue-residue sum of interaction energies. To show the important and specific residue-residue interactions in each peptide_MDM2/X structure, top interactions were illustrated in Fig. [4](#Fig4){ref-type="fig"}, Tables [2](#Tab2){ref-type="table"} and [3](#Tab3){ref-type="table"}. Accordingly, Phe3, as one of the three main residues of p53_MDM2/X binding site had the most top interaction energies with MDM2 residues of Gln72, Ile61, Val93, and Gly58. In the pDI_MDM2, the strong interactions were between Ser12 of pDI and Lys51 and Ser17 of MDM2. Ser12 of pDIm1, pDIm2, and pDIdm had also the highest interaction energies with Lys51, Ser17, and Gln18 residues of MDM2, respectively. The Ser12 interaction energy remained at the highest level in the pDIQ and pDIQm1 in complex with MDM2. The pDIQm2 peptide with no Ser12 showed the higher interaction energy in phe3 residue which might have been an important reason for total low binding energy in the pDIQm2_MDM2 complex (see the previous section). In the PMI_MDM2 structure, Ser11 indicated a strong interaction with Lys51 of MDM2. Regarding MDMX complexes (Table [3](#Tab3){ref-type="table"}), phe3 of p53 had a low level of interaction energy compared to that of p53_MDM2 complex. Modifying the p53 sequence, Tyr 6 and Trp7 of pDI with His72 and Met53 of MDMX showed the highest interaction energies in the pDI_MDMX complex. This was also the case in the pDIm1 and pDIm2 in complex with MDMX. However, in the pDIdm, the highest interaction energy was Ser12 with Lys50 of MDMX. On the other hand, Trp7 played an important role in the pDIQ and PMI interactions with MDMX. However, Glu4 and Leu11 were the critical residues of pDIQm1_MDMX and pDIQm2_MDMX complexes, respectively. The results were also depicted in Fig. [4](#Fig4){ref-type="fig"} highlighting the change of critical interacted residues during the modification of the peptide sequences from p53 to pDIdm. Given the fact that pDIdm have shown better results compared to other peptides, it can be implied that Ser12 plays a most critical role in dual inhibition of MDM2/X proteins.Figure 4200 ns snapshots of MD simulations for p53, pDI, and pDIdm in complex with MDM2 and MDMX. It illustrates the critical residues in the interactions of the peptides_MDM2/X complexes. Data are generated by gRINN tool and images have been developed using Chimera software. Table 2Top 10 residues-residue interactions in peptides_MDM2 complexes with the highest interaction energy extracted from gRINN tool. The left residue is relevant to the peptide sequence and the right one to the binding site of MDM2. If applicable, the FoldX mutations will be shown in **bold** letters.pDI-MDM2Ave total E (kcal/mol)pDIm1-MDM2Ave total E (kcal/mol)pDIm2-MDM2Ave total E (kcal/mol)pDIdm-MDM2Ave total E (kcal/mol)SER12-LYS51− 21.35SER12-LYS51− 19.03SER12-SER17− 9.43SER12-GLN18− 8.22SER12-SER17− 13.63SER12-SER17− 15.21PHE3-GLN72− 6.29PHE3-GLN72− 5.51THR11-LYS51− 9.02PHE3-GLN72− 5.59PHE3-TYR67− 2.98LEU10-ILE19− 5.07PHE3-GLN72− 5.99TRP7-LEU54− 3.99TYR6-HIS73− 2.90TRP7-LEU54− 4.77TRP7-PHE55− 3.89THR11-LYS51− 3.85TYR6-GLN72− 2.73GLN9-LYS94− 3.02TRP7-MET62− 3.13PHE3-TYR67− 2.70THR2-GLN72− 2.67GLN9-GLN18− 2.98TYR6-HIS73− 2.94TYR6-HIS73− 2.58TRP7-LEU54− 2.52**TRP4-MET62**− 2.77GLN9-SER17− 2.94THR2-GLN72− 2.40LEU10-SER17− 1.89PHE3-TYR67− 2.67LEU10-SER17− 2.89TYR6-GLN72− 2.26PHE3-MET62− 1.72SER12-PRO20− 2.60GLN9-HIS96− 2.73GLN9-SER17− 2.16**MET11-LEU54**− 1.61THR2-GLN72− 2.39pDIQ-MDM2Ave total E (kcal/mol)pDIQm1-MDM2Ave total E (kcal/mol)pDIQm2-MDM2Ave total E (kcal/mol)SER12-ALYS51− 29.41SER12-ALYS51− 25.49PHE3-GLN72− 6.61TRP6-AHIS96− 6.75SER12-AHIS96− 10.15PHE3-MET62− 2.33TRP7-LEU54− 6.19LEU11-LYS51− 6.05THR2-GLN72− 2.33PHE3-GLN72− 5.21PHE3-GLN72− 5.93TRP6-HIS73− 1.96LEU11-LYS51− 4.85TRP7-PHE55− 5.29TRP6-GLN72− 1.62THR2-GLN72− 2.37GLN9-HIS96− 3.62LEU11-LEU54− 1.45LEU10-LYS51− 2.34TRP7-LEU54− 3.06PHE3-ILE61− 1.28LEU10-LEU54− 1.95LEU10-HIS96− 2.48LEU11-PHE55− 1.25PHE3-TYR67− 1.87LEU10-LEU54− 2.36PHE3-VAL93− 1.04TRP7-PHE55− 1.69PHE3-MET62− 2.21SER12-LEU54− 0.77P53-MDM2Ave total E (kcal/mol)PMI-MDM2Ave total E (kcal/mol)PHE3-GLN72− 6.87SER11-ALYS51− 25.93PRO11-LEU54− 2.96SER11-AHIS96− 8.85THR2-GLN72− 2.78PHE3-AGLN72− 5.87LEU10-LEU54− 2.63LEU10-ALYS51− 5.41PRO11-PHE55− 2.23TRP7-APHE55− 4.09SER4-GLN72− 1.53TRP7-AMET62− 3.81PHE3-ILE61− 1.18TRP7-ALYS51− 3.70PHE3-VAL93− 1.02LEU10-ALEU54− 2.31PHE3-GLY58− 1.01TRP7-AGLY58− 2.28LEU10-VAL93− 1.00SER11-ALEU54− 2.27 Table 3Top 10 residues-residue interactions in peptides_MDMX complexes with the highest interaction energy extracted from gRINN tool. The left residue is related to the peptide sequence and the right one to the binding site of MDMX. If applicable, the FoldX mutations will be shown in **bold** letters.pDI-MDMXAve total E (kcal/mol)pDIm1-MDMXAve total E (kcal/mol)pDIm2-MDMXAve total E (kcal/mol)pDIdm-MDMXAve total E (kcal/mol)TYR6-HIS72− 4.41TYR6-HIS72− 5.40TRP7-HIS54− 7.67SER12-LYS50− 26.58TRP7-MET53− 3.12TRP7-MET53− 4.89PHE3-GLN71− 5.03ALA8-LYS50− 9.51PHE3-GLN71− 2.96PHE3-GLN71− 3.43LEU10-MET53− 2.98TRP7-MET53− 5.96TYR6-GLN71− 2.56PHE3-MET61− 2.69TYR6-HIS72− 2.83GLN9-LYS50− 4.78PHE3-TYR66− 2.55PHE3-TYR66− 2.43PHE3-MET61− 2.80**MET11-LYS50**− 4.13TRP7-ILE60− 1.84TYR6-GLN71− 2.33THR2-GLN71− 2.37LEU1-GLN71− 3.93TRP7-MET61− 1.61TRP7-LEU56− 1.77TRP7-MET61− 2.12SER12-MET53− 3.55PHE3-ILE60− 1.54PHE3-GLY57− 1.75TYR6-GLN71− 1.82PHE3-MET61− 2.92PHE3-MET61− 1.52THR2-GLN71− 1.70TRP7-MET53− 1.24**MET11-MET53**− 2.48THR2-HIS72− 1.44LEU10-PRO95− 1.61TRP7-GLY57− 1.09PHE3-TYR66− 2.10pDIQ-MDMXAve total E (kcal/mol)pDIQm1-MDMXAve total E (kcal/mol)pDIQm2-MDMXAve total E (kcal/mol)TRP7-MET53− 6.14GLU4-HIS54− 9.18LEU11-MET53− 2.25PHE3-GLN71− 3.21TRP7-MET53− 4.77TRP7-MET53− 2.12PHE3-MET61− 2.24TRP6-GLN71− 3.41PHE3-GLN71− 2.07TRP7-VAL92− 2.13PHE3-GLN71− 3.26TRP7-HIS54− 1.97LEU11-MET53− 2.08PHE3-TYR66− 2.40PHE3-MET61− 1.92PHE3-GLY57− 1.81LEU11-MET53− 2.19TRP7-VAL92− 1.62TRP7-LEU56− 1.32TRP7-GLN71− 1.63TRP7-ILE60− 1.54TRP7-ILE60− 1.23TRP7-ILE60− 1.52LEU11-HIS54− 0.96PHE3-ILE60− 1.21TRP7-HIS54− 1.38PHE3-TYR66− 0.86TRP7-GLY57− 0.86GLU4-GLY57− 1.36THR2-GLN71− 0.85P53-MDMXAve total E (kcal/mol)PMI-MDMXAve total E (kcal/mol)PHE3-GLN71− 3.19TRP7-HIS54− 8.41PHE3-MET61− 2.59PRO12-HIS54− 6.17THR2-GLN71− 1.92PHE3-GLN71− 6.04PHE3-ILE60− 1.90SER11-HIS54− 4.81LEU10-MET53− 1.16LEU10-MET53− 3.16LEU10-VAL92− 0.95TYR6-HIS72− 2.56THR2-MET61− 0.89PHE3-MET61− 2.30PHE3-GLY57− 0.89TRP7-MET61− 2.14PHE3-HIS72− 0.81TYR6-GLN71− 1.99PHE3-VAL92− 0.66SER2-GLN71− 1.89 The correlation of significant residues based on their location and interaction energies generated from the gRINN tool was shown in Figs. [5](#Fig5){ref-type="fig"} and [6](#Fig6){ref-type="fig"} for MDM2 and MDMX complexes, respectively. The dots in each graph have a color range of green to brown indicating the lower to higher residue-residue interaction energy, respectively. As shown in Fig. [5](#Fig5){ref-type="fig"}, the accumulation of interacted residues was higher in peptide_MDM2 regions. The intensity of brown dots in pDIm1, pDIm2, and pDIdm was remarkably higher compared to pDI bound to MDM2. On the other hand, in the pDIQm1_MDM2, while the dots accumulation was similar to pDIQ_MDM2, the level of green intensity was higher as well. However, the pDIQm2_MDM2 showed a decreased number of dots as well as higher brown intensity compared to the pDIQ_MDM2. Figure [6](#Fig6){ref-type="fig"} shows the residue-residue correlation in peptides_MDMX interactions. The pDIdm_MDMX graph indicated a highest mass of residue-residue dots with higher interaction energies which is in accordance with previous results.Figure 5Interaction energy matrix that displays the average interaction energies between residue pairs of two chains in the studied peptides of p53, pDI, pDIm1, pDIm2, pDIdm, pDIQ, pDIQm1, pDIQm2 and PMI in complex with MDM2. According to the color range and the number of dots, the number of interactions, and the strongest bindings are shown. Figure 6Interaction Energy Matrix that displays the average interaction energies between residue pairs of two chains in the studied peptides of p53, pDI, pDIm1, pDIm2, pDIdm, pDIQ, pDIQm1, pDIQm2 and PMI in complex with MDMX. According to the color range and the number of dots, the number of interactions, and the strongest bindings are shown. Historically, several anti-cancer peptides based on the p53_MDM2/X binding site have been developed mostly by using the phage display method. Some important ones include 12/1^[@CR16]^, PMI^[@CR17]^ and pDI^[@CR18]^. The common limitation in all of the lead peptides was the lack of suitable affinity for MDMX. pDI, as one of the promising lead anti-MDM2/X peptides, was numerously applied for kinetics and dynamics improvements. In one of the first studies, in order to increase the affinity of the pDI, a structure--activity study was utilized by Phan et al.^[@CR23]^. In this study, the MDM2/X IC~50~s of different mutant sequences of pDI were obtained and compared with p53, PMI and pDI. Ultimately, the pDIQ with the sequence of ET**F**EHW**W**SQ**L**LS was introduced as the most efficient dual inhibitory peptide with IC~50~ of 8 and 110 nM for MDM2 and MDMX, respectively (Table [1](#Tab1){ref-type="table"}). Moreover, one of the clinically promising peptides, ALRN-6924, was originally optimized based on pDI sequence. On the other hand, the suitable amount of experimental data regarding MDM2/X inhibitors over the last two decades has enabled scientists to feasibly use such data for the validation of theoretical methods. In one of the last attempts, Diller et al*.* described "CMDInventus" as a peptide design platform and a potent method to design inhibitory stapled peptides for MDM2/X^[@CR42]^. Stapled peptides are usually designed to overcome the limitations of native peptides such as short plasma half-life. However, one of the promising approaches to design an efficient stapled peptide is first to achieve a native peptide sequence (a lead peptide) as a robust inhibitor of the target protein^[@CR43]^. Afterward, synthetic modifications on the candidate peptides are likely to result in a more stable peptide-drug candidate. In the current study, it was attempted to study the p53-based lead peptides structurally in a way to guide the scientists to obtain more effective MDM2/X inhibitory peptides. In this path, the single or double mutations screening of known peptides was carried out. Mutant pDI and pDIQ, as well as, control peptides including p53, pDI, pDIQ, and PMI, went through MD simulations and corresponding analyses. Subsequently, the umbrella sampling method confirmed the experimental effectiveness pattern of the control peptides (Table [1](#Tab1){ref-type="table"}). As a result, the method was confirmed as a suitable predicting approach for new developed peptides. In the literature, the PMI is the best native dual lead peptide with *Kd* of 3.3 and 8.9 nM for MDM2 and MDMX, respectively^[@CR17]^. According to Table [1](#Tab1){ref-type="table"}, ΔG~binding~ for PMI was calculated as 25 and 19 kcal/mol for MDM2 and MDMX, respectively, which is in high agreement with the experimental data. This is also true about pDI and pDIQ. Therefore, the theoretically designed mutant peptides, especially, pDIm1, PDIm2, and pDIdm, indicated promising ΔG~binding~ in dual inhibition of MDM2/X proteins. Another key factor in suitable inhibition of MDM2/X proteins is the conformation stability of the peptides. As shown in Fig. [4](#Fig4){ref-type="fig"}, 200 ns snapshots indicated no helical conservation of p53 peptide when they are bound to MDM2/X compared to other peptides. This is strongly relevant to p53 C-terminal residues specifically Pro11. On the other hand, in the modified peptides, mutations such as Pro11Met enhanced the conformation stability in both MDM2/X complexes. Other follow-up analyses provided valid atomic-scale views of the peptide_MDM2/X interactions and their effectiveness. Accordingly, the results highlighted the importance of electrostatic forces during the interaction of peptides with MDM2/X. Besides, our data showed that Ser12 of modified peptides plays a key role in the effective MDM2/X dual inhibition. Therefore, while the last residue does not protect the helicity directly, it changes the interaction pattern in a way that the stability and affinity of the peptides increase. Moreover, Tyr6 and Trp7 are critical residues in the interaction of peptides with MDMX. These data are in agreement with those of the study that developed ATSP-7041 peptide. It shows additional Tyr6 interaction with MDMX binding pocket is an assistant to increase peptides_MDMX affinity^[@CR22]^. Overall, the data confirmed the significance of the theoretical structure-based peptide development method for robust and rational dual anti-MDM2/X peptide design. Conclusion {#Sec13} ========== MDM2 and MDMX are cooperative oncoproteins blocking the action of wild type p53 in several types of cancerous cells. Therefore, several MDM2/X inhibitory compounds have been developed over the last two decades to restore the normal proliferation of malignant cells. Some well-established lead peptides such as pDI and PMI have shown promising results in this regard. However, the peptides' common limitation is the lack of dual inhibition of both MDM2/X proteins. As a result, in this study, it was attempted to develop a theoretical structure-based method fitting to the previous experimental data. This method was capable of analyzing the p53-based peptide_MDM2/X complexes in order to develop new native peptides with a higher affinity. In this path, previously tested peptides, including p53_MDM2/X binding site, pDI, pDIQ, and PMI as well as computationally screened mutants, named pDIm1, pDIm2, pDIQm1, and pDIQm2 in complex with MDM2/X were investigated using MD simulation. Detailed secondary structure analysis indicated that the peptides' N-terminal residues contribute significantly to their helicity conservation when bound to MDM2 complexes. However, in the MDMX structures, the C-terminal residues of the peptides were more stable in the alpha-helix conformation. This contradiction makes it difficult to reach a potent, dual and stable peptide against MDM2/X proteins. Nevertheless, mutant pDI peptides, theoretically designed in this study, have shown to overcome this limitation. In addition, the ΔG~binding~ results obtained from the umbrella sampling method were in agreement with the potency of experimentally confirmed MDM2/X inhibitory peptides includingp53, pDI, pDIQ, and PMI peptides. Accordingly, the ΔG~binding~ of other screened mutant peptides, especially pDIdm, was reasonably promising. Moreover, having extracted residue by residue interaction energies, the critical residues of peptides_MDM2/X interactions were determined. Phe3 of p53 binding site was replaced with Ser12 in the modified peptides as critical interacting residue with MDM2. Lys51, Ser17, and Gln18 were the most critical residues of the interacting site of MDM2. Furthermore, since the Ser12 in pDIdm was conserved its interaction with MDMX, this case shows a high affinity to MDMX as well as MDM2. In conclusion, the result of this study highlighted the importance of atomic-scale and structure-based theoretical analyses to design potent peptides more rationally and efficiently. In conclusion, suggested rules as prediction structural markers of a promising anti-cancer p53-based peptide are including a) the presence of suitable C-terminal residues of p53-based peptides especially Glu9, Leu10, and Met11, b) the use of serine as the last residue, c) the appropriate ΔG~binding~ to MDM2/X, generated by the umbrella sampling method. Supplementary information ========================= {#Sec14} Supplementary file1 (PDF 650 kb) **Publisher\'s note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= is available for this paper at 10.1038/s41598-020-67510-8. The authors would like to acknowledge the Cancer Control Research Center, Cancer Control Foundation, Iran University of Medical Sciences, Tehran, Iran, as project No: CCF-97035 as well as the University of Tabriz for the financial supports. A.B. and E.M.A. designed the research project. N.R. performed the computational experiments. A.B. and N.R. and E.M.A. analyzed the data. A.B and N.R and E.M.A. prepared the manuscript. All authors have read, checked and approved the manuscript. The authors declare no competing interests.

Background {#Sec1} ========== Early supported discharge (ESD) with continued rehabilitation in the home has been shown to be beneficial among patients with mild to moderate stroke. The ESD- model for rehabilitation was introduced in the late 1990s and includes an interdisciplinary team with appropriate recourses that coordinates the discharge and plan, supervise and continue the rehabilitation in the home environment \[[@CR1]\]. This form of rehabilitation accelerates the discharge from hospital, reduce long term dependency and admission to institutional care \[[@CR1]--[@CR4]\]. However, the criticism has been raised that most of the randomized controlled trials on ESD services were published more than 10 years ago \[[@CR5], [@CR6]\]. Today, the majority of patients are being discharged home early after stroke, due to access of hyper-acute therapies, and implement early interventions for carotid stenosis. There is a need to evaluate efficacy and safety of ESD in current stroke care settings and to adapt ESD service to local conditions for appropriate implementation \[[@CR1], [@CR5], [@CR6]\]. In Sweden, the majority (91% in 2014) of patients with stroke are cared for at a stroke unit (<http://www.riks-stroke.org>), but there has been no major expansion of ESD. Despite recommendations in the National Guidelines for stroke care \[[@CR7]\], the proportion of patients with stroke that receive ESD after stroke unit care has varied, dramatically across Sweden. Västerbotten County (Umeå Stroke Center and Skellefteå hospital) is one of the counties with high a proportion of ESDs (<http://www.riks-stroke.org>, \[[@CR8]\]). We previously described the method, content, and outcome of ESDs according to the Umeå Stroke Center model in Västerbotten County \[[@CR9]\]. This model showed that it was possible to adopt and implement ESD for patients with stroke in Umeå. The model included important key elements for an effective ESD service \[[@CR6], [@CR8]\], such as a multidisciplinary team with experience in stroke rehabilitation, appropriate resources, periodic team meetings, and continuous evaluations of outcome with standardized measurements. Our previous results showed that ESD services reduced patient dependence in activities in daily living (ADL) and increased patient mobility, without increasing the risk of accidental falls or other injuries \[[@CR9]\]. The patients were very satisfied with the ESD-service. However, that observational implementation study did not include a control group. The present study aimed to evaluate patient-reported outcomes measures (PROMs) among patients with stroke that received modern stroke unit care, and compare PROMs between those that received or did not receive ESD. The ESD was delivered according to a previously described model \[[@CR9]\]. We hypothesized that patients that received ESD would exhibit improved PROMs regarding satisfaction with rehabilitation (primary outcome), activity in daily living (ADL), tiredness/fatigue, pain, dysthymia/depression, general health status and information about stroke (secondary outcomes) compared to controls. Methods {#Sec2} ======= For this case control, observational study, we retrieved data from the Swedish Stroke Registry, Riksstroke \[[@CR10]\], and from the Longitudinal Integration Database for Health Insurance and Labor Market Studies (LISA). Information from the Swedish Stroke Registry was linked to the LISA database through personal identification numbers. This study was approved by the Regional Ethics Review Board at Umeå University (Dnr 2012--179-32 M, 2014--273-32 M). Register {#Sec3} -------- Currently, all 72 hospitals that treat patients with acute stroke participate in the Swedish Stroke Registry, Riksstroke \[[@CR10]\], which started in 1994. The primary aim of Riksstroke is to monitor and support improvements in the quality and implementation of new methods in stroke care in Sweden. The registry includes patients with ischemic and hemorrhagic stroke and data on first-ever and recurrent strokes. The acute phase questionnaire of Riksstroke contains basic patient characteristics (age, sex, living conditions, history of previous stroke, and comorbidities), diagnosis, level of consciousness on arrival, pharmaceutical treatments, complications, and the sequence of care (type of stroke care, organization, and department). In Riksstroke the hospitals that care for patients with acute stroke have been divided into three categories: university hospitals (9 pcs), specialized nonuniversity hospitals (23 pcs) and community hospitals (40 pcs) \[[@CR11]\]. Riksstroke also includes 3-month and 12-month follow up questionnaire that describe patient-reported outcomes and rehabilitation after stroke. The 3-month and 12-month questionnaire is administrated by the hospitals and filled in by the patients. The LISA database at Statistics Sweden includes information on all Swedish citizens, starting at 16 years of age. In particular, it includes socioeconomic factors, like disposable income, education, and country of birth. Participants and setting {#Sec4} ------------------------ All patients registered in the Riksstroke registry with a first-ever diagnosis of acute stroke between 1 January, 2010 and 31 December 2013 were included in the present study, when they fulfilled the following criteria: diagnosis of ischemic or hemorrhagic stroke; mild to moderate stroke severity at admission (measured as level of consciousness on a scale of 1--3, according to the Reactive Level Scale, RLS85) \[[@CR12]\]; living at home; and independency in ADL at stroke onset. Patients that met the inclusion criteria were divided into an ESD intervention group and a control group (Fig. [1](#Fig1){ref-type="fig"}).Fig. 1Flow chart of the study inclusion procedure. ESD: early supported discharge ESD intervention group {#Sec5} ---------------------- Västerbotten County have three hospitals with different primary catchment areas that cared patients in acute and sub-acute phases of stroke. Two of these hospitals, the Umeå Stroke Center (university hospital) and Skellefteå hospital (community hospital) had similar organizations regarding stroke care~~.~~ Both hospitals had stroke unit care followed by ESD according to a previously described model \[[@CR9]\]. These hospitals provided ESD to a comparatively large proportion of patients (41.2% at Umeå Stroke Center and 57.5% at Skellefteå hospital) compared to other hospitals in Sweden. The intervention group (ESD group) consisted of consecutive patients with stroke that received modern stroke unit care, followed by the ESD-service. During the study period, ESD was delivered to 41.2% of all patients with stroke admitted to the Umeå Stroke Center and 57.5% admitted to Skellefteå. Patients with severe stroke and those who died during hospitalization were not included in the ESD intervention group. Control group {#Sec6} ------------- The control group consisted of patients treated at stroke units in hospitals that cared for patients with acute stroke. At these hospitals, a low (\< 5% of all stroke patients) proportion of patients were given ESD. Therefore, the control group, which fulfilled the same inclusion criteria as the intervention group also included some patients that received ESD (\< 5%). Patients with severe stroke and those who died during hospitalization were not included in the control group. Hospitals in Sweden that only partially implemented ESD (5--20% of stroke patients) and those who registered other models of home rehabilitation were excluded. Variables {#Sec7} --------- Patient characteristics for both groups included sex, age, stroke subtype, treatment with thrombolysis, domestic companions, mobility, hypertensive treatment, diabetes, atrial fibrillation, and smoking before stroke onset. The level of consciousness upon admission and the length of hospital stay were also considered baseline characteristics of the two groups. These variables were retrieved from the Riksstroke registry. Information on education and country of birth was retrieved from the LISA database. The level of education was classified as primary school, secondary school, or university level. Country of birth was categorized as Sweden, Nordic countries (Sweden excluded), Europe (Nordic countries excluded), or other countries. The outcome variables for this study were PROM results from the 3-months follow up recorded in Riksstroke. The primary outcome was satisfaction with the rehabilitation after discharge. Secondary outcome variables were; satisfaction with information provided about stroke, tiredness/fatigue, pain, dysthymia/depression, general health status and ADL dependence (mobility, toileting, and dressing). Data from the PROM values recorded in Riksstroke at the 3-months follow up have been validated against established measurements with the finding of accurate reliability (<http://www.riks-stroke.org>). Some questions had multiple choice responses, and for our analysis, the responses were dichotomized. For example, patients were asked the questions: "How satisfied or dissatisfied are you with the rehabilitation or training you received after your stay in the hospital?" and "How satisfied or dissatisfied are you with the stroke information provided?"; and they responded with one of the following options: very satisfied, satisfied, dissatisfied, very dissatisfied, no need of rehabilitation, needed, but did not receive rehabilitation, or I do not know. And dependent categories were taken as reference (code = ref) for variables measuring a favourable outcome (satisfaction with rehabilitation, information about stroke, general health and Adl function) When assessing outcome variables that means an unfavourable result (tiredness/fatigue, pain, depression) we used a 'positive' reference (code = ref). Similarly, the questions: "Do you feel tired/fatigue?", "Do you have any pain?", and "Do you feel depressed?" had response options; never, almost never, sometimes, often, constantly, or do not know, and they were coded as often or seldom (ref). The "often" code was assigned to: sometimes, often, constantly, and missing. The question "How would you assess your general health?" had response options: very good, somewhat good, somewhat poor, very poor, and they were coded good or poor(ref). The "poor" code was assigned to somewhat poor, very poor, and do not know. Finally, responses to the ADL questions were coded as independent and dependent (ref), as follows. The question: "How is your mobility now?" had response options: "I can get around by myself indoors and outdoors" (independent), "I can get around by myself indoors, but not outdoors" (independent), "I need assistance when I want to move around" (dependent), or "I do not know" (dependent). The question: "Do you need assistance when visiting the toilet?" had response options: "I can manage visiting the toilet by myself" (independent) and "I need assistance when visiting the toilet" (dependent). The question "Do you need assistance getting dressed and undressed?" had response option: "I can manage to get dressed and undressed by myself", (independent) and "I need assistance getting dressed and undressed" (dependent). Statistical analysis {#Sec8} -------------------- Baseline characteristics for the ESD and control groups are expressed as frequencies and proportions, for categorical data, and as the mean and standard deviation (SD), for continuous variables. Baseline values were compared between groups with the independent t-test (for continuous variables) and the chi-square test (for categorical variables). The association between PROMs and ESD were analysed in separate multivariable logistic regression models. These models assessed the probability of satisfaction with rehabilitation after discharge, satisfaction with the information provided about stroke, tiredness/fatigue, pain, dysthymia/depression, general health status, and ADL dependency (mobility, toilet hygiene, and dressing) in the ESD group compared to the control group. In addition to the group effect (ESD or control), each model included the independent factors that were significant in the baseline comparison: age (continuous), thrombolysis, smoking, atrial fibrillation, country of birth, and education. The results are presented as the odds ratio (OR) with a 95% confidence interval (95% CI). The confidence intervals and *P*-values were not adjusted for multiple testing. Sensitivity analyses were performed where the missing data (missing or unknown) were included in the reference category. Subgroup analyses were performed to compare data for Umeå Stroke Center with other university hospitals and Skellefteå hospital with community hospitals. In the subgroup analysis, missing data were included in the reference category. The analyses were performed with IBM SPSS Statistics 21 statistical software. Results {#Sec9} ======= Between 1 January 2010 and 31 December 2013, 99,833 patients whereof 74,689 with a first-ever diagnosis of acute stroke were registered in the Riksstroke. Of these patients, 30,232 were included in this study (Fig. [1](#Fig1){ref-type="fig"}). During these years, the coverage rate in The Riksstroke varies from 88 to 91%. The ESD group comprised of 1495 participants that received ESD treatment after stroke unit care at the Umeå Stroke Center (university hospital) and at Skellefteå hospital (community hospital). The control group comprised 28,737 participants treated at 3 university hospitals, 13 other large hospitals, and 39 community hospitals. The control group were treated at stroke units where a very low proportion (605 participants, 2%) received ESD. Participants in the ESD group were slightly younger, had a higher education level and were more often born in Sweden than participants in the control group. The ESD group had lower frequencies of smoking and atrial fibrillation and a higher frequency of thrombolysis than the control group (Table [1](#Tab1){ref-type="table"}).Table 1Baseline characteristics of study participants (*n* = 30,232)VariableESD groupControl group*p* value(*n* = 1495)(*n* = 28,737)Sex, n (%)0.212 Males829 (55.5)15,461 (53.8) Females666 (44.5)13,276 (46.2) Age, mean (SD)73 (12.8)74 (12.4)0.001Stroke subtype, n (%)0.244 Ischemic Stroke1314 (87.9)25,646 (89.2) Intracerebral hemorrhage159 (10.6)2741 (9.5) Undetermined22 (1.5)350 (1.2) Thrombolysis, n (%)212 (14.2)2252 (7.8)0.001 Missing1 (0.1)142(0.5) Living alone, n (%)605 (40.5)12,428 (43.2)0.091 Missing data4 (0.3)97 (0.3)Mobility, n (%)0.249 Independent indoors- and outdoors1468 (98.2)28,088 (97.7) Independent indoors27 (1.8)649 (2.3) Hypertension, n (%)817 (54.6)16,430 (57.2)0.076 Missing data3 (0.2)107 (0.4) Diabetes, n (%)250 (17.4)5254 (18.3)0.106 Missing data0 (0.0)40 (0.1) Atrial fibrillation, n (%)282 (18.9)6847 (23.8)0.001 Missing data0 (0.0)102(0.4) Smoking, n (%)157 (10.5)4331(15.1)0.001 Missing data36(2.4)2001 (7.0)Level of consciousness on admission, n (%)0.123 RLS 11391 (93.0)26,419 (91.9) RLS 2--3104 (7.0)2318 (8.1) Length of hospital stay (SU), median (Q1-Q2)5 (3--10)6 (3--13)0.920 Missing data50 (3.3)1220 (4.2)Education, n (%)0.001 Primary School527(35.3)11,902 (41.4) Secondary School590 (39.5)9930 (34.6) University259 (17.3)4428(15.4) Missing data119 (7.9)2470 (8.6)Country of birth, n (%)0.001 Sweden1431 (95.7)25,392 (88.4) Nordic countries ^a^41 (2.7)1436 (5.0) Europe ^b^11 (0.8)1234 (4.3) Other countries9 (0.6)466 (1.6) Missing data3 (0.2)209 (0.7)SU - Stroke unit^a^Except Sweden^b^Except Nordic Countries Three months after a stroke, 28,986 participants remained alive: 1432 (96%) were in the ESD group and 27,554 (96%) were in the control group. Due to differences between the ESD and control participants in the ESD groups at baseline the multivariable logistic regression models of PROMs were adjusted for age, thrombolysis, smoking, atrial fibrillation, country of birth, and education. The analysis showed that the ESD compared with the control group were more satisfied with rehabilitation after discharge (*p* \< 0.004, OR 1.78, 95% CI: 1.17--2.49), showed more independence in mobility (*p* \< 0.001, OR 1.50, 95% CI: 1.17--1.92), toileting (*p* \< 0.016, OR 1.30, 95% CI: 1.05--1.61), and dressing (*p* \< 0.03, OR 1.23, 95% CI: 1.02--1.48), and felt less dysthymia/depression (*p* \< 0.001, OR 0.68, 95% CI: 0.55--0.84) (Table [2](#Tab2){ref-type="table"}). There were no significant differences between the groups in the information received about stroke, tiredness/fatigue, pain or general health status (Table [2](#Tab2){ref-type="table"}).Table 2Multiple logistic regression of patient reported outcome variables 3 months after stroke(*n* = 28,986) Odds ratio (OR) with 95% confidence intervalVariableESD groupControl group*p* value0R95% CI(*n* = 1432)(*n* = 27,554)LowerUpperSatisfaction with rehabilitation^a^, n (%) Satisfied507 (35.4)9182 (33.3)0.0011.781.172.49 Dissatisfied47(3.3)1405 (5.1)ref No need449 (31.4)9231 (33.5) Not received44 (3.1)1501 (5.4) Missing and do not know385 (26.9)6235 (22.6)Information provided about stroke, n (%) Satisfied927 (64.7)16,820 (61.0)0.5331.080.851.37 Dissatisfied82 (5.7)1700 (6.2)ref No need87 (6.1)3256 (11.8) Missing and do not know336 (23.5)5778 (21.0)Tiredness/fatigue, n (%) Seldom780 (54.5)15,440 (56.0)ref Often410 (28.6)8595 (31.2)0.7480.980.861.11 Missing242 (16.9)3519 (12.8)Pain, n (%) Seldom956 (66.8)19,117 (69.4)ref Often217 (15.2)4745 (17.2)0.5230.950.8141.11 Missing259 (18.1)3692 (13.4)Depression, n (%) Seldom1083 (75.6)20,819 (75.6)ref Often109 (7.6)3118 (13.3)0.0010.680.550.84 Missing240 (16.8)3617 (13.1)General health, n (%) Good988 (69.0)19,163 (69.5)0.0911.150.981.35 Poor200 (14.0)4779 (17.3)ref Missing244 (17.0)3612 (13.1)Adl-independence, n (%)Mobility Independent1146 (80.0)21,722 (78.8)0.0011.501.171.92 Dependent83 (5.8)2564 (9.3)ref Missing203 (14.2)3268 (11.9)Toileting Independent1117 (78.0)21,236 (77.1)0.0161.301.051.61 Dependent112 (7.8)3247 (11.8)ref Missing203 (14.2)3247 (11.8)Dressing Independent1075 (75.1)20,394(74.0)0.0301.231.021.48 Dependent155 (10.8)3895 (14.1)ref Missing202 (14.1)3265 (11.8)^a^ After Discharge The proportions of missing responses were higher in the ESD group for all variables (14.1--26.9% missing responses) than in the control group (11.8--22.6% missing responses). In a sensitivity analysis, where missing/unknown/no need responses were included in the reference category, the ESD group remained more satisfied with rehabilitation after discharge (*p* \< 0.001, OR 1.26, 95% CI: 1.10--1.43) and felt less depressed (*p* \< 0.001, OR 0.68, 95% CI: 0.55--0.84) than controls, but the groups were not significantly different in the other variables (Table [3](#Tab3){ref-type="table"}).Table 3Patient-reported outcome variables, 3 months after stroke (n = 28,986) (sensitivity analysis)VariableESD groupControl group*p* value0R95% CI(*n* = 1432)(*n* = 27,554)LowerUpperSatisfaction with rehabilitation^a^, n (%) Satisfied507 (35.4)9182 (33.0)0.0011.261.101.43 Dissatisfied^b^925 (64.6)18,372 (67.0)refInformation provided about stroke, n (%) Satisfied927 (64.7)16,820 (61.0)0.4441.050.931.17 Dissatisfied^b^505 (35.3)10,734 (39.0)refTiredness/fatigue, n (%) Seldom780 (54.5)15,440 (56.0)ref Often and missing652 (45.5)12,114 (44.0)0.7500.980.661.11Pain, n (%) Seldom956 (66.8)19,117 (69.4)ref Often and missing476 (33.2)8437 (30.6)0.5230.950.811.11Depression, n (%) Seldom1083 (75.6)20,819 (75.6)ref Often and missing349 (24.4)6735 (24.4)0.0010.680.550.84General health, n (%) Good988 (69.0)19,163 (69.5)ref Poor and missing444 (31.0)8391 (30.5)0.0900.870.741.02Adl-independence, n (%)Mobility Independent1109 (93.0)21,722 (78.8)0.0010.780.680.90 Dependent and missing323 (22.5)5832 (21.2)refToileting Independent1132 (78.9)21,236 (77.1)0.0010.780.680.88 Dependent and missing352 (24.5)6318 (22.9)refDressing Independent1077 (75.2)20,394(74.0)0.0010.810.710.92 Dependent and missing355 (24.8)7169 (26.0)ref^a^After Discharge^b^Dissatisfied includes very dissatisfied, dissatisfied, in need, but did not received rehabilitation, I do not know and missing In the subgroup analyses of data form the Umeå Stroke Center (ESD subgroup 1) and other university hospitals (control subgroup 1), we found that the ESD group 1 was more satisfied with the rehabilitation (*p* \< 0.0001, OR2.27, 95% CI: 1.44--3.59) and stroke information provided (*p* \< 0,001, OR 1.93, 95% CI: 1.32--2.37) and experienced less depression (*p* \< 0.001, OR 0.55, 95% CI: 0.41--0.75), and were better in Adl function than the control subgroup 1 (Table [4](#Tab4){ref-type="table"}). In the subgroup analysis of data from Skellefteå hospital (ESD subgroup 2) and other community hospitals (control subgroup 2), there were no significant differences. (Table [4](#Tab4){ref-type="table"}).Table 4Subgroups analysis of patient-reported outcome variables 3 months after stroke (*n* = 18,417)University hospitalSmall hospitalsVariableESD 1 (*n* = 928)Control 1 (*n* = 4636)*p value*0R (95% CI)ESD 2 (*n* = 504)Control 2 (*n* = 12,349)*p value*0R (95% CI)Satisfaction with rehabilitation\*, n (%) Satisfied338 (36.4)1658 (35.8)0.0012.27 (1.44--3.58)169 (33.5)4024 (32.6)0.8631.04 (0.66--1.64) Dissatisfied25 (2.7)278 (6.0)ref22 (4.4)593 (4.8)ref No need267 (28.8)1342 (28.9)182 (36.1)4401 (35.6) Not received20 (2.2)215 (4.6)24 (4.8)732 (5.9) Missing and do not know278 (30.0)1143 (24.7)107 (21.2)2599 (21)Information about stroke, n (%) Satisfied590 (63.6)2626 (56.6)0.0011.93 (1.32--2.73)337 (66.9)7683 (62.2)0.060.63 (0.45--0.88) Dissatisfied38 (4.1)337 (7.3)ref44(8.7)676 (5.5)ref No need52 (5.6)595 (12.8)35 (6.9)1531 (12.4) Missing and do not know248 (26.7)1078 (23.3)88 (17.5)2459 (9, 19)Tiredness/fatigue, n (%) Seldom470 (50.6)2337 (50.4)ref310 (61.5)7256 (58.8)ref Often252 (27.2)1553 (33.5)0.1170.87 (0.74--1.04)150 (29.8)3622 (29.3)0.8600.98 (0.80--1.20) Missing206 (22.2)746 (16.1)44 (8.7)1471 (11.9)Pain, n (%) Seldom586 (63.1)3074 (66.3)ref370 (73.4)8794 (71.2)ref Often121 (13.0)756 (16.3)0.2150.87 (0.70--1.08)79 (15.7)1942 (15.7)0.9761.04 (0.78--1.29) Missing221 (23.8)806 (17.4)55 (10.9)1613 (13.1)Depression, n (%) Seldom660 (71.1)3323 (71.7)ref423 (83.9)9586 (77.6)ref Often52 (5.6)507 (10.9)0.0010.55 (0.41--0.75)35 (6.9)1104 (8.9)0.090.73 (0.51--1.05) Missing216 (23,3)806 (17.4)46 (9.1)1659 (13.4)General health, n (%) Good604 (65.1)3012 (65.0)0.0811.23 (0.98--1.54)384 (76.2)8847 (71.6)0.571.08 (0.82--1.42) Poor110 (11.9)743 (16.0)ref67 (13.3)1705 (13.8)ref Missing214 (23.1)881 (19.0)53 (10.5)1797 (14.6)Adl-independence, n (%)Mobility Independent683 (73.6)3503 (75.6)0.0011.97 (1.38--2.80)426 (84.5)9868 (79.9)0.1061.34 (0.94--1.90) Dependent45 (4.8)428 (9.2)ref38 (7.5)1214 (9.8)ref Missing200 (21.6)705 (15.2)40 (7.9)1267 (10.3)Toileting Independent670 (72.2)3451 (74.4)0.0011.68 (1.23--2.30)410 (81.3)9609 (77.8)0.5331.10 (0.81--1.49) Dependent58 (6.3)484 (10.4)ref54 (10.7)1475 (11.9)ref Missing200 (21.6)701 (15.1)1265 (10.2)40 (7.9)Dressing Independent647 (69.7)3319 (71.6)0.0011.57 (1.19--2.05)391 (77.6)9219 (74.7)0,8560.98 (0.75--1.27) Dependent80 (8.6)618 (13.3)ref75 (14.9)1854 (15.0)ref Missing201 (21.7)699 (15.1)38 (7.5)1276 (10.3) Discussion {#Sec10} ========== This case-control observational study based on the Swedish stroke registry, Riksstroke, revealed that ESD provided significant benefits to patients with stroke treated in modern stroke care units. Those that received ESD experienced more satisfaction with rehabilitation after discharge, less need for assistance with ADL, and less dysthymia/depression, compared to patients that did not receive ESD. The outcomes of this study were based on the responses to PROMs. PROMs evaluate a person's feeling: it includes symptom reporting, satisfaction with care and treatments, self-rated health, and health-related quality of life \[[@CR13]\]. In a large registry with national coverage (Riksstroke), it is important to choose simple questions that cannot be misunderstood. Rikstroke has developed simplified PROMs for dysthymia/depression, fatigue, pain, general health status and ADL (mobility, toileting and dressing). These PROMs have been validated against more established measurements (<http://www.riks-stroke.org>). The validation showed good agreement with established measurements and accurate reliability for the variables chosen in this study (<http://www.riks-stroke.org>, \[[@CR14]\]). In the current study, it was possible to identify and exclude those who did not receive ESD in the intervention group, but it was not technically possible in the control group. About 2% of the participants in the control group had received ESD, according to Rikstroke data. These participants were distributed throughout the control hospitals, and the majority of hospitals reported only a few participants/year. It is possible that these registrations are incorrect and we assumed that these patients had not received rehabilitation according to an established model for ESD and that including these patients would not affect our analysis. Patient satisfaction regarding ESD has previously been evaluated, but those studies reported conflicting results. Some studies showed improved satisfaction with ESD \[[@CR1], [@CR15]\] and others found insufficient evidence \[[@CR8]\]. Patient satisfaction may be affected by various factors, particularly homecoming experiences and patient expectations. Some studies have investigated patient experiences in homecoming and home rehabilitation in the context of ESD \[[@CR16]--[@CR18]\]. Those studies indicated that patients and their families were positive to coming home, and they expressed great satisfaction about receiving rehabilitation within the home environment. The patients felt more capable of undertaking ADL in the home environment, and they were looking forward to re-establishing meaningful participation in their chosen life roles \[[@CR16]\]. Previous research has also reported that domiciliary therapy allowed patients and their families to become more involved in decision-making, to provide more insight, and to offer motivation in rehabilitation functions \[[@CR19]\]. Another aspect of satisfaction with rehabilitation is the patient's expectations of coming home with ESD after stroke. A recent study \[[@CR20]\] on the expectations of coming home with very early supported discharge and home rehabilitation after stroke found that the participants had mixed expectations. Patients were longing to come home, but also described insecurity and fear. Despite these mixed expectations, the participants were highly confident that the ESD team would support them in achieving independence \[[@CR20]\]. Presumably, patient satisfaction was closely linked to their participation in rehabilitation. Thus, it was important that patient expectations were consistent with the intervention received. A strength of this study was the large control group, which was drawn from many different hospitals covered in Riksstroke. Another strength was the use of two registries, which provided information about both patient characteristics, education level, and country of birth. We adjusted for variables that were measured in Riksstroke and differed in the baseline comparison, but as in all observational studies, there is a possibility of unmeasured confounding. Since we expected the outcome variables to be correlated, we chose not to adjust the confidence level for multiple testing. Using Bonferroni correction, which is a conservative method, we would still report significant results for outcome with *p*-values \< 0.005 (all but toileting and dressing). A limitation in Rikstroke in general is a limited response rate of NIHSS at baseline thus excluding detailed description of stroke severity. In a previous observational implementation study describing the ESD at Umeå Stroke Center, a modified version of the NIHSS at admission that included level of consciousness, arm and leg paresis and language showed a mean value of 2 ± 2.2 (SD). In this study we aimed to model positive outcome. In general, there were more missing responses in the ESD group. In the sensitivity analysis (Table [3](#Tab3){ref-type="table"}) we included missing responses in the negative category (which corresponds to a worst-case scenario). This analysis showed similary results as the main analysis in Table [2](#Tab2){ref-type="table"}. The lack of response may be due to different causes. In this study, about 30% of the participants in the ESD group reported that they had no need of rehabilitation after discharge. However, according to information in the Riksstroke Registry, all individuals in the ESD group received rehabilitation efforts. This highlight the difficulty of collecting PROM responses at 3 months after a stroke onset. After a stroke, it is not unusual for an individual to have problems, initially, with memory, concentration, and fatigue, and all these symptoms can affect the reliability of the PROM response. The lack of response can also be due to a wish to continue life and put the stroke incidence behind. In addition, it is important to consider a recall bias: that is, by the time a patient responds to the 3-months Riksstroke follow-up survey, it may be difficult to recall their experience with rehabilitation after discharge. Difficulties to recall their experience with rehabilitation after a long time and the fact that this is a fragile group, where the health can change is the reason for not including 12-months follow up in this study. We found no differences between the ESD vs no ESD groups in pain or fatigue. This finding may be explained by the fact that a treatment recommendation was established for depression, but no consensus was established regarding the treatment of pain and/or fatigue. Our results concerning ADL dependency were consistent with previously reported results. Patients that received ESD required less ADL assistance than patients that received modern conventional care without ESD \[[@CR1], [@CR8], [@CR21]\]. Conclusions {#Sec11} =========== This case-control observational study, based on registry data of PROMs at 3-months follow up showed that patients that received ESD after stroke were more satisfied with rehabilitation after discharge and experienced less depression than patients that received other health care and/or rehabilitation care. In this study, all patients received modern stroke unit care with short hospital stays, access to hyper-acute therapies, and early carotid interventions. Therefore, our results substantiate the hypothesis that the previously shown benefits of ESD and home rehabilitation were also observed among patients treated in modern stroke care. ADL : Activities in daily living CI : Confidence Interval ESD : Early supported discharge LISA : Longitudinal Integration Database for Health Insurance and Labor Market Studies NIHSS : National Institutes of Health Stroke scale OR : Odds Ratio PROM : Patient-reported outcome measurement REF : Reference RLS85 : Reactive Level Scale SD : Standard Deviation SU : Stroke unit We thank the members of the Rikstroke Collaboration ([http://www.Rikstroke.org](http://www.rikstroke.org)). Funding {#FPar1} ======= This study was supported by the Swedish Heart and Lung foundation, the Swedish Stroke Foundation, the Northern Swedish Stroke Fund, the County of Västerbotten, and the medical faculty of Umeå University. Availability of data and materials {#FPar2} ================================== All data are available without restriction from corresponding author on reasonable request. AB, ME, E-LG and PW designed and prepared the study. AB analysed the data with advise from ME and PW. AB drafted the manuscript. ME, E-LG and PW advised on method of data analysis, presentation of results and critical revision of the manuscript. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== This study was approved by the Regional Ethics Review Board at Umeå University (Dnr 2012--179-32 M, 2014--273-32 M). Consent for publication {#FPar4} ======================= Not applicable. Competing interests {#FPar5} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Introduction ============ Bladder cancer is one of the most common malignant tumors. Most cases progress to high-grade invasive cancer despite long-standing intravesical therapies. Novel therapeutic treatment options are urgently needed to improve the overall treatment success rates for localized bladder cancer^\[^[@r1]^\]^. Therefore, stable, reliable, simple, and reproducible orthotopic animal models are critically important. Suitable animal models provide an opportunity to study the mechanism of pathogenesis and allow the research and development of novel therapeutic agents. In this study, we have successfully established a model of orthotopic bladder tumor in mice using a drift angle stylet to injure the mucous membrane of the urinary bladder. The tumor cells grew on the wall of the urinary bladder after tumor cell injection. This method is convenient, rapid, and stable for establishing bladder tumor in mice and demonstrates the growth, infiltration, and metastasis of the tumor. Materials and Methods ===================== Preparation of the drift angle stylet ------------------------------------- The stylet of the 24\# venous retention needles (Becton Dickinson Infusion Therapy Systems, Inc., America) was bent in a 5° to 7° angle at a distance of 15 mm from the needlepoint to form a Φ=2.61 mm to 3.66 mm circle when the stylet was rotated ([**Figure 1**](#f1){ref-type="fig"}). {#f1} Cell strain ----------- Mice urinary bladder transitional cancer cell line BTT-T739, which originated from inbred line T739 mice, is a carcinogen-induced \[N-butyl-N-(4-hydroxybutyl) nitrosamine\] poorly differentiated transitional carcinoma. The cell line was provided by Professor Yang Xiaofeng from Shanxi Medical University. Human urinary bladder T24 cell strain was purchased from Shanghai Institutes for Biological Science, CAS. The cells were cultured in RPMI 1640 medium supplemented with 10% FBS, 100 U/mL penicillin, and 100 µg/mL streptomycin at 37°C and 5% CO~2~. The cells were harvested by trypsin/EDTA treatment. Viability was determined using the trypan blue exclusion method. The cells were suspended in prepared PBS at a concentration of 1×10^7^/mL. Animal ------ The present study was approved by the local ethics committee and followed the guidelines of the National Research Council Guide for the Care and Use of Laboratory Animals. Female inbred line T739 mice, 4 weeks to 6 weeks old and weighing 20 g to 22 g, were purchased from Beijing HFK Bio-Technology Co. Ltd. The animal produce license number was SCXK (Jing) 2009-0004. Female Balb/C-nu-nu mice, 4 weeks to 6 weeks old and weighing 20 g to 22 g, were purchased from Vital River Laboratories. The animal produce license number was SCXK (Jing) 2006-0009. The animals were bred in the animal laboratory (license number of SYXK (Jin) 2007-0002). The temperature and humidity of the environment was kept at (26±1.5)°C and between 40% and 60%, respectively. The mice were provided with sterilized drinking water and sterile complete nutrition feed. The orthotopic transplantation of the tumor cell ------------------------------------------------ The mice in the experimental group were narcotized by sodium pentobarbital at a dosage of 60 mg/kg. The pipe casing was lubricated with liquid paraffin and inserted into the bladder cavity, and then urine was removed. The drift angle stylet was slowly inserted into the pipe casing. The pipe casing was fixed with one hand, the stylet was rotated for five rounds with the other hand, and then pulled out. Cell suspension (100 µL) of approximately 1×10^6^ cells was injected immediately. The mice in the normal control group were injected with 1×10^6^ cells directly without mucosa injury. The comparison results between the mucosa injured and uninjured urinary bladder are shown in [**Figure 2**](#f2){ref-type="fig"}. {#f2} Observation method ------------------ The tumor cell-inoculated mice were fed routinely. The activity and body weight of the mice were observed daily, and survival time was recorded. All mice that died of natural causes were dissected and checked for tumor formation on the urinary bladder and abdominal cavity and hematuria. Metastasis and wet weight of the urinary bladder were recorded after the dissection. The urinary bladder and tumor metastasized organs were fixed in 10% formaldehyde for pathological examination. The observation time of the T739 mice was 40 d, whereas that of the Balb/C-nu-nu mice was 60 d. Results ======= **Tumor developmen**t --------------------- A total of 60 T739 mice were inoculated with BTT cells in the experimental group. The results showed that the bladder tumor incidence was 100% and the average wet weight of the tumor was (0.54±0.37) g. Ten T739 mice were inoculated with BTT cells in the mucosa uninjured control group; the bladder tumor incidence was 0% ([**Table 1**](#t1){ref-type="table"}**,** [**Figure 3**](#f3){ref-type="fig"}). ###### Comparison of tumor incidence between the mucosa uninjured bladder and mucosa injured bladder by drift angle stylet. Groups Number of mice Cell inoculation amount Tumor developed amount Tumor incidence (%) ----------------------------- ---------------- ------------------------- ------------------------ --------------------- T739 without injury 10 1×10^6^ 0 0 T739 with injury 60 1×10^6^ 60 100 Balb/C-nu-nu without injury 10 1×10^6^ 0 0 Balb/C-nu-nu with injury 60 1×10^6^ 60 100 {#f3} A total of 60 Balb/C-nu-nu nude mice were inoculated with T24 cells in the experimental group. The results showed that the bladder tumor incidence was 100% and the average wet weight was (0.11±0.13) g. The bladder tumor incidence in the normal control group was 0% ([**Table 1**](#t1){ref-type="table"}**,** [**Figure 3**](#f3){ref-type="fig"}). Metastasis ---------- Metastases in the liver, kidney, mesentery, and peritoneum with bloody ascites were observed in 23% of the 60 T739 mice inoculated with BTT cells. Abdominal cavity adhesion was observed, and metastasis occurred either on several viscera (liver, kidney, mesentery, and peritoneum) or only one viscera (kidney) ([**Figure 4**](#f4){ref-type="fig"}). Only one of the 60 Balb/C-nu-nu mice inoculated with T24 cells had metastasis in the kidney ([**Figure 4**](#f4){ref-type="fig"}). {#f4} Survival time ------------- T739 mice in the experimental group began to die on the 16th day after BTT cell inoculation. The animals had serious athrepsia with hematuria. The average survival time of the mice was (26.69±9.24) d. In addition, a number of mice died from extreme athrepsia and cachexia ([**Table 2**](#t2){ref-type="table"}). Balb/C-nu-nu mice in the experimental group began to die on the 18th day after T24 cell inoculation. The average survival time of the mice was (34.59±9.8) d. The mice had extreme athrepsia and cachexia ([**Table 2**](#t2){ref-type="table"}**,** [**Figure 5**](#f5){ref-type="fig"}). ###### Average survival time and bladder weight of the mice transplanted with tumor cells. Groups Number of mice Average survival time (days) Average tumor wet weight ----------------------------- ---------------- ------------------------------ -------------------------- T739 without injury 10 \>40 0.02±0.22 T739 with injury 60 26.69±9.24 0.54±0.37 Balb/C-nu-nu without injury 10 \>60 0.02±0.32 Balb/C-nu-nu with injury 60 34.59±9.80 0.11±0.13 {#f5} Pathological examination ------------------------ The bladder mucosa of the normal control mice was intact, and no metastasis was observed ([**Figure 6A**](#f6){ref-type="fig"}). Huge solid tumors were observed under the microscope. The tumor cells infiltrated deeply into the muscular layer ([**Figure 6B**](#f6){ref-type="fig"}). Cellular atypia was observed in the pathological section of the liver and kidney ([**Figures 6C**](#f6){ref-type="fig"}**and**[**6D**](#f6){ref-type="fig"}). {#f6} Discussion ========== Orthotopic bladder transplantable tumor in mice is a practical model because it is analogous to the clinical pathological process in humans^\[^[@r1]^\]^. Four frequently used methods are classified by their difference in inoculation process, and are described as follows. The first method is intravesical instillation of tumor cell suspension^\[^[@r2]^\]^ into denuded bladders, which uses N-methyl-N-nitrosourea^\[^[@r3]^\]^ or combination of chlorhydric acid and potassium hydroxide^\[^[@r4]^,^[@r5]^\]^. This method requires no surgery and does not result in diffuse ulceration, edema, or urinary stones. The success rate of tumor implantation ranges from 28% to 97%. Moreover, the tumors appear multifocal and have an unpredictable localization. The second method is injecting tumor cells into the urinary bladder wall through the abdomen^\[^[@r6]^-^[@r10]^\]^. This method requires a skilled surgeon to complete the surgery and is time consuming. The third method is injecting weak acid, protease, or other chemicals into the urinary bladder entocoel to injure the mucous before the injection of tumor cells^\[^[@r4]^,^[@r11]^-^[@r13]^\]^. This method is also complicated and tumor formation was not stable. The fourth method is the mechanical method. The mucous membrane of the urinary bladder was injured before the injection of tumor cells^\[^[@r14]^\]^. Bisson et al.^\[^[@r14]^\]^ inserted a stylet with an outside casing into the urinary bladder through the meatus urinarius to injure the mucous. The method does not have any restrictions to the rotation degree of the stylet; thus the degree of injury to the mucous cannot be controlled. Creating an injury in the mucous of the urinary bladder is important in tumor transplantation. Hemorrhage and damage to the mucous after the injury are beneficial to the adhesion, colonization, and growth of tumor cells. An intact mucosa protects the bladder from tumor cell invasion. The drift angle stylet forms a Φ=2.61 mm to 3.66 mm circle when rotated, causing a balanced injury. Thus, the injury is controllable, and the method is more reasonable. Bisson et al.^\[^[@r14]^\]^ used a custom-made copper abrader to establish a bladder tumor. The result showed that the tumor incidence was approximately 100%. Using a mechanical damage method to induce an injury to the bladder mucosa was feasible to establish an orthotopic bladder tumor. Spotted state injury was induced by friction when rotated. Tiny tumors developed at an early stage and progressed slowly. The drift angle stylet can augment the wound surface when rotated. A number of tumor cells colonized on the mucous at an early stage, and tumors formed within a short period of time. The present model showed a high metastatic rate. Vesicoureteral reflux may cause renal metastasis. Metastasis may also have a relationship with the cell line because the metastatic rate of T24 was lower than that of BTT. Tumors that has broken through the bladder wall and invaded into the abdominal cavity are in the advanced stage, leading to high metastatic rate. The 24\# venous retention needle used in the present study is a standard medical device and is easy to obtain. The preparation of the drift angle stylet is simple, and the experiment is easy to perform. Moreover, the degree of injury is easy to control. Using drift angle stylet to injure the mucous membrane of the urinary bladder can establish a stable transplantable bladder tumor model in mice, with convenience and rapidness. No potential conflicts of interest are disclosed. The authors are thankful to Gang Tai at Shanxi Institute for the drug control and for providing the pathology sections; Professor Xiao-feng Yang at Shanxi Medical University for providing the BTT cell line; Professor Jian-zhen Zhang at Shanxi University for checking the manuscript, and Professor Jin-fen Wang at Shanxi Tumor Hospital for the pathology analysis. This study was supported by a grant from the Shanxi Science and Technology Department, China (Grant No.2010K01). [^1]: Correspondence to: Lian-sheng Ren [^2]: E-mail: <[email protected]>

1. Introduction {#sec1} =============== Osteoporosis is a prevalent bone disease that is defined by a low bone mass and an increased risk of fractures \[[@B1]\]. Adequate regulation of bone remodeling in adulthood is essential to maintain a healthy bone mass \[[@B2]\]. However, an imbalance between bone formation and bone resorption in pathological conditions or during the aging process can result in conditions that lead to osteoporosis \[[@B3]\]. In the past, the process of bone remodeling was considered to be regulated by two major cell types: bone-forming osteoblasts and bone-resorbing osteoclasts \[[@B4]\]. Currently available agents used to treat osteoporosis include estrogen, raloxifene, the bisphosphonates (e.g., alendronate and risedronate, etc.), and calcitonin. Their mechanisms are based on the inhibition of osteoclastic bone resorption to prevent further bone loss \[[@B5]\]. However, many osteoporotic patients already had lost a substantial amount of bone mass before diagnosis \[[@B6]\]. Furthermore, many side effects of antiresorptive agents have been reported, causing many patients to discontinue their use \[[@B7]--[@B9]\]. Drugs with anabolic effects have received much recent attention for osteoporosis therapy. These pharmacologic agents can ultimately stimulate new bone formation, enhance bone density, reduce bone fracture, and promote bone health. Currently, there is only one available anabolic agent on the market approved for the treatment of osteoporosis, recombinant human parathyroid hormone (hPTH). However, hPTH is limited for use in cases of severe osteoporosis \[[@B10]\], and the clinical treatment period is limited to 18 months \[[@B11]\]. Consequently, it is necessary to develop other potential anabolic agents with fewer undesirable side effects to prevent or reverse osteoporosis. The Chinese yam (*Dioscorea*) has been widely used as a herbal medicine in China for more than 2000 years. Ancient Chinese medicinal books or folk remedies have provided evidence that *Dioscorea* might have effects in regulating bone metabolism. Therefore, we initiated a project to evaluate the effects of an ethanol extract of rhizomes of *Dioscorea alata* L. cv. Phyto, Dispo85E, on bone formation and to delineate the mechanisms involved. In our study, we found that Dispo85E promoted bone formation by inducing mesenchymal stem cells (MSCs) differentiation into osteoblasts rather than adipocytes and that it also possessed antiosteoporotic activity *in vivo*. 2. Materials and Methods {#sec2} ======================== 2.1. Animals {#sec2.1} ------------ Specific pathogen-free (SPF) C57BL/6 female mice, 8 to 10 weeks of age, were obtained from the National Laboratory Animal Center (Taipei, Taiwan). Animals were housed at a constant temperature and fed with laboratory chow (PMI, Brentwood, Mo, USA) and water *ad libitum*. The protocol of the experiments was approved by the Animal Research Committee of National Yang-Ming University (Guide for Animal Experiments, National Yang-Ming University). 2.2. Preparation of Dispo85E {#sec2.2} ---------------------------- Dried and peeled tubers of *Dioscorea alata* L. cv. Phyto (2.92 kg) were extracted with 85% ethanol (EtOH) (15 L each, three times) for 24 hours. The extract was filtered through an 11 *μ*m Whatman filter, then concentrated under vacuum evaporator and lyophilized, giving a yield of 3.05%. The extract was stored at −20°C before use. 2.3. HPLC Analysis of Dispo85E {#sec2.3} ------------------------------ Twenty milligrams of dried Dispo85E was dissolved in 2 mL of dichloromethane (DCM). After centrifugation, the supernatant was evaporated to dryness, adjusted to a concentration of 5 mg/mL in DCM and then subjected to normal-phase high performance liquid chromatography (NP-HPLC) with a 100 *μ*L injection volume. The pellet was evaporated to dryness, adjusted to a concentration of 10 mg/mL in water and then subjected to reverse-phase high performance liquid chromatography (RP-HPLC) with a 100 *μ*L injection volume. RP-HPLC profiling was performed using a Mightysil RP-18 column (4.6 × 250 mm, 5 mm) at room temperature. The mobile phase was methanol and water in gradient mode (10 : 90--100 : 0 over 120 minutes). The effluent was monitored at 254 nm, and a constant flow rate was set at 0.8 mL/minute. NP-HPLC profiling was performed with a Cosmosil 5SL-II column (4.6 × 250 mm, 5 mm) at room temperature. The mobile phase was DCM and methanol (MeOH) in gradient mode as follows: 100%--98% DCM in MeOH (0--15 minutes), 98%-95% DCM (15--25 minutes), 95%--90% DCM (25--35 minutes), 90%--80% DCM (35--45 minutes), and 80%--70% DCM (45--55 minutes). The effluent was monitored at 254 nm, and a constant flow rate was set at 0.8 mL/minute. 2.4. Cell Culture {#sec2.4} ----------------- Primary mouse bone marrow cells and C3H10T1/2 cells were cultured as described previously \[[@B12], [@B13]\]. Briefly, bone marrow cells were obtained from the femoral bone of SPF-grade C57BL/6 female mice. The bone marrow cells were collected by flushing the diaphysis with *α*-minimum essential medium (*α*-MEM, Gibco BRL) through a 23-gauge needle. After flushing, the bone marrow cells were filtered through a no. 53 sterile nylon mesh to obtain a single cell suspension. The cells were cultured in osteogenic medium (*α*-MEM medium supplemented with 15% fetal calf serum (FCS, Gibco), 50 *μ*g/mL ascorbic acid (Sigma), 10 mm sodium *β*-glycerophosphate (Sigma), and 10 nm dexamethasone (Sigma)). The pluripotent murine mesenchymal cell line C3H10T1/2 was cultured in Dulbecco\'s modified Eagle\'s medium (DMEM) supplemented with 10% FCS. For the alkaline phosphatase (ALP) activity assay, primary bone marrow cells were seeded in a 96-well microplate at 2.5 × 10^5^/well and cultured in osteogenic induction medium. After two days, half of the medium was changed and the cells were incubated with Dispo85E at the indicated concentrations for 4 days. ALP activity for each cell lysate was assayed. C3H10T1/2 cells were grown to confluence in standard medium. One day after the cells reached confluence, the medium was replaced with permissive osteogenic/adipocytic medium \[[@B14]\] (DMEM with 10% FCS, 50 *μ*g/mL ascorbic acid, 10 *μ*M sodium *β*-glycerophosphate, 10 nm dexamethasone, 10 nm all-trans retinoic acid (Sigma), 10 *μ*g/mL insulin (Sigma), and 50 *μ*M isobutylmethylxanthine (IBMX, Sigma)) containing drugs at the indicated concentrations. The medium was changed every 4 days. After 12 days, ALP activity for each cell lysate was assayed. For the nodule formation assay, primary bone marrow cells were seeded in a 24-well microplate at 1 × 10^6^/well and cultured in osteogenic induction medium. After two days, half of the medium was changed and the cells were incubated with Dispo85E at the indicated concentrations. The culture medium was renewed every 4 days. After 14 days, the mineralization of bone marrow cells was analyzed. For lipid staining, C3H10T1/2 cells were grown to confluence in standard medium. One day after the cells reached confluence, the medium was replaced with adipogenic medium (DMEM with 10% FCS, 10 *μ*g/mL insulin, 1 *μ*M dexamethasone, and 0.5 mm IBMX) containing drugs at the indicated concentrations for 4 days. The cells were then incubated in a standard medium containing 5 *μ*g/mL insulin and drugs at the indicated concentrations. The medium was changed every 4 days. The oil red O staining was performed at postinduction day (PID) 12. Nile red staining was performed at PID16. 2.5. Alkaline Phosphatase Activity Assay {#sec2.5} ---------------------------------------- ALP activity was assayed at 37°C by a method modified from that of Qu et al. \[[@B15]\]. In brief, cell layers were washed twice with phosphate buffered saline (PBS) and extracted into lysis solution (10 mm Tris, 0.1% Triton X-100 buffer (pH 7.5)). Enzyme activity was determined colorimetrically using p-nitrophenylphosphate (p-NPP, Sigma) as a substrate. The reaction mixture contained 8 mm p-NPP, 2 mm mgCl~2~, and 0.5 M 2-amino-2-methyl-1-propanol, pH 10. After 10 minutes of incubation, the color change of p-NPP to p-nitrophenol was monitored at 405 nm. Cell viability was determined by resazurin assay. Fifty micromolar resazurin (Sigma) was added to the cell culture medium and incubated for 2 hours. Medium samples were collected into a 96-well plate, and fluorescence was measured with excitation at 544 nm and emission at 590 nm. 2.6. Nodule Formation Assay {#sec2.6} --------------------------- The *in vitro* mineralized nodule formation assay was performed as described previously \[[@B16]\]. In brief, cells were fixed with 10% formalin for 30 minutes at 37°C. The formalin was removed, and the cells rinsed with sterilized water three times. Next, 2% alizarin red S solution, which reacts with calcium, was added to the wells and the cells were incubated at 37°C for 10 minutes. The Alizarin solution was removed, and the cells were washed with water and dried in air. Stained cultures were photographed, and calcium deposition was quantified by extracting alizarin red S staining with 10% cetylpyridinium chloride (CPC, Sigma) and measuring the OD of the extract at 550 nm \[[@B17]\]. 2.7. Adipocyte Staining with Oil Red O {#sec2.7} -------------------------------------- After differentiation was induced, cells were stained with oil red O. Briefly, cells were washed twice with PBS and fixed with 10% formalin in PBS for 1 hour. They were then washed three times with water. Cells were stained with Oil Red O (six parts of 0.6% oil red O dye (Sigma-Aldrich) in isopropanol and four parts of water) for 30 minutes. Excess stain was removed by washing with water, and the stained cells were dried. Adipocytes stained red were observed under the phase contrast microscope with 200-fold magnification. 2.8. Quantification of Adipocyte Number by Flow Cytometry {#sec2.8} --------------------------------------------------------- The formation of mature adipocytes was quantified with flow cytometry using the lipophilic Nile red fluorescent dye as described previously \[[@B18]\]. Briefly, after the cells were induced to differentiate into adipocytes as indicated above, they were washed twice in PBS, trypsinized, pelleted by centrifugation, and fixed in 10% formaldehyde for 1 h at 4°C. A working solution of Nile red was prepared by dissolving 1 mg Nile red in 200 *μ*L DMSO and diluting to 100 mL with PBS. The cells were stained with 10 *μ*g/mL Nile red for 45 minutes at room temperature. The fluorescent emission was detected between 564 nm and 604 nm with a band-pass filter using a FACScan flow cytometer (BD Biosciences) linked with Cell-Quest 3.3 software (BD Biosciences). 2.9. Isolation and Analysis of RNA {#sec2.9} ---------------------------------- Total RNA was extracted from the bone marrow cells using the Ultraspec RNA isolation kit (Biotex laboratories INC, USA). RNA was reverse transcribed using AMV reverse transcriptase (Promega, USA) and an appropriate buffer in a final reaction volume of 40 *μ*L. The complementary DNA (cDNA) was obtained using the previous reaction solution at 42°C for 60 minutes, followed by 90°C for 5 minutes. The resultant cDNA (2.5 *μ*L) was added to 0.5 *μ*L 10 mm dNTP, 0.5 *μ*L polymerase (2 units), and 1 *μ*L of 10 *μ*M the appropriate primers, and the final volume of the reaction mixture was adjusted to 25 *μ*L. Polymerase chain reaction (PCR) was performed for 35 cycles with each cycle consisting of 45 seconds of denaturation at 94°C, 45 seconds of annealing at proper annealing temperature, and 1 minute of extension at 72°C. The primers used were as follows: OCN (5′-TCT  GAC  AAA  GCC  TTC  ATG  TCC-3′ and 5′-AAA  TAG  TGA  TAC  CGT  AGA  TGC  G-3′), IGF-1 (5′-GCT  CTT  CAG  TTC  GTG  TGT  GG-3′ and 5′-TTG  GGC  ATG  TCA  GTG  TGG-3′), BMP-2 (5′-CAT  CCA  GCC  GAC  CCT  TG-3′ and 5′-CTC  TCC  CAC  TGA  CTT  GTG-3′), ALP (5′-GCC  CTC  TCC  AAG  ACA  TAT  A-3′ and 5′-CCA  TGA  TCA  CGT  CGA  TAT  CC-3′), COL-*Ⅰ* (5′-TCT  CCA  CTC  TTC  TAG  TTC  CT-3′ and 5′-TTG  GGT  CAT  TTC  CAC  ATG  C-3′), *β*-actin (5′-GAC  TAC  CTC  ATG  AAG  ATC  CT-3′ and 5′-CCA  CAT  CTG  CTG  GAA  GGT  GG-3′), IL-4 (5′-TG  GGT  CTC  AAC  CCC  CAG  CTA  GT-3′ and 5′-GCT  CTT  TAG  GCT  TTC  CAG  GAA  GTC-3′), IL-6 (5′-ATG  AAG  TTC  CTC  TCT  GCA  AGA  GAC  T-3′ and 5′-CAC  TAG  GTT  TGC  CGA  GTA  GAT  CTC-3′), and TGF-*β* (5′-TGG  ACC  GCA  ACA  ACG  CCA  TCT  ATG  CCA  TCT  ATG  AGA  AAA  CC-3′ and 5′-TGG  AGC  TGA  AGC  AAT  AGT  TGG  TAT  CCA  GGG  CT-3′). The reaction products were analyzed by electrophoresis on a 2% agarose gel and visualized by ethidium bromide staining with ultraviolet light illumination. Gene expression levels were analyzed by IMAGEQUANTE software and normalized with *β*-actin. Real-time PCR was performed using an Applied Biosystems 7500 Real-Time PCR System (Applied Biosystems). The PCR mixture was prepared using SYBR green Mastermix (Roche). The thermal cycling conditions were 10 minutes at 95°C followed by 40 cycles of 95°C for 15 seconds and 57°C (OCN and IGF-1) for 20 seconds, with an extension period at 72°C for 40 seconds. Expression of each gene was normalized to GAPDH mRNA content. 2.10. Animal Models {#sec2.10} ------------------- There were two different sets for the *in vivo* experiment. In the first series of experiments, intact female mice (without OVX surgery) were fed a normal diet containing different concentrations (0, 40, 200, and 1000 mg/kg/day) of Dispo85E *ad libitum* for 5 days (*n* = 4 for each group). In the second set of experiments, we used an OVX mouse model to mimic osteoporosis in postmenopausal women. The mice were subjected to surgery at day 0. Sham surgery was performed by identifying the bilateral ovaries, and OVX was performed by removing the bilateral ovaries. Mice (*n* = 10-11 for each group) were initially fed a normal diet containing different concentrations (0, 40, 200, and 1000 mg/kg/day) of Dispo85E *ad libitum*. Feeding began soon after OVX surgery and continued for 42 days. We identified OVX mice by confirming uterus atrophy in OVX animals at the end of the experiment. At the end of the experiment, all mice were sacrificed by cervical dislocation. Bone marrow cells were then acquired from the femur of each mouse. The drug\'s effect on ALP activity, nodule formation, mRNA expression of the osteoblast differentiation related genes in bone marrow cells, histological change of the trabecular bone, trabecular bone mineral density (tBMD), and bone tissue microarchitecture were analyzed. To analyze the drug\'s effect on the gene expression of bone marrow cells, we extracted total RNA from bone marrow cells of the femora of the intact or OVX mice after continuous feeding with Dispo85E and examined expression through RT-PCR. To analyze the drug\'s effect on the ALP activity of bone-marrow-derived cells isolated from intact or OVX mice, the mice were sacrificed to isolate the bone marrow cells after treatment with different doses of Dispo85E. The isolated bone marrow cells were seeded in a 96-well microplate at 2.5 × 10^5^/well and cultured in osteogenic induction medium for 6 days without additional Dispo85E treatment. After 6 days of culture, the ALP activity for each cell lysate was assayed. 2.11. Histological Analysis of Trabecular Bone {#sec2.11} ---------------------------------------------- Histomorphometric measurements of trabecular bone volume were performed as described previously with minor modifications \[[@B12]\]. In brief, femora were collected 42 days after the surgical operation, fixed in Bouin\'s solution at 26°C for 24 hours, decalcified in 14% EDTA, dehydrated in progressive concentrations of ethanol, cleared in xylene, and embedded in paraffin. Five-micron-thick sections were cut and stained for hematoxylin and eosin Y. Static parameters were measured in the square, distal metaphysics of the femur within the endosteal surfaces, excluding the epiphyseal growth plate, and 1 mm distal from the end of the calcified cartilage. Histomorphometric measurements of the bone volume/total bone volume percentage (BV/TV, %) were performed at 40-fold magnification. 2.12. Analysis of Trabecular Bone Microarchitecture {#sec2.12} --------------------------------------------------- The OVX mice were sacrificed 42 days after beginning the treatment with Dispo85E. The femurs were aseptically removed, cleansed of adherent soft tissues, and deposited in a tube with 10% formalin. The trabecular bone microarchitecture of the distal femoral metaphysis was scanned by microcomputed tomography (*μ*CT, Skyscan 1076) in the region of 0.6--2.1 mm from the growth plate. The X-ray source was set at a voltage of 48 kV and a current of 200 *μ*A and filtered with a 0.5 mm aluminum filter. The scanning angular rotation was 180° with an angular step of 0.7°. The voxel size was isotropic and fixed at 8.7 *μ*m. The trabecular bone parameters were calculated using the Skyscan software CTan (Skyscan). Morphometric indices of the trabecular bone region were determined from the microtomographic data sets using direct 3D morphometry. Trabecular bone volume (BV/TV; %), trabecular thickness (Tb.Th; mm), and trabecular number (Tb.N; 1/mm) were calculated. The volumetric trabecular bone mineral density (tBMD) was also determined by *μ*CT scanning. tBMD was calculated in the conforming volume of interest described previously for the trabecular region. 2.13. Statistical Analysis {#sec2.13} -------------------------- All results are expressed as the mean and standard deviation (SD). The statistical significance was evaluated by one-way analysis of variance (ANOVA) followed by Dunnett\'s test. A level of *P* \< .05 was considered statistically significant. 3. Results {#sec3} ========== 3.1. HPLC Analysis of Dispo85E {#sec3.1} ------------------------------ Dispo85E is a mixture extracted from the traditional botanical *Dioscorea alata* L. cv. Phyto. This mixture contains both polar and nonpolar small molecule ingredients. The-reverse phase HPLC analysis showed the polar to semipolar ingredient fingerprint of Dispo85E, and the normal phase HPLC analysis showed the nonpolar ingredient fingerprint of Dispo85E ([Figure 1](#fig1){ref-type="fig"}). 3.2. Dispo85E Promotes Differentiation of Osteoblastic Cells in Bone Marrow Culture {#sec3.2} ----------------------------------------------------------------------------------- The ALP activity assay showed that bone marrow cells treated with Dispo85E had a 1.5-fold maximal increase in the activity of ALP ([Figure 2(a)](#fig2){ref-type="fig"}). In the cell viability assay, there were no significant differences between the control and Dispo85E-treated groups ([Figure 2(b)](#fig2){ref-type="fig"}). Furthermore, Dispo85E significantly stimulated mineralization of bone marrow cultures by alizarin red staining (Figures [2(c)](#fig2){ref-type="fig"} and [2(d)](#fig2){ref-type="fig"}). 3.3. Dispo85E Promotes Osteoblastogenesis and Inhibits Adipogenesis in the Mesenchymal Stem Cell Line C3H10T1/2 {#sec3.3} --------------------------------------------------------------------------------------------------------------- The real-time PCR results showed that mRNA expression of IGF-1, which could promote osteoblast differentiation and bone formation, was significantly increased after treatment with Dispo85E ([Figure 3(a)](#fig3){ref-type="fig"}). OCN (an osteoblastic marker) mRNA expression was significantly increased by approximately 2.5-fold after treatment with 1 *μ*g/mL Dispo85E ([Figure 3(b)](#fig3){ref-type="fig"}). Concomitantly, we observed increased ALP activity in C3H10T1/2 cells treated with Dispo85E ([Figure 3(c)](#fig3){ref-type="fig"}), which had no effect on cell viability ([Figure 3(d)](#fig3){ref-type="fig"}). The effects of Dispo85E on adipocyte differentiation were detected by oil red O staining ([Figure 3(e)](#fig3){ref-type="fig"}) and quantified by Nile red staining and flow cytometry (FACS) analysis ([Figure 3(f)](#fig3){ref-type="fig"}). After 12 days in adipogenic medium, C3H10T1/2 cells showed an abundance of oil-red-O-stained lipid droplets ([Figure 3(e)](#fig3){ref-type="fig"}). Dispo85E resulted in a dose-dependent decrease in cellular lipid accumulation ([Figure 3(e)](#fig3){ref-type="fig"}). Using flow cytometry and Nile red staining, the percentage of adipocytes were calculated based on the number of stained cells. After culture in adipogenic medium for 16 days, 63.6% of the cells were calculated to be adipocytes ([Figure 3(f)](#fig3){ref-type="fig"}). After treatment with 1, 10, or 100 *μ*g/mL Dispo85E, the percentages of adipocytes were 54.1%, 47.3%, and 49.8%, respectively ([Figure 3(f)](#fig3){ref-type="fig"}). 3.4. Dispo85E Promotes Osteoblastogenesis in Intact Mice (without OVX Surgery) {#sec3.4} ------------------------------------------------------------------------------ Gene ExpressionWe used the expression level of *β*-actin mRNA as a house-keeping gene. Compared to the expression of *β*-actin mRNA, the relative levels of BMP-2, TGF-*β*, and IL-4 mRNA were higher in bone marrow cells from Dispo85E-treated mice than those from untreated controls ([Figure 4(a)](#fig4){ref-type="fig"}). Alkaline Phosphate Activity AssayWe measured the specific activity of ALP of bone marrow cells isolated from Dispo85E-fed or control mice for 5 days. The ALP activity was significantly higher in all Dispo85E-treated groups ([Figure 4(b)](#fig4){ref-type="fig"}). Dispo85E at a dose of 1000 mg/kg created up to a 3-fold enhancement compared to the control group. Nodule Formation AssayTo evaluate whether Dispo85E enhanced bone mineralization, we used alizarin red staining to visualize nodule formation in bone marrow cultures ([Figure 4(c)](#fig4){ref-type="fig"}). The proportional area of alizarin-red-positive staining in the Dispo85E-treated group was larger than that in the untreated group in a dose-dependent manner ([Figure 4(d)](#fig4){ref-type="fig"}). Compared with the control group, 1000 mg/kg Dispo85E showed the strongest enhancement, with effects up to 3.4-fold above controls ([Figure 4(d)](#fig4){ref-type="fig"}). 3.5. Dispo85E Promotes Osteoblastogenesis in OVX-Induced Osteoporotic Mice {#sec3.5} -------------------------------------------------------------------------- Gene ExpressionBone morphogenetic protein-2 (BMP-2), transforming growth factor-*β* (TGF-*β*), and interleukin-4 (IL-4) are positively related to the proliferation or differentiation of bone marrow cells toward an osteoblast lineage. ALP, collagen I (COL-I), and osteocalcin (OCN) are widely used as bone formation or osteoblastic cells markers. Interleukin-6 (IL-6) is a bone resorption marker. When compared to *β*-actin, the relative levels of BMP-2, TGF-*β*, IL-4, ALP, COL-I, and OCN mRNA were higher in bone marrow cells from Dispo85E-treated mice than in untreated mice. Conversely, the bone resorption marker gene IL-6 was decreased after treatment with Dispo85E ([Figure 5(a)](#fig5){ref-type="fig"}). Alkaline Phosphate Activity AssayThe ALP activity of all Dispo85E-treated groups was significantly higher than that of the untreated group ([Figure 5(b)](#fig5){ref-type="fig"}). Nodule Formation AssayGroups treated with 200 and 1000 mg/kg/day Dispo85E showed enhanced bone mass mineralization compared to the untreated group (Figures [5(c)](#fig5){ref-type="fig"} and [5(d)](#fig5){ref-type="fig"}). Histological Analysis of Trabecular BoneHistological analysis of the trabecular bone was performed in the OVX mouse model. The animal data in the sham-operated positive control group showed a significant increase in the trabecular bone volume/total bone volume percentage (BV/TV, %) compared to the vehicle control group. There was a significant increase in BV/TV for all dosage groups compared to the vehicle control group (0 mg/kg/day). Moreover, the increase was positively correlated with dose level. The histomorphometric analysis in the OVX model revealed increases in the osteoid volume in a dose-dependent manner ([Figure 6](#fig6){ref-type="fig"}), suggesting that Dispo85E ameliorates the decreasing trabeculation of the bone marrow spaces in osteoporotic mice. Analysis of Trabecular Bone MicroarchitectureThe effects of Dispo85E on trabecular bone microarchitecture and trabecular BMD of the distal femur in OVX mice were assessed by *ex vivo μ*CT analysis. Quantification of the trabecular bone changes in the distal femoral metaphysis is shown in [Table 1](#tab1){ref-type="table"}. Compared with the sham-operated animals, the OVX control group showed a significant decrease in trabecular bone mineral density (tBMD), trabecular bone volume fraction (BV/TV), and trabecular number (Tb.N). Both 40 mg/kg/day and 1000 mg/kg/day of Dispo85E significantly improved various *μ*CT bone parameters, including tBMD, BV/TV, and Tb.N compared to the OVX control group, but trabecular thickness (Tb.Th) was not significantly altered compared to the OVX control ([Table 1](#tab1){ref-type="table"}). 4. Discussion {#sec4} ============= Several drugs have been reported to be effective in treating osteoporosis. However, most of these drugs are inhibitors of osteoclast-mediated bone resorption and present many detrimental side effects \[[@B19]\]. For example, estrogen replacement therapy has been demonstrated to prevent bone loss in postmenopausal women \[[@B20]\]. However, despite the benefits of estrogen, many studies have also shown that estrogen would increase the incidence of ovarian cancer and breast cancer \[[@B7], [@B8]\]. Furthermore, epidemiological reports have shown that hormone replacement therapy (HRT) is associated with increases in the incidence rates of breast cancer, heart attacks, stroke, and blood clot formation. Such studies have prompted the US FDA to issue safety warnings and to suggest the avoidance of HRT as a general preventative osteoporosis therapy \[[@B20], [@B21]\]. On the other hand, Teitelbaum highlighted that compounds that can stimulate new bone formation may offer safe and clinically effective prevention and reversal of osteoporosis \[[@B22]\]. Bone marrow cells have long been recognized as the source of osteoprogenitor cells \[[@B23]\]. Thus, we applied a culture model of primary bone marrow cells to evaluate the effect of Dispo85E on the differentiation of osteoblasts. Intriguingly, our observations of ALP activity and mineralization suggested that Dispo85E possessed higher efficacy in promoting osteoblast differentiation. In this assay, we measured cell viability and found no significant difference between the control and treated groups. Therefore, we suggest that Dispo85E does not influence the proliferation of stromal cells *in vitro*. Recently, it has been suggested that osteoporosis is caused by an increased ratio of adipocytes to osteoblasts and not by an imbalance between osteoclasts and osteoblasts alone \[[@B24]\]. Indeed, an increase in adipogenesis in bone marrow associated with osteoporosis is well known clinically \[[@B25]\]. Other studies have also demonstrated that increases in the number of marrow adipocytes caused bone loss in osteoporosis animal models, such as in ovariectomized or glucocorticoid-treated animals \[[@B26], [@B27]\]. It was reported that the number of osteoblasts correlated negatively with the number of adipocytes, which suggested that an inverse relationship between osteogenic and adipogenic differentiation exists \[[@B28], [@B29]\]. It is known that multipotent mesenchymal stem cells (MSCs) from bone marrow stroma can differentiate into various cells, including osteoblasts and adipocytes \[[@B30], [@B31]\]. A shift in differentiation and survival rates from an osteoblastic to adipocytic lineage can lead to an altered ratio of fat to bone cells and, ultimately, an alteration in bone mass \[[@B31]\]. Thus, it has been suggested that targeting regulatory factors to direct the fate of mesenchymal cells toward osteoblasts may provide novel therapeutic approaches for osteoporosis \[[@B32]\]. In this study, we used the C3H10T1/2 cell line, which maintains the differentiating characteristics of mesenchymal stem cells \[[@B33]\], to evaluate the effect of Dispo85E on the osteogenesis/adipogenesis balance. Our *in vitro* data showed that Dispo85E could promote the differentiation of mesenchymal stem cells into the osteoblastic lineage and reduce adipogenic differentiation. Recently, it has been reported that the number of adipocytes in bone marrow correlated inversely with the hematopoietic activity of the marrow \[[@B34]\]. We also suggest that the antiadipogenic effect of Dispo85E might have the potential for use in some hematopoietic disorders. Insulin-like growth factor I (IGF-I) is known as an important factor produced by osteoblasts to regulate bone formation and remodeling \[[@B35]\]. It can induce human adult mesenchymal stem cells (hMSCs) to differentiate into an osteogenic lineage \[[@B36]\]. Some clinical data show that IGF-I was lower in women with osteoporosis than in women with normal bone mineral density (BMD) \[[@B37]\]. In addition, low serum IGF-1 can contribute to reduced BMD, reduced osteoblast differentiation, and increased the numbers of adipocytes in bone \[[@B38]\]. Real-time PCR showed that Dispo85E increased IGF-1 mRNA expression in C3H10T1/2 cells. Thus, we suggest that one way Dispo85E regulates the differentiation of MSCs is through IGF-1 upregulation. To evaluate whether Dispo85E enhanced bone formation *in vivo*, we used two kinds of animal models. In the intact mouse model, we evaluated the ability of Dispo85E to promote osteoblastogenesis under physiological conditions. PCR showed that Dispo85E increased BMP-2, TGF-*β*, and IL-4 gene expression in bone marrow cells. BMP-2 has been known to stimulate osteoblastic maturation and collagen synthesis in osteoblastic cells \[[@B39]\], TGF-*β* has been reported to stimulate the proliferation of osteoblast precursors and increase the pool of committed osteoblasts \[[@B40]\], and IL-4 is a potent inhibitor of bone resorption \[[@B41]\]. In addition, the results of the ALP activity assay and alizarin red staining for mineralization of bone marrow showed that Dispo85E promotes osteoblast differentiation in healthy animals. Next, we used an OVX mouse model to mimic the osteoporosis seen in postmenopausal women. After 6 weeks of treatment, Dispo85E promoted osteoblastogenesis without a significant effect on uterine weight (data not shown). This result differs from the previously observed restorative effects of estrogen replacement \[[@B42]\]. Therefore, we suggest that Dispo85E might act via a different mechanism from estrogen, which affects the uterus as well as bone. In this animal study, we analyzed bone marrow cells after 42 days of OVX. Our results showed that ALP activity in bone marrow cells (cultured in osteogenic induction medium for 6 days without additional Dispo85E treatment) of Dispo85E-treated groups was higher than that of the OVX control group. These data indicated that Dispo85E-treated groups contained more osteoprogenitor cells than the untreated group in the OVX mice model. Previous studies have clearly shown a reduction of trabecular bone volume during aging and in osteoporosis patients \[[@B43]\]. In our study, histological analysis of the femora harvested from OVX mice showed an empty bone marrow space in the trabecular bone. After administration of Dispo85E for 42 days, the trabecular bone volume was increased. Additionally, the *μ*CT data confirmed that 40 mg/kg/day and 1000 mg/kg/day of Dispo85E could significantly ameliorate the deterioration of both tBMD and trabecular microarchitectural parameters in OVX mice. The 200 mg/kg/day group also showed better indices than the untreated group, although it did not reach statistical significance. This may be due to the limited number of animals in our study or inconsistencies in the mixture chemistry of the Dispo85E extract. These data indicated that Dispo85E promotes osteoblast differentiation and bone formation in an osteopenic setting without side effects on the uterus. Recent studies have shown that natural products have potential as alternative approaches to prevent and ameliorate osteoporosis through different mechanisms. For example, Xie et al. reported that *Herba epimedii* (HEP) extract can effectively suppress the OVX-induced increase in bone turnover, possibly by both an increase in osteoblastic activities and a decrease in osteoclastogenesis \[[@B44]\]. Ho et al. reported that *Flemingia macrophylla* extract can ameliorate experimental osteoporosis in OVX rats through inhibition of bone resorption \[[@B45]\]. Hidaka et al. reported that royal jelly (RJ) can prevent osteoporosis by enhancing intestinal calcium absorption \[[@B46]\]. In these studies, most of the drugs have been reported to achieve improvement in osteoporosis by regulating the balance between osteoblasts and osteoclasts. Unlike these drugs, our study has shown that Dispo85E can drive the lineage-specific differentiation of bone marrow stromal cells to differentiate into osteoblasts rather than adipocytes. Furthermore, our data showed that Dispo85E could promote the differentiation and maturation of osteoblasts in both intact and OVX mice. Therefore, we suggest that Dispo85E not only can decrease the process of bone loss but also can regulate the balance between osteoblasts and adipocytes to promote bone formation. Based on these findings, we conclude that Dispo85E regulates mesenchymal stem cell differentiation into an osteogenic lineage rather than an adipogenic lineage and ameliorates osteoporosis in a mouse model ([Figure 7](#fig7){ref-type="fig"}). Our data suggest that Dispo85E induces mesenchymal stem cell differentiation and holds promise as a novel therapeutic drug for the treatment of osteoporosis. The authors thank Grant-in-Aid (93-EC-17-A-17-I1-0034) from the Ministry of Economic, Republic of China (Taiwan) and a grant from the Ministry of Education, Aim for the Top University Plan. The authors also thank the Laboratory Animal Center, Chang Gung University for micro-CT technical support and image analysis. {#fig1} {#fig2} {#fig3} {ref-type="sec"}. Cells were stained with alizarin red S for osteoblasts and (d) the mineralized area was quantified using Meta Image software. \*\*Indicates significant difference from control group at *P* \< .01. *n* = 4 for each group.](ECAM2011-712892.004){#fig4} {ref-type="sec"}. Cells were stained with Alizarin red S for osteoblasts, and (d) the mineralized area was quantified by extracting alizarin red S staining with 10% CPC and measuring the OD of the extract at 550 nm. \*\*Indicates significant difference from OVX mice at *P* \< .01; \*Indicates significance at *P* \< .05. *n* =10-11 for each group.](ECAM2011-712892.005){#fig5} {#fig6} {#fig7} ###### Effects of Dispo85E on tBMD and morphometric parameters of distal femur in OVX model. Parameters Sham OVX + Dispo85E (mg/kg/day) ---------------- --------------- ---------------------------- ------------------ ----------------- ------------------ tBMD (g/cm^3^) 0.088 ± 0.003 0.054 ± 0.008\* 0.083 ± 0.003^‡^ 0.063 ± 0.009\* 0.081 ± 0.006^‡^ BV/TV (%) 4.34 ± 0.36 2.56 ± 0.44\* 4.17 ± 0.29^†^ 3.50 ± 0.69 4.11 ± 0.39^†^ Tb.Th (*μ*m) 60.7 ± 1.03 59.30 ± 0.71 64.36 ± 2.25 63.15 ± 2.70 63.03 ± 2.26 Tb.N (1/mm) 0.72 ± 0.06 0.43 ± 0.07\* 0.65 ± 0.04^†^ 0.55 ± 0.10 0.66 ± 0.06^†^ Values are expressed as mean ± SD; *n* = 5-6/group. tBMD: trabecular bone mineral density; BV/TV: bone volume fraction; Tb.Th: trabecular thickness; Tb.N: trabecular number. \**P* \< .05 compared with the sham operated group; ^†^ *P* \< .05 compared with OVX control group; ^‡^ *P* \< .01 compared with OVX control group.

Background {#Sec1} ========== Dancing is a very rare seizure semiology. To the best of our knowledge, there have been only a few case reports of ictal dancing or dancing-like movement. The previous reports provided either of description only or video only; or the case could hardly be called dancing. The "dancing epilepsy" was shown in a case of refractory complex partial seizures \[[@CR1]\]. However, this article did not show the electroencephalography. Therefore, we cannot confirm whether the dancing was truly ictal or only a manifestation of post-ictal confusion. A report of 'tap dancing in epilepsy' showed a video with simultaneous electroencephalography in which rhythmic leg and foot tapping \[[@CR2]\]. However, there was only slow wave activity, which indicates "post-ictal" dancing rather than "ictal". In a recent report of a patient with left temporal lobe epilepsy \[[@CR3]\], the authors provided a video with simultaneous electroencephalography but it could hardly be called dancing. Here, we present a case of dancing semiology in a patient with refractory temporal lobe epilepsy. Case presentation {#Sec2} ================= A 42-year-old right-handed woman suffered from weekly repetitions of unconscious dancing for 5 years, despite of multiple antiepileptic drugs including levetiracetam 3000 mg/d, valproate 900 mg/d, and pregabalin 300 mg/d. She was admitted to the epilepsy monitoring unit of a tertiary referral center for the feasibility of epilepsy surgery. On the initial clinical examination, she had no other symptoms or signs. The routine laboratory tests were negative. Magnetic resonance imaging showed right hippocampal atrophy (Fig. [1](#Fig1){ref-type="fig"}). Her habitual seizure was recorded during video-electroencephalography monitoring. The seizure began with the right hand automatism and ictal speech, which suggest that the ictal onset zone would be on the right side. An evolution of rhythmic delta activity was observed in the right temporal area beginning 16 s after the automatism (see Additional file 1: Video S1). As the ictal discharge spreads to the left temporal area, which means the secondary generalization, the ictal speech disappeared. After 20 s from the secondary generalization, she had rhythmical movement of her legs, similar to stepping through a dance, and the simultaneous video-electroencephalography showed regional slow waves over both of her frontal areas. When the ictal rhythm has switched to the left side, the left upper limb automatism and immobility of the right upper limb represented the rhythmic theta activity, which is still seen in the left temporal area. Taken together, we could identify the kicking and stepping like a dance as well as shaking left arm. According to her husband, the movement would have involved twirling dance, making a right turn when she was standing. However, it was not shown on the video. The dancing lasted even after the rhythmic discharge, which definitized the post-ictal dancing. We surmise that the "dancing" movement might be derived from some combination of automatism consisted of complex, rhythmic, and sequencial movement. She was completely amnestic with respect to the episode. Her ictal speech and the ictal electroencephalography imply that the right hippocampus atrophy should be the epileptogenic focus. There has been only one seizure for 3 months since a stereotactic gamma knife surgery applied to the atrophic right hippocampus.Fig. 1Magnetic Resonance Imaging. T2-weighted oblique coronal magnetic resonance imaging revealed right hippocampal atrophy Additional file 1:Simultaneous Video-electroencephalography Monitoring. Description of data: The seizure begins with right hand automatism and ictal speech, followed by an evolution of rhythmic delta activity in the right temporal area. After the secondary generalization, she showed rhythmical 2movement of legs, similar to stepping through a dance. Subsequent post-ictal paraphasia as well as confusion were documented. (WMV 12mb) The authors declare that they adhered to CARE guidelines/methodology. Conclusions {#Sec3} =========== This patient showed a semiology which can be classified as complex partial seizure followed by hypermotor seizure consisting of rhythmic movements of the extremities termed dancing \[[@CR4]\]. Although it is difficult to distinguish dancing from hypermotor semiology, the former connotes more complex, rhythmical and sequencial movement of multiple body regions. Here, we report the most plausible dancing in the ictal and post-ictal state, documented by simultaneous video and electroencephalography. The underlying mechanism and symptomatogenic zone remain to be investigated. Not applicable. Funding {#FPar1} ======= Not applicable. Availability of data and materials {#FPar2} ================================== All data analyzed during this study are included in this published article and its supplementary information files. Authors' contributions {#FPar3} ====================== KKT made substantial contributions to acquisition of data. KKT and LSK analyzed and interpreted the video-electroencephalography and magnetic resonance imaging. KKT and CK involved in drafting the manuscript and revising it critically for important intellectual content. LSK has decided to submit this manuscript for publication. All authors read and approved the final manuscript. Authors' information {#FPar4} ==================== Sang Kun Lee: Professor at Seoul national university hospital Department of Neurology, Comprehensive Epilepsy Center, Seoul national university hospital, Seoul, Republic of Korea Member of American Epilepsy Society The current Editor-in-Chief of Journal of Epilepsy Research and Journal of Korean Epilepsy Society Consulting Editor of Epilepsy Research Kon Chu, MD: Associate Professor at Seoul national university hospital Department of Neurology, Comprehensive Epilepsy Center, Seoul national university hospital, Seoul, Republic of Korea General Councilor of Korean Society for Stem Cell Research Member of Basic Science Committee of Korean Epilepsy Society Keun Tae Kim: MD Fellowship at Seoul national university hospital Department of Neurology, Comprehensive Epilepsy Center, Seoul national university hospital, Seoul, Republic of Korea Competing interests {#FPar5} =================== The authors declare that they have no competing interests. Consent for publication {#FPar6} ======================= Written informed consent for publication of the clinical details, clinical image and video was obtained from the patient. Ethics approval and consent to participate {#FPar7} ========================================== This study is in accordance with the Declaration of Helsinki. This report has been approved by the Ethics Board of Seoul national university hospital biomedical research institute. All authors agreed the publish statements of BMC Neurology.

1. Introduction {#sec1-ijerph-17-01813} =============== The current evidence-based paradigm requires that the most reliable scientific evidence is used to increase patient safety by predicting how a certain intervention might affect health and well-being. This is true both in a strictly clinical context, where doctors and patients need to evaluate the potentiality of interventions for the single case, and in public health, where policy interventions are needed at the population level. How should such evidence best be generated? This is an ongoing debate in medicine and public health, but also in philosophy of science. Recently, there has been some discussions about the epistemic primacy of statistical approaches, such as randomised controlled trials (RCTs), for establishing causality and predict the health outcomes of medical interventions. In the recent article 'Understanding and Misunderstanding Randomized Controlled Trials', for instance, Deaton and Cartwright summarise a number of arguments against the dominant reliance on RCTs over every other form of causal evidence \[[@B1-ijerph-17-01813]\]. This article has received considerable attention, partially because it is written within a growing movement that calls for evidential pluralism in the medical and health sciences \[[@B2-ijerph-17-01813],[@B3-ijerph-17-01813],[@B4-ijerph-17-01813],[@B5-ijerph-17-01813],[@B6-ijerph-17-01813],[@B7-ijerph-17-01813]\]. The main idea is that evidence from RCTs can provide useful causal information, but only when combined with evidence from other methods for causal inquiry. The purpose of causal evidence would then be to develop the theoretical knowledge and to advance the understanding of 'the hows and whys' underlying the statistical correlations. Scholars who argue that we need plural methods to establish causality use different types of arguments and have different emphasis. Some traditions, such as critical realism, are primarily critical of the ontological bias in scientific methodology; they argue that standard ways to evaluate causal evidence mistakenly rely on a positivist, Humean conception of the nature of causality \[[@B2-ijerph-17-01813],[@B8-ijerph-17-01813],[@B9-ijerph-17-01813]\]. Others are more concerned with epistemological issues of how we can know about causality using different methods, without considering the ontological question of what causality is \[[@B5-ijerph-17-01813],[@B10-ijerph-17-01813]\]. However, the ontological and epistemological traditions generally agree that what counts as 'the best causal evidence' differs depending on the research question and the research context. This stance is in opposition to a formal, probabilistic tradition which is primarily interested in the mathematical relationship between a certain hypothesis and a certain evidence, without considering the specific causal content or explanation to the evidence itself \[[@B11-ijerph-17-01813]\]. Although the formal approach is still dominant, there is increasing awareness among scientists that the evaluation of causal hypotheses cannot be restricted to purely statistical standards \[[@B12-ijerph-17-01813],[@B13-ijerph-17-01813]\]. In a recent commentary in *BMJ Evidence Based Medicine*, a large group of clinicians, medical researchers and philosophers made a common appeal to evidence-based medicine to expand its notion of 'evidence': "The rapid dominance of evidence based medicine has sparked a philosophical debate concerning the concept of evidence... The undersigned include 42 clinicians and philosophers from interdisciplinary research networks working specifically on questions related to causality in medicine worldwide. Our research has developed out of a conviction that philosophical analysis ought to have a direct impact on the practice of medicine. In particular, if we are to understand what is meant by 'evidence', what is the 'best available evidence' and how to apply it in the context of medicine, we need to tackle the problem of causality head on... Establishing causality often requires the use of multiple methods, since no single method will be universally applicable or perfect for this purpose\[[@B14-ijerph-17-01813]\] (p. 1)" How should this call for evidential pluralism be understood? How useful are the various methods available for causal inquiry? Further, how should different types of causal evidence be evaluated? Among philosophers of science, various answers to this question have been proposed. For instance, Russo and Williamson (2007) state that a causal claim must be supported both by evidence of difference-making from population studies and from evidence of the underlying mechanism, and that each of these approaches compensate for the weakness of the other \[[@B15-ijerph-17-01813]\]. Others are trying to expand formal approaches to the inclusion of different types of methods and the amalgamation of different types of evidence \[[@B16-ijerph-17-01813],[@B17-ijerph-17-01813]\]. In this paper, we propose our own version of evidential pluralism. This version is based on the philosophical idea that any type of scientific claim, including causal claims within medicine and public health, should seek to say something about a system's intrinsic properties, as well as their mutual influence and causal interaction. Our aim here is to introduce a framework to support scientists in developing appropriate methodological approaches for exploring causality. This is relevant for ensuring that all the evidence necessary for the safe treatment of individuals and populations is considered. 2. Dispositions and Science {#sec2-ijerph-17-01813} =========================== In her discussion about causal evidence and scientific methods, Cartwright urges that the whole scientific enterprise is about finding properties, or potentialities, which she calls *stable capacities* \[[@B18-ijerph-17-01813]\]. For instance, an RCT might show that, for a certain sample of population, aspirin relieves headache in more instances than a sugar pill does. This observation is interesting insofar as it allows us to make a claim about a capacity of aspirin, which is generated from its properties. Such capacity, we can say, then works as the truth-maker of causal claims: 'aspirin has a capacity to (causally) relieve headache'. What Cartwright calls 'capacities' are also commonly referred to as *causal powers*, or *dispositions*, with a very similar meaning \[[@B19-ijerph-17-01813],[@B20-ijerph-17-01813]\]. Despite some disagreements over philosophical details, proponents of capacities, powers and dispositions generally agree that these are properties or potentials of things or systems, that can become manifested under certain conditions. The philosophy of dispositions descends from an ancient tradition going back to Aristotle, but which has had a revival in recent decades. Indeed, in philosophy of science, there has been a tendency to shift the attention away from ideal and extrinsic laws of nature onto the intrinsic causal powers of things and systems \[[@B18-ijerph-17-01813],[@B21-ijerph-17-01813]\]. Shifting the attention to the dispositions, or properties, of a system corresponds to looking at the causal mechanism, the interactions and the dynamics at place. It is no longer sufficient to notice a lawlike regularity in nature, for instance, that holds under some ideal, normal or similar conditions. Rather, one now wants to understand the intrinsic dispositions and causal capacities of the system that allow such regular behaviour \[[@B20-ijerph-17-01813]\]. In other words, one is now seeking a deep causal understanding about *why* and *how* something might or might not happen. This contrasts with the investigation into *whether* and *how often* something happens, which is typically tested via statistical significance, relative frequencies, and so on \[[@B22-ijerph-17-01813]\]. There is a flourishing literature about the metaphysics of dispositions and powers (recent edited volumes include \[[@B9-ijerph-17-01813],[@B23-ijerph-17-01813],[@B24-ijerph-17-01813],[@B25-ijerph-17-01813],[@B26-ijerph-17-01813]\]). The question here is whether thinking in terms of dispositions can help the discussion about causal evidencing in medicine and public health, and if so, how? If we acknowledge that the health sciences should be concerned with understanding the underlying causal powers, capacities or dispositions of observable processes and events, then what are the specific advantages and disadvantages of different types of scientific evidence for this purpose? There has been comparative little discussion on these topics, at least to our knowledge. In what follows, we will address these questions in more detail. As the theoretical framework for our discussion, we adopt a particular idea within the literature on dispositions: the theory of causal dispositionalism, developed by Anjum and Mumford over the last decade (for a detailed overview of the theory, see \[[@B27-ijerph-17-01813]\]). The aim is to make the methodological implications of this specific philosophy of causality more explicit and show the relevance of considering philosophy of causality for medicine and public health. 3. Dispositionalism about Causality {#sec3-ijerph-17-01813} =================================== In the Anjum--Mumford dispositionalist theory, causes come from dispositions or 'causal powers'. What is a disposition in this specific view, and how exactly does it relate to causality? A disposition is, in this view, an *intrinsic* property, belonging to some particular thing, individual or process. That dispositions are intrinsic in this way is crucial for causality in medicine. For instance, an intervention cannot be said to cause an effect unless it has an intrinsic disposition toward the recovery. Dispositional properties can exist unmanifested. A woman can be fertile without ever becoming pregnant, but once fertility is manifested in pregnancy, causality has happened. From this philosophical perspective, causality requires more than a single disposition. In order to manifest its effect, the disposition must interact with a number of other dispositions, or 'mutual manifestation partners' \[[@B28-ijerph-17-01813]\]. These are reciprocal partners for the manifestation, such as the dispositions of the sperm, ovum, uterus, placenta, and so on, to initiate and maintain the pregnancy \[[@B29-ijerph-17-01813]\]. All dispositions will also have a degree of tendency, which can range from very weak to very strong. Something can be more or less robust, soluble, toxic, vulnerable, and so on. The workings of intrinsic properties could for example give a statistically weak tendencies, such as the tendency of oral contraception to produce thrombosis. But in some individuals, this disposition might be very strong. Still, no matter how strong the tendency is, a disposition cannot manifest alone or in isolation. Causal production is then, in this view, always a matter of complex interaction of multiple dispositions, many of which are represented by the background conditions. For instance, a patient must be an appropriate manifestation partner for the medical intervention to work, or else be a non-responder. But rather than seeing the patient as the background condition for the intervention to do its causal work, which is the standard view of causality, causal dispositionalism takes all contextual factors that influence the outcome as causes themselves. For instance, the outcome of behaviour-changing techniques in improving lifestyle of individuals will depend not only on the type of technique and the single treated individual, but also on what is already there in the person's context, such as her current life phase, health, occupation, social status and family situation. A cancer survivor might be more disposed to lifestyle changes and thus a better mutual manifestation partner for the intervention than someone with a good general health or low motivation. Individuals, and even whole populations, can be thought of as mutual manifestation partners for an intervention to do its causal work---for instance, when a social intervention is tested. Because the effect is a result of the complex interaction of different dispositions, the causal process will also be highly sensitive to context, according to this theory. What a disposition does in one context, together with a certain combination of mutual manifestation partners, will be different from what it does in another combination. For a causal dispositionalist, therefore, the default expectation should be that the same disposition will tend to produce different effects depending on which dispositions it interacts with. A biomedical example is the way in which thalidomide tends to cause teratogenicity, but only if taken when the embryo is in the first trimester in a pregnant woman \[[@B30-ijerph-17-01813]\]. A famous public health example is given by Cartwright, showing that the intervention of teaching nutrition to mothers tends to cause an improvement in the children's health. This did not however work in communities where the main person responsible for planning and preparing the family meals was not the mother, but rather the mother-in-law \[[@B31-ijerph-17-01813]\]. To sum up, causal dispositionalism defines causality as the result of the interaction of multiple intrinsic properties, or dispositions. 4. Detecting Dispositions {#sec4-ijerph-17-01813} ========================= If we understand causality dispositionally, as briefly described here, then how should this affect the way we study, establish, and evaluate causality in medicine? In dispositionalist terms, to establish a causal link between an intervention *I* and an effect *E* corresponds to establishing whether *I* has intrinsic dispositions that, in combination with other dispositions, can eventually produce *E*. Different types of methods will have some strengths and some limitations for the purpose of finding and establishing intrinsic dispositions, as we now go on to show. The methods discussed are listed in [Table 1](#ijerph-17-01813-t001){ref-type="table"}. Note that this is not meant as an exhaustive list of methods, but as representative enough to demonstrate that the dispositionalist theory suggests that we need more than one type of method to establish causality. Further, the purpose of this discussion is not to set some quality standards for the individual methods, but rather to offer a more general framework to show how the various methods can complement and inform each other in the causal inquiry. This paper is then presenting an ontologically informed argument in support of methodological pluralism, motivated by a philosophical theory of what causality is, rather than a purely epistemological point about how we can establish causality empirically \[[@B2-ijerph-17-01813]\]. 4.1. Experimentation with Lab Models {#sec4dot1-ijerph-17-01813} ------------------------------------ Lab experimentation is commonly used to investigate the causal mechanisms underlying correlations, clinical reports and various other types of observations. This is done in order to evaluate the biological plausibility of a causal claim. For instance, if a certain genetic mutation is correlated with increased vulnerability to a class of chemicals, one can investigate whether there is a biological explanation for such correlation by setting up experiments with lab models. This includes a large variety of techniques and models: in vivo, in vitro and ex vivo experimentation with animals, cell culture, microscopy, genetic manipulation and post-mortem examinations, to name only some. Experimentation with lab models has certain strengths for establishing dispositions. Specifically, they allow us to study the causal processes in detail and over time, which can give insights into what a particular disposition is and how it interacts with different manifestation partners and under a variety of conditions and contexts. On the other hand, lab experiments have some epistemic limitations from the dispositionalist perspective. Since a disposition will produce different effects in combination with different sets of other dispositions, we should not expect that an intervention will produce exactly the same effect under some controlled lab conditions as they do in the clinic. However, these contextual differences can be harnessed in an attempt to understand more about a disposition and its various potential manifestation partners for a particular effect. Again, this requires a reciprocal collaboration between basic research and the clinic \[[@B22-ijerph-17-01813],[@B32-ijerph-17-01813]\]. Some examples will help with illustrating these advantages and disadvantages for establishing dispositions. In colorectal cancer, the receptor for epidermal growth factor (EGFR) is highly expressed \[[@B33-ijerph-17-01813]\]. In lab experiments, the monoclonal antibody cetuximab was shown to bind EGFR with high affinity, which suggested that cetuximab might have a disposition to slow down the progress of colorectal cancer (by blocking EGFR) \[[@B34-ijerph-17-01813]\]. However, this disposition does not manifest in the clinic in patients with a mutation in the KRAS protein \[[@B35-ijerph-17-01813]\]. This indicated that KRAS is a necessary mutual manifestation partner for cetuximab to do its causal work, which again contributed to expand our causal knowledge of the molecular mechanisms by which cetuximab blocks the molecular pathways downstream of EGFR. Once we know why the effect did not happen without this disposition of the KRAS protein, this different set of dispositions can be modelled in the lab---for instance, using cells expressing mutated KRAS. The dispositional strength of lab experimentation is not restricted to the biological realm, but it can sometimes expand to causal mechanisms of a social type. One example is the investigation of the mechanisms underlying addiction. Lab animals helped clarify that social isolation, especially during early life stages, is a factor that can contribute to increased addictive behaviour in adult life \[[@B36-ijerph-17-01813]\]. 4.2. Patient Narrative {#sec4dot2-ijerph-17-01813} ---------------------- Narrative-based medicine was defined as 'a fundamental tool to acquire, comprehend and integrate the different points of view of all the participants having a role in the illness experience' \[[@B37-ijerph-17-01813]\]. The attention to patient narrative as an essential tool for medicine has been increasingly emphasised in the last 30 years \[[@B38-ijerph-17-01813]\]. The main message of this movement is that medical care should shift its primary focus away from problem solving and toward understanding, but that this can only be done by considering patient's stories. One thing is that listening to patient narratives is considered essential to understand their needs and values \[[@B39-ijerph-17-01813]\]. Another point, more relevant for causal inquiry, is that patients can provide essential elements of the causal story, through the subjective narration of their experience \[[@B40-ijerph-17-01813],[@B41-ijerph-17-01813]\]. By listening to the story that a patient tells, and to the way in which she chooses to tell it, one can map detailed contextual information that could contribute to explaining the patient's current condition. How does this work? Let us illustrate this with an example from general practice. A patient with a long history of unsolved chronic muscle cramps, symptoms of epilepsy, eating disorders, anxiety attacks and self-harm, was further harmed by adverse effects from long-term medical treatments and surgery. For decades, medical examinations could not uncover the aetiology of her complex condition, until analysis of her narrative, previously overlooked, indicated an important role of childhood trauma, violence and abuse \[[@B42-ijerph-17-01813]\]. This type of patient-based evidence has been key to acknowledging cumulative childhood stress as a plausibly strong disposition toward immune dysfunction and other chronic conditions in adulthood \[[@B43-ijerph-17-01813],[@B44-ijerph-17-01813],[@B45-ijerph-17-01813],[@B46-ijerph-17-01813]\]. Because of this and many similar stories, there is a growing interest in the way that patient experiences can serve as causal evidence and in how this evidence should be evaluated. We see some clear dispositional strength in considering evidence from patient narratives. Rich patient information allows the detection of relevant dispositions that are uniquely combined in this patient. This could be information of a biomedical or biographical nature or concerning lifestyle or life situation. By considering these, one might get a better idea of the causal story leading up to their condition, as well as causally relevant information about what might be the best treatment for this particular individual. On the other hand, there are some limitations from the dispositionalist perspective. It is not easy to establish which of the many elements in the narrative that represent relevant dispositions for the illness or treatment. Indeed, if one does not have additional theoretical knowledge about what factors could influence the condition or treatment in general, one might not even know what to notice in the narrative as potentially informative. This is a problem for all the methods based on in-depth analyses of one or a few instances. An example can be found in the practice of pharmacovigilance: the post-marketing monitoring of secondary effects of drugs. The aim of pharmacovigilance is to build a hypothesis of causality between a drug and a symptom, based on spontaneous reports of patients' experiences. The WHO Collaborating Centre for International Drug Monitoring, Uppsala Monitoring Centre, has the world's largest database of spontaneous reports of suspected adverse drug reactions. They use this database to assess whether the reported reactions were in fact caused by an intrinsic disposition of the drug. Although pharmacovigilance is largely based on statistical analysis of such databases, a key step in the generation of a causal hypotheses is to assess the likelihood of a causal relationship based on a single or a few patient reports. In the words of pharmacovigilance expert, Rebecca Chandler, '*it is logical that clinical stories contained in adverse event reports \[...\] are integral to the development of hypotheses of drug safety concerns*' \[[@B47-ijerph-17-01813]\] (p. 23). In order to catch which of the details in the patient stories point to relevant causal dispositions, however, the clinical narration must be read through the lenses of some kind of theoretical knowledge. This could, for instance, be knowledge about the drug's components, their mechanism of action, side effects of similar drugs or pathological mechanisms underlying the reported symptom. The practice of pharmacovigilance is thus an excellent example of how using databases of patient narratives, in combination with theoretical knowledge and clinical expertise, can be effective for detecting causal dispositions. 4.3. Case Studies {#sec4dot3-ijerph-17-01813} ----------------- A case study is the detailed description and analysis of one single instance of a certain type of phenomenon. In medicine, it typically consists of the analysis of one medical case that is of particular interest---for instance, if it is new, rare and unusual, or somehow contributes to the advancement of medical understanding. For the purpose of public health, case studies of a particular context are often used to gain insights into the mechanisms underlying observed correlations. For instance, in order to understand the causal relationship between loneliness, social isolation and general health in older adults, case studies were conducted in specific contexts \[[@B48-ijerph-17-01813]\]. In medicine and public health, there is disagreement about the epistemic role of such a type of study. One view, which is dominant in the evidence-based paradigm, is that a case study provides qualitative knowledge only about the analysed context and can therefore not be used to derive general causal claims. In contrast, others argue that a detailed case study might have the potential for generalisation across similar instances. According to this view, depending on the research question, generalising from the detailed understanding of one instance can be equally justified (or equally wrong) as generalising from the statistical analysis of a few parameters in a larger population \[[@B49-ijerph-17-01813],[@B50-ijerph-17-01813]\]. This latter point is in line with dispositionalism: since dispositions are intrinsic properties, they tell us something general about the potentiality of a thing or a system, even if they are rarely manifested. Accordingly, if one knows how the disposition manifests (in which contexts and in combination with which manifestation partners), this knowledge can potentially be applied beyond the single case. For the purpose of uncovering intrinsic dispositions, case studies are important tools. Indeed, they typically include a variety of scientific methodologies to gain a full overview of the context under scrutiny. A well-conducted case study can converge medical examinations, lab analyses, medical history, qualitative interviews, analysis of social context and of lived experiences, and can therefore be analysed from the perspectives of different experts. This detailed and complex analysis aims to *understand* how the dispositions at place interact with each other. Moreover, it allows the description of rare or unexpected effects in a particular individual or context, suggesting previously unknown dispositions or mutual manifestation partners. This in turn can contribute to theory development of causal mechanisms. On the other hand, case studies have some limitations for the purpose of detecting dispositions. Specifically, the information in the case study is already defined, filtered and interpreted, restricting the causal information gathered from the instance analysed. This is especially relevant in medicine, where case studies translate the narrative and perspective of the patient into medical terms and categories. In this process, the researcher filters the information that is considered relevant, in light of their knowledge, interest and research question. When we miss the plurality of perspectives, we might also miss important information about the dispositions at place in the single case. This can be exemplified with a case study in the field of medical anthropology, aimed at investigating the causal relationship between the high incidence of illness in a low income African-American neighbourhood and pollution from surrounding industries \[[@B51-ijerph-17-01813]\]. With this deep analysis, the author not only draws a general conclusion about the causal relationship between the race and class experience and the susceptibility to environmental risk. She also points out that, without a deep understanding of the single case, we risk missing and misinterpreting crucial causal evidence: "...*in collecting surface soil samples, testers had actually sampled new dirt that residents had imported and put over their old, contaminated dirt... 'They sent out some people to do that testing out here and they scooped a little bit of dirt with spoons on the ground. Hey, I done put dirt on top of dirt trying to get rid of the floods and things we been having out here for years'*.\[[@B51-ijerph-17-01813]\] (p. 117)" There are, however, also case studies that are more biomedical in nature and that can potentially be of great importance for discovering new and unexpected dispositions. Consider, for instance, the infamous case reports of malformations after embryonic exposure to the drug thalidomide. The study of embryonic malformation in cases of maternal exposure to thalidomide suggested that certain tissues of the developing embryo were, contrary to previous assumptions, manifestation partners for the drug. These reports led to investigations that ultimately resulted in new theories of teratogenicity, safe placental barrier, and sensitivity in different species. This in turn motivated mechanistic studies, uncovering, for instance, the disposition of thalidomide to inhibit angiogenesis and, consequently, to treat myeloma \[[@B52-ijerph-17-01813]\]. 4.4. Case Control and other Retrospective Studies {#sec4dot4-ijerph-17-01813} ------------------------------------------------- Case control and retrospective studies are commonly used in medicine and epidemiology. In case control studies, one considers a group of people with a similar condition and tries to find a factor which is common to all the participants' medical history, but absent in a control group. This means that one is looking for a common cause of social, medical or toxicological type, that might explain the difference in outcome between the test group and the control. The dispositionalist advantage of this study design is that, while looking for a factor that supposedly made a difference between two samples of the same population, it allows a detailed causal study of small samples, such as outlier cases or very rare conditions. In this way, one might detect dispositions and manifestations that are otherwise too rare to show up through other statistical approaches. Such studies also acknowledge the temporal aspect of causality---that dispositions might take a long time to manifest as a detectable outcome. This could in turn contribute to discover a causal mechanism. The downside, on the other hand, is that these studies do not specifically look for manifestation partners that might contribute to the outcome but are not common to all the participants. Indeed, they are designed to find common dispositions across different contexts. We also need to point out that the output of these studies are correlations and not dispositions. Without a further theory of mechanism that can plausibly account for how the common factor brings about the common outcome, we cannot claim to have found an intrinsic property. This point is not only valid for this experimental design, but for all controlled experiments that are primarily designed for detecting correlations between variables at a statistical level. There is, for instance, a correlation between the exposure to paracetamol and incidence of asthma in children \[[@B53-ijerph-17-01813]\]. This provides nothing more than a suggestion that paracetamol disposes toward asthma. The correlation could be due to the fact that exposure to paracetamol and asthma have a common cause---for instance, chronic airway inflammation and infections. As others have already noted \[[@B1-ijerph-17-01813]\], such risk of confounding can be reduced with certain experimental designs, but never eliminated. Even a correlation produced by the most sophisticated population trial provides an indication, but never the ultimate evidence, for an intrinsic disposition. An example of the discovery of a causal disposition from a case control study is the following. Between 1966 and 1969 at the Vincent Memorial Hospital, Massachusetts, eight young women were found to have a rare form of vaginal adenocarcinoma. This observation triggered a retrospective comparative study, which showed that seven of the patients had been exposed to the exogenous hormone stilbestrol during embryonic life \[[@B54-ijerph-17-01813]\]. This suggested that the drug disposed toward an anomaly of the vaginal epithelium, which in time could result in cancer. The evidence for this disposition was then strengthened with lab experimentation, and provided new knowledge about the mechanism by which the drug affects the development of vaginal epithelium \[[@B55-ijerph-17-01813]\]. 4.5. Cohort Studies and other Prospective Studies {#sec4dot5-ijerph-17-01813} ------------------------------------------------- In cohort studies, a large group of people is observed over a period of time, in which relevant medical events and exposures of every participant are recorded in databases. Such data are then used to answer different research questions about existing correlations between possible risk factors or susceptibility factors and health outcomes, at the population level. Like retrospective studies, cohort studies also acknowledge the temporal aspect of causality, which is central if one thinks in terms of intrinsic dispositions and their manifestations. One could argue that, since prospective studies record events as they happen over time, rather than being based on memories and existing records, such studies account better for the temporal dimension than retrospective studies. It is, however, not easy to determine how much time an unknown disposition will take to manifest. In order to make a good estimate of this, one must have a great deal of background knowledge and mechanistic understanding of the processes in place, so that the study is not discontinued before it is possible to make the relevant observations. Some dispositions might take generations to develop, as was the case with stilbestrol and vaginal cancer, while others can manifest within seconds. In addition to the temporal dimension, another dispositional advantage of cohort studies is that they have the potential to address causal complexity and the interactions of different manifestation partners. By broadly monitoring development in a cohort over a long time, one could detect the interdependency of a certain health outcome with numerous factors. In principle, these observations could allow the study of causal complexity by pointing to intrinsic dispositions (e.g., genetics), extrinsic dispositions (e.g., environmental stressors) or even their mutual interactions. Again, since these observations are made at the population level, they just indicate causal complexity, but provide per se no understanding of it: of how and why intrinsic properties interact and manifest. Still, in combination with other approaches, such as deep contextual analyses of single cases, the evidence produced in cohort studies can potentially be a good starting-point for the detection of dispositions. The dispositionalist downside of this experimental design is that when resources are limited, priority tends to be given to large numbers of people over large numbers of factors monitored. This takes away a great potential of this study design for uncovering causal complexity of dispositions. Normally, research on databases from cohort studies tends to point to the interrelation of a few rather than many factors, normally just two or three. Cohort studies can be used to discover both biomedical and psychosocial dispositions. For instance, some cohort studies found a positive association between an increased consumption of processed meat and an increased incidence of colorectal cancer \[[@B56-ijerph-17-01813]\]. However, this evidence is not conclusive and is only suggestive of an intrinsic disposition of processed meat to contribute to cancer development until further evidence is available. For instance, experimental studies in animals might indicate dose relatedness, reversibility or the biological mechanisms underlying this correlation \[[@B57-ijerph-17-01813]\]. In turn, results from animal studies can inform further clinical studies. Another cohort study is the Norwegian Trøndelag Health study (HUNT), which so far has had over 125,000 participants since 1984 (<https://www.ntnu.edu/hunt>). In this case, a large range of factors were monitored, with an initial aim to study risk factors of cardiovascular conditions, metabolic conditions and quality of life. Because of this unusually broad approach, researchers have been able to use the study database to uncover a considerable correlation between self-reported childhood trauma and adult comorbidity \[[@B46-ijerph-17-01813]\]. That this correlation points to a disposition is supported by mechanistic evidence of how allostatic load can lead to systemic chronic inflammation \[[@B58-ijerph-17-01813]\]. Other supporting evidence comes from qualitative, phenomenological studies using patient narratives \[[@B42-ijerph-17-01813]\]. 4.6. Randomised Controlled Trials (RCTs) {#sec4dot6-ijerph-17-01813} ---------------------------------------- The purpose of a randomised controlled trial is to compare the outcome of one or more randomly assigned clinical intervention(s). This is considered the gold standard for causal evidencing in medicine and evidence-based decision making. The reason for this is that it is specifically designed to look for difference makers, while controlling for confounding factors. This means that any difference in outcome between the test group and the control group should be caused by the tested interventions, since all other differences should be homogenously distributed between the two groups. The exclusive reliance on RCTs has been critically debated in recent decades, both from philosophy of science and from the research community (see, for instance, the special issue of *Social Science and Medicine* 'Randomized Controlled Trials and Evidence-based Policy: A Multidisciplinary Dialogue', August 2018). Here, we only consider the strengths and limitations of RCTs for establishing intrinsic dispositions and how they interact with other mutual manifestation partners. An obvious strength of RCTs from the perspective of causal dispositionalism is that dispositions tend to make a difference to the outcome \[[@B2-ijerph-17-01813]\]. This means that comparative methods, and particularly RCTs, are best suited for picking out this feature of dispositions. Statistically significant results from an RCT could indicate that the intervention played a causal role for the outcome, either as an intrinsic disposition or as a necessary background condition (a sine qua non). The latter would be where something that was necessary for the effect, although it did not as such cause the effect. Hypothetically speaking, if one had no understanding of the underlying biological mechanisms, one might, for instance, find that hysterectomy significantly reduces the risk of unwanted pregnancy and take this to mean that the uterus is the cause rather than a necessary condition for pregnancy. This shows how what is the advantage of RCTs from a dispositionalist perspective is also the reason why they cannot produce dispositional evidence on their own. Moreover, RCTs are less suited for identifying mutual manifestation partners for a disposition, since the focus is on one or a few particular interventions for which there can be a control or comparison, and on one or a few outcomes. The experimental design is intended to minimise complexity---for instance, through strict inclusion and exclusion criteria \[[@B2-ijerph-17-01813],[@B59-ijerph-17-01813],[@B60-ijerph-17-01813]\]. From a dispositionalist perspective, heterogeneity of the outcome could point to important interactions and manifestation partners \[[@B22-ijerph-17-01813]\]. This type of causal knowledge must thus be found via other methods, such as lab experiments, case control studies, or observational studies. An example can illustrate this. Large randomised controlled trials showed a convincing association between the use of statins as a pharmacological treatment of hypercholesterolemia and a lower incidence of cardiovascular mortality \[[@B61-ijerph-17-01813]\]. These studies, combined with the known mechanism by which statins interfere with cholesterol biosynthesis and thereby with lipid metabolism, provided good evidence that statins have an intrinsic disposition to prevent heart disease by lowering the concentrations of lipids in the blood. However, the response to this type of therapy was highly heterogeneous \[[@B62-ijerph-17-01813]\], which points to a number of unknown manifestation partners, and unknown dispositions of the drug. Understanding such complexity requires a different type of evidence---for instance, pharmacogenetics observational studies analysing the genetic underpinning to the heterogeneous drug response \[[@B63-ijerph-17-01813]\]. 4.7. N-of-1 Trials {#sec4dot7-ijerph-17-01813} ------------------ In an N-of-1 clinical trial, several different interventions are tested in random sequence in the same patient or context, and the outcomes are compared. The aim of this type of experiment is to find which treatment is correlated to the best outcome in the single case, especially for complex, context-sensitive and multi-factorial treatments and conditions. For instance, the correlation between type of diet and symptoms of an illness can be tested with N-of-1 trials, with the advantage that the patient works as her own control. Having a perfectly matching control is important, since the response to a dietary change is interrelated to a number of other known and unknown individual variables. N-of-1 trials are good tools for establishing difference making in a unique context. Difference making, we said, can be a good indicator of dispositions, since dispositions tend to make a difference. This is because of their comparative design and because repeatable interventions can be randomly allocated, sometimes even blindly. N-of-1 trials therefore have some epistemic advantages for establishing dispositions, especially because they allow us to monitor the effects of different interventions within the same patient. This means that these studies can help reveal how different dispositions manifest in interaction with a particular patient and their unique combination of dispositions, even if this specific combination of mutual manifestation partners is never repeated in another patient. There are, however, also some epistemic limitations. Further, this type of experiment, taken in isolation, only establishes a correlation. To move from this to claiming that the correlation is an expression of an intrinsic disposition, we will need additional theoretical knowledge and high quality patient-based evidence. An N-of-1 trial could, for instance, be used in cases where we seek to understand which factors could influence the intensity and frequency of symptoms, such as migraine attacks, epileptic seizures, cramps or episodes of memory loss. Although different people might share the same disposition toward migraine, the triggers and the vulnerability to these will typically vary from one individual to another. Such individual variation, which is essential in causal dispositionalism, is not a problem in these trials, since they already consider the most relevant 'subpopulation'. Still, any such individual outcomes will remain correlations until a further plausible mechanism of action is provided. For example, N-of-1 trials were used in single patients with osteoarthritis to compare the effectiveness of paracetamol and ibuprofen to tailor treatment to the individual patient \[[@B64-ijerph-17-01813]\]. While this was useful for choosing the best treatment (the correct manifestation partner) for the patient's unique set of dispositions, patients were also encouraged to monitor other factors that, in interaction with the drugs, seemed to affect the health outcome (eating habits, etc.). This is valuable for clarifying whether the therapeutic disposition belongs to the drug or to some other contextual factor, or both. 5. Combining Evidence for Establishing Dispositions {#sec5-ijerph-17-01813} =================================================== We now summarise the advantages and disadvantages of the reviewed research methods for the purpose of detecting dispositions and manifestation partners, as well as indicating the corresponding recommendations, in the following table ([Table 2](#ijerph-17-01813-t002){ref-type="table"}). Looking at the strengths and limitations of each of these methods, we can see that it is necessary to combine them for the purpose of causal enquiry and evidencing. That said, our analysis also shows why the different methods would have to play different roles at different stages in this process. Specifically, we have seen that, while all types of evidence contribute to detect dispositions and their manifestation partners, some types of evidence play an indispensable role for establishing their intrinsicality. This allows us to support a pluralist methodology while at the same time offer a way to prioritise which methods to use at different stages. In light of the dispositionalist theory, as outlined above, we thus propose the following approach to causal inquiry and evidencing in medicine and public health ([Figure 1](#ijerph-17-01813-f001){ref-type="fig"}): In [Figure 1](#ijerph-17-01813-f001){ref-type="fig"}, we identify three stages of the causal discovery process (blue boxes): (i) observed phenomenon; (ii) hypothesis for new dispositions and manifestation partners; (iii) established dispositions and manifestation partners. In order to proceed from one stage to the next, certain types of evidence are necessary but not sufficient in every type of causal inquiry (all the green boxes). Among all types of causal evidence, these are the most useful for establishing intrinsicality. Moreover, some additional supporting evidence is still required to establish a disposition (some but not necessarily all of the yellow boxes). Different types of supporting evidence can be more suitable in different cases, depending on what is most relevant for the type of research question and causal hypothesis and on what knowledge is already available. In the process described in [Figure 1](#ijerph-17-01813-f001){ref-type="fig"}, we first start from an observed phenomenon (stage 1), either in an individual or in a population. To arrive at a new hypothesis (stage 2), one needs to first gain a deeper understanding of the observed phenomenon in its context. This can be done by collecting evidence from qualitative research (e.g., ethnographic investigation) or case studies. Additionally, one must collect the existing knowledge, for example, through literature review, expert consultations or communication with other stakeholders. The hypothesis should include a disposition and its manifestation partners, which can then be tested with experimentation or further observation, or both. Testing with experimentation could be done in lab models or in populations (e.g., randomised controlled trials, N-of-1 trials). Testing with further observations can be done in population studies (e.g., case control, cohort studies) or by increased clinical alert (e.g., patient narrative). Once a disposition and its manifestation partners are established, or at least corroborated, causal inquiry is still not finished. Indeed, some of the most important information for broadening the causal understanding will come from observed cases of causal failure, in which the disposition does not manifest in the expected way. By collecting and investigating such unexpected outcomes and outlier cases, new manifestation partners and new dispositions might be detected. We have described some such cases above, in [Section 4](#sec4-ijerph-17-01813){ref-type="sec"}. 6. Concluding Remarks {#sec6-ijerph-17-01813} ===================== From a dispositionalist perspective, establishing that an intervention causes an effect requires three forms of understanding, which should ultimately increase patient safety. First, we need to understand whether the intervention actually has an intrinsic disposition to produce the effect. Second, we need to understand which other dispositions or mutual manifestation partners played or could play a causal role in the process, either contributing to or counteracting the effect. Finally, we need to understand how these different dispositions interacted. This means that causal investigation must aim at theory development and mechanistic knowledge. In conclusion, we wish to anticipate one possible objection to our proposal. Such objection resonates an argument typically used in defence of the current evidence-based paradigm and can be summarised as follows. Is it feasible to prioritise the advance of theoretical knowledge, given the scarcity of resources for medical research, the huge costs and the amount of practical challenges that need to be addressed? Should medicine not be satisfied with finding solutions to problems, rather than using resources to investigate how and why these solutions work? This reasoning is built on the assumption that it is possible to make good medical research, aimed to answer the 'whether' and 'how often' questions, without the knowledge of 'how' and 'why'. Such an assumption has been extensively addressed and challenged \[[@B2-ijerph-17-01813],[@B10-ijerph-17-01813],[@B13-ijerph-17-01813]\]. Theoretical understanding of medical observations and correlations is necessary to plan research, design experiments, interpret results, evaluate and weight experimental designs, and to use research findings in clinical decision making. A trade-off that de-prioritises theory building and understanding of phenomena, therefore, would ultimately result in an overall decrease in quality of medical research in general, and in an inefficient use of resources. Thus, the correct use of tools such as statistical significance, analysis of frequencies and statistical models ultimately depends on our theoretical understanding of human physiology, pathology, health and illness (see \[[@B13-ijerph-17-01813]\] for a detailed argument). We have here tried to show why such knowledge cannot be obtained by using one single methodological approach, but that different methods are useful for picking out different aspects of causality in medicine and public health. The authors contributed equally to the conception, development and drafting of this work. All authors have read and agreed to the published version of the manuscript. This work was financed by the WHO Collaborating Uppsala Monitoring Centre for International Drug Monitoring. The authors declare no conflicts of interest. {#ijerph-17-01813-f001} ijerph-17-01813-t001_Table 1 ###### A list of the research methods reviewed here, and an overview of their general purposes for the practice of causal inquiry. Research Method Purpose -------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Experimentation with lab models** To find causal mechanisms underlying observations **Analysis of patient narrative** To map detailed contextual information from the patient's story that could contribute to explaining the patient's current condition **N-of-1 trials** To test and compare the effects of different interventions within the same patient or context **Case studies** To analyse a medically interesting case or case series in sufficient detail, including relevant biographical data, social conditions, symptoms, diagnosis, management, and outcomes **Case control and other retrospective studies** To find a common cause in a population with a similar condition by comparing recorded medical data against a control **Cohort studies and other prospective studies** To find correlations between possible risk factors or susceptibility factors and health outcomes by observing a group of people over a period of time **Randomised controlled trials** To compare the outcome of one or more randomly assigned clinical intervention(s) ijerph-17-01813-t002_Table 2 ###### Summary of the advantages and disadvantages of the different research methods for the purpose of uncovering causal dispositions and manifestation partners, with corresponding recommendations. ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Research Method Advantages for Detecting Dispositions Disadvantages for Detecting Dispositions Recommendations ---------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Experimentation with lab models Allow a study of causal processes in detail over time, including many dispositions and manifestation partners.\ The dispositions and manifestation partners studied in the lab are likely to interact differently in the context of application (the clinic). Couple experimentation in lab models with clinical observations (especially where a disposition fails to manifest as expected) to gain understanding about a disposition and its various manifestation partners for a particular effect. Can inform causal mechanism by offering insight into interactions of manifestation partners under various controlled conditions. Analysis of patient narrative Allow the detection of relevant dispositions that are uniquely combined in the patient.\ As with all in-depth analyses of one or few instances, it is difficult to narrow down which of the many elements in the patient narrative represent relevant dispositions for the illness or treatment. Couple analysis of patient narratives with population studies and theoretical knowledge about what factors could influence the condition or treatment in general. Facilitate the understanding of the causal story leading up to a condition, allowing a detection of various relevant dispositions and manifestation partners. Case studies Help understand how the dispositions at place interact with each other, by gaining a detailed overview of the context under scrutiny.\ Plurality of perspectives is missed because information is organised into medical categories, filtered and interpreted in light of researchers' knowledge, interest and research question.\ Include patient narratives for a qualitatively rich description of the case in order to include all the potentially important causal information.\ Allow the description of rare or unexpected effects in a particular individual or context, suggesting previously unknown dispositions or mutual manifestation partners. Information about the dispositions at place in the single case might be lost. Couple cases studies with existing theoretical knowledge and experimentation in order to exclude irrelevant dispositions. N of 1 trials Help reveal how different dispositions manifest in interaction with one particular patient and their unique combination of dispositions and manifestation partners. Establish only a correlation and not an intrinsic disposition. Couple N of 1 trials with theoretical knowledge and high quality patient-based evidence in order to investigate whether the correlation might point to an intrinsic disposition. Case control and other retrospective studies Allow detection of rare dispositions through a detailed causal study of small samples, such as outlier cases or very rare conditions.\ Look only for common dispositions across different contexts and not for context-specific manifestation partners.\ Couple retrospective studies with experimentation in lab models and review existing theoretical knowledge to investigate whether the correlation might point to an intrinsic disposition. Acknowledge the temporal aspect of causality, which in turn contribute to discover a causal mechanism. As with all statistical studies in populations, these studies point at correlations which are not necessarily intrinsic dispositions. Cohort studies and other prospective studies Acknowledge the temporal aspect of causality.\ Difficult to predict how long a disposition will take to manifest at population level.\ Couple prospective studies with case studies, patient narratives and mechanistic evidence to investigate whether the correlation might point to an intrinsic disposition. Can potentially address causal complexity and the interactions of different manifestation partners. When resources are limited, priority is given to large numbers of people over large numbers of factors monitored, hindering the potential for uncovering causal complexity. Randomised controlled trials Well-suited to detect a difference in outcome at population level, which dispositions tend to do (but not always). Identify all difference-makers irrespectively of whether they are causes or simply necessary conditions for the outcome.\ Couple RCTs with lab experimentation, case control studies and case studies to investigate causal interactions and manifestation partners. Not well-suited to identify multiple manifestation partners (causal complexity). -----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Background ========== Schizophrenia is a severe mental disorder with grave personal and social costs \[[@B1]\]. Approximately 1% of the population develops schizophrenia during their lifetime. Over the years, many genes have been reported to be responsible for the susceptibility to schizophrenia \[[@B2]\]. In general, schizophrenia is considered to be a complex disease with multiple genetic and environment etiological factors. Linkage analysis, association and positional cloning studies and candidate gene approaches \[[@B3]\] have been successful in identifying risk genes. The way in which multiple genes, each possibly having a small individual contribution, leads to vulnerability and then the pathophysiology, remains to be elucidated. In order to figure out the relationship among those genes, we should investigate not only in gene-gene interaction level but also a whole picture at the protein level. Recent works to map the protein-protein interaction (PPI) in human to curate human metabolism and regulatory networks offer the relationships among different disease genes \[[@B4],[@B5]\]. The protein clusters in the network may represent the modules with biological functions \[[@B6]\]. It is also reported that if the disease candidate genes are treated as a phenotype, these genes are likely to be function together in the normal cell \[[@B7]\]. In this study, we provided a novel strategy by taking advantages of PPI to discover the regulatory mechanisms among disease candidate genes. We speculated that disease candidate genes may cluster together in a functional network at a protein level. Protein complexes interact with preferred partners to form a biological module serving a specific collective function \[[@B8]\]. When using a network-clustering method by calculating the pairwise distance in the protein interaction network \[[@B6]\], two major protein clusters were found which were involved in synaptic transmission and signal transduction protein cluster. We proposed a model to explain the interaction between NRG1 and CACNG2 which not only fell into the synaptic transmission cluster at protein interaction level but also associated at the gene-gene interaction level. Recent molecular studies implicate neuregulin1 (NRG1) as the most promising risk factor for schizophrenia \[[@B9],[@B10]\]. Liu and colleagues also found suggestive linkage evidence of schizophrenia to loci near NRG1 on chromosome 8p21 in an ethnically distinct Taiwanese sample \[[@B11]\]. There is also evidence that this genetic risk is elevated when accompanied by genetic changes in the gene for ErbB4, one of neuregulin\'s binding partners. NRG1-mediated ErbB signalling has important roles in neural development \[[@B12]-[@B14]\], as well as in the regulation of neurotransmitter receptors thought to be involved in the pathophysiology of schizophrenia \[[@B15]\]. Hahn and colleagues suggest that enhanced endogenous NRG1-ERBB4 signalling may be responsible for N-methyl-D-aspartate receptors (NMDARs) hypofunction of the disease state \[[@B16]\]. NMDA receptors are a major subtype of glutamate receptors and mediate slow excitatory postsynaptic potentials (EPSPs). Glutamate is the major excitatory neurotransmitter in the brain, and it has been proposed that disruption in glutamate signalling may underlie many of the symptoms of schizophrenia \[[@B17]\]. NRG1 reduces the tyrosine phosphorylation of NMDA receptors, a modification that is triggered by the binding of NMDA or glutamate. NMDAR hypofunction may contribute to the symptomatic features of schizophrenia \[[@B18]\]. ERBB4 associates with NMDAR via DLG4 (also called PSD95), and the binding to DLG4 is probably involved in the enhanced activation of ERBB4. This association provides a physical link between ERBB4 and the NMDAR. These findings add to our basic understanding of glutamatergic transmission, which has been implicated in the pathogenesis of schizophrenia. CACNG2, also known as stargazin, was found to interact directly with AMPA receptor and allow interaction of the receptor with the scaffold proteins of the postsynaptic density, such as DLG4 \[[@B19],[@B20]\]. In a previous linkage study of schizophrenia that included Taiwanese samples, CACNG2 was also reported as a vulnerability gene for neuropsychologically defined subgroups of schizophrenic patients \[[@B21]-[@B23]\]. Bats and colleagues found that a mutation in PDZ domain of CACNG2 will increase AMPA receptor diffusion. CACNG2 regulates trafficking of AMPA-type glutamate receptors and stabilizes them at the postsynaptic density when neurotransmitters are received \[[@B20]\]. Results and discussion ====================== Products of candidate disease genes form two major clusters in a schizophrenia-related protein interaction sub-network ---------------------------------------------------------------------------------------------------------------------- There are two major types of reactions which are complex formation and covalent modification in the signalling pathway. Both types of reactions have protein-protein interactions (PPI), which can be detected by high throughput methods. It has been shown that proteins which are involved in the same pathway, are likely to cluster together in the PPI network \[[@B8]\]. If the candidate genes may increase the risk of acquiring a disease synergistically, it implies that these genes are likely to work together in the normal cell \[[@B7]\]. Taking these two observations together, it is likely that the products of candidate genes may cluster together in a protein network. Therefore, we have collected 36 reported candidate genes from the literature and used them as a query set to retrieve the nearest neighbours of the candidate proteins. There were in total 831 human proteins retrieved from 6 major PPI databases (see Materials and Methods section). Interestingly, the retrieved interactions linked the gene products of these candidate genes in a big cluster even though these genes were found by a different approach (Figure [1](#F1){ref-type="fig"}). This result implies that the products of these candidate genes may be important candidates and they will work together in the cell. {#F1} In order to cluster proteins that are close to one another, the distances between every pair of these 831 proteins were computed by using a standard algorithm based on shortest-path of network topology \[[@B6]\]. On the basis of these pairwise distances, two major clusters (Figure [2](#F2){ref-type="fig"}) were found by using a visualization tool, called Generalized Association Plot (GAP) \[[@B24]\]. By examining the enriched gene ontology terms for the members of these two protein clusters \[[@B25],[@B26]\], the possible function of these clusters was identified. As shown in figure [1](#F1){ref-type="fig"}, the small cluster (cluster 1), which contains PPP3CC, NOTCH4, RASD2 and BMP6 genes, may be involved in signal transduction. The large cluster (cluster 2), which contains the NRG1 and CACNG2 genes, may be mainly involved in synaptic transmission and sometimes in neural development. NRG1 and CACNG2 were both reported as vulnerability genes from an association study of schizophrenia that included Taiwanese samples \[[@B11],[@B22]\]. These two genes were found to have strong interaction on the basis of linkage and association studies (unpublished data). Thus, it would be interesting to see how these functionally unrelated genes act synergistically in the development of schizophrenia. {#F2} Discovering the protein interactions between NRG1 with CACNG2 according to biological interpretationIn order to explore the detailed relation between NRG1 and CACNG2, the cluster 2 sub-network was extracted from the original large network for further study. This cluster contains 7 gene products of candidate genes, which content DPYSL2, CRTC1, DISC1, CACNG2, KCNJ12, PTK2B and NRG1, and 204 interacting proteins; part of this cluster is shown in Figure [3](#F3){ref-type="fig"}. The NRG1 protein is connected to CACNG2 protein via the ERBB and DLG protein families, which are known to be involved in glutamatergic signalling process \[[@B12]\]. Since cluster 2 proteins were retrieved by the nearest neighbor approach, this subset of proteins will definitely lose some interacting proteins excluded from the sub-network in the disease forming process. Hence, the second or even third neighbors may be needed to propose a biologically plausible mechanism. This step was done manually and the goal is to recover the proteins that may affect synaptic transmission. Therefore, membrane receptors that may link the function of NRG1 and CACNG2 were added to this sub-network. Two major protein families were added, which were the NMDA receptor subunits and the AMPA receptor subunits, respectively. Both receptors are calcium channels that are triggered by glutamate \[[@B27]\]. Thus, they have the potential to act synergistically. DLG4 protein is an important intermediate between NRG1 and CACNG2, because DLG4 protein is interacting with ERBB4, which may receive a signal from NRG1. On the other hand, DLG4 is interacting with both the NMDA and AMPA receptors. As a result, both receptors may receive the NRG1 signal from neighboring neuron cells. {#F3} Since functionally related genes are usually co-regulated, we went further to check whether these genes were co-regulated in the brain tissue. If disease is considered as a perturbation to the normal state, different brain tumors may perturb a given gene to different extent. The GSE4271 microarray data set deposited in Gene Expression Omnibus at National Center for Biotechnology Information contains 100 samples from 15 assigned subsets \[[@B28]\]. These gene expression data have been used to compute the correlation of the gene expression for each pair of genes and to establish a relevance network \[[@B29]\]. Since disease is treated as a perturbation, this disease sample-derived network actually represents the co-expression in relation to normal cells. As shown in Figure [4](#F4){ref-type="fig"}, NRG1, DLG4 and NMDA receptor (GRIN complex) genes showed strong correlation (correlation coefficients above or equal to 0.64). The fact that these gene pairs are correlated implies that their expressed proteins should also be functionally coordinated. The DLG4 gene has a reasonable correlation (coefficient = 0.45) with the CACNG2 and AMPA receptor (GRIA complex) genes. The correlation of gene expression between NRG1 and CACNG2 is less strong (coefficient = 0.36), because CACNG2 may not be the rate-limiting component in the downstream pathway. {#F4} A working hypothesis for interpreting the interaction between NRG1 and CACNG2 genes ----------------------------------------------------------------------------------- As described previously, DLG4 appears to be a hub, which receives the NRG1-ERBB4 signal and then relays the signal to the NMDA receptor and the CACNG2. Hahn and colleagues suggest that schizophrenia is marked by increased NRG1-ERBB4 signalling and may lead to further suppression of NMDA receptor function by reducing NMDAR tyrosine phosphorylation \[[@B16]\]. Yau and his colleagues demonstrated that the NMDAR in the synaptic sites of ErbB4-deficient mice were more abundant than that of wild-type control and releasing NRG to activate ERBB4 signalling will decrease synaptic activation of NMDA receptors \[[@B30],[@B31]\]. They suggested that this activated ERBB4 signalling will stimulate internalization of NMDA receptors in the synaptic sites. As a result, an increase in the NRG1-ERBB4 signal may either reduce the tyrosine phosphorylation of NMDA receptor or enhance the internalization of NMDA receptor in the postsynaptic neuron (Figure [5A](#F5){ref-type="fig"}). It is not clear whether these two mechanisms are exclusive at this point. The hypo-phosphorylation event may decrease the glutamate binding and consequently decrease the calcium influx. Alternatively, the internalization of receptor could be mediated by the enhanced interaction between DLG4 and internalized NMDA receptors. Therefore, less NMDA receptor will be available on the membrane and the capacity for cation influx (mainly calcium influx) will decrease. ![**A glutamatergic synapse focusing on the organization of DLG4 protein**. A) The increased NRG1-ERBB4 signaling may lead to hypophosphorylation of GRIN complex \[[@B16]\] or internalization of GRIN complex \[[@B30]\] in postsynaptic neuron. Less NMDA receptor will be available on the membrane and the capacity for cation influx (mainly calcium influx) will decrease. B) A direct interaction between CACNG2 and DLG4 mediates the synaptic delivery of GRIA complex \[[@B38]\]. The anchored CACNG2 may recruit the GRIA complex to the synaptic region \[[@B20]\] and increase the cation influx. Phosphorylation of DLG4 will release it from the cytoskeleton \[[@B32]\] and fail to recruit the GRIA complex efficiently. C) Taken together, DLG4 links two mechanisms to decrease the cation influx at the synaptic area. The presence of gene variations in both NRG1 and CACNG2 may thus create synergistic effect to affect the influx of Ca^2+^.](1471-2105-9-S12-S23-5){#F5} On the other hand, the interaction between CACNG2 and DLG4 may anchor CACNG2 on the cytoskeleton. The anchored CACNG2 may recruit the AMPA receptor to the synaptic region \[[@B20]\] and increase the cation influx. Phosphorylation of DLG4 will release it from the cytoskeleton \[[@B32]\] and fail to recruit the AMPA receptors efficiently (Figure [5B](#F5){ref-type="fig"}). Although the calcium-dependent tyrosine kinase PTK2B is interacting with the DLG4 protein (see Figure [3](#F3){ref-type="fig"}), it is not clear whether PTK2B is catalyzing this reaction. This enzyme has been shown to regulate the activation of calcium channels \[[@B33]\]. Because there are less AMPA receptors in the synaptic area, the capacity for cation influx (mainly calcium influx) will also decrease. Taken together, DLG4 links two mechanisms to decrease the cation influx at the synaptic area. The presence of gene variations in both NRG1 and CACNG2 may thus create a synergistic effect to affect the influx of Ca^2+^(Figure [5C](#F5){ref-type="fig"}). It has been shown that the schizophrenia patient has more NRG1-ERBB4 complex in the synaptic area than controls. This model specifically predicts that schizophrenia patients will have less cation influx in the synaptic area. Because both NMDA and AMPA receptors are triggered by glutamate, this model also predicts that glutamate may play an important role in pathogenesis. Many of the evidence for the glutamate hypothesis of schizophrenia implicate the NMDA-type glutamate receptor. But the glutamate role may be more complex because there are hints that AMPA receptors also contribute to schizophrenia symptoms, both independently or via effects on NMDA receptors \[[@B34]\]. Conclusion ========== DLG4, which receives the NRG1-ERBB4 signal and then relays the signal to the NMDA receptor and the CACNG2, links two mechanisms to decrease the cation influx at the synaptic area. On the basis of the protein interaction network, the NRG1-triggered NMDAR protein internalization and the CACNG2 mediated AMPA receptor recruiting may act together in the glutamatergic signalling process. Since both the NMDA and AMPA receptors are calcium channels, this process may regulate the influx of Ca^2+^. Ca^2+^is necessary for transmission at the neuromuscular junction and other synapses. Reducing the synaptic calcium influx due to variants of NRG1 and CACNG2 might explain the basis of schizophrenia. This PPI network analysis approach combined with the support from co-expression analysis may provide an efficient way to propose disease mechanisms for various highly heritable diseases. Materials and methods ===================== Constructing a protein network ------------------------------ Protein-protein interaction data were obtained from Integrated Protein Interaction Resource (IPIR, <http://ymbc.ym.edu.tw/ipir>). IPIR has integrated protein-protein interaction information from BIND, DIP, HPRD, MINT, MIPS and IntAct databases. In this case, we chose brain, cerebellum, cerebrum and nervous as tissue filter. By using 36 candidate proteins as a data set to look for its primary protein neighbours and secondary protein neighbours, there were 831 proteins retrieved from databases. This network is displayed by Cytoscape which provides basic functionality for a visual representation of the graph and integrated data \[[@B35]\]. Pairwise distance matrix by generalized association plots (GAP) --------------------------------------------------------------- For each biological network investigated, relevant proteins (nodes) and the interaction among them (edges) were assembled as follow. Each edge in the network was assigned a length of one. A pairwise distance matrix contains the length of shortest path between every pair of proteins in the network. Each distance in the matrix was shown as an \"association\", defined as 1/*d*^2^, where *d*is the shortest path distance. Generalized Association Plots (GAP) \[[@B24]\], which is a graphical environment for matrix visualization and information mining, were used to view this results. We have used the Gene Ontology annotations to assign functional category labels to the proteins of the PPI network \[[@B25]\]. GoMiner, a tool for biological interpretation of gene sets, was used to annotate the enriched gene ontology terms of these protein clusters \[[@B26]\]. GoMiner used the Gene Ontology (GO) to identify the biological processes, functions and components represented in gene lists. Calculating the correlation values among genes ---------------------------------------------- The microarray data was obtained from Gene Expression Omnibus (GEO) \[[@B36]\], and the accession number for the data series is GSE4271 \[[@B28]\]. Robust Multichip Average (RMA) normalization was performed to compute gene expression values for Affymetrix data and to carry out quality assessment using probe-level metrics \[[@B37]\]. After normalizing the microarray data, we used the Pearson\'s correlation, performed by a Perl module called \"Statistics::RankCorrelation\", to represent the correlation coefficient of each pair of probe sets. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= Ueng-Cheng Yang has designed the approaches and made a preliminary analysis. Pei-Chun Hsu has extended the analysis and jointly proposed the working hypothesis with Ueng-Cheng Yang. Kuan-Hui Shih has performed the co-expression analysis for brain tissue by using public microarray data. Hai-Gwo Hwu and his research team have provided the disease candidate genes and the domain knowledge for schizophrenia. Acknowledgements ================ We thank to Dr. Hsin-Chou Yang in the Institute of Statistical Science, Academia Sinica for helpful discussion and Dr. Chia-Huei Lee in National Health Research Institute for providing array-CGH data. This work was supported by grant NSC-96-3112-B-010-015 from National Research Program for Genomic Medicine, National Science Council (Taiwan). This article has been published as part of *BMC Bioinformatics*Volume 9 Supplement 12, 2008: Asia Pacific Bioinformatics Network (APBioNet) Seventh International Conference on Bioinformatics (InCoB2008). The full contents of the supplement are available online at <http://www.biomedcentral.com/1471-2105/9?issue=S12>.

1. Introduction =============== Many attempts have been made to use breeding or molecular biological methods to modify the ability to produce secondary metabolites in medicinal plants. Among the challenges being addressed, manipulations of the morphine biosynthesis in the opium poppy (*Papaver somniferum* L.), particularly the conversion of narcotic morphine to codeine, which is of high importance as an antitussive and a synthetic source of dihydrocodeine, or to thebaine, which is also an important starting material for the semi-synthesis of the analgesic oxycodone, will contribute to the control of narcotics, and to the supply of useful alkaloids for the production of pharmaceuticals. The gradual elucidation of enzymology of the alkaloid biosynthesis in *P. somniferum* led to genetical engineering of alkaloid biosynthetic pathway using native genes. The first report was on the introduction of a gene encoding berberine bridge enzyme (BBE) to *P. somniferum* in antisense orientation \[[@B1-pharmaceuticals-05-00133]\]. To date, several reports on metabolic engineering of *P. somniferum* have appeared, such as RNAi-mediated gene silencing of codeinone reductase (COR) \[[@B2-pharmaceuticals-05-00133]\], overexpression of COR \[[@B3-pharmaceuticals-05-00133]\], overexpression and antisense co-suppression of (*S*)-*N*-methylcoclaurine-3\'-hydroxylase (CYP80B3) \[[@B4-pharmaceuticals-05-00133]\], overexpression and RNAi-mediated gene silencing of salutaridinol-7-*O*-acetyltransferase (SalAT) \[[@B5-pharmaceuticals-05-00133]\], and RNAi-mediated gene silencing of SalAT \[[@B6-pharmaceuticals-05-00133]\]. Mutant poppy *top1* \[[@B7-pharmaceuticals-05-00133]\] which accumulates thebaine and oripavine as major alkaloids instead of morphine was also established by the treatment of mutagen (ethyl methanesulphonate) and screening of progeny plants. The T-DNA insertional mutant clone of *P. somniferum* PsM1-2, which we developed by the infection of the *Agrobacterium rhizogenes* strain MAFF03-01724, regenerated shoots from embryogenic callus that lacked the ability to produce morphine. Codeine was detected as a major alkaloid in this *in vitro* shoot culture \[[@B8-pharmaceuticals-05-00133]\]. By the improvement of the alkaloid analysis and proceeding studies on this mutant, thebaine (*ca.* 55 μg/g dry weight) and codeine (*ca.* 20 μg/g dry weight) were found to be the major opium alkaloids in the *in vitro* regenerated shoots \[[@B9-pharmaceuticals-05-00133]\]. The information provided from this mutant, which shows an altered alkaloid composition, might make an important contribution to the further modification of alkaloid production in *P. somniferum*, and therefore we carried out genetic and phenotypic analyses on this mutant. Recently, long unidentified enzymes involved in the two demethylation steps in the conversion of thebaine to morphine were successfully identified as non-heme dioxygenases \[[@B10-pharmaceuticals-05-00133]\]. These two enzymes, namely, thebaine 6-*O*-demethylase (T6ODM) and codeine *O*-demethylase (CODM), represent the first known 2-oxoglutarate/Fe(II)-dependent dioxygenases that catalyze *O*-demethylation. The altered alkaloid composition in the PsM1-2 mutant may be due to the genetic mutation in the conversion steps from thebaine to morphine. In the present study, an expression analysis of these two enzymes together with selected genes involved in the morphine biosynthesis was carried out to reveal the molecular mechanism of the mutation. 2. Results and Discussion ========================= 2.1. Morphological Characteristics of the PsM1-2 Mutants -------------------------------------------------------- The days to flowering, number of petals, appearance of split on the boundary of the petal, and height of the aerial part at the seed-filling stage of soil-cultivated T~0~ mutant and selfed progenies are summarized in [Table 1](#pharmaceuticals-05-00133-t001){ref-type="table"}. The T~0~ primary mutant showed delay of flowering and dwarfness. In addition, a deep split was observed on the boundary of the petal ([Figure 1](#pharmaceuticals-05-00133-f001){ref-type="fig"}). Delay of flowering was consistantly observed in the progenies. The number of petals, which was not altered at the T~0~, varied in the T~1~, T~2~ and T~3~ progenies. A deep split at the boundary of the petal was observed in 45% of T~1~ plants, 33% to 83% of T~2~ plants, and 8.3% and 10% of T~3~ plants. pharmaceuticals-05-00133-t001_Table 1 ###### Summary of the morphological characteristics of PsM1-2 T~0~ mutant, selfed progenies, and WT plant. Progenies Lines Number of Plants Days to Flowering (Mean ± SD) (days) Number of Petals: Percentage (%) Split on Petal Boundary (%) Plant Height (Aerial Part) (Mean ± SD) (cm) ----------------- ------- ------------------------------ --------------------------------------------------- ---------------------------------- ----------------------------- --------------------------------------------- T~0~ WT 1 47 \*^1^ 4: 100 0 60.0 T~0~ 1 71 \*^1^ 4: 100 100 38.0 T~1~ WT 6 53.5 ± 4.8 4: 100 0 42.4 ± 5.8 T~1~ 60 100.6 ± 14.6 ^\#\#\#\#^ 3: 1.7, 4: 41.7, 5: 35.0, 6: 16.7, 7: 3.3, 8: 1.7 45.0 52.1 ± 8.5 ^\#\#^ T~2~ WT 12 53.3 ± 4.0 3: 25.0, 4: 75.0 8.3 36.0 ± 7.6 \#1-27(HT) 15 90.8 ± 12.6 ^\#\#\#\#^ 5: 60.0, 6: 33.3, 10: 6.7 60.0 44.7 ± 5.4 ^\#\#^ \#2-17(HT) 6 79.8 ± 2.5 ^\#\#\#\#^ 5: 50.0, 6: 33.3, 8: 16.7 83.3 45.1 ± 3.4 ^\#^ \#2-1(LT) 12 83.3 ± 6.8 ^\#\#\#\#^ 5: 66.7, 6: 16.7, 7: 8.3, 8: 8.3 33.3 35.6 ± 7.8 \#2-6(LT) 10 76.4 ± 3.6 ^\#\#\#\#^ 5: 10.0, 6: 40.0,7: 30.0, 8: 10.0,12: 10.0 80.0 39.4 ± 3.1 T~3~ WT 6 109.4 ± 0.9 \*^2^ 4: 100 0 80.3 ± 5.8 \#1-27(HT)L\#2 10 129.2 ± 11.9 \*^3,\ \#\#\#^ 4: 40.0, 5: 50.0, 6: 10.0 10.0 45.8 ± 7.9 ^\#\#\#\#^ \#2-17(HT)\#2-1 12 131.1 ± 7.3 \*^4,\ \#\#\#\#^ 3: 8.3, 4: 75.0, 5: 16.7 8.3 47.0 ± 13.4 ^\#\#\#\#^ \*^1^: Days after transplanting; \*^2^: n = 5; \*^3^: n = 9; \*^4^: n = 11; ^\#^*p* \< 0.05; ^\#\#^*p* \< 0.01; ^\#\#\#^*p* \< 0.005; and ^\#\#\#\#^*p* \< 0.001 *vs.* WT. 2.2. Alkaloid Composition in the PsM1-2 Mutants ----------------------------------------------- The soil-cultivated PsM1-2 T~0~ primary mutant accumulated 16.3% (% dry weight) of thebaine as a major opium alkaloid in the latex, which was not detected in the WT ([Figure 2](#pharmaceuticals-05-00133-f002){ref-type="fig"}; [Table 2](#pharmaceuticals-05-00133-t002){ref-type="table"}). The morphine content in the mutant was 1.3%, which was *ca.* one tenth of that in the WT, and the codeine content was 4.2% in the mutant, *vs*. 1.3% in the WT. {#pharmaceuticals-05-00133-f001} {#pharmaceuticals-05-00133-f002} The alkaloid compositions in the dried opium of selected progenies are summarized in [Table 2](#pharmaceuticals-05-00133-t002){ref-type="table"}, and the morphine and thebaine contents of the T~1~, T~2~ and T~3~ plants are plotted on a scatter diagram ([Figure 3](#pharmaceuticals-05-00133-f003){ref-type="fig"}). The HPLC chromatograms of the representative lines of the T~1~ plants, WT plant, and authentic standards are shown in the [Supplementary Figure 1](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"}. pharmaceuticals-05-00133-t002_Table 2 ###### Opium alkaloid contents in PsM1-2 T~0~ mutant and selfed progenies. Progenies Lines Number of plants Morphine Codeine Thebaine Papaverine Noscapine ----------------------- ---------------- -------------------- -------------------- --------------------- -------------------- --------------------- ------------ T~0~ WT 1 10.9 1.3 nd \* 2.0 9.2 T~0~ 1 1.3 4.2 16.3 2.3 10.2 T~1~ WT 6 11.2 ± 4.0 2.6 ± 2.1 0.3 ± 0.2 2.4 ± 0.7 11.6 ± 4.3 T~1~ 60 6.3 ± 4.6 ^\#^ 3.8 ± 1.5 11.1 ± 6.1 ^\#\#\#^ 1.6 ± 0.5 7.9 ± 2.1 ^\#\#\#^ Selected lines (T~1~) \#1-27(HT) \- 4.3 5.1 23.1 2.3 7.2 \#2-17(HT) \- 5.5 3.7 24.4 1.9 6.8 \#2-1(LT) \- 23.0 1.3 0.3 1.8 8.4 \#2-6(LT) \- 13.6 2.1 1.0 1.5 6.9 T~2~ WT 11 18.4 ± 3.3 1.5 ± 0.9 0.4 ± 0.2 2.7 ± 1.0 18.4 ± 4.3 \#1-27(HT) 15 7.0 ± 4.1 ^\#\#\#^ 5.8 ± 1.6 ^\#\#\#^ 19.1 ± 7.3 ^\#\#\#^ 2.2 ± 0.4 9.6 ± 2.4 ^\#\#\#^ \#2-17(HT) 6 9.8 ± 8.1 ^\#^ 6.0 ± 0.5 ^\#\#\#^ 14.5 ± 6.1 ^\#\#\#^ 3.0 ± 0.4 9.2 ± 1.7 ^\#\#\#^ \#2-1(LT) 12 7.6 ± 3.3 ^\#\#\#^ 5.8 ± 1.1 ^\#\#\#^ 15.9 ± 7.2 ^\#\#\#^ 2.8 ± 0.8 8.0 ± 2.0 ^\#\#\#^ \#2-6(LT) 10 7.6 ± 6.8 ^\#\#\#^ 4.6 ± 1.3 ^\#\#\#^ 13.4 ± 6.7 ^\#\#\#^ 2.7 ± 0.5 11.7 ± 2.7 ^\#\#\#^ Selectedlines (T~2~) \#1-27(HT)L\#2 \- 4.9 6.1 29.6 2.4 9.4 \#2-17(HT)\#2-1 \- 3.7 6.5 20.0 3.1 10.4 \#2-1(LT)\#2-4 \- 5.3 4.3 29.4 3.1 9.1 \#2-6(LT)\#2-2 \- 3.1 4.9 21.1 2.7 13.2 T~3~ WT 6 11.1 ± 4.1 1.5 ± 0.5 3.3 ± 2.1 1.9 ± 0.5 4.2 ± 1.6 \#1-27(HT)L\#2 10 2.5 ± 0.6 ^\#\#\#^ 4.3 ± 0.4 ^\#\#\#^ 7.7 ± 1.9 ^\#\#^ 1.2 ± 0.1 ^\#\#\#^ 5.1 ± 0.7 \#2-17(HT)\#2-1 12 1.8 ± 0.5 ^\#\#\#^ 2.9 ± 0.5 ^\#\#\#^ 8.1 ± 2.3 ^\#\#\#^ 1.1 ± 0.2 ^\#\#\#^ 4.1 ± 0.8 Mean value of the alkaloid content (% dry weight) with standard deviation (mean ± SD) for each line and the alkaloid content of selected lines are summarized. nd \*: Not detected; ^\#^*p* \< 0.05; ^\#\#^*p* \< 0.005; and ^\#\#\#^*p* \< 0.001 *vs.* WT. The thebaine content in T~1~ plants varied widely, from 0.3% to 26.5%. From these plants, two high thebaine lines, \#1-27(HT) (thebaine content: 23.1%) and \#2-17(HT) (24.4%), and two low thebaine lines, \#2-1(LT) (0.3%) and \#2-6(LT) (1.0%), were selected and subjected to analysis of the T~2~ progeny. Interestingly, most of the progeny plants from both the HT and LT lines showed the high thebaine phenotype. From the T~2~ lines, two lines, \#1-27(HT)L\#2 (thebaine content: 29.6%) and \#2-17(HT)\#2-1 (20.0%), were selected for the analysis of T~3~ progeny. The thebaine content in T~3~ plants ranged from 4.2% to 10.0% in \#1-27(HT)L\#2 and from 3.7% to 10.9% in \#2-17(HT)\#2-1. The average thebaine content in T~3~ plants (two lines combined) was 2.4-fold of that in the WT; in contrast, the average morphine content decreased to *ca.* one fifth of that in the WT ([Figure 4](#pharmaceuticals-05-00133-f004){ref-type="fig"}). {#pharmaceuticals-05-00133-f003} ![Morphine, codeine, and thebaine contents in T~3~ progeny. Mean value of six (WT), 10 \[\#1-27(HT)L\#2\], and 12 \[\#2-17(HT)\#2-1\] plants. Bars indicate standard deviation. \* *p* \< 0.005 and \*\* *p* \< 0.001 *vs.* WT.](pharmaceuticals-05-00133-g004){#pharmaceuticals-05-00133-f004} 2.3. T-DNA Insertion Loci Analysis by IPCR and AL-PCR ----------------------------------------------------- The genomic DNA regions adjacent to the inserted T-DNA borders were analyzed by the IPCR and AL-PCR methods. The obtained DNA fragments are summarized in [Supplementary Table 1 and Figure 5](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"} along with the PCR methods, the combination of template circular or adaptor-ligated genome DNA libraries, and the primer sets. Sequence analysis of the amplified products revealed that the fragments were classified into three types, (A) T-DNAs connected with *P. somniferum* genome DNA, (B) T-DNAs connected in tandem, and (C) T-DNAs connected with T-DNA internal fragments, as shown in [Figure 5](#pharmaceuticals-05-00133-f005){ref-type="fig"}. Type (A) includes four types of genome DNA fragments adjacent to T-DNA LB, and six types of genome DNA fragments adjacent to T-DNA RB. Of these fragments, LB1g and RB2g, LB3g and RB6g were confirmed to be both ends of single genomic loci, by PCR over the LB and RB genomic regions. Although the tally of paired border for other fragments were not found, at least eight independent T-DNA integrated sites, namely RB1, RB2 (LB1-RB2), RB3, RB4, RB5, LB2, LB3 (LB3--RB6), and LB4 were estimated to exist in the T~0~ plant. A DNA fragment homologous (59% identity at the amino acid level) to the WRKY4 transcription factor (DDBJ/EMBL/GenBank accession no. AF425835) from *A. thaliana* was found in the LB1g region, at 695--952 bp 5\' upstream of the junction. The DNA sequence of LB1g, which included a *WRKY*-like gene, was deposited in the DDBJ/EMBL/GenBank under accession No. AB574419. No other gene with significant homology was found by the BLAST search tool in the genomic DNA regions adjacent to the inserted T-DNA. {#pharmaceuticals-05-00133-f005} Type (B) includes six types of DNA fragments. RB and LB were connected in a tail-to head manner at different junctions, or short DNA fragments were sandwiched. The fragment "Tandem 3" was found only in the genome direct amplification of T-DNA borders. Type (C) consists of four DNA fragments. Two of the fragments were made up of short partial fragments of T-DNA riorf13 attached to an LB, and the other two were made up of RB attached to a fragment of T-DNA riorf13. In summary, T-DNA border fragments found were at eight independent sites of T-DNA integration, six borders of T-DNAs connected in tandem, and four borders connected with T-DNA internal fragments. As for the copy numbers, types (A) and (B) corresponded to eight and six copies of T-DNAs, respectively. In the case of type (C), LB and RB connected with riorf13 were possibly borders of independent T-DNAs or borders of the same T-DNA. Therefore, the copy number can be estimated as two at minimum to four at maximum. Finally, the T-DNA copy number in the PsM1-2 T~0~ primary mutant could be estimated as 16 to 18. 2.4. Analysis of the Heredity Manner of T-DNA Inserted Loci ----------------------------------------------------------- The PCR analysis over T-DNA border and the adjacent genomic DNA found in the IPCR and AL-PCR analyses revealed that several T-DNA inserted loci were eliminated by selfing ([Figure 6](#pharmaceuticals-05-00133-f006){ref-type="fig"}). {#pharmaceuticals-05-00133-f006} In the high thebaine line \#1-27(HT), of the eight loci that were suggested to be independent T-DNA integration sites, RB3 and RB4 were eliminated in T~1~ progeny; in addition, in the high thebaine line \#2-17(HT), RB2, RB4 were eliminated in T~1~ progeny and the additional elimination of LB4 was observed in T~2~ progeny. On the other hand, with respect to the LT lines that showed low thebaine content at T~1~ progeny, sites RB2, RB4, and LB4 were eliminated in \#2-1(LT), and sites RB3, RB4, RB5, and LB3 were eliminated in \#2-6(LT). Notably thebaine content in these LT lines increased again in the T~2~ progeny to 29.4% in \#2-1(LT)\#2-4 and to 21.1% in \#2-6(LT)\#2-2 ([Table 2](#pharmaceuticals-05-00133-t002){ref-type="table"}) without a change in the T-DNA insertion pattern ([Figure 6](#pharmaceuticals-05-00133-f006){ref-type="fig"}). These results imply that none of the eight T-DNA integrated loci were indispensable for the high thebaine phenotype. 2.5. T-DNA Copy Number Analysis by Real-Time PCR ------------------------------------------------ Standard curves for the quantification of the T-DNA copy number in T~0~ and selected progenies were prepared for each target region, LB1g, LB1j and orf2. The formulae and correlation coefficients were as follows: LB1g: y = −1.39ln(x) + 23.82 (r^2^ = 0.991); LB1j: y = −1.44ln(x) + 23.38 (r^2^ = 0.994); and orf2: y = −1.43ln(x) + 23.75 (r^2^ = 0.997). The relative abundances of each region in the samples were calculated by these formulae from the value of Delta Rn. The relative abundances in whole numbers, when the abundance of LB1g was set as 2, were LB1g:LB1j:orf2 = 2:1:15 in T~0~. And for T~1~\[\#1-27(HT)\] and its progenies, the abundances were as follows (in the order of LB1g:LB1j:orf2): T~1~\[\#1-27(HT)\], 2:2:6; T~2~\[\#1-27(HT)L\#2\], 2:2:7; and T~3~\[\#1-27(HT)L\#2\#1\], 2:2:7. And for T~1~\[\#2-17(HT)\] and its progenies, the values were as follows: T~1~\[\#2-17(HT)\], 2:nd:10; T~2~\[\#2-17(HT)\#2-1\], 2:nt:10; and T~3~\[\#2-17(HT)\#2-1\#1\], 2:nt:7 (nd: not detected; nt: not tested). These results are summarized in [Figure 7](#pharmaceuticals-05-00133-f007){ref-type="fig"}. {#pharmaceuticals-05-00133-f007} For the abundance of LBj and orf2 in the \#1-27(HT) series, the LB1-RB2 T-DNA insertion locus was estimated to become homozygous at the T~1~ progeny, as indicated by the doubled abundance of LB1j in T~1~. And the T-DNA copy number, estimated by the abundance of the orf2 region, was drastically decreased from 15 to six in T~1~, then increased to seven in T~2~ and kept at seven in T~3~. These data imply that more than half of the total T-DNA copies were eliminated in the first selfing. For the \#2-17 series, the LB1j region was not detected in T~1~, which was consistent with the elimination of the LB1-RB2 T-DNA insertion loci in T~1~ revealed by the T-DNA insertion loci analysis ([Figure 6](#pharmaceuticals-05-00133-f006){ref-type="fig"}). The abundance of orf2 was decreased from 15 to 10 in T~1~, and then decreased again from 10 to seven in T~3~, which implies that more than half of the T-DNA copies in the \#2-17(HT) series were also eliminated by repeated selfing. 2.6. Expression Analyses on Morphine Biosynthetic Genes by RT-PCR ----------------------------------------------------------------- Firstly we tried to apply realtime-PCR for the expression analysis of morphine biosynthetic genes including *T6ODM*, *CODM* using the primer sequence reported by Hagel and Facchini \[[@B10-pharmaceuticals-05-00133]\]. However, prior to run the realtime-PCR, we found that PCR with these primers using our cDNA as a template gave multiple products. Although we have designed several primers, they could not make the PCR product as a single band. It may be attributed to the relatively high sequence homology of coding region among *T6ODM*, *CODM*, and *DIOX2*. Therefore we hired the semi-quantitative RT-PCR method for the expression analysis. To distinguish RT-PCR products between *T6ODM* and *CODM*, primers were designed to give different product size, *i.e.*, 549 bp for *T6ODM* and 411 bp for *CODM*. Expression analysis on selected morphine biosynthetic genes downstream of (*S*)-*N*-methylcoclaurine revealed that the expression of *CODM* was completely diminished in PsM1-2 ([Figure 8](#pharmaceuticals-05-00133-f008){ref-type="fig"}). On the other hand, the expression of *T6ODM* seemed to be slightly up-regulated in the PsM1-2 compared with the WT plant. Specific amplification of these two genes was confirmed by the comparison of the size of the bands and their calculated amplicon size. In PsM1-2, the expression levels of *CYP80B3* and *SalAT* seem to be slightly higher than WT, whereas *4\'OMT* seems to be down-regulated. No significant difference in the expression level of genes was observed between PsM1-2 and WT for *Cor1-1* or *Cor2-1*. ![The morphine biosynthetic pathway downstream of (*S*)-*N*-methylcoclaurine with the results of expression analysis of selected morphine biosynthetic genes, *CYP80B3*, *4\'OMT*, *SalAT*, *T6ODM*, *COR* (alleles *Cor1-1* and *Cor2-1*), and *CODM* by RT-PCR. Actin was used as an experimental control. Presumably, the pathway via oripavine (dotted pathway) does not exist in the *P. somniferum* Japanese cultivar "Ikkanshu" which we have used in this study \[[@B11-pharmaceuticals-05-00133],[@B12-pharmaceuticals-05-00133]\].](pharmaceuticals-05-00133-g008){#pharmaceuticals-05-00133-f008} 2.7. Discussion --------------- Morphological abnormalities, such as varied numbers of petals and splits on the boundary of petals, were frequently observed in the selfed progenies in the present study. However, no clear correlation was found between these morphological abnormalities and altered alkaloid compositions. Therefore, these findings were thought to be independent of the mutation in the secondary metabolism. At the T~2~ generation, difference between high thebaine line and low thebaine line which was obvious at T~1~ generation, has disappeared. If the high thebaine phenotype is cause by the single mutation of the locus by T-DNA insertion, low thebaine phenotype should be dominant in the progeny plants. However, as observed in [Figure 3](#pharmaceuticals-05-00133-f003){ref-type="fig"}, most of the progeny plants of low thebaine T~1~ lines have gained high thebaine phenotype again, which indicates that the multiple loci are responsible for the high thebaine phenotype. For the reason of this phenomenon, it is also possible that, methylation or suppression has occurred in unstable manner on the alkaloid biosynthesis related genes caused by the multiple T-DNA insertion events. The content of thebaine, which was the major alkaloid in the latex of the mature plants of the mutants, varied widely in the T~1~ progeny. But by repeated selfing, in the T~3~ progeny, although the maximum content of thebaine (10.9%) was not particularly high, the range of thebaine content was much narrower than that in the T~1~ and T~2~ progenies. When the value of CV (coefficient of variation: standard deviation/average value) for the thebaine content was compared among T~1~, T~2~, and T~3~ progenies, it was 0.54 in T~1~, 0.40 in T~2~ (two HT lines combined), and 0.26 in T~3~ (two lines combined). And the CV for the morphine content was 0.74 in T~1~, 0.70 in T~2~ (two HT lines combined), and 0.28 in T~3~ (two lines combined). These lines of evidence indicate that the high thebaine (2.4-fold and 2.5-fold of WT in T~3~ \#1-27(HT)L\#2 and \#2-17(HT)\#2-1, respectively) and low morphine (0.2-fold of WT in both T~3~ \#1-27(HT)L\#2 and \#2-17(HT)\#2-1, respectively) phenotypes were stabilized by repeated selfing. Analyses of the T-DNA integration sites and T-DNA copy number on the primary T~0~ mutant revealed that at least eight integration sites exist and as many as 18 copies of T-DNAs were estimated to be integrated into the genomic DNA in a highly complicated manner. Considering the complexity of the T-DNA integration, the IPCR, AL-PCR, and real-time PCR methods employed in this study can be considered as the most suitable methods for T-DNA insertional analysis, and more suitable than Southern blotting, whose signals may be beyond interpretation in this context. The number of T-DNA copies in PsM1-2 was too large for the transgenes integrated by genetical transformation. The presence of high numbers of transgenic insertions can lead to poor expression of transgenes through silencing. In this study, we tried to simplify the T-DNA integration structure and stabilize the high thebaine phenotype, and then to gain insight into the genetic factors for the altered alkaloid composition by obtaining selfed progenies. The T-DNA integration sites in PsM1-2 were paired to be homozygous or dropped off by selfing, and finally became half of the T~0~ in the selected T~3~ progenies. Although it is possible that other T-DNA copies were not detected, no correlation was found between any of the T-DNA integration sites and the altered alkaloid composition, by considering these data together, a reduction in the T-DNA copy number seems to have resulted in the stabilization of the high thebaine phenotype. Although it is hard to confirm, there is also a possibility that genome reorganization independent of T-DNA insertion has occurred during shoot regeneration or long term maintenance of *in vitro* culture. As we have accomplished the stabilization of high thebaine phenotype by selfing up to T~3~ generation, backcross experiment utilizing these selfed progeny plants is in progress. In this study, the only gene homologous to the known gene found at the T-DNA integration loci was the *AtWRKY4* gene homologue found in the 5\' upstream region of LB1g. As some type of WRKY transcription factor may function as a transcriptional regulator of benzylisoquinoline alkaloid biosynthesis in *Coptis japonica* Makino \[[@B13-pharmaceuticals-05-00133]\], the contribution of this locus to the altered alkaloid composition in the mutant was suspected. However, analysis of the T-DNA heredity manner indicated that the T-DNA insertion at LB1-RB2 region was not essential for the high thebaine phenotype. The expression analyses on selected morphine biosynthetic genes, including two novel demethylases, *T6ODM* and *CODM*, between the *in vitro* shoot culture of the PsM1-2 mutant and seedlings of the WT plant revealed that the expression of *CODM* was fully suppressed in the mutant. Although the correlation between the transcript level of biosynthetic genes in young organs, such as seedling or *in vitro* shoot culture, and the alkaloid composition in the latex of mature plant needs to be clarified, the observed differences between the wild type plant and the mutant can be correlated to the alkaloid composition difference in them (morphine was detected in the WT, however, almost no morphine in the mutant \[[@B8-pharmaceuticals-05-00133]\]). Kinetic studies on recombinant T6ODM and CODM from *P. somniferum* \[[@B10-pharmaceuticals-05-00133]\] have revealed that oripavine is the most preferred substrate of T6ODM, followed by thebaine, while codeine is not accepted as a substrate. On the other hand, CODM showed a higher preference for codeine than thebaine. Considering the substrate preference of these two demethylases, thebaine can be accumulated solely only under the condition that the expression of both *T6ODM* and *CODM* is suppressed, and the suppression of *CODM* may result in accumulation of codeine. In actuality, however, a large amount of thebaine with a smaller amount of codeine is accumulated in the latex from mature plants of PsM1-2 mutants. This pattern of compounds detected in the mutant is similar to that of the *T6ODM*-silenced transformant by virus-induced gene silencing \[[@B10-pharmaceuticals-05-00133]\]. In contrast, the *CODM*-silenced transformant accumulates mainly codeine, together with smaller amounts of thebaine and morphine \[[@B10-pharmaceuticals-05-00133]\]. Although the alkaloid productivities of those transformants cannot be simply compared with PsM1-2, as the alkaloid composition varies highly even among the cultivars \[[@B14-pharmaceuticals-05-00133]\], it is assumed that suppression of CODM did not simply lead to the thebaine accumulation in PsM1-2. And it is also possible that in a Japanese cultivar that does not have the pathway from thebaine to morphine via oripavine \[[@B11-pharmaceuticals-05-00133],[@B12-pharmaceuticals-05-00133]\], the substrate preferences of T6ODM and CODM differ from those of oripavine-producing cultivars. As the regulation of opium alkaloid production in *P. somniferum* is highly complicated and varies among cultivars---and even among the developmental stages \[[@B15-pharmaceuticals-05-00133],[@B16-pharmaceuticals-05-00133]\] or individual parts of a single plant \[[@B17-pharmaceuticals-05-00133]\]---further detailed studies on the molecular regulation of alkaloid production, such as expression analyses of *T6ODM* and *CODM* in the latex-producing capsule of PsM1-2, are required. 3. Experimental Section ======================= 3.1. Plant Materials -------------------- The wild type (WT) plant of *P. somniferum* L. used was the Japanese cultivar "Ikkanshu", and the *A. rhizogenes* strain MAFF03-01724 T-DNA insertion mutant line was PsM1-2 \[[@B8-pharmaceuticals-05-00133]\]. The *in vitro* culture of PsM1-2 used in this experiment was previously subjected to a single round of cryopreservation and regenerated to plantlet on Murashige-Skoog (MS) solid media \[[@B18-pharmaceuticals-05-00133]\] by the method described previously \[[@B19-pharmaceuticals-05-00133],[@B20-pharmaceuticals-05-00133]\] with slight modifications. 3.2. Maintenance and Cultivation of Plant Materials --------------------------------------------------- The WT plant seeds were obtained from the field-grown plants at the Research Center for Medicinal Plant Resources, Division of Tsukuba. The PsM1-2 T~0~*in vitro* shoot culture was maintained on MS solid media at 20 °C under a 14 h light/10 h dark condition and then transplanted in soil in a 9 cm diameter pot and acclimatized in a phytotron in 60% relative humidity under a cycle of 16 h light at 20 °C and 8 h dark at 17 °C. Seeds of T~1~ plant obtained from the soil-cultivated plant of the PsM1-2 T~0~ primary mutant were sown on the soil in a 15 cm diameter pot and cultivated in a greenhouse under a 16 h light/8 h dark cycle at 20 °C and 60% relative humidity. Plants were fertilized with 500-fold diluted Hyponex^®^ (Hyponex Japan, Osaka, Japan) once a week. T~2~ seeds from the two lines of T~1~ plants that showed high thebaine content and had abundant mature seeds were selected for cultivation of T~2~ progeny. The cultivation conditions were the same as for T~1~ plants. T~3~ seeds from two lines of T~2~ plants with high thebaine content were germinated on rock wool with fertilization with 2,000-fold diluted Hyponex^®^ in a greenhouse under a 16 h light/8 h dark cycle at 20 °C and 60% relative humidity. After one month, seedlings were transplanted onto the soil in a 9 cm diameter pot, and grown in the growth chamber under a 12 h light/12 h dark condition (short day condition) at 20 °C and 60% relative humidity. *Ca.* 80 days after sowing, the lighting was changed to a long day condition of 16 h light/8 h dark at 20 °C and 60% humidity for flowering. After transplanting, plants were fertilized with 500-fold diluted Hyponex^®^ once a week. For each experiment, WT plants were grown together as an experimental control. All self-pollination events were performed manually. 3.3. Phenotypic Observation of the PsM1-2 Mutants ------------------------------------------------- Phenotypic parameters such as days to flowering, number of petals, appearance of splitting on the boundary of the petal, and the height of the aerial part at the seed-filling stage, were observed on each plant. 3.4. HPLC Analysis of Alkaloid Content in the Latex --------------------------------------------------- The opium alkaloid content in the latex was analyzed by HPLC. Latex was collected from the capsule of either the WT or mutant *P. somniferum ca.* two weeks after flowering, by incising the capsule surface. Collected latex was dried at 50 °C. Approximately 5 mg of dried latex was measured accurately and subjected to alkaloid extraction by adding 5 mL of methanol followed by 30 min of sonication and mixing thoroughly using a tube mixer. After centrifugation at 20,000× *g* for 1 min, supernatant was applied to an Ultrafree-MC spin column (Millipore, Bedford, MA, USA) and centrifuged at 20,000× *g* for 1 min, and then 5 μL of the flow through was injected into an HPLC column. The HPLC conditions were as follows. HPLC instruments: Waters Alliance PDA System (separation module: 2795; photodiode array detector: 2996) (Waters, Milford, MA, USA). Column: TSK-GEL ODS100V (pore size 5 μm, φ4.6×250 mm) (Tosoh, Tokyo, Japan). Solvent system: CH~3~CN (A), 10 mM sodium 1-heptanesulphonate (pH 3.5) (B). Solvent gradient (A%): 0 min 28%, 15 min 34%, 25 to 39 min 40%, 40 min 28%. Detection: UV 200 to 400 nm (spectrometric identification of compounds), UV 284 nm (quantitative analysis). Column temperature: 30 °C. Flow rate: 0.7 mL/min. HPLC data were collected and analyzed by an Empower system (Waters). Alkaloid components were identified by the comparison of retention time and the UV spectra with authentic standards. Morphine hydrochloride and codeine phosphate were purchased from Takeda Pharmaceutical Company Limited (Osaka, Japan). Oripavine was a gift from Einar Brochmann-Hanssen (University of California, San Francisco, CA, USA). Magnoflorine iodide and jateorrhizine were gifts from Akira Ikuta (Science University of Tokyo, Japan). Reticuline and columbamine were gifts from Fumihiko Sato (Kyoto University, Japan). Isothebaine was isolated from *Papaver pseudo-oriental*e (Fedde) Medw. by our group. Thebaine was a gift from Ruri Kikura-Hanajiri (National Institute of Health Sciences, Japan). Papaverine hydrochloride, noscapine hydrochloride, coptisine chloride, sanguinarine chloride, and berberine chloride were purchased from Wako Pure Chemical Industries (Osaka, Japan). Alkaloid contents were calculated as a weight percent of the dried latex (opium). 3.5. Genomic DNA Preparation from P. somniferum ----------------------------------------------- Genomic DNA was prepared from *ca.* 100 μg of fresh leaves of selfed plants grown in the growth chamber, or from *ca.* 100 μg of whole *in vitro* plantlet of the PsM1-2 T~0~ mutant, which mainly consisted of leaves and stems, by using a DNeasy Plant Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. 3.6. Analysis of T-DNA Insertion Loci by IPCR and AL-PCR -------------------------------------------------------- The inverse-PCR (IPCR) method \[[@B21-pharmaceuticals-05-00133],[@B22-pharmaceuticals-05-00133]\] and adaptor ligation PCR (AL-PCR) method were used for the analysis of the flanking unknown genome DNA sequence, adjacent to the inserted T-DNA. In this study, the Vectorette PCR method \[[@B23-pharmaceuticals-05-00133],[@B24-pharmaceuticals-05-00133],[@B25-pharmaceuticals-05-00133]\], an improved method of AL-PCR, was employed to reduce non-specific amplicons. The genomic DNA library for each PCR method was constructed by digestion of genomic DNA by the appropriate restriction enzymes and self-ligation to form a circular DNA library, or an adaptor linker attached genome DNA library. 3.7. Genomic Library Construction for IPCR ------------------------------------------ Genomic DNA was digested with the restriction enzymes *Bam*HI, *Eco*RV, *Hae*III, *Kpn*I, *Pvu*II, *Ssp*I, or *Stu*I. Completely digested DNA was ligated by using a Fastlink^®^ DNA Ligation Kit (AR Brown, Tokyo, Japan) to form a circular genome DNA library. 3.8. Genomic Library Construction for AL-PCR -------------------------------------------- The sequences of the adaptor oligo DNA and adaptor specific primers used in this study are listed in [Supplementary Table 2](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"}. Two complementary oligo DNAs, AP-LS and AP-SS, were annealed to form an adaptor unit. Genomic DNA was digested with the restriction enzymes *Eco*RV, *Hae*III, *Pvu*II, *Ssp*I, or *Stu*I, which produce blunted ends. The completely digested DNA was ligated with adaptor units by using a Fastlink^®^ DNA Ligation Kit to form an adaptor ligated genome DNA library. 3.9. IPCR and AL-PCR -------------------- Amplification of the target region was performed by the nested PCR method using TaKaRa Ex Taq^TM^ DNA polymerase (Takara Bio, Shiga, Japan) under the following conditions. The combinations of PCR methods, template genome DNA libraries and primer sets are listed in [Supplementary Table 1](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"}. T-DNA-specific primers were designed based on the DNA sequence of the T-DNA region of the *A. rhizogenes* plasmid pRi1724 (DDBJ/EMBL/GenBank accession no. AP002086). The first PCR conditions were as follows: primary denaturation at 94 °C for 5 min; followed by 30 cycles of 94 °C for 1 min, 42 °C for 2 min, and 72 °C for 3 min; with a final extension at 72 °C for 10 min. After PCR, the solution was held at 4 °C. The first PCR reaction solution was applied to a SUPREC^TM^-02 filter (Takara Bio) to eliminate the primers and then used as a template for the second PCR. Second PCR conditions were as follows: primary denaturation at 94 °C for 5 min; followed by 30 cycles of 94 °C for 1 min, 48 °C for 2 min, and 72 °C for 3 min; with a final extension at 72 °C for 10 min. After PCR, the solution was held at 4 °C. The product of the second PCR was gel purified and cloned into the sequencing vector pT7-Blue^®^ (Novagen, Madison, WI). Propagated plasmid DNA was subjected to DNA sequencing using a BigDye^®^ Terminator v3.1 Cycle Sequencing Kit and ABI PRISM^®^ 3100---Avant Genetic Analyzer (Applied Biosystems Japan, Tokyo, Japan). Homology search was performed on T-DNA flanking genome DNA sequences with the BLAST tool at NCBI. 3.10. Direct Amplification of T-DNA Borders Connected in Tandem --------------------------------------------------------------- PCR was performed on uncut genome DNA of T~0~ to amplify the border region of T-DNAs connected in tandem. The primers used are listed in [Supplementary Table 1](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"}. The PCR conditions were the same as for the IPCR. 3.11. Analyses of T-DNA Insertion Loci and Heredity Manner by PCR ----------------------------------------------------------------- The PCR method was employed to confirm the T-DNA integration loci on *P. somniferum* genome DNA and to analyze the heredity manner in the selfed progenies. To find out the tally of the paired genomic regions found adjacent to the T-DNA left borders (LBs) and right borders (RBs) revealed by IPCR and AL-PCR analyses, PCR amplification was performed with the pair of genomic region-specific LB and RB (e.g*.*, LB1g *vs*. RB2g) primers listed in [Supplementary Table 3](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"}, under the following PCR conditions: primary denaturation at 94 °C for 5 min; followed by 30 cycles of 94 °C for 30 s, 58 °C for 30 s, and 72 °C for 1 min; with a final extension at 72 °C for 10 min. After PCR, the solution was held at 4 °C. TaKaRa Ex Taq^TM^ was used as the PCR polymerase. The PCR product was separated on agarose gel. The paired genomic regions, which gave a PCR product was judged as the single T-DNA integrated locus. To judge whether or not the T-DNA integration loci were present in the selfed progenies, PCR amplification was performed between the genome region-specific primers listed in [Supplementary Table 3](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"} and T-DNA LB- or RB-specific primers (MAFF-226A or MAFF-14963S). PCR was performed under the following PCR conditions: primary denaturation at 94 °C for 5 min; followed by 30 cycles of 94 °C for 30 s, 58 °C for 30 s, and 72 °C for 1 min; with a final extension at 72 °C for 10 min. After PCR, the solution was held at 4 °C. GoTaq^®^ Green Master Mix (Promega, Madison, WI, USA) was used as the PCR polymerase. The PCR product was separated on agarose gel. 3.12. T-DNA Copy Number Analysis by Real-time PCR ------------------------------------------------- The T-DNA copy number was analyzed by the quantitative real-time PCR method \[[@B26-pharmaceuticals-05-00133],[@B27-pharmaceuticals-05-00133]\]. The strategy used for estimating the T-DNA copy number is as follows. When information of one of the integrated T-DNA sites was provided by T-DNA integrated loci analysis, and also, when the T-DNA integrated site was a single copy, the copy number of the integrated T-DNA could be calculated as a multiple of the relative abundance of standard DNA fragments, as shown in [Figure 9](#pharmaceuticals-05-00133-f009){ref-type="fig"}. This estimation method can be enacted under the hypotheses that (1) the genome DNA of PsM1-2 is diploid (2n = 22) \[[@B28-pharmaceuticals-05-00133]\], (2) all of the T-DNA is integrated into the host genome DNA in a heterozygous manner, and (3) one of the integrated T-DNAs for which both the LB and RB borders are known (e.g*.*, the LB1-RB2 locus) is a single copy. Under these hypotheses, by comparing the relative abundance of the T-DNA internal region (in this case, orf2), the T-DNA---*P. somniferum* genome junction region (LB1j), and the *P. somniferum* genome region (LB1g), we can calculate the inserted T-DNA copy number by fixing the abundance of LB1g as two. {#pharmaceuticals-05-00133-f009} In our experiment, one of the T-DNA-integrated sites, LB1-RB2, which will be described in the Results section, was set as a standard. And for the quantification standard plasmid DNA, we constructed pLB1, which included, the LB1g, LB1j, and orf2 regions of the T-DNA. A DNA fragment with these three regions was amplified by PCR with the primers LB1-orf2-S (5\'-CTC ATA AGC AGT GGT ATT GCT C-3\') and LB1-orf2-A (5\'-CGC ATT CAT GCG GTT ATG GAG-3\') and KOD-Plus-DNA polymerase (Toyobo, Osaka, Japan) under the following PCR conditions: primary denaturation of 94 °C for 2 min; followed by 35 cycles of 94 °C for 15 s, 62 °C for 30 s, 68 °C for 90 s. After PCR, the solution was held at 4 °C. The amplified product was cloned into the pT7-Blue^®^ vector (Novagen) and then propagated in *E. coli*. The quantitative standard plasmid DNA pLB1 and genome DNA prepared from the primary T~0~ mutant and selected T~1~, T~2~, and T~3~ progenies of the PsM1-2 mutant were diluted serially with the dilution buffer supplied with the real-time PCR reagent SYBR^®^ Premix Ex Taq^TM^ II (Perfect Real Time; Takara Bio). Real-time PCR was run using the target region-specific primers listed in [Supplementary Table 4](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"} with the real-time PCR reagent on an ABI PRISM 7000 Sequence Detection System (Applied Biosystems Japan). The obtained data were analyzed using the supplied software (Applied Biosystems Japan) and the relative abundance of each target region was deduced from each Delta Rn value using standard curves. Standard curves for each target region were plotted with the plasmid concentration (fg/μL) on the x-axis and the Delta Rn on the y-axis. The curves showed good correlations. The relative abundances of each target region were calculated so that the abundance of the LB1g region was 2, and then rounded off to a whole number. 3.13. Actin Gene Amplification from P. somniferum ------------------------------------------------- A fragment of actin cDNA was amplified by degenerate PCR using the forward primer 5\'-AAR GCN AAY MGN GAR AAR ATG AC, and the reverse primer 5\'-CCR TAN ARR TCY TTN CKD ATR TC, which were designed from the completely conserved regions of the amino acid sequences of other actins, such as *Arabidopsis thaliana* (*actin-1*: DDBJ/EMBL/GenBank accession No. M20016), *Nicotiana tabacum* (*actin*: X63603), and *Zea mays* (*Maz56*: U60514). cDNA synthesized from the total RNA of young seedlings of *P. somniferum* was used as a template for PCR. The manual hot-start procedure was used for the amplification. TaKaRa Ex Taq^TM^ DNA polymerase was added after primary denaturation at 94 °C for 5 min, and then the following protocol was carried out in a GeneAmp2400 thermal cycler (Applied Biosystems Japan): 30 cycles of 94 °C for 1 min, 48 °C for 2 min, and 72 °C for 3 min; with a final extension at 72 °C for 10 min. After PCR, the solution was held at 4 °C. The amplified fragment was cloned into the pT7-Blue^®^ vector followed by DNA sequencing. Two representative actin cDNA sequences, whose deduced amino acid sequences showed 92% and 95% identity to the *Arabidopsis actin-1*, were named *PsACT1* (AB574417) and *PsACT2* (AB574418), respectively. 3.14. Expression Analysis of the Morphine Biosynthetic Genes ------------------------------------------------------------ The expression levels of selected morphine biosynthetic genes downstream of (*S*)-*N*-methylcoclaurine, *CYP80B3* (DDBJ/EMBL/GenBank accession no. AF134590 \[[@B29-pharmaceuticals-05-00133]\]), (*R*,*S*)-3\'-hydroxy-*N*-methylcoclaurine 4\'-*O*-methyltransferase (*4\'OMT*; AY217333 \[[@B15-pharmaceuticals-05-00133]\]), *SalAT* (AF339913 \[[@B30-pharmaceuticals-05-00133]\]), *T6ODM* (GQ500139 \[[@B10-pharmaceuticals-05-00133]\]), *COR* (allele *Cor1-1*: AF108432; allele *Cor2-1*: AF108438 \[[@B31-pharmaceuticals-05-00133]\]), and *CODM* (GQ500141 \[[@B10-pharmaceuticals-05-00133]\]) in the WT plant and the PsM1-2 mutant were analyzed and compared by the reverse transcription PCR (RT-PCR) method. Total RNA was prepared from the whole plants of two-week-old seedlings of field-grown WT *P. somniferum*, or from whole *in vitro* plantlet of the PsM1-2 T~0~ mutant, which mainly consisted of leaves and stems, by using an RNeasy Plant Mini Kit (Qiagen) according to the manufacturer's instructions. One microgram of total RNA samples was subjected to single-stranded cDNA synthesis by reverse-transcription with oligo-(dT) primer (RACE32: 5\'-GAC TCG AGT CGA CAT CGA TTT TTT TTT TTT TT-3\') \[[@B32-pharmaceuticals-05-00133]\] using Superscript® II Reverse Transcriptase (Life Technologies, Carlsbad, CA, USA) according to the manufacturer's instructions. Synthesized ss-cDNA was used as a template for PCR with the gene-specific primers listed in [Supplementary Table 5](#pharmaceuticals-05-00133-s001){ref-type="supplementary-material"}. The PCR conditions were as follows: primary denaturation 94 °C for 5 min; followed by 30 cycles of 94 °C for 30 s, 58 °C for 30 s, and 72 °C for 1 min; with a final extension at 72 °C for 10 min. After PCR, the solution was held at 4 °C. PCR products were separated on 1.0% agarose gel and signal intensities were observed. The actin gene *PsACT1* from *P. somniferum* was used as an experimental control. 3.15. Statistical Analysis -------------------------- Values were expressed as the mean ± standard deviations (SD) and were analyzed by the Tukey-Kramer multiple comparison test using the statistical analysis system "R" software package \[[@B33-pharmaceuticals-05-00133]\]; a *p* value of less than 0.05 was considered significant. 4. Conclusions ============== By combining genetic and phenotypic analyses of the T-DNA insertional mutant PsM1-2 with selfing, we have succeeded in stabilizing the high thebaine phenotype in coordination with a reduction in the number of inserted T-DNA copies. Although the genetic mode of CODM suppression in *in vitro* plantlet and of the accumulation of thebaine still remain unknown, studies on this mutant and its progenies may provide new insights into the molecular basis of morphine biosynthesis, and could ultimately allow us to manipulate the biosynthesis of this compound at will. We thank Naoko Tanaka and Naoko Onodera for their technical assistance. This study was supported in part by a grant from the Ministry of Health, Labor and Welfare of Japan. The authors declare no conflict of interest. ###### PDF-Document (PDF, 200 KB) ###### Click here for additional data file.

Learning pointsManagement decisions appropriate for a non-athlete might be inappropriate in an athlete as they may result in disqualification, financial loss, or put the athlete at additional risk of complications.Cardiologists with expertise in sports cardiology should be involved in the management of athletes as it is often complex and requires a holistic approach.Left atrial appendage occlusion devices can play an important role in reducing stroke risk in selected cases. Introduction ============ Caring for athletes with cardiac disease requires an approach that caters to the specific needs of the individual. Often athletes require their care to fit around training and competition requirements and this can come into conflict with the best care their clinicians feel they can offer. Medications and interventions with proven symptomatic and prognostic benefit may affect athletes' performance and lead to poor adherence. Moreover, they may result in disqualification from competitive sports which are likely to carry both personal and financial consequences. In some individuals, engaging in demanding physical activity and competitive sports against medical advice may carry significant health risks. Therefore, shared decision-making is vital and alternative management strategies are often warranted to ensure appropriate adherence to prescribed treatment. This case illustrates this conflict in a professional rugby player with a cardiomyopathy and atrial fibrillation (AF) on anticoagulation who wished to continue to play and discusses how an alternative approach was able to optimize his care. Timeline ======== Date Events --------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- November 2017 Admission with decompensated heart failure and new diagnosis of atrial fibrillation (AF).Transthoracic echocardiogram (TTE) showed evidence of left ventricular (LV) dilatation and severe LV systolic dysfunction.Discharged on heart failure medications and anticoagulation for stroke prevention. January 2017 Unobstructed coronary arteries on angiography.Early recurrence of AF following direct current cardioversion. February 2017 Left ventricular systolic function remained severely impaired on outpatient TTE. April 2018 Cardiac magnetic resonance showed dilated cardiomyopathy with ejection fraction (EF) 37%. No scaring or fibrosis seen on late gadolinium enhancement. July 2018 Right arm weakness and paraesthesia in keeping transient ischaemic attack.Screening for connective tissue disorders, HIV, syphilis, and Fabry disease was negative. October 2018 Left ventricular systolic function returned to 'near normal' (EF 50%). Euvolaemic.Advised not to play rugby due to high bleeding risk on anticoagulation. November 2018 Playing rugby on anticoagulation.Referred for consideration of left atrial appendage occlusion (LAAO) device. July 2019 Successful LAAO device implantation.Anticoagulants stopped. Case presentation ================= A 27-year-old male professional rugby player was admitted to his local district general hospital with a 2-day history of chest tightness and breathlessness. He had no other significant past medical history and was not taking any regular medications. He admitted to regularly taking cocaine and performance-enhancing steroids. He was haemodynamically stable with normal saturation. The main findings on physical examination were an irregularly irregular pulse and bibasal crackles. His admission electrocardiogram (ECG) showed AF and his chest X-ray findings were in keeping with pulmonary oedema. Initial bloods tests were within normal range. Transthoracic echocardiogram (TTE) revealed bi-atrial dilatation and a moderately dilated left ventricle (left ventricular end diastolic diameter 7 cm) with mild concentric left ventricular (LV) hypertrophy with an ejection fraction (EF) 35--40%. He was acutely managed with intravenous diuretics and initiated on evidence-based heart failure medications including beta-blockers and angiotensin-converting enzyme inhibitors. In view of his drug history, a working diagnosis of drug-induced dilated cardiomyopathy (DCM) was made. Once stabilized, he was discharged on bisoprolol 2.5 mg and ramipril 2.5 mg. In anticipation of a direct current cardioversion (DCCV), he was started on rivaroxaban 20 mg. He was counselled not to participate in any competitive sports. To further investigate his cardiomyopathy, he underwent an outpatient coronary angiogram which revealed unobstructed coronary arteries. In addition, a cardiac magnetic resonance (CMR) confirmed a dilated left ventricle with globally impaired systolic function and a calculated EF of 37%. There was no evidence of scarring or fibrosis on delayed enhancement images. He was unable to maintain sinus rhythm following DCCV and relapsed back into persistent AF. His CHA~2~DS~2~-VASc score was 1 (LV dysfunction) which does not strictly mandate anticoagulation; however, he made an informed decision to continue rivaroxaban. On a follow-up TTE performed 3 months later, LV systolic function remained unchanged. His ramipril was increased and he was initiated on eplerenone. Left ventricular function gradually improved on optimal medical therapy and, at 8 months of follow-up, had returned to near-normal (EF 50--55%). He remained in AF and experienced a brief episode of left arm weakness and paraesthesia suggestive of a transient ischaemic attack (TIA) despite being compliant with anticoagulation. As his CHA~2~DS~2~-VASc score increased to 3 (TIA, LV dysfunction) he now had a clear indication for anticoagulation. Connective tissue disorders, syphilis, and Fabry disease screening were negative. At 1-year follow-up, he was asymptomatic (New York Heart Association 1) but remained in AF and continued to participate in competitive rugby, whilst on oral anticoagulation despite counselling regarding the high bleeding risk. He was reluctant to terminate his professional rugby career prematurely and sought alternative stroke prevention strategies that obviated the need for continuous anticoagulation. He was referred to a Tertiary Cardiology Centre for further management of his AF and consideration of a left atrial appendage occlusion (LAAO) device. A 27 mm Watchman Flx (Boston Scientific, MA, USA) device was successfully deployed under general anaesthetic in the left atrial appendage with a good seal and no leaks ([*Figures 1*--*4*](#ytz242-F1){ref-type="fig"}, [Supplementary material](#sup1){ref-type="supplementary-material"}). He was discharged home on a 6-week course of aspirin and clopidogrel therapy with a follow-up transoesophageal echocardiogram to assess LAAO device position and guide cessation of antiplatelet strategy. {#ytz242-F1} {#ytz242-F2} {#ytz242-F3} {#ytz242-F4} Discussion ========== Recommendations regarding participation in competitive sports should be given following comprehensive evaluation of the athlete's disease characteristics and a thorough risk assessment. The work-up includes a 12-lead ECG, echocardiography, CMR, 24-h Holter monitor, and cardiopulmonary exercise testing. Disqualification from competitive sports is likely to carry both personal and financial consequences for athletes.[@ytz242-B1] Ensuring that the athlete is involved in the decision-making process is therefore paramount. This athlete was strongly advised not to engage any competitive sports, in line with the European Association of Preventive Cardiology (EAPC) recommendations. The EAPC position paper states that athletes with DCM should not participate in competitive sports if any of the following are present[@ytz242-B1]: symptomatic, orEjection fraction \<40%, orextensive late gadolinium enhancement (i.e. \>20%) on CMR, and/orfrequent/complex ventricular tachyarrhythmias on ambulatory ECG monitoring and exercise testing, orhistory of unexplained syncope. Once his LV systolic function had improved there was no further restriction on exercise from a cardiomyopathy perspective. On admission, his CHA~2~DS~2~-VASc score was 1 (LV dysfunction) which is not a strict indication to initiate oral anticoagulation \[Class IIA, level of evidence (LOE B)\] as the evidence supporting a net clinical benefit of oral anticoagulation in patient with a single stroke risk factor (excluding gender) is limited.[@ytz242-B2] Oral anticoagulation was started in anticipation of a DCCV and, after appropriate counselling, he made an informed decision to continue on rivaroxaban. However, his CHA~2~DS~2~-VASc increased to 3 (TIA, LV dysfunction) and he had definitive indication to continue term-anticoagulation (Class I, LOE A).[@ytz242-B2] As a professional rugby player, he was susceptible to repeated trauma. Avoidance of playing rugby competitively was warranted due to the high risk of bleeding whilst on anticoagulation. If, despite counselling, athletes continue to participate in full-contact sports, advice should be offered in order to mitigate the bleeding risks. In this case, the most viable management strategy that would enable simultaneously adequate stroke prevention and a return to his professional career was an LAAO device. An LAAO device may be considered for stroke prevention in patients with AF with contra-indications to long-term anticoagulation (Class IIb, LOE B).[@ytz242-B2] The most widely used catheter-based devices are the Watchman (Boston Scientific, MA, USA) and AMULET (St. Jude Medical, MN, USA). They are all self-expanding devices deployed via a percutaneous, endocardial approach.[@ytz242-B5] Alternatively, if the left atrial appendage anatomy is deemed unsuitable, the Lariat (SentreHeart, CA, USA) combines an epicardial and endocardial technique to ligate the left atrial appendage. [@ytz242-B6] In the last decade, a large body of evidence of their efficacy and safety has emerged mainly through large multicentre global registries. Only the Watchman has been compared to vitamin K antagonists (VKAs) in two non-inferiority, randomized controlled trials, PROTECT AF, and PREVAIL.[@ytz242-B7]^,^[@ytz242-B8] In both trials, the Watchman was non-inferior to VKA for the composite primary endpoint of stroke, systemic embolism, and cardiovascular or unexplained death. Further supporting these findings, a meta-analysis combining the data of PROTECT AF and PREVAIL with two registries showed an 80% reduction in the risk of haemorrhagic stroke and a 50% reduction in the risk of cardiovascular/unexplained death when compared with VKA.[@ytz242-B9] The risk of AF in athletes appears to have a U-shaped dose--response curve to exercise, from being protective in low-intensity training to a marked increase in high-intensity endurance athletes.[@ytz242-B2]^,^[@ytz242-B10]^,^[@ytz242-B11] The pathophysiology is unclear but likely results from a complex interaction of atrial remodelling, inflammation, and increased vagal tone.[@ytz242-B12] The recommended initial approach is to assess response to a period of deconditioning for 2 months; an approach not always acceptable to athletes.[@ytz242-B13] A rhythm control strategy, including catheter ablation, may be pursued to mitigate significant AF related symptoms or to preclude the use of antiarrhythmic drugs which may impair performance or be prohibited.[@ytz242-B2] This athlete had an early recurrence of AF following DCCV and deconditioning that did not reduce his AF burden. A catheter ablation was considered but not indicated as he was asymptomatic, beta-blockers were not affecting his performance and there was no evidence of tachycardia-associated cardiomyopathy. Catheter ablation has not been shown to lower the risk of stroke (CABANA trial) and current international guidelines highlight that it is a treatment for symptoms and cannot be used a means to stop OAC in patients with a high-risk profile.[@ytz242-B2]^,^[@ytz242-B14] Conclusion ========== Management of athletes presenting with cardiomyopathy and AF is often challenging. Expert opinion should be sought, and guidance should be individualized. Management options are aimed to minimize the risks to the athletes if they choose to return to competitive sports. It is reasonable to consider an LAAO device in athletes with AF and a risk profile that would normally warrant oral anticoagulation who are competing in contact sports. Lead author biography ===================== {#ytz242-F5} Dr Andre Briosa e Gala graduated from Charles University of Prague in 2011. He completed his Foundation and Core Medical Training in the Oxford Deanery, attaining membership of the Royal College of Physicians (UK) in 2015. In 2016, in started his Cardiology training in the University Hospital of Southampton. He is currently an electrophysiology clinical research fellow in the Oxford University Hospitals with a research interest in atrial fibrillation. Supplementary material ====================== [Supplementary material](#sup1){ref-type="supplementary-material"} is available at *European Heart Journal - Case Reports* online. **Slide sets:** A fully edited slide set detailing this case and suitable for local presentation is available online as [Supplementary data](#sup1){ref-type="supplementary-material"}. **Consent:** The author/s confirm that written consent for submission and publication of this case report including image(s) and associated text has been obtained from the patient in line with COPE guidance. **Conflict of interest:** none declared. Supplementary Material ====================== ###### Click here for additional data file.

Full text ========= A new term, \'selective estrogen receptor modulator\'; (SERM), has infiltrated the estrogen receptor (ER) literature lately \[[@B1]\]. It is nothing more than the reaffirmation of an old fact, namely that different estrogens have different effects, in different tissues. The major natural estrogens -estradiol, estriol and estrone - bind ERs with differing affinities, hence variations in their tissue distribution and concentrations influence the extent of their estrogenic effects. Studies with synthetic estrogens have focused on antiestrogenic ligands, which bind ERs and interefere with the actions of the natural estrogens. Tamoxifen is the prototypical antiestrogen, and newer second-generation antagonists, such as raloxifene, are in various stages of clinical trials \[[@B2]\]. Both tamoxifen and raloxifene are SERMs, because their antiestrogenic effects are restricted to only certain tissues. Tamoxifen has been used for more than 20 years to treat ER-positive breast cancers \[[@B3]\]. It was first demonstrated to be effective in advanced disease, later in adjuvant settings, and most recently as a breast cancer preventant in women at high risk. Thus, in various settings tamoxifen is an inhibitory ER ligand in the breast, and this property explains both its efficacy and its widespread use. The picture is not all rosy, however. True to its SERM nature, tamoxifen is not antiestrogenic in all tissues. For example, in the uterus tamoxifen is a potent estrogen, where, like estradiol (when unopposed by progestins), it induces epithelial hyperplasia and endometrial cancers \[[@B4]\]. The excitement surrounding raloxifene stems from the fact that, like tamoxifen, it is an antagonist in the breast, but, unlike tamoxifen, it lacks estrogenic activity in the uterus \[[@B2],[@B5]\]. In summary, tamoxifen can be either an agonist or an antagonist in normal tissues. Unfortunately, the same duality of function operates in malignant tissues, including breast cancers. Almost without exception, breast cancers that initially respond well to tamoxifen by growth cessation or regression eventually resume growing despite the continued presence of the antagonist. How can this \'acquired resistance\'; be explained? Most tamoxifen-resistant tumors continue to express ER \[[@B6]\], suggesting that resistance is not simply due to outgrowth of a nonresponsive, ER-negative sub-population. Indeed, tamoxifen-resistant tumors remain responsive to growth inhibition by pure antiestrogens (but clinical data are sparse) and other hormonal therapies \[[@B3],[@B5]\]. Paradoxic reports of tumor stasis and even regression after tamoxifen withdrawal in resistant patients \[[@B7]\] suggest that in at least some resistant tumors the antagonist has switched to an agonist. Thus, for several years the notion has been advanced that the term \'resistance\' inappropriately describes such tumors, and that tamoxifen is not simply inactive (as implied by the term \'resistance\'), but, instead, that it has switched to an agonist, and actively stimulates tumor growth \[[@B8],[@B9]\]. That the same ligand can have opposing transcriptional and biologic effects has long been puzzling, but recent advances in our understanding of the molecular biology of steroid receptors has shed light on this paradox. We now know that transcriptional regulation by liganded, DNA-bound receptors is influenced by their association with multiprotein activator or repressor complexes. Detailed analyses of the identity and function of the constituent \'coregulatory\' proteins in these complexes are being carried out in many laboratories. They break down into two classes - coactivators and corepressors - and involve proteins with a variety of functions, including the following: enzymes such as acetylases, deacetylases, methyltransferases, ubiquitin ligases, proteases, ATPases and kinases; proteins with activator or repressor domains that stabilize or destabilize protein-protein interactions; scaffolding proteins involved in the assembly of multiprotein complexes; and even nonpeptide factors such as the steroid receptor RNA activator \[[@B10],[@B11]\]. What does this have to do with tamoxifen? It turns out that the activity of the tamoxifen-ER complex can be exquisitely modulated by the nature of the associated coregulatory proteins. Binding of corepressors, such as the silencing mediator for retinoid and thyroid receptors or nuclear receptor corepressor, suppresses the partial agonist activity of tamoxifen. At least one antagonist-specific coactivator, the L7 switch protein for antagonist, enhances the partial agonist activity of tamoxifen \[[@B9]\]. As a result of these basic molecular studies, there is now intense interest in correlating tamoxifen resistance in breast cancer with the underexpression of corepressors or the overexpression of coactivators. These proteins could clearly represent the next targets for therapeutic interventions. Additionally, although we have learned a great deal about steroid receptor coregulatory proteins in recent years, most investigators believe that only a minor subset have been identified to date. This is because the many subtle structural variations in the conformation of receptors that result from the binding of different ligands yield multiple subtly different targets on the receptor\'s surface for the binding of a variety of coregulators. It is this variability that can, in part, explain the tissue specificity and paradoxic agonist activity of ligands like tamoxifen. The hunt is therefore also on to identify the large number of endogenous coregulatory proteins that are probably lurking in tissues, and, additionally, to synthesize their pharmacologic equivalents with a view to manipulating the functional direction of ligand-receptor complexes. In a recent paper, Norris *et al* \[[@B12]\] described a novel method to define an array of synthetic peptides that interact specifically with estradiol- or tamoxifen-occupied ER, and regulate their transcriptional activity. Several methods have recently been developed to select members of random peptide libraries based on their binding affinity to known protein targets \[[@B13]\]. In the method of phage-display, a library of phage, each displaying a different cloned peptide sequence on its surface, is exposed to a plastic plate coated with the target protein. Specifically bound phage are eluted, the phage are amplified, and the process is repeated for several rounds, after which the selected clones of interest are isolated from the phage, the DNAs are sequenced, and the peptides they encode are deduced. Norris *et al* \[[@B12]\] used tamoxifen- or estradiol-occupied ER as the target protein bound to the plate, and they ensured that the receptors would be in the appropriate DNA-bound structural conformation by precoating the plastic with DNA containing estrogen response elements. The screen led to the isolation of several, 15 amino acid peptides, representing three major classes: α /β I, which interacts with estradiol-occupied ER; α /β III or V, which interact with tamoxifen-occupied ER; and α II, which interacts with ER in the presence of either ligand, in the presence of a pure antiestrogen, and even in the absence of ligand. The α /β I peptide SSNHQSSRLIELLSR interacts with ER only in the presence of estradiol, and not in the presence of SERMs like tamoxifen, raloxifene, GW7604, idoxifene, nafoxidene or the pure antiestrogen ICI182,780. In the presence of agonists, it also interacts with the progesterone receptor B-isoform, and glucocorticoid receptors. When overexpressed, α /β I and α II peptides reduce the transcriptional activity of estradiol, whereas α /β III or V have no effect, which is consistent with their inability to bind ER in the presence of the agonist. On the other hand, peptides α /β III or V are quite tamoxifen-specific for ER, but also bind antagonist-occupied progesterone receptors. Six peptides of the α /β V class were isolated, that had the consensus sequence (S/M)X(D/E)(W/F)(W/F)XXXL. α /β III or V, as well as α II, inhibit the partial agonist effect of tamoxifen, but do not alter transcription by estradiol-occupied ER. The inhibitory activity of these synthetic peptides thus resembles that of the natural corepressors SMRT and N-CoR \[[@B9]\]. It would be of interest to determine whether the complementary DNAs encoding these synthetic peptides could be used as probes to isolate additional endogenous corepressors from complementary DNA libraries. At present the list of known corepressors is much smaller than that of known coactivators \[[@B11]\], and it is unclear whether this discrepancy represents a true cellular condition, or whether it is an artifact due to the technical complexity of screening for corepressors. Norris *et al* \[[@B12]\] speculated that each class of peptides recognizes different protein contact sites on the ER protein; contact sites that are generated specifically by the class of ligand bound to the receptors. They postulated that these contact sites could be targets for drug discovery. Analogous suggestions have previously been made for the use of corepressor or coactivator-occupied receptors to screen for new ligands \[[@B9]\]. The studies of Norris *et al* \[[@B12]\], along with those of others cited herein, indicate that we are at the brink of important insights into the molecular mechanisms by which ER and their ligands regulate hormone dependence and resistance in breast cancers. These insights will bring completely new approaches to treating these tumors, and if their promise is confirmed they will allow us to predict, and pehaps even prevent or reverse, development of resistance. It is an exciting time to be studying the roles of steroid hormones in breast cancer!

Introduction {#s1} ============ The assessment of human daily physical activity in population studies requires accurate, cheap, and feasible measurement technology [@pone.0061691-Corder1], [@pone.0061691-Wareham1], [@pone.0061691-Wong1]. Accelerometers are increasingly being used for physical activity assessment and most of the accelerometers that have been used in population studies express their output in proprietary units usually referred to as "counts" [@pone.0061691-Hagstromer1], [@pone.0061691-Colley1]. Accelerometer devices, based on acceleration sensors which allow for raw data storage expressed in g-units or SI units at a relatively high sampling frequency have been used in gait analysis [@pone.0061691-Brandes1], [@pone.0061691-MoeNilssen1] and ambulant activity classification [@pone.0061691-Aminian1], [@pone.0061691-Veltink1] for a number of years. The output of raw accelerometers is not summarized by the monitor allowing for increased control over data processing by the end-user in contrast to the traditional accelerometers. Technological developments in recent years have made raw accelerometry feasible for population research, allowing weeklong data collection. A measured acceleration signal consists of a gravitational component, a movement component, and noise [@pone.0061691-Veltink1]. During static conditions or conditions of steady state non-rotational movement, the gravitational component is visible as the offset of one or more sensor axes and can then be used for detection of the sensor orientation relative to the vertical plane [@pone.0061691-Veltink1]. The separation of the gravitational component from the acceleration signal is complicated by the fact that in the presence of rotational movements the frequency domains of the movement-related component and the gravitational component can overlap, thus making simple frequency-based filtering inappropriate for perfect separation. The first two studies that identified the challenge of separating the components of acceleration lacked a comparison against a reference method [@pone.0061691-Redmond1], [@pone.0061691-VanSomeren1]. Studies by Bouten et al. and Bourke et al. used a reference method, but were limited to laboratory experiments that may not generalise to accelerometer data collected under real life conditions [@pone.0061691-Bourke1], [@pone.0061691-Bouten1]. None of the studies as mentioned above systematically evaluated how metric accuracy varies across magnitudes and frequencies of acceleration. Characterisation of the latter may be important to gain insight into metric performance under real-life conditions. The use of gyroscopes in addition to acceleration sensors could be regarded as the solution for separating the gravitational component from the acceleration signal [@pone.0061691-Roetenberg1], [@pone.0061691-Sabatini1], [@pone.0061691-Yun1]. However, these devices do not yet meet feasibility requirements for use in large scale observational research. Raw accelerometry has been applied in various epidemiological studies since it became sufficiently feasible in the period 2008--2010. Most of these studies are not published yet, but already amount to over ten thousand participants. None of these datasets include gyroscopic data and therefore require an accelerometer-specific solution. The main objective of the present study was therefore to evaluate the ability of different methods (metrics) of processing acceleration signals to remove the gravitational component of acceleration by comparison against a reference method under a range of standardised kinematic conditions. A second objective was to assess the shared variance between these metrics in human physical activity data collected during daily life and the impact of metric selection on the accuracy with which daily energy expenditure can be estimated. Methods {#s2} ======= Ethics Statement {#s2a} ---------------- Ethical approvals were obtained from the Cambridgeshire research ethics committee, Cambridge (UK) and from the Regional Ethical Review Board in Umeå (Sweden). Study Design {#s2b} ------------ The main experiment in this study was done with a robot and did not involve testing of human participants. Two additional sets of experiments were performed, the first to test the degree to which metrics convey similar information when applied to wrist and hip signals, and the second to assess the implication of such differences for estimation of daily physical activity-related energy expenditure. Robot Experiment {#s2c} ---------------- An industrial robot (TX90, Stäubli Tec-Systems GmbH, Bayreuth, Germany; see [**Figure 1**](#pone-0061691-g001){ref-type="fig"}) was used to rotate accelerometers (GENEA, Unilever Discover, Sharnbrook Bedfordshire, UK) in the vertical plane following a general minimum-jerk oscillatory motion (single plane). The motion was applied to establish a standardized alternating contribution of gravity to the accelerometer output. The robot consists of an articulated arm with six joints from which the fifth joint counted from the base of the robot was used in this study. The oscillating motion was continuous (non-damping) around a single horizontal axis. The trajectory was programmed using a 7th order polynomial function with kinematic constraints **([Supporting Information S1](#pone.0061691.s001){ref-type="supplementary-material"})**. A high order function was needed to reduce the natural vibrations transmitted between the robot and its own base [@pone.0061691-Piazzi1], [@pone.0061691-Kyriakopoulos1]. An example of the angular position over time for one experimental condition is given in [**Figure 2**](#pone-0061691-g002){ref-type="fig"}. {#pone-0061691-g001} {#pone-0061691-g002} The frequency of oscillation, the radius of rotational movement (shortest distance to centre of rotation), and the angular range of motion were systematically varied. The range of frequency conditions was limited by the maximal amount of mass moment of inertia and torques that could be absorbed by the robot and supporting frame. For all frequencies ranging from 0.05 Hz to 1.2 Hz, eighteen tri-axial accelerometers were positioned along the length of a 70 cm bar mounted to the flange of the robot at 10 cm from the centre of rotation. The application of eighteen accelerometers in parallel allowed for assessment of the relationship between metric output and the radius of movement. To reduce mass moment of inertia at the higher frequencies of oscillation (\>1.1 Hz) a shorter bar (20 cm) was used, see [**Figure 1**](#pone-0061691-g001){ref-type="fig"}. The shorter bar provided space for the attachment of only five accelerometers. The torque can be further reduced by reducing the range of angular rotation; some experimental conditions were defined by this constraint. For reference purposes, all eighteen accelerometers were also tested under static conditions (no robot movement) at angles 0° and 22.5°. Each experimental condition was done for three minutes. An overview of all experimental conditions is shown in [**Table 1**](#pone-0061691-t001){ref-type="table"}. For monitoring potential vibrations, a source of experimental error, one additional accelerometer was attached to the base of joint 5 for all experimental conditions. The base of joint 5, i.e. the robotic with its joint 1 up to joint 4, should in theory not move during these experiments. 10.1371/journal.pone.0061691.t001 ###### Experimental conditions of the robot setup. {#pone-0061691-t001-1} Frequencies Angle range\* Number of accelerometers (range in position relative to axis of rotation) -------------------------------------------------------- --------------- --------------------------------------------------------------------------- 0 Hz 0° and 22.5° 18 (0.13--0.78 m) 0.05 to 0.55 Hz (steps of 0.05) 0--90° 18 (0.13--0.78 m) 0.60, 0.70, and 0.80 Hz 0--45° 18 (0.13--0.78 m) 0.90, 1.00, and 1.10 Hz 0--20° 18 (0.13--0.78 m) 1.20 and 1.30 Hz 0--45° 5 (0.13--0.29 m) 1.4 to 2.6 (steps of 0.1), 2.8, 3.0, 3.2, 3.6 and 4 Hz 0--20° 5 (0.13--0.29 m) \[\*for 0° the bar is in horizontal position and for 90° the bar is pointing upwards relative to the axis of rotation\]. Human Experiments {#s2d} ----------------- In order to facilitate the interpretation of the robot experiment in the context of human daily (free-living) physical activity, we asked 47 men and 50 women (healthy, aged 22--65 yrs) to wear accelerometers on their wrist and on their hip for seven days during free-living as previously described [@pone.0061691-vanHees1]. We also re-analysed wrist acceleration signals obtained during free-living conditions from 65 healthy women (aged 20--35 yrs) as previously described [@pone.0061691-vanHees1]. In this latter sample, physical activity-related energy expenditure (PAEE) was assessed using the doubly labelled water method in combination with resting energy expenditure measured by indirect calorimetry [@pone.0061691-vanHees1]. For both human studies, objectives and procedures were explained in detail to the participants, after which they provided written and verbal informed consent. Accelerometer {#s2e} ------------- The accelerometer comprised a tri-axial STMicroelectronics accelerometer (LIS3LV02DL) with a dynamic range of ±6 g (1 g = 9.81 m·s^−2^), as described elsewhere [@pone.0061691-van1]. The acceleration was sampled at 80 Hz and data were stored in g units for offline analyses. In the robot experiment, the accelerometer was aligned by two aluminium strips on each side of the bar (insert, [**Figure 1**](#pone-0061691-g001){ref-type="fig"}) and covered by duck-tape on top, see [**Figure 1**](#pone-0061691-g001){ref-type="fig"}. The radius length, i.e. the distance from the axis of rotation to the accelerometer chip, was assessed by measurement tape to the closest mm. The position of the accelerometer chip inside the accelerometer packaging was obtained from the manufacturer. In the human experiment, the accelerometers were attached to the wrist with a nylon weave strap and to the hip with an elastic belt. Participants were instructed to wear the accelerometer on the wrist continuously for 24 hours per day throughout the whole observation period and to remove the hip accelerometer during sleeping hours. The manufacturer calibration of all acceleration sensors was tested under static conditions (no movement, vector magnitude = 1 g) and adjusted if necessary. Metrics {#s2f} ------- For the robot analyses three metrics for the estimation of acceleration related to movement were evaluated: (i) the Euclidean norm (vector magnitude) of the three raw signals minus 1, referred to as ENMO; (ii) the application of a high-pass frequency filter (4^th^ order Butterworth filter with ω~0~ = 0.2 Hz) to each raw signal, after which the Euclidean norm was taken from the three resulting signals, , referred to as HFEN, and; (iii) metric HFEN plus the Euclidean norm of the three low-pass filtered raw signals (4^th^ order Butterworth with ω~0~ = 0.2 Hz) minus 1 g, referred to as HFEN~+~. The third metric has not been described previously. The motivation for metric HFEN~+~ is as follows: In the absence of rotational movement the Euclidian norm of the three low-pass filtered raw signals (LFEN) is equal to 1 g. In the presence of rotation, however, LFEN may be different to 1 g due to imperfect separation; there we add this difference (positive or negative) to HFEN. A low frequency component above 1 g may result from low-frequency accelerations perpendicular to the direction of rotation, e.g. the centripetal force when sitting on a swing. A low frequency component below 1 g could indicate that part of the gravitational component is still contained in the high-frequency content, e.g. rotations in the vertical plane as a result of which gravity is an alternating component in the signal. A further elaboration on the motivation for metric HFEN~+~ can be found in **[Supporting Information S1](#pone.0061691.s001){ref-type="supplementary-material"}**. For some of the metrics described above the output could in theory be negative. To gain insight into when this happens, negative values were not corrected for the robot experiment. However, for the accelerometer data collected in daily human movement, negative metric output was rounded off to zero before further analysis. The filter cut-off frequency of 0.2 Hz for metrics HFEN and HFEN~+~ was chosen on the presumption that most of daily acceleration related to movement for most human body parts occurs at frequencies higher than 0.2 Hz. n the robot experiment, the exact absolute value of this filter cut-off frequency (0.2 Hz) was considered of minor relevance as this experiment intends to investigate frequency of rotation and frequency of filtering on a relative scale. For the human part of our study, both a cut-off frequency of 0.2 Hz and 0.5 Hz were evaluated to assess the effect of threshold selection in relation to human movement. Additionally, the human part of our study was extended with the application of a band-pass frequency filter version of HFEN (4^th^ order Butterworth filter with ω~0~ = 0.2--15 Hz), referred to as BFEN, to assess the effect of high-frequency noise removal. Finally, the Euclidean norm of the three raw acceleration signals (EN) without subtraction of gravity was added to the evaluations in human data to assess the relevance of attempting to remove the gravitational component from an applied perspective. To sum up, metrics evaluated in this investigation include Euclidian Norm (EN), Euclidian Norm Minus One (ENMO), Bandpass-Filtered followed by Euclidian Norm (BFEN), Highpass-Filtered followed by Euclidian Norm (HFEN), and Highpass-Filtered followed by Euclidian Norm Plus difference between 1 g and low-pass-filtered component (HFEN~+~). Analysis {#s2g} -------- Reference values for robot acceleration were calculated based on forward kinematics of the robot arm using the radius length () of each accelerometer relative to the axis of rotation and the robot arm's angle , angular velocity , and angular acceleration over time. Although the robot recorded the joint angle at 250 Hz, this information was not used due to known issues of numerical noise in the derivation of angular velocity and angular acceleration. Instead, the angular velocity and angular acceleration were derived analytically by taking the first and second derivative of the input command equations describing the angular motion as used for controlling the robot. Next, equation I was used to calculate reference acceleration . Here, represents the tangential acceleration and represents the centripetal acceleration, which when taken together as the vector magnitude add up to the overall acceleration of the accelerometer. The average metric output and reference values were calculated over an integer number of oscillating periods in the middle two minutes of each experimental condition (3 minutes), after which absolute and relative measurement errors were expressed. Relative errors were calculated as (Estimated -- Reference)/Reference. For reference purposes, all analyses were repeated based on simulated acceleration signals using the equations as in equation II and equation III. Here, refers to the acceleration signal perpendicular to the length of the bar which captures the tangential acceleration combined with the effect of the gravitational component and refers to the acceleration signal in parallel to the length of the bar which captures the centripetal acceleration combined with the gravitational component. The centre of rotation is assumed to not change position. Metrics ENMO, HFEN, HFEN~+~, BFEN and EN were applied to the raw data collected on the wrist and hip (7 days) after which metric output was averaged over consecutive non-overlapping 1 minute time windows. Further, metrics ENMO, HFEN, HFEN~+~, BFEN and EN were applied to the raw data collected in the human participants where PAEE reference data was available. Here, metric output was averaged per person. A detailed description of the detection of monitor non-wear periods and signal clipping are provided in **[Supporting Information S1](#pone.0061691.s001){ref-type="supplementary-material"}**. Fifteen minute blocks that were classified as non-wear or clipping were replaced by the average of blocks at the same time periods of the day (from the other days in each individual record). If no data was collected for a certain part of the day then it was imputed by 1 g for metric EN and by 0 g for all other metrics. All signal processing and statistics were performed in R (<http://cran.r-project.org>). Statistics {#s2h} ---------- Means and (relative) differences were computed for the data resulting from the robot experiment. In order to evaluate whether differences between metrics resulted in different measures of free-living human movement, repeated measures ANOVA was used to assess the within- and between-individual explained variance between metrics, stratified by wrist and hip placement. Analyses were performed for all data points excluding non-wear time segments and repeated including imputed data for non-wear time segments. The most important difference is that this would either include or exclude hip accelerometer values for sleeping hours. Results were very similar, and we only report results excluding non-wear time for these analyses. Average and standard deviation of metric output are reported based on imputed data to facilitate the comparison between this study population with future study populations. For the PAEE analyses, participant inclusion criteria were identical to our previous work [@pone.0061691-vanHees1]: more than 50% detected monitor wear time and at least one day of valid data. Linear regression analysis was used to assess how much of the variation in daily PAEE, expressed in MJ/day, can be explained by each metric in combination with body weight. Additionally, we tested the additive value of metrics by adding combinations of metrics to the regression model. Results {#s3} ======= Robot conditions and corresponding reference acceleration are presented in [**Figure 3**](#pone-0061691-g003){ref-type="fig"}. The accelerometer attached to the base of joint 5, which in theory should not move, recorded a magnitude of acceleration (vibration) beyond the sensor's noise level (SD: 2.6 mg = 0.0026 g) for most experimental conditions. On average the acceleration of the robot joint was 4% to 5% of the average acceleration of the accelerometers on the bar attached to the flange, see [**Table 2**](#pone-0061691-t002){ref-type="table"}. The highest value of 76% for ENMO was the result of computed acceleration being close to zero (−5.13 mg). {#pone-0061691-g003} 10.1371/journal.pone.0061691.t002 ###### Average (mg) and relative (%) acceleration of the base of joint 5 (should ideally be zero) by experimental condition and metric. {#pone-0061691-t002-2} Metrics ------------ ------------- --------- ------ ------ 0.05--0.2 0--90 −3.9 13.4 9.4 76.0% 7.1% 6.3% 0.25--0.55 0--90 −4.9 14.2 9.2 −12.3% 2.2% 2.4% 0.6--0.8 0--45 −2.9 18.9 15.7 −8.1% 3.7% 3.7% 0.9--1.1 0--20 0.9 21.5 22.0 6.7% 5.8% 6.4% 1.2--1.3 0--45 1.5 9.3 10.8 2.0% 1.4% 1.9% 1.4--2.0 0--20 0.1 35.9 35.2 0.4% 7.8% 7.9% 2.1--3.0 0--20 1.4 17.1 18.3 0.7% 1.9% 2.1% 3.2--4.0 0--20 2.3 74.8 73.5 0.2% 4.3% 4.3% **Average** 0.7 25.6 24.3 8.2% 4.3% 4.4% Relative values are expressed as percentage of average metric output for all accelerometers attached to the bar as fixed to the flange. The metric output for each accelerometer attached to the bar was compared against the reference acceleration. Metric HFEN~+~ was more accurate compared to metric HFEN with an average difference in absolute measurement error of respectively, 90 mg and 109 mg. Measurement error was lowest for metric HFEN~+~ in all but one experimental conditions based on oscillation frequencies higher than 0.2 Hz. On the contrary, metric ENMO outperformed the other metrics for frequencies of oscillation below 0.2 Hz, see [**Table 3**](#pone-0061691-t003){ref-type="table"}. For all metrics, except ENMO, relative and absolute measurement error was lower for higher radius settings, see [**Table 3**](#pone-0061691-t003){ref-type="table"}. 10.1371/journal.pone.0061691.t003 ###### Evaluation of metrics using empirically recorded acceleration signals. {#pone-0061691-t003-3} Freq.(Hz) Angle (°) Radius (m) Acc. (mg) ENMO HFEN HFEN~+~ ------------ ----------- ------------ ----------- ------------ ------------ ------------ 0\* 0 0.1--0.3 0 −9 4 −5 0\* 0 0.3--0.6 0 0 6 6 0\* 0 0.6--0.8 0 −3 9 6 0\* 22.5 0.1--0.3 0 −4 3 0 0\* 22.5 0.3--0.6 0 −11 5 −4 0\* 22.5 0.6--0.8 0 −11 7 −4 0.05--0.2 0--90 0.1--0.3 14 −16(−173) 167 (1427) 132 (1184) 0.05--0.2 0--90 0.3--0.6 31 −38 (−162) 155 (619) 112 (447) 0.05--0.2 0--90 0.6--0.8 48 −55 (−144) 152 (442) 107 (343) 0.25--0.55 0--90 0.1--0.3 129 −122 (−98) 435 (498) 212 (272) 0.25--0.55 0--90 0.3--0.6 281 −251 (−93) 364 (194) 89 (76) 0.25--0.55 0--90 0.6--0.8 434 −354 (−86) 308 (108) −3 (24) 0.6--0.8 0--45 0.1--0.3 161 −153 (−97) 206 (149) 141 (102) 0.6--0.8 0--45 0.3--0.6 351 −328 (−95) 152 (49) 57 (21) 0.6--0.8 0--45 0.6--0.8 541 −465 (−87) 118 (24) 9 (3) 0.9--1.1 0--20 0.1--0.3 134 −128 (−99) 93 (78) 83 (67) 0.9--1.1 0--20 0.3--0.6 293 −292(−100) 73 (27) 44 (17) 0.9--1.1 0--20 0.6--0.8 451 −419 (−93) 68 (16) 35 (8) 1.2--1.3 0--45 0.1--0.3 508 −432 (−87) 160 (35) 63 (14) 1.4--2.0 0--20 0.1--0.3 390 −364 (−95) 72 (22) 54 (16) 2.1--3.0 0--20 0.1--0.3 832 −618 (−79) 47 (7) 22 (3) 3.2--4.0 0--20 0.1--0.3 1700 −779 (−50) 45 (3) 14 (1) Values are average absolute differences in mg (average relative error % in brackets §) between each metric output and the actual acceleration related to movement for various sections of the experiment. \[Acc, average reference acceleration; \*zero movement condition; § Relative measurement error was calculated per experimental condition and then averaged across each section of the experiment\]. Replication of the analyses with simulated acceleration signals confirmed the empirical findings as described above. A detailed overview of the results based on simulated acceleration signals are included in **[Supporting Information S1](#pone.0061691.s001){ref-type="supplementary-material"}**. Data and R-scripts related to the robot experiments are available on our website: <http://www.mrc-epid.cam.ac.uk/research/resources>. When metrics were applied to human wrist and hip acceleration signals collected during free-living conditions, repeated measures ANOVA showed that the shared within- and between-individual variances (r-squared) varied between metric pairs and body locations, see [**Table 4**](#pone-0061691-t004){ref-type="table"} **and** [**Table 5**](#pone-0061691-t005){ref-type="table"}. Lowest shared variance was found for metric-pairs involving metric EN; for example, this metric shared 54 and 11% of the within- and between--individual variance, respectively, with metric BFEN for hip acceleration, see [**Table 5**](#pone-0061691-t005){ref-type="table"}. Highest shared variances were observed between the filter-based metrics. For example, metrics HFEN and BFEN as well as versions of HFEN with different cut-off frequencies were all highly correlated both within and between individuals and for both hip and wrist data (r-square values \>0.96), see [**Table 4**](#pone-0061691-t004){ref-type="table"} **and** [**5**](#pone-0061691-t005){ref-type="table"}. A difference between wrist and hip worth noting was the shared variance between ENMO and the filter-based metrics HFEN, BFEN and HFEN+. Here, the shared variance within individuals was highest for the hip (0.92 vs. 0.87 on average), while the shared variance between individuals was highest for the wrist (0.87 vs. 0.62 on average), see [**Table 4**](#pone-0061691-t004){ref-type="table"} **and** [**Table 5**](#pone-0061691-t005){ref-type="table"}. 10.1371/journal.pone.0061691.t004 ###### Explained variance (r^2^) within (above diagonal) and between (below diagonal) individual wrist accelerometer data for all combinations of data processing metrics. {#pone-0061691-t004-4} ω~0~ (Hz) EN ENMO BFEN HFEN HFEN HFEN~+~ HFEN~+~ --------------------------------- ----------- -------- --------- ----------- -------- --------- --------- --------- **ω~0~ (Hz)** **−** **−** *0.2--15* *0.2* *0.5* *0.2* *0.5* **EN** **−** **−** 0.91 0.61 0.62 0.71 0.75 0.80 **ENMO** **−** 0.92 **−** 0.80 0.81 0.89 0.91 0.95 **BFEN** *0.2--15* 0.58 0.80 **−** 0.99 0.96 0.96 0.93 **HFEN** *0.2* 0.60 0.82 1.00 **−** 0.98 0.97 0.94 **HFEN** *0.5* 0.64 0.88 0.98 0.99 **−** 0.98 0.98 **HFEN~+~** *0.2* 0.74 0.91 0.97 0.97 0.98 **−** 0.99 **HFEN~+~** *0.5* 0.77 0.95 0.94 0.95 0.98 0.99 **−** *Mean (sd) acceleration \[mg\]* 1016(9) 32(10) 114(25) 118(26) 93(22) 110(25) 94(23) \[ω~0~: cut-off for frequency filter\]. 10.1371/journal.pone.0061691.t005 ###### Explained variance (r^2^) within (above diagonal) and between (below diagonal) individual hip accelerometer data for all combinations of data processing metrics. {#pone-0061691-t005-5} ω~0~ (Hz) EN ENMO BFEN HFEN HFEN HFEN~+~ HFEN~+~ --------------------------------- ----------- -------- -------- ----------- -------- -------- --------- --------- ω~0~ (Hz) − − *0.2--15* *0.2* *0.5* *0.2* *0.5* EN − − 0.77 0.54 0.55 0.58 0.61 0.63 ENMO − 0.75 − 0.89 0.90 0.92 0.94 0.95 BFEN *0.2--15* 0.11 0.46 − 1.00 0.99 0.99 0.98 HFEN *0.2* 0.10 0.46 1.00 − 0.99 0.99 0.98 HFEN *0.5* 0.11 0.48 0.98 0.98 − 0.97 0.99 HFEN~+~ *0.2* 0.52 0.85 0.78 0.78 0.75 − 0.99 HFEN~+~ *0.5* 0.54 0.89 0.76 0.75 0.76 0.99 − *Mean (sd) acceleration \[mg\]* 1007(15) 18(16) 46(15) 48(15) 42(14) 50(21) 45(20) \[ω~0~: cut-off for frequency filter\]. For the modelling of PAEE, HFEN~+~ outperformed metrics ENMO, HFEN, BFEN and EN, explaining 36% of the variance in daily PAEE, see [**Table 6**](#pone-0061691-t006){ref-type="table"}. When pairs of metrics were added to the regression model, no significant additive value was found (p\>0.05 corresponding with increases in model r^2^ of less than 0.01). 10.1371/journal.pone.0061691.t006 ###### Overview of regression models for predicting PAEE (MJ day^−1^) based on N = 63 women. {#pone-0061691-t006-6} Model input ω~0~ (Hz) SE R^2^ Equation ------------- ----------- ------ ----------------------------------------- --------------------------------------- EN − 0.99 0.26[\*](#nt109){ref-type="table-fn"} −56.146 + BW × 0.023 + EN × 57.093 ENMO − 0.94 0.34[\*\*](#nt108){ref-type="table-fn"} −0.172 + BW × 0.025 + ENMO × 0.057 BFEN 0.2--15 0.97 0.30[\*\*](#nt108){ref-type="table-fn"} −0.913 + BW × 0.021 + BFEN × 0.023 HFEN 0.2 0.97 0.30[\*\*](#nt108){ref-type="table-fn"} −0.905 + BW × 0.021 + HFEN × 0.023 HFEN 0.5 0.95 0.32[\*\*](#nt108){ref-type="table-fn"} −0.769 + BW × 0.022 + HFEN × 0.027 HFEN~+~ 0.2 0.93 0.36[\*\*](#nt108){ref-type="table-fn"} −1.114 + BW × 0.023 + HFEN~+~ × 0.025 HFEN~+~ 0.5 0.93 0.36[\*\*](#nt108){ref-type="table-fn"} −0.805 +BW × 0.023 + HFEN~+~ × 0.026 \[SE: Residual standard error; : p\<.001; : p\<.01; ω~0~: cut-off for frequency filter; BW = body weight (kg)\]. Discussion {#s4} ========== The present study demonstrates that the choice of signal processing technique for summarising accelerometer data can have a substantial impact on the accuracy with which acceleration related to movement is measured. Subsequently, the choice of signal processing technique impacts on the summary measures of human acceleration data and criterion-related validity for estimating daily PAEE. In the past, physical activity researchers did not have the opportunity to select a metric; the metric decision was made by the manufacturer of the accelerometer [@pone.0061691-Plasqui1], [@pone.0061691-Bonomi1], [@pone.0061691-Corder2], [@pone.0061691-Assah1], [@pone.0061691-Rothney1]. The first and main part of this paper evaluated metrics under a range of standardised kinematic conditions in order to gain insight into how the accuracy of metric output relates to the kinematics of movement. No single metric outperformed all other metrics for all experimental conditions. Metric HFEN~+~ resulted in less measurement error compared to metric HFEN. This result may indicate that HFEN~+~ manages to retrieve some of the non-gravitational acceleration in the lower frequency range and/or remove gravitational acceleration from the frequency range above the filter threshold in contrast to metric HFEN. Metric HFEN~+~ outperformed metrics ENMO and HFEN for the experimental conditions based on oscillating frequencies higher than the cut-off frequency as used by its frequency filter (0.2 Hz), while the ENMO metric outperformed metrics HFEN and HFEN~+~ for experimental conditions based on oscillating frequencies below this cut-off frequency. This difference between HFEN, HFEN~+~ and ENMO may partly be explained by the fact that metrics HFEN and HFEN~+~ aim to remove the gravitational component by making assumptions on its representation in the frequency content of an acceleration signal, while ENMO aims to remove the gravitational component based on assumptions with regard to its magnitude. Metric HFEN~+~ could be seen as a hybrid version of the two approaches as it relies on both an assumption about the representation of gravity in the frequency domain and an assumption about the magnitude of gravity. The mutual assumption by metrics ENMO and HFEN~+~ that gravity is measured as 1 g would not hold true if acceleration sensors are not accurately calibrated and would therefore result in biased metric output. Further, metric ENMO has one additional limitation: For a signal with an offset of 1 g (e.g. containing the gravitational component) and an amplitude of less than 1, taking the square will increase the amplitude. On the contrary, if the square is taken from a signal with no offset (e.g. no gravity) and the amplitude is less than one, then taking the square will decrease the amplitude. Therefore, taking the square of three orthogonal signals like in metric ENMO will result in a stronger contribution of vertical accelerations that alternate around 1 g to the resulting summary measure compared with horizontal accelerations that alternated around 0 g. The reference acceleration as used for the evaluation of the metrics may not have been exactly equal to the true acceleration that the accelerometers were exposed to; imprecision in accelerometer positioning and system vibrations are possible sources of error. In theory, the acceleration of a rotating and non-translating object is proportional to the distance from its centre of rotation, the radius length. A discrepancy of 5 mm (plausible) in the assessment of accelerometer position would represent 0.6% for the accelerometer farthest away and 3.7% for the accelerometer closest to the axis of rotation. This would translate into a similar degree of error in the calculated reference acceleration (0.6--3.7%). Secondly, vibrations of the whole robot during operation may have resulted in the true acceleration exposure being higher than what we calculated it to be. The accelerometer attached to the base of joint 5 did record acceleration beyond the sensor's noise level likely resulting from the movement of the robot system itself. We believe that robot movement was caused by the supporting frame that vibrated towards the extreme experimental conditions; the robot itself has a high stiffness. The accelerometers attached to the bar mounted on the flange have been exposed to these vibrations as well as those intended by the experimental design. The replication of the robot analyses with simulated acceleration signals confirmed the empirical findings, indicating that environmental vibrations had no significant impact. As for the analyses conducted on data collected during human daily life, the shared within-individual variances were all above 80% between metrics which make some attempt at removing the gravitational component, indicating the pattern within an individual is picked up quite similarly between those metrics. The between-individual shared variances, which is a measure of the metrics' ability to rank individuals similarly, showed some differences between hip and wrist positions, most notably lower similarity between ENMO and frequency-filtered metrics for hip than wrist. Whether this reflects differences in monitoring protocols (24-hr vs. non-sleep time), differences in signal to error ratio and/or differences in frequency characteristics of the gravitational component as measured by triaxial accelerometry at these two positions is difficult to conclude from our data. However, it should be noted that shared variances only indicate to what extent metrics are similar in describing variance on a relative level but not what the shared variance represents; it will also include any correlated measurement error and should therefore be interpreted with caution. Physical activity-related energy expenditure and body acceleration are only distally related to each other. As a consequence, differences in explained variance in daily PAEE does not serve as direct evidence for a metric's ability to remove the gravitational component. HFEN~+~ outperformed HFEN when using daily PAEE as a reference, which confirms the findings from the higher frequency conditions in the robot experiment. Further, ENMO turned out to be a good alternative for HFEN~+~. The correspondence between the strong performance of ENMO in explaining variance in PAEE in the current analysis with the strong performance of ENMO in the lower frequency range of the robot experiment might indicate that wrist acceleration in daily life is dominated by translational accelerations and/or accelerations resulting from low frequency rotations. A second explanation for the strong performance of metric ENMO may be its higher sensitivity to vertical accelerations (vertical acceleration is amplified) as explained above. The latter would indicate that vertical wrist accelerations are the stronger determinant of daily PAEE compared with accelerations in the horizontal plane. A third and final explanation could be that ENMO is more accurate at measuring translational acceleration compared with some of the other metrics, as the signal is never deformed by frequency filtering in ENMO. The subtraction of one in ENMO has a constant effect on all the metric output and would in theory be perfectly correlated with EN, which should therefore correlate the same with PAEE. However, there is one additional difference between the two metrics, namely the replacement of negative values by zero in ENMO, which explains why metric ENMO outperforms metric EN for the prediction of PAEE. The truncation of negative values to zero could be hypothesized to be an effective correction mechanism for errors in the subtraction of the gravitational component. Filter settings for HFEN and HFEN~+~ were briefly evaluated indicating that a 0.5 Hz filter cut-off frequency may perform slightly better than a 0.2 Hz filter cut-off frequency for predicting PAEE. A more thorough optimization of filter settings could lead to further improvement but also introduces the risk of over-fitting filter configurations to one study population, which may not generalise to others. One previous study investigated the need for removing the gravitational component using metabolic energy expenditure as reference method and concluded that attempting to remove the gravitational component is not worth the effort [@pone.0061691-Bouten1]. In that particular study, body segment position and orientation over time were derived from a 2D optical system and used to simulate acceleration sensor output [@pone.0061691-Bouten1]. The validity of these simulations was only assessed for the lower back position and not for the five other simulated sensor positions, complicating the interpretation of study results. Our own results indicate that attempting to remove the gravitational component is worth the effort for estimating daily PAEE in humans based on wrist accelerometry as ENMO, HFEN and HFEN~+~ clearly outperformed metric EN. Additional research is needed to explore the potential of combining metrics in a fashion that the best metric is chosen depending on the kinematic conditions. It should be noted that all PAEE-related results apply to the wrist placement and cannot be generalized to other body locations. Future research is therefore also needed to explore the importance of metric selection for other body locations, in particular commonly used positions at the lower back and the hip. Conclusions {#s4a} ----------- In conclusion, none of the metrics as evaluated systematically outperformed all other metrics across a wide range of standardised kinematic conditions. However, choice of metric explains different degrees of variance in daily physical activity. Supporting Information {#s5} ====================== ###### **Additional information on signal processing and replication of robot findings with simulated data.** (DOC) ###### Click here for additional data file. We thank those who participated in the study. We acknowledge Unilever Ltd for loan of GENEA monitors. Finally, we would like to thank Antony Wright and Les Bluck (MRC Human Nutrition Unit) for their involvement in the analysis of isotopic enrichment of urine samples. [^1]: **Competing Interests:**Vincent van Hees, who led on this manuscript, was funded by a BBSRC industry-CASE studentship. This studentship came with funding from both the BBSRC and an industry partner, Unilever Discover Ltd in this case (<http://www.bbsrc.ac.uk/web/FILES/Guidelines/studentship_handbook.pdf>). Unilever Discover Ltd had no involvement in the study as presented and was only informed about progress and final results. This does not alter the authors\' adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: VVH LG ECDL ME MP ST UE FR PWF AH SB. Performed the experiments: VVH LG ME ECDL FR. Analyzed the data: VVH. Contributed reagents/materials/analysis tools: VVH ECDL ME. Wrote the paper: VVH SB.

Introduction {#Sec1} ============ Prion diseases, such as bovine spongiform encephalopathy (BSE) in cattle, scrapie in sheep, and Creutzfeldt-Jakob disease (CJD), kuru and Gerstmann-Sträussler-Scheinker (GSS) syndrome in humans, are a group of neurodegenerative disorders caused by prions, self-replicating β-sheet-rich infectious polymeric assemblies of misfolded host-encoded cellular prion protein (PrP^C^)^[@CR1]--[@CR4]^. Whilst rare, prion diseases are an area of intense research interest, as it is increasingly recognised that other degenerative brain diseases, such as Alzheimer's and Parkinson's diseases, also involve the accumulation and spread of aggregates of misfolded host proteins through an analogous process of seeded protein polymerisation^[@CR2],[@CR5]--[@CR8]^. Consequently, study of 'prion-like' mechanisms has been recognised to have much a wider relevance to the understanding of neurodegenerative disorders^[@CR9]--[@CR11]^. PrP^C^ is a cell surface, predominantly α-helical, glycosylphosphatidylinositol (GPI)-anchored glycoprotein that is sensitive to protease treatment and soluble in detergents^[@CR1]^. In contrast, prions may acquire protease-resistance and are classically designated as PrP^Sc^ (refs. ^[@CR12],[@CR13]^[@CR13]. PrP^Sc^ is found only in prion-infected tissue and is β-sheet-rich aggregated material, partially resistant to protease treatment, and insoluble in detergents^[@CR14]^. Transmission experiments to transgenic mice provide strong supporting evidence that alternative conformers or assembly states of PrP^Sc^ encode multiple prion strains, which differ in their pathogenic properties^[@CR2],[@CR15]^. Transgenic mice expressing only human PrP with either valine or methionine at residue 129 have shown that this common human polymorphism constrains the propagation of distinct human prion conformers, and the occurrence of associated patterns of neuropathology consistent with the conformational selection model of prion propagation^[@CR16]--[@CR20]^. Heterozygousity at codon 129 is thought to confer resistance to prion disease by inhibiting homologous protein--protein interactions essential for efficient prion replication with the presence of methionine or valine at residue 129 controlling the propagation of distinct human prion strains^[@CR2],[@CR21]^. Biophysical measurements suggest that this powerful effect of residue 129 on prion strain selection is likely to be mediated via its effect on the conformation of the disease-associated PrP^Sc^ form, or its precursors or on the kinetics of their formation, as it has no measurable effect on the structure, folding or stability of PrP^C[@CR22]^. The acquired prion disease kuru, which was epidemic amongst the Fore linguistic group of the Papua New Guinea highlands when first studied in the 1950′s, and which was transmitted during mortuary feasts, imposed strong genetic selection on the Fore, essentially eliminating residue 129 homozygotes^[@CR23]^. A novel variant of prion protein, V127, unique to the affected population in the epicentre of the kuru epidemic, was also identified^[@CR24]^. In this variant, the glycine at residue 127, which is fully conserved amongst vertebrate PrP primary structures, is substituted by valine. The V127 polymorphism was found on one copy of the *PRNP* gene in unaffected individuals within the population, suggesting that this polymorphism conferred resistance to prion disease, having been selected for in response to the kuru epidemic^[@CR23],[@CR24]^. The protection afforded by this polymorphism was modelled using transgenic mice expressing human PrP^[@CR25]^, and showed that heterozygous mice expressing both alleles containing glycine and valine at residue 127 (G/V127), echoing the human resistance genotype, exhibited profoundly reduced susceptibility to infection with kuru and classical CJD prions. Most importantly, however, and in complete contrast to the protective effect of the residue 129 polymorphism, homozygous mice expressing human PrP with solely valine at residue 127 (V127), showed total resistance to all inoculated human prion strains. A comparison of the incubation periods between hemizygous mice expressing wild-type G127 human PrP only, with heterozygous mice expressing both G127 and V127 PrP, indicated a dose-dependent dominant-negative inhibitory effect of V127 PrP on prion propagation, resulting in prolonged incubation periods and variable attack rates in heterozygotes^[@CR25]^. These data indicated that V127 PrP is intrinsically resistant to prion propagation and can inhibit propagation involving wild-type (WT) G127 PrP. In essence, this single amino acid substitution, at a residue completely conserved in vertebrate evolution, has as potent a protective effect on the host as a null mutation. Consequently, the structural and mechanistic basis of the protective effect of the V127 mutation is of keen interest as it may provide key insights into the mechanism of prion conformational conversion and recruitment. As a first step in characterising the effect of this protective polymorphism on PrP, we undertook a detailed investigation of the effect of the residue 127 polymorphism on the biophysical properties of the native cellular PrP^C^ conformation using a combination of X-ray crystallography, NMR and equilibrium unfolding. We show that this mutation imposes local changes in backbone conformation which facilitate formation of intermolecular hydrogen bonds between native-state dimers and imposes conformational restrictions on this region of the protein. In addition, it significantly alters millisecond timescale conformational rearrangements in regions of PrP proposed to be important in prion transmission^[@CR26]--[@CR28]^. These effects may modulate the conversion of native PrP^C^ to a disease-associated form or on pathway intermediates relevant to the disease process, and provide a mechanistic explanation for the protective effect of this mutant. Results {#Sec2} ======= Choice of PrP variants studied {#Sec3} ------------------------------ Persons who were exposed to kuru and survived the epidemic were predominantly heterozygotes at PrP residue 129^[@CR23]^. The V127 protective polymorphism in human PrP was always present on an M129 allele^[@CR24]^, consequently our main interest was with the V127/M129 PrP variant. However, we took the opportunity, given the known biological effect of the residue 129 polymorphism to also study the V127 variant with valine at residue 129 (V127/V129), and both forms of wild-type PrP (G127/M129 and G127/V129) with the aim of dissecting the effects of both of these protective polymorphisms. V127 PrP structures closely resemble wild-type G127 PrP {#Sec4} ------------------------------------------------------- To determine whether the overall structure of PrP^C^ was affected by the protective V127 variant we crystallised recombinant human PrP (residues 119--231), with valine at residue 127, (V127/M129 and V127/V129), complexed with the Fab fragment of the anti-PrP antibody ICSM18, as performed previously with G127/M129 PrP (Supplementary Table [1](#MOESM1){ref-type="media"} and Supplementary Fig. [1](#MOESM1){ref-type="media"})^[@CR29]^. The crystal structures of both V127 variants (V127/M129, 2.3 Å resolution, pdb 6SV2 and V127/V129, 2.5 Å resolution, pdb 6SUZ) closely resembled that of WT G127/M129 (pdb 2W9E, Fig. [1a](#Fig1){ref-type="fig"} and Supplementary Fig. [2](#MOESM1){ref-type="media"})^[@CR29]^. The structured C-terminal domain (residues 125--225) comprises three α-helices (α1--α3) and a short, two-stranded, anti-parallel β-sheet (Fig. [1](#Fig1){ref-type="fig"} and Supplementary Fig. [3](#MOESM1){ref-type="media"}). Residue 127 immediately precedes the first β-strand of the β-sheet whereas residue 129 lies within it. The residues surrounding 127 and 129 are well defined in both crystal structures (Figs. [2](#Fig2){ref-type="fig"} and [3](#Fig3){ref-type="fig"}) and show that the side-chains of both residues are predominantly located on the protein surface. Neither the 127 nor 129 polymorphisms substantially perturb the backbone or sidechain positions, or hydrogen bonding, of residues within the β-sheet (Fig. [1b](#Fig1){ref-type="fig"} and Supplementary Fig. [2a--c](#MOESM1){ref-type="media"}). Both circular dichroism (CD) and heteronuclear NMR spectra (Supplementary Figs. [4](#MOESM1){ref-type="media"}--[6](#MOESM1){ref-type="media"}) are consistent with the crystal structures accurately reflecting the solution structure of the proteins. The global stability and unfolding behaviours of the V127/M129 and V127/V129 variants (Supplementary Fig. [7](#MOESM1){ref-type="media"} and Supplementary Table [2](#MOESM1){ref-type="media"}) are also not significantly affected by the substitution of valine for glycine at position 127, reflecting the lack of major structural perturbation.Fig. 1Effect of the V127 polymorphism on the structure of human PrP^C^.**a** V127/M129 (PDB 6SV2 -- green) and wild-type G127/M129 human PrP (PDB 2W9E -- light blue^[@CR29]^) crystal structures, superimposed in cartoon representation. Residues 125--223 are shown. The r.m.s. deviations of backbone heavy atoms are less than 0.44 Å between these structures. The sidechains of V127 (red) and R164 (blue) are shown as sticks. This figure and the other structural figures were prepared using *PyMOL* (PyMOL Molecular Graphics System, Schrödinger, LLC). **b** Side chain packing in the V127/M129 (green) and WT G127/M129 (light blue) β-sheets. The PrP backbone immediately preceding residue 127 in V127/M129 PrP is displaced due to the bulkier valine sidechain at residue 127. The sidechain and backbone positions of residues in the β-sheet are very similar, with the exception of the sidechain of arginine 164 (R164), which due to its close proximity to residue 127 is displaced in the V127 variant. This perturbation (see also Fig. [8](#Fig8){ref-type="fig"}) is observed in solution by a marked chemical shift change in the Nε peak arising from the R164 sidechain group in NMR HSQC spectra (Supplementary Fig. [6](#MOESM1){ref-type="media"}). **c** Four-stranded intermolecular anti-parallel β-sheet formed between neighbouring V127/M129 PrP molecules (in green and lime green). **d** Intermolecular β-sheet contacts in V127/M129 PrP (green) and WT G127/M129 PrP (light blue). The amino acid sidechains of residues found in the intermolecular β-sheet are shown in stick representation, with the residue 127 and 129 polymorphisms in red and yellow respectively. **e**, **f** Intermolecular β-sheet hydrogen bonding in V127/M129 (**e**) and G127/M129 PrP (**f**). Hydrogen bonds stabilising the intermolecular β-sheet are shown as blue dotted lines, between the amide (blue) and carbonyl (red) groups of the denoted amino acids, with the corresponding distances in Å. The β-sheet interface in the V127/M129 PrP crystal is stabilised by an additional pair of hydrogen bonds between the carbonyls of G126 and amides of A133 (**e**). The additional hydrogen bond pair between G126 and A133 is not formed in WT G127/M129 PrP as the hydrogen bond distance is too long (7.7 Å) (**f**).Fig. 2The quality of the electron density maps for PrP in the V127/M129 PrP - ICSM-18 Fab complex at 2.3 Å resolution.Residues from the PrP β-sheet and the V127 polymorphism are shown; 2F~O~ --F~C~ map contoured at 1σ. V127 polymorphism restricts PrP backbone conformation {#Sec5} ----------------------------------------------------- Despite the crystal structures being mostly unperturbed by the V127 polymorphism, a number of localised differences were identifiable. The most significant area of variation is found immediately N-terminal of residue 127 (residues 125--127). This region adopts an essentially identical conformation in both the V127/M129 and V127/V129 structures, which differs significantly from WT G127/M129 PrP (Fig. [1b](#Fig1){ref-type="fig"} and Supplementary Fig. [2a--c](#MOESM1){ref-type="media"}). In particular, the reduction in the conformational plasticity of the backbone due to the valine/glycine substitution at position 127 leads to a very different conformation at this point (V127 Phi angle = −70.5°, c.f. G127 = +106.9°), as the WT backbone conformation is in a disallowed region of conformational space for valine. Consequently, the Cα of G126 in V127/M129 PrP is displaced by 2.9 Å, and the Cα of L125 by 2.2 Å (equivalent Cα atoms of most other surrounding residues are displaced by 0.2--0.3 Å). These Cα atom positions are well defined in both V127 structures (Figs. [2](#Fig2){ref-type="fig"} and [3](#Fig3){ref-type="fig"}). Furthermore, the V127 polymorphism appears to reduce conformational variability at residue 127, and concomitantly increases structural definition of the β-sheet, as implied by a comparison of relative B-factors of this region in V127/M129 PrP when compared with WT G127/M129 PrP. These lower B-factors extend from L125 to A133, beyond the end of the first strand of the β-sheet (Fig. [3](#Fig3){ref-type="fig"}). In V127/M129 PrP, the average Cα B-factors for both the N-terminus (residues 126--131; 30 Å^2^), and β-strand 1 (residues 128--131; 27 Å^2^) are lower than the average B-factor for the core secondary structure elements (31 Å^2^). In contrast, in wild-type G127/M129 PrP, the corresponding values for both the N-terminus (46 Å^2^) and β-strand 1 (42 Å^2^) are higher than the average B-factor (39 Å^2^). As the crystal structures are all isomorphous with the same crystal packing, we suggest that the reduction in B-factors is likely due to conformational restriction introduced by the valine sidechain, and by additional intermolecular hydrogen bonding found in the V127 crystals, described below.Fig. 3Thermal parameter (B-factor) distribution in human PrP.(**a**) V127/M129 PrP (**b**) G127/M129 PrP shown as "putty" representation, as implemented by PyMOL. The V127/M129 PrP Cα atom B-factors range from 22.7 Å^2^ to 96.9 Å^2^ with average values of 38.3 Å^2^ for the whole protein, and 30.4 Å^2^ for the core secondary structure elements (residues 128--131 (β-strand 1), 144--154 (α-helix 1), 160--164 (β-strand 2), 174--186 (α-helix 2) and 202--220 (α-helix 3)). The Cα B-factors are depicted on the structure in dark blue (lowest B-factor) through to red (highest B-factor), with the radius of the ribbon increasing from low to high B-factor. The lowest B-factor is observed in the region of α-helix 2 (α2) and α-helix 3 (α3) where the disulphide bridge links the two α-helices at residues 179 and 214 (dark blue), with the antibody-binding epitope spanning α-helix 1 also displaying lower than average B-factors, consistent with the antibody contacts stabilising this region of PrP relative to the overall structure. The largest B-factors are observed in the loop region linking helices α2 and α3 (red) (α2- α3 loop; residues 191--199), where the electron density clearly shows more disorder than elsewhere in the structure. In contrast, the B-factors for residues in close proximity to the V127 polymorphism are not unusually high, and all of these residues are clearly observed in the electron density (see also Fig. [2](#Fig2){ref-type="fig"}). V127 polymorphism extends PrP intermolecular β-sheet {#Sec6} ---------------------------------------------------- Notably, dimers between crystallographically-related PrP molecules are observed in the crystals (Fig. [1c](#Fig1){ref-type="fig"}). Association is mediated by a short segment of the anti-parallel β-sheet with hydrogen bonds formed between the first β-strand (residues 128--131) of each molecule^[@CR29]^. This results in the formation of a four-stranded intermolecular β-sheet between the existing anti-parallel β-sheets of each PrP molecule, involving close homotypic contacts at L130 (Fig. [1e, f](#Fig1){ref-type="fig"}). Similar intermolecular interactions are also observed in the non-isomorphous crystal structures of sheep^[@CR30]^, rabbit^[@CR31]^ and human PrP^[@CR32]^ in the absence of antibody, and in different crystallographic space groups (Fig. [4](#Fig4){ref-type="fig"}). This suggests that this interaction is not a crystal packing artefact, and may reflect a greater biological significance for prion propagation, especially as residue 129 is protective and crucial to the aetiology and neuropathology of prion disease, and residue 127, which is in close proximity to the dimer interface, can completely prevent prion propagation.Fig. 4The interaction of PrP molecules in various PrP crystal structures.**a** Superposition of human V127 (green), ovine^[@CR30]^ (pink), rabbit^[@CR31]^ (grey) and human D178N^[@CR32]^ (yellow) PrP dimers from their respective crystal structures. Unlike V127, the other structures were obtained from apo-crystals in the absence of antibody. The ICSM18 antibody-binding epitope consists of α-helix 1 which is remote from the PrP dimer interface (see **b** and Supplementary Fig. [1](#MOESM1){ref-type="media"}). **b** Close up view of the β-sheet dimer interface common to the crystal dimers. The relative orientation of the two interacting PrP molecules in each structure differs depending on the intermolecular hydrogen-bonding patterns. The residue 129 polymorphism is accommodated within the dimer interface without significant perturbations of surrounding amino acids (Supplementary Fig. [2c, d](#MOESM1){ref-type="media"}). In contrast, substitution by valine at residue 127 results in the formation of an additional pair of intermolecular hydrogen bonds in both V127 structures, between the backbone carbonyl and amide groups of G126 and A133 respectively (Fig. [1e](#Fig1){ref-type="fig"}, Supplementary Fig. [2d](#MOESM1){ref-type="media"}), due to the alteration in backbone conformation. This orients the G126 carbonyl group towards the dimer interface, and its hydrogen bond acceptor A133. In the WT G127/M129 PrP dimer, the corresponding G126 CO -- A133 N^H^ distance is 7.7 Å, as the carbonyl group of G126 points away from the dimer interface (Fig. [1f](#Fig1){ref-type="fig"}). This additional hydrogen bonding with V127 extends the β-sheet dimer interface to residues 126--133, thereby encompassing V127, whereas G127 is not involved in dimer contacts in WT G127/M129 PrP. The hydrogen bond distances for these additional H-bonds in the V127 structures (2.8--2.9 Å) indicate a strong interaction. Also, the hydrogen bonds involving L130 at the centre of the intermolecular β-sheet interface are shortened (3.0 Å compared with 3.3 Å) (Fig. [1e, f](#Fig1){ref-type="fig"}). Thus, rather than preventing PrP dimerisation through disruption of intermolecular hydrogen bonding^[@CR33],[@CR34]^, the V127 polymorphism appears to increase native-state dimer hydrogen bonding. Increased conformational variability in V127 PrP structures {#Sec7} ----------------------------------------------------------- Intriguingly, altered conformational variability is observed in key regions distant from the site of the V127 polymorphism, in particular the loop linking the second strand of the β-sheet and helix 2 (β2-α2 loop; residues 165--172; Fig. [3](#Fig3){ref-type="fig"}). This region, which has been shown to affect prion cross-species transmissibility^[@CR26]--[@CR28],[@CR35],[@CR36]^, is adjacent to the β-sheet, packing against residues N-terminal to the first β-strand (including residue 127), the β-sheet itself, the C-terminus of helix 3, and is in close proximity to the disease-associated residue D178^[@CR37]^. The Cα B-factors for residues 169--172 within this loop are higher than the average for the rest of the protein in both V127 structures (50 vs. 38 Å^2^ for V127/M129 & 56 vs. 45 Å^2^ for V127/V129). This contrasts with WT G127/M129, where these residues are better defined than the average, according to their B-factors (40 vs. 42 Å^2^). The B-factors for the V127 structures are consistent with an alteration in the degree of conformational exchange in this loop region of the protein compared to WT PrP, possibly compensating for the reduced conformational variability observed in the β-sheet region. The remaining regions of the V127 PrPs display B-factors that are comparable to WT values. Altered conformational variability also seen in solution {#Sec8} -------------------------------------------------------- Crucially, the altered conformational variabilities are also observed in solution. The effects of the V127 and V129 polymorphisms on the dynamics of PrP were investigated using NMR relaxation data (Supplementary Fig. [8](#MOESM1){ref-type="media"}), coupled with *Modelfree* (Fig. [5](#Fig5){ref-type="fig"}) and reduced spectral density analyses (Fig. [6](#Fig6){ref-type="fig"}, Supplementary Fig. [9](#MOESM1){ref-type="media"})^[@CR38],[@CR39]^. The former uses order parameters (*S*^2^) to report internal sub-nanosecond (ns) motions. *S*^2^ values range from 0 for highly flexible to 1 for rigid systems. The β-sheet and helical regions of all PrP variants exhibit *S*^2^ values of 0.8--0.9, typical of structured regions of folded proteins (Fig. [5](#Fig5){ref-type="fig"}). However, a number of residues in structured regions, for example E168 in V127/M129 and D178 in G127/M129 PrP display anomalously low *S*^2^ values. These are subject to millisecond (ms) conformational dynamics described below (Fig. [6](#Fig6){ref-type="fig"}).Fig. 5Degree of order in PrP variants.Order parameters (*S*^2^) for G127 (G127/M129), V127 (V127/M129) and V129 (G127/V129) PrP. Residues of the β1-α1 (residues 134--144) and α2--α3 loops (residues 194--199) display slightly reduced *S*^2^ values (0.6--0.8), reflecting increased flexibility, commonly observed in loop regions of globular proteins^[@CR22],[@CR71],[@CR73]^, and in previous studies of PrP^C^. The N-terminus (residues 119--124) preceding the β-sheet is mobile and disordered, with low *S*^2^ values (0.15--0.5) and a lack of electron density in the crystal structures. These order parameters are mapped onto the structures of the PrP variants in Supplementary Fig. [10](#MOESM1){ref-type="media"}.Fig. 6Conformational dynamics in the PrP variants.**a** Reduced spectral density function J(0), describing the amplitude of zero frequency motions in the PrP variants at 800 MHz. Uncharacteristically large J(0) values, such as those exhibited by β-sheet residues (128--131 and 160--164), and G131 and R164 in V127/M129 PrP in particular, indicate ms -- µs dynamics. The dotted lines in the J(0) graphs are two standard deviations greater than the mean J(0) for the N-terminus (residues 200--210) of helix 3, of the respective variants. **b** Effect of V127 and V129 polymorphisms on the amplitude of zero frequency J(0) motions at 800 MHz. J(0) changes relative to G127/M129 PrP. Residues which experience significant changes in J(0) due to the V127 substitution include G131, R164 and E168. V129 results in altered J(0) motions for residues 129 and 131, within the first β-strand, M166, and residue 178, located in helix 2. The observed changes are due to differential ms conformational dynamics (see **c**). **c** PrP ms dynamics (*R*~ex~) modelled in the Relax Modelfree analysis. The V127 polymorphism increases ms dynamics (*R*~ex~) within the β-sheet (G131/R164) and β2-α2 loop (E168/Q172), and diminishes those at the C-terminus of helix 3. The V129 polymorphism also increases ms dynamics in the first β-strand (V129/G131), and the C-terminus of helix 3. Residues 166 and 172 at either end of the β2-α2 loop are also perturbed. These *R*~ex~ values are mapped onto the structure of PrP^C^ (See Fig. [7](#Fig7){ref-type="fig"}. and also Supplementary Figs. [10](#MOESM1){ref-type="media"} and [11](#MOESM1){ref-type="media"}). The *Modelfree* approach also allows a general separation for each residue of ms conformational dynamics (*R*~ex~ values) from ns and sub-ns motions. These ms timescale motions are often associated with large-scale co-operative conformational changes and highlight residues that populate low-free energy alternative conformations. For each of the PrP variants, a number of residues exhibited *R*~ex~ values (Fig. [6c](#Fig6){ref-type="fig"}). These are concentrated in a spatially close region, involving the β-sheet (V129/G131/R164), the β2-α2 loop (M166/E168/Q172) and the C-terminus of helix 3 (I215 T216/Y218/E219/E221; Supplementary Figs. [3](#MOESM1){ref-type="media"}, [10](#MOESM1){ref-type="media"} and [11](#MOESM1){ref-type="media"}). The line-broadening of resonances D167, Y169, S170 and N171 beyond detection in the HSQC spectra of all three variants also likely reflect ms dynamics. The observed conformational dynamics are consistent with a proposed interconversion of the β2-α2 loop between a more populated 3~10~-helix and a type I β-turn^[@CR27],[@CR30],[@CR32],[@CR40],[@CR41]^ (Supplementary Fig. [12](#MOESM1){ref-type="media"}). The V127 polymorphism results in large increases in ms dynamics for residues G131 and R164 in the β-sheet, and E168 and Q172 in the β2-α2 loop, but decreases in the C-terminus of helix 3 (215--221; Figs. [6](#Fig6){ref-type="fig"} and [7](#Fig7){ref-type="fig"}). In a number of WT PrP crystal structures the sidechain of R164 forms a pair of hydrogen bonds with the carboxyl group of E168 (2.5 and 3.1 Å in PDB 2W9E^[@CR29]^). The introduction of the bulkier valine sidechain at residue 127 appears to sufficiently perturb the side-chain position of R164 such that the interactions with E168 are essentially removed (the equivalent distances are 3.2 and 3.8 Å; Fig. [8](#Fig8){ref-type="fig"}). Significantly, NMR chemical shift changes in the N^ε^ signal from the R164 sidechain in V127/M129 PrP reflect this alteration in side-chain orientation, with the N^ε^ being perturbed by the change in its proximity to the aromatic ring of Y128 and the change in hydrogen bonding of R164 N^H1^. (Supplementary Fig. [6](#MOESM1){ref-type="media"}). The loss of these interactions is a likely source of the increase in the ms dynamics of both residues, which appears to be disseminated along the rest of the β2-α2 loop, as residue Q172, the other visible resonance in the β2-α2 loop, also displays a marked increase.Fig. 7Effect of V127 polymorphism on the amplitude of PrP ms dynamics (*R*~ex~).The sidechains of residues which experience altered ms dynamics in V127/M129 PrP, relative to G127/M129 PrP are shown (Fig. [6c](#Fig6){ref-type="fig"}), with varying width of backbone and colour. Residues showing increased *R*~ex~ values in the V127 variant, such as G131, R164, E168 and Q172, are coloured red, with those showing a reduction, such as Y218, are coloured blue. Those residues for which a comparison is not possible, due to absence of data are not coloured. The orientation of R164 and E168 sidechains in G127/M129 PrP are shown in yellow^[@CR29]^, illustrating the loss of hydrogen bonding in V127/M129 PrP, caused by a steric clash between V127 and R164 sidechains in the V127 variant. Also shown is the hydrogen bond between Y169 and D178, showing the close association between the β2-α2 loop and another residue which has a key effect on the aetiology of human prion disease.Fig. 8Perturbation of the R164 sidechain by V127 in V127 PrP crystals.Comparison of residue 127, R164 and E168 side-chain positions in WT G127/M129 (cyan), V127/M129 (green) and V127/V129 PrP (yellow). Glycine 127 is coloured bright red, with the valine sidechains of residue 127 in V127/M129 and V127/V129 PrP coloured dark red. R164 and E168 in wild-type G127/M129 PrP are coloured dark blue, and lighter blue in both V127 variants. In both V127 variants the sidechain of R164 is sufficiently displaced from its position in the wild-type protein to significantly weaken the specific, strong (2.5 Å) hydrogen bonding interaction with E168 observed in wild-type G127/M129 PrP. Similarly, the V129 polymorphism also affects ms dynamics, however, different residues are affected. V129 increases *R*~ex~ values for G131 and itself (Fig. [6c](#Fig6){ref-type="fig"}, Supplementary Figs. [10](#MOESM1){ref-type="media"} and [11](#MOESM1){ref-type="media"}). This alteration in G131 exchange dynamics has recently been observed in mouse PrP^[@CR42]^. In addition, we also observe that the *R*~ex~ values of D178 are markedly reduced in the V129 polymorph. This is illustrated by the reduction of line-broadening of D178 observed in the HSQC spectra of G127/V129 PrP (Supplementary Fig. [13](#MOESM1){ref-type="media"}). This is notable as the residue 129 M/V polymorphism affects the disease phenotype of the pathogenic D178N mutation which causes inherited prion disease. D178N is associated with the clinico-pathological phenotype Fatal Familial Insomnia (FFI) when residue 129 is methionine, and CJD when it is valine^[@CR37]^. In V127/M129 PrP the D178 HSQC resonance cannot be observed directly as it is heavily overlapped with that of V127, but an analysis of the intensity of signals in V127/M129 3D HNCO NMR spectra indicates that D178 does indeed experience ms dynamics, to a similar extent as wild-type G127/M129 PrP. This suggests that the V129 polymorphism alters PrP conformational variability independently of the 127 mutation. A number of residues at the C-terminus of helix 3 (I215, Y218, E219, R220, E221 and S222) experience altered ms dynamics in V127/M129 compared with G127/V129 PrP. Residues I215/Y218/E221/S222 are on a face of the helix that interacts with residues in the β2-α2 loop (Fig. [7](#Fig7){ref-type="fig"}, Supplementary Fig. [11](#MOESM1){ref-type="media"}). For example, residues Y218 and S222 closely interact with M166, while I215 and Y218 interact with Q172. In particular, residues Y218 and E221 are subject to marked increases in *R*~ex~ values in G127/V129 PrP, whereas in V127/M129 PrP there is a reduction at the C-terminus. V127 polymorphism does not perturb PrP stability {#Sec9} ------------------------------------------------ To test whether the variations in dynamics have a substantial effect on local stability, hydrogen/deuterium exchange rates were obtained on V127/M129 PrP. The observed rates of hydrogen/deuterium exchange allow the determination of amide protection factors, which indicate the extent to which hydrogen bonding and burial prevents solvent access. The hydrogen/deuterium exchange data indicate that the protection factors and stabilities of the PrP secondary structure elements are indistinguishable between V127/M129 and G127/M129 PrP (Supplementary Fig. [14](#MOESM1){ref-type="media"})^[@CR22],[@CR43]^. The protection factors of the secondary structure elements, with the notable exception of the first strand of the β-sheet and in the vicinity of the disulphide bond, reflect the equilibrium constant between native and unfolded states of the protein (*K*~F/U~)^[@CR43]^. The majority of residues that display observable protection factors therefore exchange from the globally unfolded state. This is also the case with G127/V129 PrP^[@CR22]^. The V127 and V129 polymorphisms thus do not induce any alternatively folded states in which the core of the protein is destabilised. This is noteworthy as the first strand of the β-sheet (where the residue 127 and 129 polymorphisms lie) displays anomalously low protection factors corresponding to reduced stability (\~30 times less than the other secondary structure elements), suggesting that its stability might be affected more readily by the protective polymorphisms. Notably, a number of regions that are subject to the ms conformational dynamics affected by both protective polymorphisms are in areas that do not display measureable hydrogen protection, for example the β2-α2 loop and the C-terminus of helix 3. Effect of V127 polymorphism on PrP in vitro fibrilisation {#Sec10} --------------------------------------------------------- The lack of major structural perturbation or altered stability of the V127 variant in comparison to WT PrP^C^ suggests that the polymorphism may act by primarily affecting the efficiency of conversion of PrP^C^ to its disease-associated aggregated form. To assess this we firstly examined the ability of the V127 variant to fibrilise under partially denaturing conditions. When agitated in 2 M GuHCl, PrP can be induced to form amyloid. Binding of the fluorescent thiazole dye thioflavin T to these β-sheet-rich fibrillar structures reports their formation, allowing a quantitative analysis of the kinetics of fibril formation^[@CR44]^. We found that although V127/M129 PrP can be induced to fibrilise within the time scale of the experiment, it did so with a significantly longer lag-time than WT G127/M129 PrP (Fig. [9](#Fig9){ref-type="fig"}). This is particularly interesting as substitutions to valine, and other bulky hydrophobic residues typically promote β-sheet formation and self-association required for amyloid formation^[@CR45]^. These data are however consistent with previously published data which modelled the effect of the V127 mutation on a mouse PrP background, and which indicated that the V127 variant is inherently more resistant to fibrilisation than WT PrP^[@CR46]^.Fig. 9Quantitative analysis of the effect of V127 on PrP fibril formation.**a** Formation of amyloid fibrils as reported by increasing Thioflavin T (ThT) fluorescence. The lines superimposed on the data are non-linear curve fits to Eq. ([2](#Equ2){ref-type=""}), as described in the Methods section. **b**, **c** Fibrillogenesis of V127/M129 PrP occurs with significantly longer mean half- and lag times in comparison to WT G127/M129 PrP (*P* ≤ 0.01, paired *t*-test). Centre lines show the medians for each data set; box limits indicate the 25th and 75th percentiles as determined by R software (<http://shiny.chemgrid.org/boxplotr/>); whiskers extend 1.5 times the interquartile range from the 25th and 75th percentiles, outliers are represented by dots. *N* = 5 sample points for each data set. Amplification of protease-resistant PrPSc seed (PMCA) {#Sec11} ----------------------------------------------------- Although fibrillar material can be generated using these partially denaturing conditions, the material generated in such reactions has not been shown to be reliably infectious. In contrast, the protein misfolding cyclic amplification (PMCA) technique^[@CR47]^ has been shown to amplify infectious and PK-resistant material with high fidelity. PMCA is a cyclical process where periods of conversion of substrate PrP^C^ by small amounts of PrP^Sc^ "seed" are interspersed with bursts of sonication. We performed PMCA reactions using brain homogenate from mice overexpressing WT G127/M129 PrP (Tg35) WT G127/V129 PrP (Tg152) or V127/M129 PrP (Tg183). Both WT PrP substrates allowed amplification of PK-resistant material. In contrast there was no amplification using V127/M129 as substrate (Fig. [10](#Fig10){ref-type="fig"}). These results are consistent with the observed disease characteristics of the vCJD strain type which propagates most readily with WT PrP with methionine at residue 129, and which failed to generate protease-resistant PrP or cause disease in transgenic mice expressing solely V127/M129 PrP^[@CR25]^.Fig. 10Western blot indicating the presence of PK-resistant material in PMCA reaction (+) and non-sonicated control (−) samples of PrP^Sc^ amplified with Tg35 (huPrP G127/M129), Tg152 (huPrP G127/V129) and Tg183 (huPrP V127/M129) brain homogenate as substrate.In each case, a small amount of seed can be detected in the non-sonicated control samples with varying levels of amplification observed in the reactions with different substrates. Of note is the lack of amplification with V127/M129 PrP (Tg183) as substrate. Discussion {#Sec12} ========== This structural and biophysical study was stimulated by the remarkable effect of the V127 polymorphism on human prion propagation. Transgenic mouse transmissions show that V127 PrP is incapable of supporting prion transmission and propagation, consistent with the human clinical resistance data^[@CR24]^, and is even able to inhibit heterologous propagation of wild-type protein containing glycine at residue 127^[@CR25]^. This differs from the residue 129 polymorphism, where similar studies suggest the importance of homologous protein interactions in prion propagation^[@CR17],[@CR18],[@CR20],[@CR48]^, and the preferential selection of different prion strains by PrP molecules with different primary structure as a result of conformational selection^[@CR2],[@CR16]^. Unlike the residue 129 polymorphism, no strain switching or strain mutation was observed with V127, even in the homozygous state, indicating that it confers complete resistance via the variant protein itself^[@CR25]^, leading to the hypothesis that an altered PrP^C^ fold may be the cause of resistance to prion disease^[@CR34]^. The profound effects that these mutations have on human prion disease pathogenesis may provide key insights into the mechanism of prion conversion, and have a wider relevance to other templated protein mis-folding diseases, where changing a single amino acid could have a similar dramatic effect, with potential significance for therapeutic strategies^[@CR9],[@CR10]^. The structural consequences of the glycine to valine substitution at residue 127 on PrP^C^ is therefore of major and wide potential interest. Here, we have shown that there is a close similarity in overall structure between both V127 variants studied (V127/M129 and V127/V129), and wild-type G127/M129 PrP. Solution spectra (CD and NMR) confirm that the crystal structures faithfully reflect the solution structures of the proteins, allowing detailed analysis of the structural effect of the V127 polymorphism on PrP^C^. We find little evidence for a major structural change in the β-sheet, or α-helices, in contrast to a recent NMR structure which identified unique features caused by the V127 polymorphism^[@CR34]^. In particular we do not observe any displacement of amino acid side chains within the β-sheet, which are well defined by the electron density, and note that this crystal structure satisfies the inter-residue β-sheet NOE distance constraints used for the NMR structural study within 0.25 Å^[@CR34],[@CR49]^, apart from two which are satisfied within 0.43 Å and 1.03 Å (both to the Hε of Y162). However, distinct perturbations of key regions which affect prion transmission and propagation are observed. Specifically, the V127 substitution reduces the conformational variability of the protein backbone immediately preceding the first strand of the β-sheet and radically alters the local backbone conformation. This facilitates the formation of an additional two intermolecular hydrogen bonds, which stabilise the native-state dimer association observed in the crystals. This dimeric association has been observed in a number of different PrP structures crystallised in the absence of antibody^[@CR30]--[@CR32]^ (Fig. [4](#Fig4){ref-type="fig"}). In these, the PrP β-sheet interface is composed of two intermolecular hydrogen bonds, as in WT G127/M129 PrP^[@CR29]^. The V127 dimer interface presented here is unique in both the length of the β-sheet interface and number of hydrogen bonds (4), and argues against the proposal that V127 disfavours native-state dimerisation, by reducing main-chain hydrogen-bond interactions^[@CR33],[@CR34]^. Formation of the intermolecular β-sheet has been proposed as a possible initiation point for β-sheet-mediated oligomerisation to explain the genetic susceptibility and prion strain selection determined by the polymorphic residue 129 in human prion disease^[@CR23],[@CR29],[@CR30],[@CR32]^. If β-strand interactions in this region of the protein mediate PrP interactions during PrP^Sc^ formation, then the packing and geometry of this segment of the chain would have a strong selective effect on conformation, and also productive prion propagation^[@CR29]^. No displacement of the protein backbone or stabilisation of the PrP^C^ dimer interface is caused by the residue 129 polymorphism^[@CR22],[@CR32]^, which may explain the marked effect of the V127 polymorphism on prion pathogenesis. The PrP^C^ dimer interaction does not appear to be a crystal packing artefact as endogenous PrP^C^ dimers have been detected in N2a cells and purified brain fractions^[@CR50],[@CR51]^ with the dimerisation region mapped to the hydrophobic domain of PrP (residues 112--133)^[@CR52]^. PrP^C^ dimerisation inhibits PrP^Sc^ accumulation and prion replication^[@CR53],[@CR54]^, and has a dominant-negative inhibitory effect on the conversion of monomeric PrP^C[@CR54]^. These findings suggest that it may be possible to halt prion formation by stabilising PrP^C^ dimers. Given the strengthened dimer association seen in the V127 crystals this may be one aspect of the protective mechanism of the V127 mutant. The conformational restriction imposed by the V127 polymorphism may also be sufficient to inhibit homotypic protein--protein contacts in heterodimers of V127 and G127 PrP, or prevent the formation of extended β-sheet structure required to convert the PrP N-terminal unstructured region into protease-resistant β-enriched forms^[@CR41],[@CR42],[@CR46],[@CR55]^. Alternatively, the marked alteration in local backbone conformation and increased stability of intermolecular β-sheet interactions may prevent PrP folding into a thermodynamically permissible prion assembly^[@CR3],[@CR4],[@CR56],[@CR57]^. As V127 PrP also inhibits the generation of infectious assemblies of wild-type PrP, this would suggest that it must be either capping nascent prion assemblies or structurally weakening infectious prion assemblies on incorporation. The PMCA data presented here indicates that V127 PrP is not a permissive substrate for amplification of protease-resistant PrP^Sc^ disease seed. It is possible that the polymorphism introduces a protease cleavage site, which would lead to the destruction of a polymer formed of the homomeric V127 protein. PrP V127 incorporation would also dope, with a dose-dependent effect, heteromeric polymers composed of both G127 and V127 PrP variants. Their reduced stability could increase cellular clearance, which would also explain the "dominant negative" effect of V127 on prion propagation^[@CR58]^. In addition to these local structural perturbations, long-range sequence interactions between the protective residue 127 and 129 polymorphisms affect the conformational distribution of a spatially distinct region including the β-sheet, β2-α2 loop, the C-terminus of helix 3 and the disease-associated residue D178^[@CR37]^. These conformationally variable elements have been shown to be key determinants of prion transmission and cross-species prion susceptibility, and a number are associated with inherited forms of prion disease, for example G131V, D167N, V210I, E211Q, Q212P and Q217R^[@CR26],[@CR28],[@CR37],[@CR59]--[@CR64]^. In particular, the β2-α2 loop (residues 165--172) has been proposed to be a key modulator of prion transmission and disease-associated PrP misfolding^[@CR26]--[@CR28],[@CR35],[@CR36],[@CR58]^. V127 alters the structural flexibility of the β-sheet and conformational dynamics of the β2-α2 loop by disrupting the electrostatic interaction between R164 of the β-sheet and E168 within the adjacent β2-α2 loop. Loss of this interaction would disrupt hydrogen bonding and close packing between Y169, F175 and D178 and destabilise the dominant 3~10~--helical conformation of the β2-α2 loop^[@CR27],[@CR36],[@CR40]^ (Supplementary Figs. [11](#MOESM1){ref-type="media"} and [12](#MOESM1){ref-type="media"}). Notably, residue 168 (human numbering) is polymorphic in sheep PrP, in which either glutamine or arginine can be accommodated^[@CR65]^. The arginine polymorph, which would also diminish the electrostatic interaction with R164, is associated with resistance to scrapie, and potently inhibits prion conversion^[@CR60],[@CR62]^. These observations support the notion that increased conformational variability associated with the loss of charge interaction between R164 and E168 increases resistance to prion disease. Substitution of arginine for glutamine at the totally conserved Q172, the dynamics of which are significantly altered in the V127 polymorph also potently inhibits in vitro prion infectivity^[@CR66]^. The recent NMR structural study of V127 PrP also identified significant alterations in the conformational dynamics in the β-sheet and α-helix 2 as a result of the polymorphism^[@CR34]^. Although there are a number of residues perturbed in both studies, the marked effect that we observe on the dynamics of the β2-α2 loop was not observed in the previous study. The differences in protein dynamics may be ascribed to the different solution conditions for data acquisition. In this study PrP was crystallised at pH 8.0, with the NMR solution dynamics acquired at pH 5.5, whereas the previous NMR study used pH 4.5, which may cause glutamate and aspartate side chains to adopt an unphysiological protonated state^[@CR34]^. The calculated p*K*~a~′s of E168 and D167 in V127/M129 PrP are 4.7 and 4.4, respectively^[@CR67]^. The hydrogen-bonding and charge interactions of the β2-α2 loop involving the side-chains of these residues will be weakened through increased protonation. This will likely cause altered conformational dynamics in the β2-α2 loop and adjacent regions. For example, the lack of assigned resonances for R164 by Zheng et al.^[@CR34]^ may be ascribed to intermediate timescale exchange broadening these signals beyond detection. Given the p*K*~a~′s of interacting residues, the pH at which structural studies are carried out might thus be very significant. It is of great interest that both the V127 and V129 polymorphisms have long-range effects on the conformational distribution of these regions of the protein and also the β-sheet. The correlation between these conformationally variable regions of PrP and its propensity to form disease-related isoforms suggest that these regions of the protein are important in prion assembly-PrP^C^ interactions, determining efficient binding and conversion^[@CR27],[@CR36]^. Indeed, relatively subtle variations in the strength and orientation of monomer docking can dramatically affect the productivity of fibrillogenic interactions and determine barriers to amyloid formation^[@CR68]^. Given the effect of the V127 polymorphism on the PrP β-sheet backbone geometry and intermolecular association of PrP monomers observed here, it is tempting to speculate that dimerisation via the formation of the intermolecular PrP β-sheet may be a critical event in PrP oligomerisation and prion propagation and thus explain the exceptional effect of the residue 127 polymorphism on human prion disease. Methods {#Sec13} ======= Recombinant PrP and antibodies {#Sec14} ------------------------------ Recombinant human PrP containing residues 119--231 (PrP^119--231^) was produced and purified as previously described^[@CR69]^. This length of construct was chosen as the PrP N-terminus up to approximately residue 125 is unstructured in full-length (residues 23--231) and truncated (residues 91--231) PrP and compromises the NMR dynamics characterisation due to its effect on the rotational tumbling of the structured domain^[@CR70],[@CR71]^. Removal of the N-terminal tail does not affect the structure or local structural fluctuations of the PrP structured globular domain^[@CR72],[@CR73]^. PrP containing valine at residue 127 (V127/M129 and V127/V129), and wild-type PrP with glycine at residue 127 on both 129 methionine and valine backgrounds (G127/M129 and G127/V129) were expressed and purified for biophysical analysis. ICSM 18 was purchased from D-Gen Limited. The Fab fragment of ICSM 18 was prepared by limited papain digest of the mature antibody followed by purification using gel filtration chromatography. Crystallisation conditions {#Sec15} -------------------------- ICSM 18-Fab and PrP were mixed at a molar ratio of 3:1 for preparation of the complex prior to crystallisation, and incubated at room temperature for 30 min before buffer-exchanging the complex into 50 mM Tris, 150 mM NaCl, pH 8.0 and filtering through a 0.22 µm membrane prior to crystallisation. Crystals of the complex were obtained by using the sitting-drop vapour diffusion technique; droplets containing 5--6 mg/mL PrP in 0.4 M and 0.75 M ammonium sulphate, 0.05 M Tris (pH 7.5 and 8.0) were equilibrated over wells containing 0.8 M and 1.5 M ammonium sulphate, 0.1 M Tris (pH 7.5 and 8.0). Crystals, round in shape, grew in 6 months to 0.05--0.2 mm diameter. In situ data collection and analysis {#Sec16} ------------------------------------ Data were collected at room temperature in situ on beamline I03 at Diamond Light Source, with the crystallisation plates sealed in biohazard bags. Multiple wedges of data were collected from different parts of the same crystal, and from different crystals, and scaled together to provide a complete dataset. We typically collected 15° of data from each crystal in 0.3° oscillations. Data were integrated with *XDS*^[@CR74]^ and then subsequently *BLEND*^[@CR75]^ was used to analyse how well the different wedges of data scaled together and the results used to decide which datasets should be scaled and merged with *AIMLESS*^[@CR76]^. Structure determination and refinement {#Sec17} -------------------------------------- The structures were solved by molecular replacement using PHASER^[@CR77]^ with the heavy and light chains of the Fab fragment of antibody ICSM18 and the PrP molecule used as search models (protein databank accession code 2W9E). Electron density maps were inspected and the models built using *COOT*^[@CR78]^ followed by refinement with *REFMAC5*^[@CR79]^. Data collection and final refinement statistics are summarised in Supplementary Table [1](#MOESM1){ref-type="media"}. Ramachandran statistics for V127/M129 and V127/V129 structures (in parentheses) are as follows; residues in most favoured region: 96.7% (96.7%), residues in additionally allowed regions: 3.3% (3.3%) and residues in disallowed regions: 0.0% (0.0%)^[@CR80]^. The final coordinates of the V127/M129 and V127/V129 structures have been deposited in the Brookhaven Protein Data Bank (<http://www.rcsb.org>), with accession numbers 6SV2 and 6SUZ respectively. NMR sample preparation and spectroscopy {#Sec18} --------------------------------------- For the NMR study ^15^N & ^13^C/^15^N-labelled samples of PrP were prepared. Following purification, protein samples were either (A) buffer-exchanged into 20 mM sodium acetate, containing 1.5 mM sodium azide (NaN~3~), pH 5.5 through dialysis, then concentrated in Vivaspin 20 centrifugal concentrators to protein concentrations of 0.8--1.2 mM or (B) dialysed against deionised water then lyophilised and resuspended into 20 mM sodium acetate, 1.5 mM NaN~3~, pH 5.5. 10% D~2~O (v/v) was added to the NMR samples to provide the lock signal, together with TSP as the chemical shift reference to 1 mM final concentration. NMR samples were placed in Sigma FEP NMR sample tube liners (Z286397--1EA), held within Wilmad PP-528 NMR tubes for NMR data acquisition. Assignment spectra for V127/M129 PrP were acquired at 303 K on Bruker DRX-600 and DRX-800 spectrometers, with ^15^N-relaxation measurements for V127 and V129 PrP acquired at 298 K on Bruker Avance III 500 and 800 MHz spectrometers, all equipped with 5 mm ^13^C/^15^N/^1^H triple-resonance probes. Sensitivity-enhanced ^1^H-^15^N HSQC^[@CR81],[@CR82]^ and standard triple-resonance experiments^[@CR83]^ with uniformly ^13^C/^15^N-labelled protein (HNCA, HNCACB, CBCA(CO)NH and HNCO) were used to obtain V127/M129 backbone resonance assignments. Proton chemical shifts were referenced to TSP. ^13^C & ^15^N chemical shifts were calculated relative to TSP, using the gyromagnetic ratios of ^13^C, ^15^N and ^1^H (^15^N/^1^H = 0.101329118; ^13^C/^1^H = 0.251449530). For residue 127, when comparing ^13^C chemical shifts, the difference in residue type was compensated for by subtracting residue-specific random coil shifts (glycine/valine) to generate secondary chemical shifts, which were then subtracted^[@CR84]^. NMR data were processed and analysed using Felix 2007 (Accelrys, San Diego), *Topspin (*v 3.2, Bruker*)* and CCPN Analysis (*v. 2.3.1*)^[@CR85]^ software. Spin relaxation measurements {#Sec19} ---------------------------- Spin relaxation measurements (T~1~, T~2~ and ^15^N(^1^H)-NOE) were acquired on 1 mM ^15^N-labelled PrP^119--231^ WT (G127/M129) PrP, V127 (V127/M129) PrP and V129 (G127/V129) PrP as described in Yip et al.^[@CR86]^. Briefly, by using this methodology, heating compensation was improved by the incorporation of a compensation block based on the relaxation block, followed by a pre-scan ^1^H saturation sequence and constant length recovery period. The *T*~1~ data were obtained using ^15^N relaxation delays of 50, 100, 200\*, 300, 500, 800\*, 1000, 1500, 2000, 3000, 4000 and 5000 ms. The *T*~2~ data were obtained using ^15^N relaxation delays of 8.5, 17.0, 33.9\*, 50.9, 67.8, 101.8\*, 135.7, 186.6, 254.4 ms (500 MHz) and 7.8, 15.7, 31.4\*, 47.0, 62.7\*, 94.1, 125.4, 172.5 and 235.2 ms (800 MHz; the asterisks denote duplicate measurements). *T*~1~ and *T*~2~ datasets were recorded as pseudo 3D experiments, with randomised order of time increments. Two separate nitrogen offsets were used to reduce build-up of off-resonance artefacts during the CPMG block of the *T*~2~ measurements. For the ^15^N-^1^H-NOE measurement, two two-dimensional spectra were acquired with a relaxation delay of 6 s between scans. Spectra were collected with *t*~1~ acquisition times of 94.7 ms (500 MHz)/59.2 ms (800 MHz) and *t*~2~ (direct) acquisition times of 127.8 ms (500 MHz)/91.8 ms (800 MHz). Errors for time-series *T*~1~ and *T*~2~ data were calculated from the overall standard deviation for duplicate data points in the series. Errors for the NOE data were estimated from measurements of the root mean-square deviation of the base-plane noise in those spectra. Non-linear least-squares (Levenberg-Marquardt) fitting of two-parameter exponential functions to decay data was performed using in-house routines using Numerical Python. *Modelfree* analysis {#Sec20} -------------------- Protein dynamics were analysed by *Relax* (*v. 3.3.1*)^[@CR38],[@CR39]^, using the *T*~1~, *T*~2~ and ^15^N{^1^H}-NOE spin relaxation data. Reduced spectral density mapping analysis, as implemented by the *Relax* default *J*~(w)~ mapping script mode was used to obtain *J*~(ω)~ values for each given field strength. A full *Modelfree* analysis^[@CR87]^ was carried out using the "d'Auvergne" protocol within *Relax*. Extended order parameters (*S*~2~, *S*^f^~2~, *S*^s^~2~), the effective correlation time for fast internal motions (*t*~e~) and intermediate exchange broadening contribution (*R*~ex~) values were obtained using this protocol. Amide exchange protection experiments {#Sec21} ------------------------------------- Hydrogen-deuterium exchange rates (*k*~ex~) were determined by adding 260 µl 20 mM sodium acetate, 1 mM sodium azide, pH 4.5, dissolved in 100% (v/v) D~2~O to lyophilised PrP samples, to obtain final protein concentrations of 1 mM. A series of sensitivity-enhanced ^1^H-^15^N HSQC spectra^[@CR81],[@CR82]^ were acquired at 293 K on a Bruker DRX-800 spectrometer. The decay curves of the ^1^H-^15^N HSQC cross-peaks were fitted to single exponential decays with offset, and protection factors (*k*~ex~/*k*~int~) for observable amides were determined using intrinsic amide exchange rates^[@CR88]^ (*k*~int~). Acquisition of the first experiment began \~5 min after mixing, setting a lower limit on the detection of protection factors of \~5. Circular dichroism {#Sec22} ------------------ Circular dichroism was measured at 25 °C with a Jasco J-715 spectropolarimeter, using a 0.1 cm pathlength quartz cuvette. The sample temperature was controlled with a circulating water bath. Far-UV (amide) CD spectra were recorded between 180 nm and 300 nm with 20 μM protein (2 nm bandwidth; Data Pitch 0.5 nm). In all, 10--50 spectra were averaged. Equilibrium unfolding measurements {#Sec23} ---------------------------------- For equilibrium unfolding experiments, 6 μM protein was incubated in 10 mM HEPES, 25 mM NaCl pH 7.5, and increasing concentrations of GuHCl denaturant. Molecular ellipticity (\[θ\], degree M^−1^ cm^−1^) was recorded at 222 nm (5 nm bandwidth; 20 s integration time). The denaturation profile for each protein was measured in three separate experiments. Conversion to molar denaturant activity {#Sec24} --------------------------------------- To allow more accurate extrapolation of data to calculate folding parameters in the absence of denaturant and the free energy change of protein folding (Δ*G*), denaturant concentration (\[GuHCl\]) was converted to molar denaturant activity (*D*), as described in Parker et al.^[@CR89]^, using *C*~0.5~ = 7.5. Equilibrium constant between folded and unfolded states {#Sec25} ------------------------------------------------------- For the two-state equilibrium unfolding transitions, data were fitted to the following equation, where *K* and *K*~(W)~ are equilibrium constants between the folded and unfolded states at a given denaturant activity (*D*) and in water, respectively, and *m* describes the sensitivity of the equilibrium to denaturant activity^[@CR89]^.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{K}} = {{K}}_{{\mathrm{(W)}}}{\mathrm{exp}}\,\left( {m.{\mathrm{D}}} \right)$$\end{document}$$ For visual representation of the data shown, data were converted to proportion folded, *α*~F~, using the following, *α*~F~ = (*K*/(1 + *K*)). Data fitting was carried out using *GraFit* (Erithacus software). The significance of the differences in free energy for folding and *m* values between the three variants characterised were determined by paired two-tailed Student's *t* test. Quantitative analysis of the kinetics of PrP fibril formation {#Sec26} ------------------------------------------------------------- Recombinant V127/M129 and WT G127/M129 PrP (residues 119--231) was dialysed into 20 mM sodium acetate, 2 mM sodium azide, pH 6.0, and then denatured by the addition of GuHCl to a final concentration of 6 M. Denatured PrP was then diluted to a final concentration of 10 µM in 20 mM sodium acetate, 2 M GuHCl, 10 mM EDTA, 100 µM Thioflavin T (ThT) and pH 6.0. All solutions were filtered through a 0.22-µm filter to remove particulates. In all, 200 µl aliquots were placed in silanised Greiner 96-well flat-bottomed plates (\#655077) containing four 0.5-mm diameter zirconium ceramic beads in each well to assist agitation. The plates were incubated at 37 °C with constant agitation in a Tecan Infinite F200 Microplate Fluorimeter. Fibril formation was monitored through the increase in ThT fluorescence (excitation 430 nm, emission 485 nm), with readings acquired every 600 s. Five replicates were used for each PrP sample. To determine the half- and lag-times for fibril formation, data were fitted to an empirical function described by Nielsen et al.^[@CR90]^.$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${\mathrm{Fi}} + {\mathrm{Ff}}/\left\{ {1 + {\mathrm{exp}}\left[ { - \left( {{{t}}-{{t}}_{\mathrm{m}}} \right)/\tau } \right]} \right\}$$\end{document}$$where Fi is the initial fluorescence reading, Ff is the final fluorescence reading, *t* is time, *t*~m~ is the time taken to half maximal fluorescence and *τ* is the reciprocal of the propagation rate during the rise phase \[1/*k*~(apparent)~\]. Lag-time is defined as *t*~m~ − 2*τ*. Formation of protease-resistant PrP by PMCA amplification {#Sec27} --------------------------------------------------------- PMCA substrate homogenates were prepared from mice that had been perfused with PBS containing 5 mM EDTA at the time of death. PrP-null (*Prnp*^*o/o*^), Tg35 (homozygous for huPrP G127/M129), Tg152 (homozygous for huPrP G127/V129) or Tg183 (homozygous for huPrP V127/M129)) mouse brains^[@CR25]^ were homogenised in cold conversion buffer (PBS containing 150 mM NaCl, 1.0% (v/v) Triton X‐100, 4 mM EDTA and 1× Complete protease inhibitor (Roche Applied Science)), using a Duall tissue grinder to give a 10% (w/v) homogenate. Substrates were seeded with a 1/100 dilution of vCJD (I4618) 10% brain homogenate in PBS. Each reaction mixture was divided in two prior to PMCA with one half stored at −70 °C as a minus PMCA control. PMCA consisted of 96 cycles of 30 s sonication every 30 min in a Misonix S3000 at 75% power output (Misonix, Farmingdale, NY), reactions were carried out with 40 µl substrate in 200-µl thin-walled PCR tubes at 35 °C. Samples were digested with 50 µg ml^−1^ proteinase K (PK) for 1 h at 37 °C. The reaction was stopped by the addition of AEBSF in SDS-loading buffer and samples were boiled for 10 min before running on 16% Tris-glycine gels. Western blotting was carried out according to Unit protocol, using 3F4 (Merck Inc, N.J., U.S.A) as the primary antibody and goat anti-mouse IgG conjugated to alkaline phosphatase (Sigma A2179) as the secondary antibody. Statistics and reproducibility {#Sec28} ------------------------------ In the reported experiments, each protein sample was identically-engineered. The sample size (*n*) of each experiment is provided in the corresponding figure captions in the main manuscript and supplementary information files. Sample sizes were chosen to support meaningful conclusions. All in vitro folding experiments were replicated at least three times. In vitro fibrillisation assays were replicated five times. *T*~1~ and *T*~2~ NMR data were recorded with randomised order of time increment and each included one duplicate dataset. Replicate experiments were successful. Investigators were not blinded during experimental measurements or data analysis. Reporting summary {#Sec29} ----------------- Further information on research design is available in the Nature Research [Reporting Summary](#MOESM7){ref-type="media"} linked to this article. Supplementary information ========================= {#Sec30} ###### Supplementary Information ###### Supplementary Data 1 ###### Supplementary Data 2 ###### Supplementary Data 3 ###### Peer Review File ###### Description of Additional Supplementary Files ###### Reporting Summary ###### PDB validation report 1 ###### PDB validation report 2 **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** is available for this paper at 10.1038/s42003-020-01126-6. We thank Richard Newton for the preparation of figures. This work was supported by the Medical Research Council. We are grateful to the staff at Diamond Light Source for access to X-Ray diffraction data collection facilities. We gratefully acknowledge the longstanding and major contribution of the late Anthony Clarke to our structural studies. L.L.P.H., R.C., D.S., M.J.C., A.M.H., K.M., R.L.B., G.S.J., J.P.W. and J.C. designed research; L.L.P.H., R.C., D.S., M.B., E.B.S., S.F., M.J.C., A.M.H., K.M., R.L.B. and J.P.W. performed research; D.S. and K.M. contributed new analytic tools; L.L.P.H., R.C., M.J.C., A.M.H., K.M., R.L.B., J.P.W. and J.C. analysed data; L.L.P.H., R.C., K.M., R.L.B., J.B., J.P.W. and J.C. wrote the paper. The atomic coordinates for the crystal structures described in this paper have been deposited in the Brookhaven Protein Data Bank (<https://www.rcsb.org/>) (accession nos. 6SV2 (V127/M129 PrP) & 6SUZ (V127/V129 PrP)). The data that support the findings of this study are available from the corresponding author upon reasonable request. G.S.J. and J.C. are shareholders, and J.C. a director of D-Gen Limited, an academic spin-out company in the field of prion disease diagnosis, decontamination, and therapeutics, which provided the ICSM18 monoclonal antibody used in this study. The remaining authors share no competing interests.

Key Points {#Sec1} ========== CBD has been reported to exert a number of physiological, biochemical, and psychological effects that have the potential to benefit athletes.The available evidence is preliminary, at times inconsistent, and largely based on preclinical studies involving laboratory animals.Rigorous, controlled investigations clarifying the utility of CBD in the sporting context are warranted. Introduction {#Sec2} ============ *Cannabis sativa* contains numerous chemical compounds with potential bioactive effects, including at least 144 cannabinoids \[[@CR56], [@CR76]\]. The most studied of the cannabinoids are Δ^9^-tetrahydrocannabinol (Δ^9^-THC), renowned for its distinctive intoxicating effects \[[@CR73], [@CR123]\], and cannabidiol (CBD)---a non-intoxicating cannabinoid that is particularly enriched in industrial hemp cultivars grown for seed and fibre \[[@CR61]\]. CBD was first isolated in 1940 and initially considered to be biologically inactive, with no apparent therapeutic or "subjective" drug effects \[[@CR1]\]. However, in 1973, Carlini et al. \[[@CR27]\] demonstrated anticonvulsant effects of CBD in a preclinical model, which were later mirrored in humans suffering from intractable epilepsy \[[@CR46]\]. A subsequent rise in research into CBD \[[@CR206]\] has uncovered interactions with numerous molecular targets \[[@CR92]\] and a range of potential therapeutic applications \[[@CR138]\]. Following successful phase 3 clinical trials \[[@CR53], [@CR54], [@CR172]\], the oral CBD solution, Epidiolex®, has also recently gained Food and Drug Administration approval as a regulated prescription medication to treat certain forms of paediatric epilepsy. Recently, interest in CBD has intensified among the general population as evidenced by an exponential rise in internet searches for 'CBD' in the United States (USA) \[[@CR108]\]. Some professional athletes (e.g. golfers, rugby players) also appear to be using CBD (e.g. 'Team cbdMD' <https://www.cbdmd.com/>), despite there being no published studies demonstrating beneficial effects on sport or exercise performance. In many jurisdictions, including the USA and Europe, access to regulated, prescription CBD (i.e. Epidiolex®) is limited to patients with intractable epilepsy. However, a wide range of low dose (e.g. 5--50 mg·d^−1^) CBD-containing "nutraceuticals" (primarily in oil or capsule form) have become readily available online and over-the-counter (e.g. pharmacies, health food stores) \[[@CR20], [@CR125]\]. This includes some varieties that are marketed specifically to recreational and elite athletes (e.g. cbdMD, fourfivecbd). The use of these products is likely to become even more widespread if the World Health Organization's recommendation that CBD no longer be scheduled in the international drug control conventions is adopted by the United Nations member states \[[@CR201]\]. Cannabis has been prohibited in all sports during competition since the World Anti-Doping Agency first assumed the responsibility of establishing and maintaining the list of prohibited substances in sport 15 years ago \[[@CR89]\]. In 2018, however, CBD was removed from the Prohibited List \[[@CR199]\], presumably on the basis of mounting scientific evidence that the cannabinoid is safe and well-tolerated in humans \[[@CR16], [@CR169]\], even at very high doses (e.g. 1500 mg·day^−1^ or as an acute dose of 6000 mg) \[[@CR170]\]. While several recent reviews have described the impact of cannabis on athlete health and performance \[[@CR99], [@CR176], [@CR188]\], the influence of CBD alone has yet to be addressed. The aim of this narrative review was to explore evidence on the physiological, biochemical, and psychological effects of CBD that may be relevant to sport and/or exercise performance and to identify relevant areas for future research. Given the absence of studies directly investigating CBD and sports performance, this review draws primarily on preclinical studies involving laboratory animals and a limited number of clinical trials involving non-athlete populations. Cannabidiol (CBD): Molecular Targets, Pharmacokinetics and Dosing {#Sec3} ================================================================= Molecular Targets {#Sec4} ----------------- The distinctive intoxicating effects of Δ^9^-THC (as well as some of its therapeutic effects) involve the activation of CB~1~R (the cannabinoid type 1 receptor) \[[@CR12]\]. This ubiquitous receptor is expressed throughout the central nervous system, the peripheral nervous system, and in the cardiovascular system, gastrointestinal (GI) tract, skeletal musculature, liver, and reproductive organs \[[@CR205]\]. Unlike Δ^9^-THC, CBD is not an agonist of CB~1~R, although it may act as a negative allosteric modulator (NAM) at this site (i.e. decreasing the potency and/or efficacy of other ligands without activating the receptor itself) \[[@CR92], [@CR106]\]. Δ^9^-THC also acts as an agonist at CB~2~R (the cannabinoid type 2 receptor) \[[@CR12]\] and there is emerging evidence of CBD functioning as a partial agonist at this site \[[@CR171]\]. CB~2~R is primarily located on immune system cells but is also expressed in the cardiovascular system, GI tract, bone, liver, adipose tissue, and reproductive organs \[[@CR205]\]. CBD may also influence the endocannabinoid system indirectly via the inhibition of fatty acid amide hydrolase (FAAH), a key enzyme involved in the degradation of the principle endocannabinoid signalling molecule, anandamide (AEA) \[[@CR92], [@CR110]\]. The inhibition of FAAH is predicted to lead to an increase in brain and plasma concentrations of AEA, which acts as a partial agonist at CB~1~R and CB~2~R, thereby increasing endocannabinoid tone \[[@CR92], [@CR110]\]. Increases in endocannabinoid tone may also occur as a result of CBD inhibiting AEA transport via effects on fatty acid-binding proteins (and this mechanism may have more relevance than FAAH inhibition in humans) \[[@CR57]\]. CBD also interacts with many other non-endocannabinoid signalling systems \[[@CR92]\]. Briefly, at concentrations ≤ 10 μM, CBD has been reported to interact with the serotonin 1A \[5-HT~1A~\] receptor, the orphan G protein-coupled receptor 55, as well as the glycine, opioid, and peroxisome proliferator-activated receptors, various ion channels (e.g. the transient potential vanilloid receptor type 1 channel \[TRPV1\] and other transient potential vanilloid channels) and various enzymes (e.g. cyclooxygenase (COX)1 and COX2, cytochrome P450 enzymes) \[[@CR11], [@CR92]\] (see Ibeas et al. \[[@CR92]\] for review). CBD also possesses antioxidant properties \[[@CR92]\]. It is important to recognise that the molecular targets of CBD are still being established, with many of those identified in in vitro cellular assays still to be validated as occurring in vivo. As such, the functional relevance of many of these interactions remains to be established. Pharmacokinetics {#Sec5} ---------------- CBD is often consumed orally as oil; however, it can also be ingested in other forms (e.g. gel capsules, tinctures, beverages, and confectionery products) and applied topically \[[@CR20], [@CR125]\]. High concentration CBD "vape oils" (i.e. for use in e-cigarette devices) are also available in some countries, as are some CBD-dominant forms of cannabis (sometimes known as "light cannabis") that can be smoked or vaporised \[[@CR20], [@CR125]\]. Pure, synthetic, crystalline CBD was also vaporised in a recent laboratory study \[[@CR160]\]. Taylor et al. \[[@CR170]\] recently conducted a comprehensive analysis of oral CBD oil pharmacokinetics in healthy participants. When administered as a single, oral dose (1500--6000 mg), the time to reach peak plasma concentrations (*t*~max~) was \~4--5 h and the terminal half-life was \~14--17 h. Although *t*~max~ did not increase dose-dependently in this investigation \[[@CR170]\], another study \[[@CR19]\], involving a much lower oral dose of CBD (300 mg), did indicate a shorter *t*~max~ (i.e. \~2--3 h). Peak plasma concentrations (*C*~max~) were \~0.9--2.5 μM in Taylor et al. \[[@CR170]\], but increased \~4.9-fold when CBD was administered with a high-fat meal (i.e. \~5.3 μM at 1500 mg dose) \[[@CR170]\]. Both studies observed a large amount of inter-individual variation in pharmacokinetic responses \[[@CR19], [@CR170]\]. The pharmacokinetics of inhaled CBD are yet to be well characterised. However, smoked "light cannabis" (with a lower Δ^9^-THC and higher CBD content than other varieties) has been reported to elicit high serum CBD concentrations at 30 min post-treatment (that decline over time) \[[@CR146]\]. A recent study in which participants vaporised 100 mg of CBD likewise observed high blood CBD concentrations 30 min post-treatment \[[@CR160]\]. As neither study collected blood samples within \< 30 min of CBD administration, *t*~max~ and *C*~max~ are unknown \[[@CR146], [@CR160]\]. CBD is metabolised by several cytochrome P \[CYP\] 450 enzymes (e.g. CYP3A4, CYP2C9, CYP2C19) which convert it to a number of primary and secondary metabolites (e.g. 7-OH-CBD, 6-OH-CBD, and 7-COOH-CBD) \[[@CR177]\]. Complex pharmacokinetic interactions may occur when CBD is co-administered with other drugs (e.g. Δ^9^-THC) and dietary constituents (e.g. caffeine) that also utilise these enzymes \[[@CR6], [@CR163]\]. Interspecies Dose Conversions {#Sec6} ----------------------------- Given the number of preclinical studies involving animal models that will be discussed in this review, it is important to consider interspecies dose equivalence (Table [1](#Tab1){ref-type="table"}). The USA Food and Drug Administration \[[@CR30]\] recommend the following approach to interspecies dose conversion: $$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathrm{HED}\left(\mathrm{mg}\cdot {\mathrm{kg}}^{\hbox{-} 1}\right)={\mathrm{Dose}}_{\mathrm{Animal}}\left(\mathrm{mg}\cdot {\mathrm{kg}}^{\hbox{-} 1}\right)\times \left(\frac{{\mathrm{Km}}_{\mathrm{Animal}}}{{\mathrm{Km}}_{\mathrm{Human}}}\right) $$\end{document}$$ Where HED is the human equivalent dose and Km is a correction factor estimated by dividing the average body mass (BM) of the species (60, 0.020 and 0.150 kg for 11 humans, mice and rats respectively) and by its surface area (see: Nair, et al. \[[@CR134]\] for 12 further details). Differences between systemic and oral dosing should also be considered \[[@CR9]\]. Intraperitoneal (i.p.) dosing is often used in animal studies and has been reported to elicit *C*~max~ values \~7-fold higher than oral dosing in mice \[[@CR52]\]. Thus, an "oral equivalent dose" can be approximated by multiplying the i.p. dose by seven \[[@CR9]\] (Table [1](#Tab1){ref-type="table"}). Intravenous (i.v.) dosing will produce even higher plasma CBD concentrations; however, the authors are not aware of any published data that would facilitate conversion between i.v. and oral dosing in rodents. Please note that these values are intended as a guide only and subject to limitations (e.g. interspecies differences in drug potency and receptor expression/configuration). Table 1Oral human equivalent CBD doses from mouse and rat intraperitoneal dosesMouse to Human CBD Dose ConversionRat to Human CBD Dose ConversionMouse Dose\ (mg·kg^-1^, i.p.)HED\ (mg, p.o.)Rat Dose\ (mg·kg^-1^, i.p.)HED\ (mg, p.o.)13416851705340103411068120681201362301021302043602043604086Each HED is based on a body mass of 60 kg and calculated as per the methods described in 2.3 *Dose Conversions*. The highest documented acute oral CBD dose in humans is 6000 mg; the highest documented chronic oral CBD dose in humans is 1500 mg \[169\]. HED: Human Equivalent Dose; i.p.: Intraperitoneal; p.o.: Oral Cannabidiol (CBD) in Sport and Exercise Performance {#Sec7} =================================================== Literature Search Methodology {#Sec8} ----------------------------- The clinical and preclinical literature was reviewed to identify studies investigating the effects of CBD that might be relevant within a sport and/or exercise context. The online databases PubMed (MEDLINE), Web of Science (via Thomas Reuters), and Scopus were searched between April and October of 2019 using terms such as: 'cannabinoid' 'cannabidiol', 'CBD' and 'cannabis'. This review focuses primarily on effects that have been demonstrated in vivo and generally avoids attempting to predict functional effects on the basis of target-oriented in vitro data, given the numerous molecular targets of CBD \[[@CR92]\] and the fact that exercise itself induces complex biochemical changes. Nonetheless, some potential interactions are noted. As our intent was to summarise evidence on a range of potentially relevant topics, rather than provide a detailed assessment of the literature, the reader will be directed to more focused reviews, where appropriate. All doses described are oral and acute (single), unless otherwise stated. Exercise-Induced Muscle Damage---Muscle Function, Soreness, and Injury {#Sec9} ---------------------------------------------------------------------- Exercise, particularly when strenuous, unfamiliar, and/or involving an eccentric component, can cause ultrastructural damage to skeletal muscle myofibrils and the surrounding extracellular matrix \[[@CR36], [@CR59]\]. This exercise-induced muscle damage (EIMD) impairs muscle function and initiates an inflammatory response \[[@CR59]\]. While inflammation is integral to EIMD repair, regeneration, and adaptation \[[@CR59]\], excessive inflammation may contribute to prolonged muscle soreness and delayed functional recovery \[[@CR7], [@CR158]\]. CBD modulates inflammatory processes \[[@CR21]\]. In preclinical models of acute inflammation, CBD has been reported to attenuate immune cell accumulation (e.g. neutrophils, lymphocytes macrophages) \[[@CR102], [@CR130], [@CR149], [@CR186]\], stimulate production of anti-inflammatory cytokines (e.g. interleukin (IL)-4, IL-10) \[[@CR190], [@CR191], [@CR23]\] and inhibit production of pro-inflammatory cytokines (e.g. IL-1β, IL-6, IL-8, tumour necrosis factor (TNF)-α) \[[@CR10], [@CR50], [@CR55], [@CR62], [@CR63], [@CR113], [@CR130], [@CR149], [@CR154], [@CR186]\] and reactive oxygen species \[[@CR62], [@CR130], [@CR186]\]. Models demonstrating such effects have included lung injury induced by chemical treatment \[[@CR149]\] and hypoxic--ischemia (HI) \[[@CR10]\]; liver injury induced by ischemia-reperfusion \[[@CR63], [@CR130]\] and alcohol feeding \[[@CR186]\]; myocardial \[[@CR55]\] and renal \[[@CR62]\] ischemia-reperfusion injuries; surgically induced oral lesions \[[@CR102]\]; chemically induced osteoarthritis \[[@CR145]\]; spinal cord contusion injury \[[@CR113]\], and colitis \[[@CR23], [@CR50], [@CR154]\] (see Burstein \[[@CR24]\] for review). Anti-inflammatory effects are generally observed at higher CBD doses in vivo (e.g. ≥ 10 mg·kg^−1^, i.p.); although, lower doses (e.g. \~1.5 mg·kg^−1^, i.p.) have indicated efficacy in some studies \[[@CR145]\]. Research investigating the effects of CBD on inflammation in humans is limited and inconclusive \[[@CR94], [@CR133]\]. In terms of muscle-specific inflammation, one preclinical study has investigated the effect of high-dose CBD (i.e. 60 mg·kg^−1^·d^−1^, i.p.) on transcription and synthesis of pro-inflammatory markers (i.e. IL-6 receptors, TNF-α, TNF-β1, and inducible nitric oxide synthase) in the gastrocnemius and diaphragm of dystrophic MDX mice (a mouse model of Duchenne muscular dystrophy) \[[@CR91]\]. In this investigation, CBD attenuated mRNA expression of each marker and reduced plasma concentrations of IL-6 and TNFα. Improvements in muscle strength and coordination, as well as reductions in tissue degeneration, were also reported at this dose. Lower, but still relatively high, CBD doses (20--40 mg·kg^−1^·day^−1^, i.p.) had no functional benefits \[[@CR91]\]. Of course, it is important to recognise that EIMD and muscular dystrophy differ in their pathophysiology, and so the effects observed in MDX mice may involve mechanisms less relevant to EIMD (e.g. skeletal muscle differentiation, autophagy) \[[@CR91]\]. While CBD could potentially aid in muscle recovery, other anti-inflammatory agents, such as ibuprofen (a non-steroidal anti-inflammatory drug \[NSAID\]) have been reported to attenuate exercise-induced skeletal muscle adaptation \[[@CR120]\]. The precise mechanism(s) underpinning these effects have not been fully elucidated, although it may be that the prevention of inflammation inhibits angiogenesis and skeletal muscle hypertrophy \[[@CR120]\]. Human trials also suggest that ibuprofen may not influence EIMD, inflammation, or soreness \[[@CR144], [@CR175]\]. Thus, if CBD exerts its effects via similar mechanisms, it could possibly attenuate the benefits of training without influencing muscle function or soreness. Future studies investigating this are clearly warranted to clarify such issues and elucidate the potential benefits of CBD. Neuroprotection---Concussion and Subconcussion {#Sec10} ---------------------------------------------- Recent estimates suggest that 6--36% of high school and collegiate athletes in the USA have experienced more than one concussion \[[@CR72]\], potentially predisposing them to long-term neurodegenerative diseases \[[@CR72]\] and an increased risk of suicide \[[@CR64]\]. Concussion is a distinct form of mild traumatic brain injury (TBI) in which a biomechanical force temporarily disrupts normal brain functioning causing neurological--cognitive--behavioural signs and symptoms \[[@CR97]\]. Similar injuries that do not produce overt (acute) signs or symptoms are termed "subconcussions" \[[@CR97]\]. In TBI, the primary injury occurs as a result of the biomechanical force; secondary injury is then sustained through a complex cascade of events, including HI, cerebral oedema, increased intracranial pressure, and hydrocephalus \[[@CR203]\]. These processes are, in turn, related to a number of detrimental neurochemical changes, including glutamate excitotoxicity, perturbation of cellular calcium homeostasis, excessive membrane depolarisation, mitochondrial dysfunction, inflammation, increased free radicals and lipid peroxidation, and apoptosis \[[@CR203]\]. While the primary injury may not be treatable, interventions that attenuate secondary sequelae are likely to be of benefit \[[@CR203]\]. Only one study \[[@CR14]\] has investigated the biochemical and neuropsychological effects of CBD in an animal model of TBI. Here, C57BL/6 mice were given chronic CBD treatment (3 μg·day^−1^, oral) 1--14 and 50--60 days post- (weight drop) brain insult. CBD attenuated the behavioural (e.g. anxious and aggressive behaviour, depressive-like behaviour, impaired social interactions, pain-related behaviours) and some of the cortical biochemical abnormalities were observed. Specifically, CBD tended to normalise extracellular glutamate, [d]{.smallcaps}-aspartate, and γ-aminobutyric acid concentrations in the medial prefrontal cortex, suggesting a reduction in excitotoxicity. However, neuronal damage was not measured directly in this study \[[@CR14]\]. Other preclinical studies have investigated the impact of CBD on different animal models of acute neuronal injury, in particular, acute cerebral HI \[[@CR4], [@CR13], [@CR31], [@CR68], [@CR69], [@CR80], [@CR81], [@CR83], [@CR100], [@CR105], [@CR127], [@CR129], [@CR142], [@CR143], [@CR153]\]. Studies administering a single (acute) dose of CBD shortly post-HI (e.g. ≤1 h) have produced inconsistent results. For instance, while Garberg et al. \[[@CR68], [@CR69]\] found no effect of CBD (1 or 50 mg·kg^−1^, i.v.) on HI-induced neuronal damage in piglets, others observed neuroprotection at similar doses (e.g. 1 mg·kg^−1^, i.v \[[@CR105], [@CR143]\]., 1 mg·kg^−1^, s.c \[[@CR127], [@CR142]\]., and 5 mg·kg^−1^, i.p \[[@CR31]\].) in piglets and rats. When given chronically, or repeatedly within a short timeframe proximal to the HI event, however, CBD appears to be neuroprotective. Effective dosing strategies have varied and included initiating treatment several days pre-HI (e.g. 100 or 200 μg·day^−1^, intracerebroventricular 5 days; Wistar rats \[[@CR100]\]), shortly pre- and/or post-HI[1](#Fn1){ref-type="fn"}, and up to 3 days post-HI (e.g. 3 mg·kg^−1^·day^−1^, i.p. 12 days; ddY mice \[[@CR80]\]). Thus, chronic CBD treatment may be more effective than acute intervention. While "pre-incident" dosing might also be beneficial, it is noted that in practice, this would require humans at risk of TBI to use CBD chronically as a prophylactic. The precise mechanism(s) underpinning the neuroprotective effects of CBD are not completely understood (see Campos et al. \[[@CR25]\] for review), but may involve decreased inflammation, oxidative stress, and excitotoxicity \[[@CR142], [@CR143]\] and increased neurogenesis \[[@CR129]\]. Preclinical studies have also demonstrated beneficial effects of CBD in other animal models of neurodegeneration (e.g. transgenic model of Alzheimer's disease \[[@CR34], [@CR35]\], brain iron-overload \[[@CR47], [@CR48]\]). Collectively, these data suggest that research investigating the utility of CBD in ameliorating the harmful long-term effects of repeated sports concussions is warranted. Nociceptive and Neuropathic Pain {#Sec11} -------------------------------- Persistent pain is common in athletes \[[@CR74]\]. Nociceptive pain, which includes inflammatory pain, typically occurs with tissue damage; whereas neuropathic pain typically results from a lesion or disease in the somatosensory nervous system \[[@CR74]\]. Neuropathic pain is common among para-athletes with spinal cord injuries and can also arise with surgery (e.g. to treat an existing injury) or if there is repetitive mechanical and/or inflammatory irritation of peripheral nerves (e.g. as in endurance sports) \[[@CR74]\]. Clinical trials investigating the combined effects of Δ^9^-THC and CBD (e.g. Sativex®) on chronic neuropathic pain have yielded promising initial results \[[@CR87], [@CR114], [@CR151], [@CR156]\]. However, the therapeutic effects of CBD administered alone have received limited clinical attention. Preclinical (in vivo) studies investigating the effects of CBD on neuropathic and nociceptive pain are summarised in Table [2](#Tab2){ref-type="table"}. Despite some methodological inconsistencies (e.g. the pain model, period of treatment, route of delivery), most preclinical studies appear to have observed a significant analgesic effect of CBD \[[@CR29], [@CR39]--[@CR41], [@CR51], [@CR70], [@CR75], [@CR78]\], albeit somewhat less pronounced than the effects of Δ^9^-THC \[[@CR29], [@CR78]\] (e.g. Hedges' *g* = 0.8 vs. 1.8 \[[@CR78]\]) or of gabapentin (e.g. Hedges' *g* = 2.0 \[[@CR78]\]), a commonly used agent for treating neuropathic pain. Capsazepine co-treatment has also been reported to attenuate CBD-induced analgesia, suggesting that the effect may be mediated, at least in part, by the TRPV~1~ channel \[[@CR40], [@CR41], [@CR51]\]. This mechanism is noteworthy as studies have implicated the TRPV~1~ in the development of mechanical hyperalgesia induced by muscle inflammation \[[@CR66], [@CR140]\]. Table 2Preclinical studies investigating the effect of CBD on neuropathic and nociceptive pain in vivoCitationAnimalModelTreatment(s)Treatment Effect**Neuropathic Pain** De Gregorio et al., (2019) \[[@CR51]\]Wistar ratsSNI5 mg·kg^-1^·d^-1^, s.c. 7 dCBD sig. decreased mechanical allodynia on Tx day 7. Casey et al., (2017) \[[@CR29]\]C57BL/6 miceCCI30 mg·kg^-1^, s.c.CBD sig. decreased mechanical allodynia 2 h, but not 0.5, 1, 4 or 6 h, post-Tx compared to baseline.0.01, 0.1, 1, 10 or 100 mg·kg^-1^, s.c.CBD dose-dependently decreased mechanical and cold allodynia. King et al., (2017) \[[@CR101]\]C57BL/6 miceCT (Paclitaxel)0.625--20 mg·kg^-1^, i.p. 15 min prior to CT on days 1, 3, 5 and 71 and 20 mg·kg^-1^ CBD sig. attenuated the development of mechanical allodynia measured on Tx days 9 and 14, but not 21.CT (Oxaliplatin)1.25--10 mg·kg^-1^, i.p. 15 min prior to CT on days 1, 3, 5 and 71.25--10 mg·kg^-1^ CBD sig. attenuated the development of mechanical allodynia measured on Tx days 2, 4, 7 and 10.CT (Vincristine)1.25--10 mg·kg^-1^, i.p. 15 min prior to CT on days 1, 3, 5 and 7CBD did not attenuate the development of CT-induced mechanical allodynia measured on Tx days 5, 10, 15 and 22. Harris et al., (2016) \[[@CR78]\]C57BL/6 miceCT (Cisplatin)2 mg·kg^-1^, i.p.CBD sig. decreased tactile allodynia 1 h post-Tx.0.5, 1 or 2 mg·kg^-1^, i.p. 30 min prior to CT every second day for 12 dCBD did not attenuate the development of CT-induced tactile allodynia measured on Tx days 6, 10 and 12. Ward et al., (2014) \[[@CR187]\]C57BL/6 miceCT (Paclitaxel)2.5 or 5 mg·kg^-1^·d^-1^, i.p. 15 min prior to CT on days 1, 3, 5 and 72.5 and 5 mg·kg^-1^·d^-1^ CBD attenuated the development of CT-induced mechanical allodynia. Toth et al., (2010) \[[@CR174]\]CD1 miceSTZ Diabetes0.1, 1 or 2 mg·kg^-1^·d^-1^, i.n. 3 months1 and 2 mg·kg^-1^·d^-1^ CBD sig. attenuated the development of thermal and tactile hypersensitivity compared to 0.1 mg·kg^-1^·d^-1^ CBD.2 mg·kg^-1^·d^-1^, i.n. 1 monthCBD did not alleviate developed thermal or tactile hypersensitivity.1, 10 or 20 mg·kg^-1^·d^-1^, i.p. 3 months20 mg·kg^-1^·d^-1^ CBD sig. attenuated the development of thermal and tactile hypersensitivity compared to 1 mg·kg^-1^·d^-1^ CBD.20 mg·kg^-1^·d^-1^, i.p. 1 monthCBD did not alleviate developed thermal or tactile hypersensitivity. Costa et al., (2007) \[[@CR41]\]Wistar ratsCCI2.5, 5, 10 or 20 mg·kg^-1^·d^-1^, oral 7 d5, 10 and 20 mg·kg^-1^·d^-1^ CBD sig. decreased thermal and mechanical hyperalgesia on Tx day 7.**Nociceptive (Inflammatory) Pain** Genaro et al., (2017) \[[@CR70]\]Wistar ratsIncision0.3, 1, 3, 10 or 30 mg·kg^-1^, i.p.3 mg·kg^-1^ CBD sig. decreased mechanical allodynia between 30- and 150-min post-Tx; 10 mg·kg^-1^ CBD sig. decreased mechanical allodynia 60 min post-Tx, only. Hammell et al., (2016) \[[@CR75]\]Sprague-Dawley ratsFCA0.6, 3.1, 6.2 or 62.3 mg·kg^-1^·d^-1^, t.c. 4 d6.2 and 62.3 mg·kg^-1^ CBD sig. decreased pain-related behaviour on Tx day 4 and thermal hyperalgesia on Tx days 2, 3 and 4. Costa et al., (2007) \[[@CR41]\]Wistar ratsFCA20 mg·kg^-1^·d^-1^, oral 7 dCBD sig. decreased thermal and mechanical hyperalgesia on Tx day 7. Costa et al., (2004) \[[@CR39]\]Wistar ratsCarrageenan5, 7.5, 10, 20 and 40 mg·kg^-1^, oral5, 7.5, 10, 20 and 40 mg·kg^-1^·d^-1^ CBD sig. decreased thermal hyperalgesia 1--5 h post-Tx. Costa et al., (2004) \[[@CR40]\]Wistar ratsCarrageenan10 mg·kg^-1^, oralCBD sig. decreased thermal hyperalgesia 1 h post-Tx.The 'Treatment Effects' described are in comparison to a vehicle condition, unless otherwise stated*CBD* Cannabidiol, *CCI* Chronic Constriction Injury, *CT* Chemotherapy, *FCA* Freund's Complete Adjuvant, *i.n.* Intranasal, *i.t.* Intrathecal, *s.c.* Subcutaneous, *SNI* Spared Nerve Injury, *STZ* Streptozotocin, *t.c.* Transcutaneously, *Tx* Treatment It is important to recognise that the analgesic effect of CBD likely depends on several factors, including the treatment dose and the type of pain involved. Indeed, low doses of CBD (e.g. ≤ 1 mg·kg^−1^, i.p.) do not consistently attenuate pain \[[@CR29], [@CR41], [@CR70], [@CR75], [@CR101]\]; while higher doses are sometimes found to be more \[[@CR29]\], and other times, less \[[@CR70]\], efficacious than moderate doses in preclinical studies (Table [3](#Tab3){ref-type="table"}). This highlights the importance of determining a therapeutic dose for CBD in analgesia. Data from King et al. \[[@CR101]\] also demonstrate the selectivity of the response, indicating that CBD only effective in attenuating the development of neuropathic pain induced by certain chemotherapeutic agents (i.e. paclitaxel and oxaliplatin but not vincristine). Thus, placebo-controlled trials of CBD in treating pain in clinical populations and athletes are warranted. Table 3Clinical trials investigating the effect of CBD on subjective anxietyCitationStudy designParticipantsTreatment(s)ContextTreatment effect**Healthy participants**Zuardi et al. (1993) \[[@CR207]\]DB (PC); BSDI: 10; 23 yC: 10; 23 yOral 300 mgPublic speaking (90 min post-tx)CBD sig. reduced SA post-public speaking. SA prior to, in anticipation of, and during speaking were unaffected.Crippa et al. (2004) \[[@CR45]\]DB (PC); BSDI: 5 M; 30 ± 5 yC: 5 M; 30 ± 5 yOral 400 mgLow stressCBD sig. reduced SA 60 and 75 min post-tx.Bhattacharyya et al. (2010) \[[@CR18]\]DB (PC); WSD15 M; 27 ± 6 yOral 600 mgLow stressSA was unaffected 1, 2 and 3 h post-tx.Martin-Santos et al. (2012) \[[@CR123]\]DB (PC); WSD16 M; 26 ± 5 yOral 600 mgLow stressSA was unaffected 1, 2 and 3 h post-tx.Hindocha et al. (2015) \[[@CR86]\]DB (PC); WSD48; 22 yVaporised 16 mgLow stressSA was unaffected 2, 30, 60. 90 and 120 min post-tx.Zuardi et al. (2017) \[[@CR208]\]DB (PC); BSDI: 12 (6 M); 23 ± 3 yI: 11 (5 M); 23 ± 3 yI: 12 (6 M); 23 ± 3 yC: 12 (6 M); 22 ± 2 yI: Oral 100 mgI: Oral 300 mgI: Oral 900 mgPublic speaking(150 min post-tx)300 mg CBD sig. reduced SA compared to 900 mg CBD (but not placebo) during public speaking. SA prior to, in anticipation of, and post-speaking were not affected by any treatment.Linares et al. (2019) \[[@CR116]\]DB (PC); BSDI: 15 M; 24 ± 3 yI: 15 M; 25 ± 3 yI: 12 M; 23 ± 3 yC: 15 M; 25 ± 4 yI: Oral 150 mgI: Oral 300 mgI: Oral 600 mgPublic speaking(90 min post-tx)300 mg CBD (but not 150 or 600 mg) sig. reduced SA during public speaking. SA prior to, in anticipation of, and post-speaking were not affected by any treatment.**Clinical populations**Crippa et al. (2011) \[[@CR44]\]DB (PC); WSDSAD Participants10 M; 24 ± 4 yOral 400 mgLow stressCBD sig. reduced SA 60, 75 and 140 min post-Tx.Bergamaschi et al. (2011) \[[@CR15]\]DB (PC); BSDSAD ParticipantsI: 12 (6 M); 23 ± 2 yC: 12 (6 M); 23 ± 2 yOral 600 mgPublic speaking(90 min Post-Tx)CBD sig. reduced SA during public speaking. SA prior to, in anticipation of, and post-speaking were not affected.Hundal et al. (2017) \[[@CR90]\]DB (PC); BSDHigh Trait ParanoiaI: 16 (8 M); 26 ± 9 yC: 18 (8 M); 25 ± 8 yOral 600 mgVirtual reality(130 min Post-Tx)SA was unaffected.Masataka (2019) \[[@CR124]\]DB (PC); BSDSAD ParticipantsI: 17 (8 M); 18--19 yC: 20 (8 M); 18--19 yOral 300 mg·d^-1^ (28-days)Low stressCBD sig. reduced SA 4-weeks post-tx.*BSD* between-subject design, *C* control group, *CBD* cannabidiol, *DB* double-blind, *I* intervention group, *M* male subjects, *PC* placebo-controlled, *SAD* social anxiety disorder, *SA* subjective anxiety, *SB* single-blind, *Tx* treatment, *WSD* within-subject designThe "Treatment effects" described are in comparison to a placebo condition, unless otherwise stated Exercise-Induced Gastrointestinal (GI) Damage {#Sec12} --------------------------------------------- While strenuous exercise increases blood supply to the active skeletal muscles, cardiopulmonary system and skin---other organs and tissues, including the GI tract, experience reduced oxygen and nutrient delivery \[[@CR180]\]. If exercise is prolonged (e.g. \>40 min), this "GI ischemia", as well as the inflammation and oxidative stress that accompanies reperfusion, can compromise epithelial integrity \[[@CR180]\]. Such effects may negatively influence exercise performance and post-exercise recovery due to GI distress (e.g. nausea, vomiting, abdominal angina, bloody diarrhoea) and impaired nutritional uptake \[[@CR180]\]. CBD has demonstrated some effects that may be relevant to the management of exercise-induced GI damage. For instance, preclinical studies have shown that CBD (e.g. 0.01--10 mg·kg^−1^, i.v. or 10 mg·kg^−1^, i.p.) can attenuate tissue damage (e.g. reduce necrosis, blood concentrations of tissue damage markers and inflammation) induced by acute, peripheral ischemia-reperfusion (e.g. kidney, myocardium, liver) \[[@CR55], [@CR60], [@CR62], [@CR63], [@CR130], [@CR185]\] and colitis \[[@CR23], [@CR50], [@CR154]\] in vivo; benefits that have generally been attributed to its reported antioxidant and anti-inflammatory effects (see also section "Exercise-Induced Muscle Damage---Muscle Function, Soreness, and Injury") \[[@CR50], [@CR55], [@CR60], [@CR62], [@CR63], [@CR130], [@CR154], [@CR185]\]. Also, of interest is that CBD (1--100 μM) has been reported to restore intestinal permeability in vitro following exposure to *Clostridium difficile toxin A,* ethylenediaminetetraacetic acid and pro-inflammatory stimuli (e.g. interferon-gamma, TNF-α) \[[@CR2], [@CR3], [@CR71]\]. Of course, it is important to recognise that evidence to support a therapeutic effect of CBD on GI damage in humans is currently lacking. In fact, two placebo-controlled, double-blinded clinical trials, one investigating the effect of CBD (10 mg·d^−1^; 56 days) on symptom severity in Crohn's disease (*n* = 20) \[[@CR133]\] and the other examining the impact of a "CBD-rich botanical extract" (250 mg·day^−1^ \[4.7% THC\]; 56 d) on the likelihood of remission in ulcerative colitis (*n* = 60) \[[@CR94]\], have so far been unable to demonstrate a protective effect of CBD (above placebo) on disease markers, including C-reactive protein, faecal calprotectin, and pro-inflammatory cytokines (e.g. IL-2, IL-6, TNF-α). While CBD could potentially attenuate exercise-induced GI damage, it is important to note that other anti-inflammatory agents, such as the NSAID, ibuprofen, have been reported to exacerbate exercise-induced GI damage and impair gut barrier function \[[@CR181]\]. The precise mechanism(s) underpinning these effects have not been fully elucidated. However, NSAIDs have been suggested to augment GI ischemia by inhibiting the COX1 and COX2 enzymes and interfering with nitric oxide production \[[@CR180]\]. Some in vitro research similarly suggests that CBD partially inhibits COX1 and COX2, although this effect has only been reported at supraphysiological concentrations (e.g. 50--500 μM CBD) \[[@CR92]\]. Thus, the effect of CBD on exercise-induced GI damage warrants clarification. Bone Health {#Sec13} ----------- While the beneficial effects of high-impact exercise on bone health are well established \[[@CR38]\], other factors within the sporting context (e.g. traumatic injuries, low energy availability \[[@CR117]\]) may cause or contribute to reduced bone health and the development of fractures in athletes. A small number of preclinical studies have investigated the effects of CBD on bone structure and function \[[@CR103], [@CR112], [@CR135]\]. While most have used animal models that are limited in their direct relevance to sport and/or exercise performance (e.g. periodontitis, systemic skeletal degeneration due to spinal cord injury) one investigation \[[@CR103]\] did report that CBD improved the healing of femoral fractures in Sprague-Dawley rats. Specifically, chronic CBD treatment (i.e. 5 mg·kg^--1^·day^--1^, i.p.) decreased callus size 4-weeks post-fracture and enhanced the biomechanical properties of the bone at 8-weeks (i.e. maximal force, work-to-failure on a 4-point bending test) \[[@CR103]\]. While the mechanism(s) underlying this effect require clarification, CBD may act to inhibit expression of RANK and RANK-L (i.e. indicative of an effect to suppress osteoclastogenesis, and thus, bone resorption) and decrease the production of pro-inflammatory cytokines (e.g. IL-1β, TNF-α) at the site of injury \[[@CR103], [@CR135]\]. Evidence that activation of CB~2~R induces bone matrix deposition has also recently emerged \[[@CR150]\] alongside data suggesting that CBD may have partial agonist effects at this site \[[@CR171]\]. These initial findings suggest that further research investigating the effect of CBD on acute skeletal injuries is worthwhile. Cardiovascular (CV) and Metabolic Functions {#Sec14} ------------------------------------------- A number of studies have measured CV responses to CBD (100--1200 mg) in humans and, overall, it appears that resting HR is unaffected (see Sultan et al. \[[@CR165]\] for review). However, some evidence does suggest that CBD (600 mg) reduces resting systolic BP (e.g. --6 mmHg) \[[@CR95], [@CR164]\]. Preclinical studies have likewise shown that CBD influences vascular function \[[@CR139], [@CR161], [@CR162], [@CR193], [@CR194]\]. Briefly, in vitro CBD treatment (i.e. ≤ 2 h exposure to 1--10 μM) has been reported to induce vasorelaxation \[[@CR139]\] and potentiate vasorelaxation to acetylcholine \[[@CR162], [@CR193]\] in isolated (pre-constricted) arteries of rats \[[@CR139], [@CR162], [@CR193]\]. A recent study \[[@CR161]\] also found that in vitro CBD treatment (i.e. ≤ 2 h exposure to 10 μM) induced \~40% vasorelaxation in isolated (pre-constricted) mesenteric arteries of humans with various clinical conditions (e.g. cancer, inflammatory bowel disease, type Z diabetes mellitus \[T2DM\]). In addition to "resting" CV parameters, a recent meta-analysis (of largely preclinical studies) found that CBD attenuated "stress-induced" (e.g. via fear-conditioning or physical-restraint) increases in HR and BP (BP −3.5 mmHg; 95% CI −5.2, −1.9; *I*^2^ = 73%; HR −16 mmHg; 95% CI −26, −6; *I*^2^ = 92%) \[[@CR165]\], that said, most studies measuring CV responses to CBD (150--600 mg) under "stress-inducing" conditions in humans (e.g. public speaking) find no effect on HR or BP \[[@CR15], [@CR116], [@CR207]\]. One placebo-controlled, double-blinded (single-dose) crossover trial of healthy males (*n* = 9) \[[@CR95]\] did report that CBD (600 mg) increased HR in the presence of certain stressors (i.e. a mental arithmetic test, an isometric contraction on a hand-grip dynamometer, and cold exposure); and, at times, reduced systolic and diastolic BP. However, these differences were apparent at baseline (pre-stress) and the data were not standardised to account for this, making interpretation difficult. Taken together, these findings suggest that CBD has the potential to influence CV function. However, the implications of these effects in regard to exercise performance are unclear. Studies investigating the effect of CBD on exercise-induced CV responses are therefore required to clarify its utility within the sport and exercise context. One final observation to note is that some initial data suggest CBD might influence mitochondrial function. Indeed, in vivo CBD treatment has been reported to increase the activity of mitochondrial complexes \[[@CR48], [@CR77], [@CR130], [@CR178]\] (30--60 mg·kg−^1^, i.p. acute or chronic 14 days Wistar rats; 3 and 10 mg·kg^−1^, i.v. C57BL/6 J mice; 10 mg·kg^−1^·day^−1^, i.p. 5 days C57BL/6 J mice; and 10 mg·kg^−1^·day^−1^, i.p. 14 days Wistar rats) in various tissues and models (i.e. healthy brain, myocardium following doxorubicin treatment, brain following neonatal iron-overload, hepatic ischemia-reperfusion injury) and increase mitochondrial biogenesis \[[@CR77]\] (10 mg·kg^−1^·day^−1^, i.p. 5 days C57BL/6 J mice; myocardium following doxorubicin treatment). Such effects could have implications for energy metabolism during exercise. Thermoregulation {#Sec15} ---------------- Heat loss mechanisms play a pivotal role in the maintenance of homeostasis during exercise; any treatment or condition that alters core body temperature (T~C~), therefore has the potential to impact exercise performance \[[@CR192]\]. The effect of CBD on T~C~ has been investigated in rodents \[[@CR65], [@CR82], [@CR83], [@CR85], [@CR96], [@CR118], [@CR166], [@CR182]\]. The most recently published study found that CBD (100 mg·mL^−1^, 30 min vaporised; Wistar rats) reduced T~C~ (−1.0 °C) 60 and 90 min following inhalation in resting animals \[[@CR96]\]. In contrast, Long et al. \[[@CR118]\] reported a hyperthermic effect (+2.0 °C) 30 min post-treatment (1 and 10 mg·kg^−1^, i.p.; C57BL/6JArc mice) during a chronic-dosing experiment, although, this response was only observed intermittently during a 21-day protocol (e.g. \~8% of total measurements). The fact that CBD affected T~C~ in these experiments is difficult to explain, since although other cannabinoids (e.g. Δ^9^-THC, AEA) have demonstrated a capacity to moderate T~C~ when administered exogenously (e.g. low doses of Δ^9^-THC may sometimes induce hyperthermia \[[@CR168]\] and high doses cause hypothermia \[[@CR65], [@CR82], [@CR83], [@CR85], [@CR96], [@CR118], [@CR166], [@CR182]\]), these effects occur via a CB~1~R-mediated mechanism \[[@CR42], [@CR166], [@CR196]\]. In addition to this, no other studies appear to have detected changes in T~C~ with CBD administration \[[@CR65], [@CR82], [@CR83], [@CR85], [@CR166], [@CR182]\]. Overall, despite some inconsistencies, the available data suggest that CBD is unlikely to have a major influence on T~C~ or thermoregulatory processes. In any case, it seems that with the exception of self-reported feelings of "coldness" \[[@CR33], [@CR88]\], exogenous cannabinoids do not typically induce the same overt, significant effects on T~C~ in humans \[[@CR104], [@CR183]\] as are seen in rodents. Still, it should be acknowledged that the thermoregulatory response to heat stress (i.e. passive or metabolic), specifically, has not been studied. Dietary Intake and Feeding {#Sec16} -------------------------- An adequate intake of energy and nutrients is essential to support optimal athletic training, recovery, and performance \[[@CR173]\]. Various preclinical studies have investigated the effect of CBD on feeding behaviour in rodents \[[@CR58], [@CR93], [@CR155], [@CR159], [@CR195]\], with results suggesting that higher doses may influence food intake several hours post-treatment. Indeed, while CBD, at doses of 3--100 mg·kg^−1^, i.p. (IRC mice) \[[@CR195]\] and 1--20 mg·kg^−1^, i.p. (Wistar rats) \[[@CR155]\], failed to influence food intake during a 1 h ad libitum feeding period, moderate to high doses of CBD (4.4 mg·kg^−1^, i.p. \[[@CR58]\]. and 50 mg·kg^−1^, i.p. \[[@CR159]\]) suppressed food intake (in rats) during longer ad libitum feeding periods (i.e. 4--6 h). In line with these results, Ignatowska-Jankowska et al. \[[@CR93]\] found that chronic CBD treatment (2.5 and 5 mg·kg^−1^·day^−1^, i.p. 14 days) attenuated BM gains in growing Wistar rats. A recent systematic review of human trials also reported that individuals with epilepsy receiving CBD (5--20 mg·kg^−1^·day^−1^) were more likely to experience decreased appetite than those receiving placebo (i.e. \~20 vs. 5% of patients) \[[@CR107]\]. That said, a mechanistic understanding of these effects of CBD on feeding behaviour remains to be established. Other cannabinoids with CB~1~R agonist effects (e.g. Δ^9^-THC, AEA, cannabinol) reliably induce hyperphagia when administered exogenously \[[@CR58], [@CR197], [@CR198]\]; but CBD lacks such an effect. Ignatowska-Jankowska et al. \[[@CR93]\] did report that the selective CB~2~R antagonist, AM630, prevented CBD-induced BM changes; however, CB~2~R has not generally been linked to feeding behaviour, and if CBD is indirectly increasing endocannabinoid tone (i.e. via AEA) \[[@CR92]\], this might be expected to promote feeding behaviour (via indirect CB~1~R agonist effects) \[[@CR197]\]. A role for GI side effects in affecting appetite therefore cannot be ruled out \[[@CR107]\]. Further preclinical research appears to be required to clarify the mechanisms underlying these functional effects on feeding. Controlled trials are also needed to determine whether CBD influences appetite and dietary behaviour in humans, particularly during the pre- and post-exercise period, where nutrient provision is critical. Illness and Infection {#Sec17} --------------------- Some research suggests that athletes experience a decrease in immunity and are at increased risk of developing acute illnesses (particularly upper respiratory tract infections) during periods of heavy training and competition \[[@CR184]\]. This phenomenon has been attributed to various factors such as increased psychological stress, poor sleep, long-haul travel, exposure to extreme environments (e.g. altitude), and low energy availability \[[@CR184]\]. A recent review of online content identified a number of webpages claiming benefits of using CBD for the treatment of viral illnesses, including cold and flu \[[@CR167]\]. However, research supporting such "protective effects" of CBD is extremely limited. In fact, the authors identified only two (in vitro) studies reporting anti-microbial effects \[[@CR5], [@CR179]\] and two (in vitro) studies reporting anti-viral effects \[[@CR119], [@CR122]\] (see Tagne et al. \[[@CR167]\] and Nichols et al. \[[@CR136]\] for reviews). In the former, CBD demonstrated anti-microbial activity against various strains of *Staphylococcus aureus* \[[@CR5], [@CR179]\], as well as *Staphylococcus pyogenes*, *Staphylococcus milleri*, and *Staphylococcus faecalis* \[[@CR179]\] at minimum concentrations of \~3.2--15.9 μM; however, one study also found that these effects were virtually abolished when the original media (a nutrient broth agar) was replaced with one containing 5% blood (increasing the minimum concentration to \~160 μM CBD) \[[@CR179]\]. In the latter, CBD indicated anti-viral activity against the hepatitis C virus (EC~50~ = 3.2 μM) and the Kaposi's sarcoma-associated herpesvirus (EC~50~ = 2.1 μM), but not the hepatitis B virus \[[@CR119], [@CR122]\]. While these findings hint at some promise, others caution that CBD could potentially weaken host defence against invading pathogens because of its tendency to modify the function of various immune cells (see also section "Exercise-Induced Muscle Damage---Muscle Function, Soreness, and Injury") \[[@CR136], [@CR147]\]. Importantly, a systematic review of studies investigating the safety of CBD in individuals with intractable epilepsy found that upper respiratory tract infections were similarly infrequent in participants who received the active treatment (5--20 mg·kg^−1^·day^−1^) and placebo (approx. 10% of individuals) \[[@CR107]\]. Further research to develop a better understanding of "if" and "how" CBD influences the development and progression of illness and infection in both athlete and non-athlete populations would be useful. Sports Performance Anxiety (SPA) {#Sec18} -------------------------------- High levels of pre-competition stress, or sports performance anxiety (SPA) \[[@CR141]\], can be detrimental to athletic performance \[[@CR43]\]. This impairment has been attributed to both the direct (i.e. anxiogenic) and indirect (e.g. decreased nutritional intake, increased energy expenditure, loss of sleep) effects of SPA \[[@CR28]\]. While behaviour therapies (e.g. cognitive behavioural therapy) are the preferred treatment, a combination of pharmaceutical and psychological interventions may be indicated in some cases \[[@CR141]\]. A number of (small) clinical trials have investigated the effect of CBD on subjective anxiety in healthy individuals \[[@CR18], [@CR45], [@CR86], [@CR116], [@CR123], [@CR207], [@CR208]\] and in individuals with social anxiety disorder (SAD) \[[@CR15], [@CR44], [@CR124]\] and high trait paranoia \[[@CR90]\] under both standard (i.e. "low stress") \[[@CR18], [@CR44], [@CR45], [@CR86], [@CR123], [@CR124]\] and "stress-inducing" (e.g. simulated public speaking) \[[@CR15], [@CR90], [@CR116], [@CR207], [@CR208]\] conditions (Table [3](#Tab3){ref-type="table"}). Overall, results suggest that CBD has little influence on anxiety under "low stress" conditions in healthy participants \[[@CR18], [@CR86], [@CR123]\]. However, several studies have demonstrated anxiolytic effects of CBD (300--600 mg) under "stress-inducing" conditions in both healthy participants and those with SAD \[[@CR15], [@CR44], [@CR116], [@CR124], [@CR207]\]. In fact, CBD (300 mg) had comparable efficacy to the anxiolytic 5-HT~1A~ agonist drug, ipsapirone (5 mg), during a simulated public speaking test in one study \[[@CR207]\]. On the other hand, some other clinical investigations (involving similar "stress-inducing" stimuli) have found no effect of CBD \[[@CR90], [@CR208]\] and Hundal et al. \[[@CR90]\] observed a non-significant trend (*p* \< 0.10) towards an anxiogenic effect with 600 mg CBD in a high trait paranoid group (*n* = 32). Inconsistencies could be due to inter-individual differences in baseline anxiety levels and the magnitude of the stress-response to the stressor imposed, small sample sizes, and differences in the dose and formulation of CBD provided. Indeed, Linares et al. \[[@CR116]\] observed an inverted U-shaped dose-response relationship between acute CBD treatment and subjective anxiety, indicating that 300 mg (Hedges' *g* = 1.0) had a stronger anxiolytic effect than 150 mg (Hedges' *g* = 0.7) or 600 mg (Hedges' *g* = 0.6). Taken together, it appears that moderate doses of CBD may be anxiolytic in stressful situations and in individuals with SAD. Thus, studies investigating the effect of CBD (in conjunction with behaviour therapies) on pre-competition anxiety, as well as nutritional intake, energy expenditure, symptom perception during exercise (e.g. ratings of perceived exertion), and sleep in athletes who are negatively impacted by SPA are warranted. Sleep {#Sec19} ----- The importance of adequate sleep in facilitating optimal athletic performance and recovery is increasingly recognised \[[@CR121]\]. Yet, athletes often sleep less (e.g. \~6.5--6.7 h·night^−1^) and experience poorer quality sleep than non-athletes \[[@CR79], [@CR109], [@CR152]\]. Factors that contribute to poor sleep among athletes include evening competitions and training sessions, pre-competition anxiety, use of caffeine, and long-haul travel (e.g. jet lag, travel fatigue) \[[@CR121]\]. Several studies have investigated the effect of CBD on sleep in humans \[[@CR26], [@CR32], [@CR115], [@CR157]\]. The first placebo-controlled, double-blinded (single-dose) crossover trial \[[@CR26]\] found that 160 mg CBD (but not 40 or 80 mg) increased self-reported sleep duration in individuals with insomnia (*n* = 15); although time to sleep onset, number of sleep interruptions, and likelihood of experiencing "good sleep" were unchanged. Two case studies also indicated the benefits of CBD \[[@CR32], [@CR157]\]. Specifically, Chagas et al. \[[@CR32]\] observed a reduction in symptoms of rapid eye movement sleep-behaviour disorder in four individuals with Parkinson's disease (75--300 mg·day^−1^; 42 day) and Shannon et al. \[[@CR157]\] found that CBD (\~25 mg·day^−1^) improved subjective sleep quality in a young girl with post-traumatic stress disorder (PTSD). Of course, these studies \[[@CR26], [@CR32], [@CR157]\] are limited in that they rely on subjective measures of sleep and involve small sample sizes; the improvements observed could also be due to CBD attenuating other sleep-impairing comorbid conditions (e.g. anxiety, PTSD). Indeed, a recent placebo-controlled, double-blinded (single-dose) crossover trial \[[@CR115]\] found no effect of CBD (300 mg) on sleep architecture measured via polysomnography in healthy adults (*n* = 27). While CBD seems unlikely to directly influence sleep in healthy humans \[[@CR115]\] (and may be "sleep-promoting" in those with certain comorbid conditions) \[[@CR26], [@CR32], [@CR157]\], a small number of rodent studies suggest that the cannabinoid could actually be "wake-inducing" \[[@CR128], [@CR132], [@CR204]\]. One placebo-controlled, double-blinded (single-dose) crossover trial of healthy individuals (*n* = 8) \[[@CR137]\] also found that low-dose CBD (15 mg) counteracted some of the sedative effects of co-administered Δ^9^-THC (15 mg), i.e. increasing overnight wakefulness; although, this effect could be due to CBD acting as a NAM of CB~1~R, thereby attenuating Δ^9^-THCs effects on that receptor \[[@CR92]\]. Differences in the doses of CBD administered might partly explain this inconsistency. Indeed, Monti \[[@CR128]\] observed a biphasic effect of CBD on sleep in Wistar rats, such that a lower dose decreased, and a higher dose increased (20 vs. 40 mg·kg^−1^, i.p.), slow-wave sleep latency. However, inter-species differences are also a consideration as nocturnal animals appear to exhibit different circadian patterns of endocannabinoid signaling compared to humans \[[@CR84], [@CR131]\]. Collectively, the current evidence on CBD and sleep endorses the need for further research in clinical populations and athletes. Cognitive and Psychomotor Function {#Sec20} ---------------------------------- A small number of clinical trials have investigated the effects of CBD on cognitive and psychomotor function in healthy individuals (Table [4](#Tab4){ref-type="table"}). Overall, results suggest a minimal influence of CBD on cognitive or psychomotor function. While one early investigation \[[@CR98]\] reported that CBD (15 or 60 mg) caused participants to under-/over-estimate the duration of a 60-s interval on several occasions throughout a repetitive testing protocol (i.e. as is indicative of impaired concentration), the magnitude of error was small (e.g. \<5 s; particularly, in comparison to that observed with Δ^9^-THC, e.g. \~10--25 s \[[@CR98]\]) and the effect was inconsistent, suggesting it may be an artefact of the between-subject design and small sample size. A more recent investigation \[[@CR86]\] observed an improvement in emotion recognition with CBD (16 mg vaporised); however, this "ability" may have limited relevance to the sporting context. More applicable, perhaps, is Dalton et al. \[[@CR49]\] finding no effect of CBD (150 μg·kg^−1^ vaporised) on balance or coordination (a product of both cognitive and motor function). A more recent investigation likewise found no effect of oral or vaporised CBD (100 mg) on cognitive performance between 30 min and 8 h post-treatment \[[@CR160]\] (Table [4](#Tab4){ref-type="table"}). Thus, while only a narrow range of doses and relatively few discrete cognitive functions have been studied, the available data suggest that CBD is unlikely to impact cognitive or psychomotor function on healthy individuals. Table 4Clinical trials investigating the effect of CBD on cognitive and psychomotor function in healthy individualsCitationStudy designParticipantsTreatment(s)Cognitive tasksTreatment effectKarniol et al. (1974) \[[@CR98]\]DB (PC); BSDI: 5 M; 21--34 yI: 5 M; 21--34 yI: 5 M; 21--34 yC: 5 M; 21--34 yI: Oral 15 mgI: Oral 30 mgI: Oral 60 mgTime estimation task^a^15 mg CBD sig. altered the ETI (without FB) at 45 and 95 min. 60 mg CBD sig. altered the ETI (without FB) at 95 and 180 min. No other sig. differences were observed.Dalton et al. (1975) \[[@CR49]\]DB; WSD15 M; 21--24 yVaporised 150 μg·kg^-1^Wobble boardPursuit meterDelayed auditory feedback"Pegs"Standing steadiness, hand-eye coordination, cognition and manual coordination were unchanged from baseline 5--85 min post-txLeweke et al. (2000) \[[@CR111]\]SB; WSD9 M; 26--35 yOral 200 mgBinocular depth inversionDepth inversion scores were unchanged from baseline 1, 2, 3, 4, 5, 6 and 24 h post-tx.Bogwardt et al. (2008) \[[@CR22]\]DB (PC); WSD15 M; 27 ± 6 yOral 600 mgGo/No-Go TaskReaction time and accuracy were unaffected between \~1 and 2 h post-tx.Bhattacharyya et al. (2009) \[[@CR17]\]DB (PC); WSD15 M; 27 ± 6 yOral 600 mgVerbal memory taskWord recall was unaffected \~1--2 h post-tx.Palo Fusar-Poli et al. (2009) \[[@CR67]\]DB (PC); WSD15 M; 27 ± 6 yOral 600 mgFacial emotion recognitionReaction time and accuracy were unaffected \~1--2 h post-tx.Hindocha et al. (2015) \[[@CR86]\]DB (PC); WSD48 (34 M); \~22 yVaporised 16 mgFacial emotion recognitionCBD sig. improved emotion recognition at 60% intensity 10 min post-tx. Recognition at 20, 40, 80 and 100% intensity were unaffected.Hundal et al. (2017) \[[@CR90]\]DB (PC); BSDI: 16 (8 M); 26 ± 9 yC: 18 (8 M); 25 ± 8 yOral 600 mgSymbol codingDigit span (forward and reverse)Verbal learning taskDigit-symbol recoding, forward digit span, reverse digit span, immediate recall and delayed recall were unaffected 135 min post-tx.Arndt & de Wit (2017) \[[@CR8]\]DB (PC); WSD38 (19 M); 24 ± 4 yOral 300, 600and 900 mgEmotional stroopFacial emotion recognitionAttentional bias taskReaction time, emotion recognition and attentional bias were unaffected 3--4 h post-tx.Birnbaum et al. (In Press) \[[@CR19]\]No blind; WSD8 (6 M); 49 yOral 300 mgPhonemic and semantic fluencyDigit spanTrail making test (A & B)Symbol--digit modality taskTest scores were unchanged from baseline 2.5 h post-tx (test outcomes not specified)Spindle et al. (In Press) \[[@CR160]\]DB (PC); WSD18 (9 M); 31 ± 6 yOral 100 mg and Vaporised 100 mgDigit symbol substitutionDivided attention taskPaced serial addition taskSpeed and accuracy of digit symbol substitution and paced serial addition, as well as speed, accuracy and tracking on the divided attention task were unaffected between 0.5 and 8 h post-tx.*BSD* between-subject design, *C* control group, *CBD* cannabidiol, *DB* double-blind, *ETI* estimated time interval, *FB* feedback, *I* intervention group, *M* male subjects, *PC* placebo-controlled, *SB* single-blind, *Tx* treatment, *WSD* within-subject designThe "Treatment effects" described are in comparison to a placebo condition, unless otherwise stated^a^Time estimation task: participants were asked to produce a 60 s interval 10 times. The first 5 estimations were without feedback and the remaining 5 were with feedback. This task was repeated 45, 95. and 180 min post-treatment Other Considerations {#Sec21} ==================== CBD "Nutraceutical" Products {#Sec22} ---------------------------- Over-the-counter CBD-containing "nutraceuticals" are now readily available in many countries and of increasing interest to the community \[[@CR108]\]. However, it is important to be aware that these products are often limited in that they typically contain low levels of CBD (e.g. \~10-50 mg·mL^−1^), \[[@CR20]\] making it difficult to achieve the higher doses often used in the studies described in this review. Additionally, over-the-counter CBD nutraceuticals are not always manufactured to the same pharmaceutical standards as regulated, prescription CBD products (e.g. Epidiolex®). Indeed, a recent study \[[@CR20]\] reported that only \~31% of CBD extracts sold online (*n* = 84) were "labelled accurately" (i.e. a measured CBD content within 10% of the labelled value); nearly half underestimated, and a quarter overestimated, their CBD content---approximately one-fifth of samples also contained detectable levels of Δ^9^-THC. A similar investigation \[[@CR148]\] of nine CBD "e-liquids" (used for vaporising) found that two contained Δ^9^-THC, four contained 5-fluoro MDMB-PINACA (a potent synthetic cannabinoid receptor agonist with powerful psychoactive effects), and one contained dextromethorphan. Thus, individuals using these products are at risk of over-/under-dosing, adverse health effects and/or possibly, recording a positive drug test result. This suggests that until such time as better manufacturing standards are imposed, athletes in competition might wish to avoid using non-regulated CBD-containing nutraceuticals, or, at least carefully investigating their quality control and provenance before using them. Conversion of CBD to Δ^9^-THC {#Sec23} ----------------------------- In vitro *studies* have shown that CBD can undergo conversion to Δ^9^-THC with prolonged exposure to simulated gastric fluid \[[@CR126], [@CR189]\]. However, this effect has not been observed in vivo. When investigating the phenomenon in minipigs (30 mg CBD·kg^−1^·day^−1^, oral 5 days), which, like humans, have omnivorous diets and other GI similarities (e.g. pH, transit time, drug absorption), Wray et al. \[[@CR202]\] found no detectable (i.e. \<0.5 ng·mL^−1^) Δ^9^-THC or 11-OH-THC in any plasma or GI tract samples collected. Consroe et al. \[[@CR37]\] also found no detectable Δ^9^-THC (i.e. \<0.5 ng·mL^−1^) in human plasma following chronic high-dose CBD treatment (700 mg·day^−1^; 30 days). These data suggest that pure CBD is unlikely to produce a positive drug test result (i.e. indicated by a urinary THC--COOH level \> 150 ng·mL^−1^ \[[@CR200]\]) or any intoxicating (ergolytic) effects due to conversion to Δ^9^-THC. To date, however, no studies have directly investigated whether CBD can elicit a positive drug test result in athletes. Conclusions {#Sec24} =========== CBD has been reported to exert a number of physiological, biochemical, and psychological effects, that have the potential to benefit athletes. For instance, there is preliminary supportive evidence for anti-inflammatory, neuroprotective, analgesic, and anxiolytic actions of CBD and the possibility it may protect against GI damage associated with inflammation and promote the healing of traumatic skeletal injuries. However, it is important to recognise that these findings are very preliminary, at times inconsistent, and largely derived from preclinical studies. Such studies are limited in their generalisability to athletes (and humans in general), and often administer high doses of CBD that may be difficult to replicate in humans. The central observation is that studies directly investigating CBD and sports performance are lacking, and until these are conducted, we can only speculate in regard to its effects. Nonetheless, this review suggests that rigorous, controlled investigations clarifying the utility of CBD in the sporting context are clearly warranted. Δ^9^-THC : Δ^9^-tetrahydrocannabinol 5-HT~1A~ : Serotonin 1A receptor AEA : Anandamide CV : Cardiovascular CBD : Cannabidiol CB~1~R : Cannabinoid CB~1~ receptor CB~2~R : Cannabinoid CB~2~ receptor *C*~max~ : Maximum plasma concentrations COX : Cyclooxygenase CYP : Cytochrome P EIMD : Exercise-induced muscle damage HI : Hypoxic ischemia FAAH : Fatty acid amide hydrolase GI : Gastrointestinal GPR : G protein-coupled receptor IL : Interleukin i.p. : Intraperitoneal i.v. : Intravenous NAM : Negative allosteric modulator NSAID : Non-steroidal anti-inflammatory drug p.o. : Oral SPA : Sports performance anxiety SAD : Social anxiety disorder TBI : Traumatic brain injury T~C~ : Core body temperature *t*~max~ : Time to reach maximum plasma concentrations TRVP~1~ : Transient potential vanilloid receptor type 1 channel TNF : Tumour necrosis factor USA : United States of America E.g. 10--30 mg·kg^−1^·day^−1^, i.p. 3 days \[[@CR129], [@CR153]\], 1 mg·kg^−1^·day^−1^, i.v. 3 days \[[@CR13]\], 1--3 mg·kg^−1^, i.p. pre- and 3 h post-HI \[[@CR81], [@CR83]\] or 0.1 mg·kg^−1^, i.v. 15 and 240 min post-HI \[[@CR4]\]. **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Not applicable. All authors were involved in the conception and design of this review. DM was responsible for collating the relevant manuscripts and data. All authors contributed to the drafting and revising of the article, and the final approval of the published version of the manuscript. Authors' information {#FPar1} ==================== Not applicable. Danielle McCartney, Melissa J Benson, Anastasia S Suraev, and Iain S McGregor receive salary support from the Lambert Initiative for Cannabinoid Therapeutics, a philanthropically funded centre for medicinal cannabis research at the University of Sydney. No other sources of funding were used to assist in the preparation of this article. Not applicable. Not applicable. Not applicable. Danielle McCartney, Melissa J Benson, Ben Desbrow, Christopher Irwin, and Anastasia S Suraev have no potential conflicts of interest with the content of this article. Iain S McGregor currently acts as a consultant to Kinoxis Therapeutics and is named as an inventor on several patents relating to novel cannabinoid therapeutics, none of which relate to sports physiology or medicine.

###### Strengths and limitations of this study - This study is the first to evaluate the risk of bias in the randomised controlled trials referenced in the guidelines for cardiopulmonary resuscitation. - A detailed protocol for risk of bias assessments was used to ensure reproducibility and transparency in the evaluation. - Various subgroup analyses were performed after stratification according to topics, impact factor, and the years of introduction and update of the Consolidated Standards of Reporting Trial statement. - The risk of bias assessed using the Cochrane Collaboration's tool can be subjective. - We did not contact authors to resolve the unclear information when judging the risk of bias. Introduction {#s1} ============ Randomised controlled trials (RCTs) provide the most reliable evidence for the impacts of medical interventions.[@R1] However, RCTs can be biased by faults in the design, performance, analyses and reporting. Bias is a systemic error that underestimates or overestimates the true effects of an intervention.[@R2] Bias can invalidate the results of RCTs, potentially leading to patients receiving non-beneficial or harmful treatments.[@R3] The Cochrane Collaboration's tool for assessing the risk of bias in randomised trials was developed to clarify the trial process and increase accuracy.[@R2] The Cochrane Collaboration's tool has been used to assess the risk of bias in RCTs conducted in various fields, including paediatrics, orthopaedics, urology, neurology and ophthalmology.[@R3] Recently, the risk of bias in 20 920 RCTs included in the Cochrane Review was reported.[@R8] In the emergency medicine area, the Cochrane Collaboration's risk of bias tool was used to evaluate RCTs that assessed simulation-based medical education.[@R9] However, to our knowledge, no study has evaluated the risk of bias in the RCTs referenced in the guidelines for cardiopulmonary resuscitation (CPR). Clinical guidelines are an important tool for knowledge transfer and affect millions of clinicians and patients.[@R10] The main advantage of guidelines is that they reduce unjustified variations in patient care, but biased guidelines are potentially harmful and ineffective for the patient.[@R11] Although the Cochrane Collaboration's tool for assessing the risk of bias was used to conduct an evidence evaluation of RCTs in the 2015 American Heart Association (AHA) guidelines for CPR and emergency cardiovascular care (ECC),[@R12] the assessment of individual items of bias or overall features was not presented. The purpose of this study was to identify the risk of bias in the RCTs referenced in the 2015 AHA guidelines update for CPR and ECC,[@R12] which is used worldwide and considered as the basis for resuscitation. Methods {#s2} ======= We identified the RCTs cited as references in the 2015 AHA guidelines update for CPR and ECC. Articles containing the search term 'random' in the title or abstract were extracted. After reviewing the contents of the abstract and text, the studies that were included in the actual randomisation process were confirmed. We included clinical studies in which participants were human patients and excluded animal studies. Studies that analysed existing RCTs and RCTs published in a letter format were also excluded. We analysed the RCTs using the Cochrane Collaboration's tool for assessing the risk of bias in randomised trials.[@R1] The following six domains of the Cochrane Collaboration's tool were selected to evaluate the risk of bias: random sequence generation and allocation concealment for selection bias, blinding of participants and personnel for performance bias, blinding of outcome assessment for detection bias, incomplete outcome data for attrition bias and selective reporting for reporting bias. We did not prespecify other sources of bias that are not identified by the above six domains described because we were not able to simply define the other sources of bias in RCTs of various topics included in the guidelines. The risk of bias for each domain was reported as 'low', 'unclear', or 'high'. The criteria for assessing the risk of bias are shown in online [supplementary appendix 1](#SP1){ref-type="supplementary-material"}. As shown in the study by Zhai *et al*, real-time randomisation was added to the domain of allocation concealment.[@R5] It was considered low risk when the treatment was assigned at the time of randomisation and allocation concealment was guaranteed. If participants do not know the study objectives, the domain of blinding of participants and personnel was judged as a low risk of bias. In addition, studies with objective/permanent end-points, such as mortality or laboratory data, were evaluated as having a low risk of bias for the blinding of outcome assessment. Two independent reviewers (YC and CK) scored the RCT articles in each domain, and a third reviewer (BK) resolved the discrepancies. Kappa values for the inter-rater agreement between the two reviewers were calculated for each of the six domains. 10.1136/bmjopen-2018-023725.supp1 We identified the number of RCTs in 5-year intervals. The percentages of RCTs with high, unclear and low risks of bias were determined for each of the six Cochrane domains. We examined the risk of bias in RCTs by grouping them into seven topics (basic life support \[BLS\], adult advanced cardiovascular life support \[ACLS\], postcardiac arrest care, acute coronary syndrome, neonatal and paediatric resuscitation, education, and others). We also investigated the risk of bias in RCTs based on the type of intervention (drug trial or non-drug trial). The journal's impact factor (IF) in 2017 and the Journal Citation Report (JCR) categories were identified. We designated journals without an IF as having an IF of zero. The risk of bias in the six domains was identified in the following journal IF groups: IF\<5 (low IF), 5≤IF\<10 (intermediate IF) and IF≥10 (high IF). The risk of bias for each Cochrane domain was evaluated in RCTs divided into three periods (≤1995, 1996--2009 and ≥2010) based on the year of introduction and the update of the Consolidated Standards of Reporting Trials (CONSORT) statement. We used R V.3.4.0 ([www.R-project.org](www.R-project.org)) to plot the numbers and proportions of RCTs over 5-year intervals. Microsoft Excel was used for data collection and creating the graphs showing the risk of bias. Patient and public involvement {#s2a} ------------------------------ No patients were involved in setting the research question or the outcome measures, nor were they involved in developing plans for the design or implementation of the study. No patients were asked to advise on data interpretation or writing the manuscript describing the results. The results of the research are not planned to be disseminated to study participants or the relevant patient community. Results {#s3} ======= Three hundred RCTs referenced in the 2015 AHA guidelines were identified. After the exclusion of 27 articles, 273 RCTs were selected for analyses. Studies were excluded because they were animal studies (n=23), studies that analysed existing RCTs (n=3), and RCTs published in a letter format (n=1) (online [supplementary appendices 2, 3](#SP1){ref-type="supplementary-material"}). The RCT articles included in this study were published from 1980 to 2015. The number of RCTs has increased during this period ([figure 1](#F1){ref-type="fig"}), and 90.5% (247/273) of RCTs were published after 1996. A total of 42.1% (115/273) of the RCTs included in this study were published after 2010. {#F1} [Table 1](#T1){ref-type="table"} shows the baseline characteristics of the RCTs. The median number (IQR) of participants randomised and the number of participants analysed were 140 (59--372) and 130 (54--335), respectively. Education (33.0%, 90/273) was the most common topic cited in the guidelines, followed by neonatal and paediatric resuscitation (18.3%, 50/273) and acute coronary syndrome (14.7%, 40/273). The most common JCR category was critical care medicine (27.5%, 75/273), followed by medicine, general and internal (22.0%, 60/273). ###### Characteristics of the randomised controlled trials referenced in the 2015 American Heart Association guidelines update for cardiopulmonary resuscitation and emergency cardiovascular care (n=273) Characteristics Values ------------------------------------------------- --------------- Number of participants randomised, median (IQR) 140 (59--372) Number of participants analysed, median (IQR) 130 (54--335) *Topics, n (%)*  BLS 19 (7.0)  ACLS 32 (11.7)  Postcardiac arrest care 21 (7.7)  Acute coronary syndrome 40 (14.7)  Neonatal and paediatric resuscitation 50 (18.3)  Education 90 (33.0)  Others 21 (7.7) *Type of intervention, n (%)*  Drug trial 37 (13.6)  Non-drug trial 236 (86.4) *JCR category, n (%)*  Critical care medicine 75 (27.5)  Medicine, general and internal 60 (22.0)  Paediatrics 41 (15.0)  Cardiac and cardiovascular systems 32 (11.7)  Emergency medicine 21 (7.7)  Anaesthesiology 11 (4.0)  Education 8 (2.9)  Nursing 4 (1.5)  Clinical neurology 3 (1.1)  Infectious diseases 2 (0.7)  Obstetrics and gynaecology 2 (0.7)  Peripheral vascular disease 2 (0.7)  Pharmacology and pharmacy 2 (0.7)  Medicine, research and experimental 1 (0.4)  Substance abuse/psychiatry 1 (0.4)  Surgery 1 (0.4)  Not listed in JCR 7 (2.6) ACLS, adult advanced cardiovascular life support; BLS, basic life support; JCR, Journal Citation Reports. The median IF in 2017 was 5.9 (IQR: 3.7--16.8) ([table 2](#T2){ref-type="table"}). Seven RCTs were published in journals without an IF. Approximately one-third of the RCTs (95/273, 34.8%) included in the present study were published in journals with 5≤IF\<10, and 28.6% (78/273) were published in journals with an IF≥10. *The New England Journal of Medicine* had the highest IF and was the most cited journal among the RCTs included in this study. Among the included topics, the highest median IF was observed for trials investigating acute coronary syndrome (18.9, IQR: 5.2--53.3) and the lowest was observed for studies investigating neonatal resuscitation (3.7, IQR: 2.3--5.5). Based on the type of intervention, the median IF of drug trials (18.9, IQR: 4.7--79.3) was much higher than that of non-drug trials (5.9, IQR: 3.6--10.1). ###### Impact factor of the randomised controlled trials referenced in the 2015 American Heart Association guidelines update for cardiopulmonary resuscitation and emergency cardiovascular care (n=273) Characteristics Values ------------------------------------------------------ ------------------- Journal IF (2017), median (IQR) 5.9 (3.7--16.8) Journal IF (year of publication), median (IQR) 3.6 (1.6--6.4) *Journal IF (2017), n (%)*  IF\<5 100 (36.6)  5≤IF\<10 95 (34.8)  IF≥10 78 (28.6) *Journal IF (year of publication), n (%)*  IF\<5 181 (66.3)  5≤IF\<10 33 (12.1)  IF≥10 59 (21.6) *Top 20 high-IF journals, n (%), IF (2017)* 158 (57.9)  *New England Journal of Medicine* 32 (11.7), 79.258  *The Lancet* 11 (4.0), 53.254  *Journal of the American Medical Association, JAMA* 8 (2.9), 47.661  *European Heart Journal* 4 (1.5), 23.425  *British Medical Journal, BMJ* 2 (0.7), 23.295  *Archives of Internal Medicine* 1 (0.4), 19.989  *Circulation* 9 (3.3), 18.880  *Journal of the American College of Cardiology* 5 (1.8), 16.834  *Intensive Care Medicine* 4 (1.5), 15.008  *JAMA Pediatrics* 2 (0.7), 10.769  *JACC: Cardiovascular Intervention* 1 (0.4), 9.881  *Clinical Infectious Disease* 2 (0.7), 9.117  *Chest* 1 (0.4), 7.652  *Neurology* 1 (0.4), 7.609  *Critical Care Medicine* 12 (4.4), 6.630  *Anesthesiology* 2 (0.7), 6.523  *Circulation: Cardiovascular Intervention* 1 (0.4), 6.504  *Stroke* 1 (0.4), 6.239  *Addiction* 1 (0.4), 5.953  *Resuscitation* 58 (21.2), 5.863 *Topics, IF, median (IQR)*  BLS 15.0 (5.9--79.3)  ACLS 5.9 (4.7--59.8)  Postcardiac arrest care 15.0 (6.6--47.7)  Acute coronary syndrome 18.9 (5.2--53.3)  Neonatal and paediatric resuscitation 3.7 (2.3--5.5)  Education 5.9 (2.6--5.9)  Others 5.9 (5.4--18.9) *Type of intervention, IF, median (IQR)*  Drug trial 18.9 (4.7--79.3)  Non-drug trial 5.9 (3.6--10.1) ACLS, adult advanced cardiovascular life support; BLS, basic life support; IF, impact factor; JCR, Journal Citation Reports. The kappa values for inter-rater agreement of the individual domains of the Cochrane Collaboration's tool were 0.86 for random sequence generation, 0.89 for allocation concealment, 0.89 for the blinding of the participants and personnel, 0.84 for the blinding of outcome assessment, 0.76 for incomplete outcome data and 0.71 for selective reporting. The proportion of trials with each risk of bias rating for the six domains is shown in [figure 2](#F2){ref-type="fig"}. A large proportion of trials displayed a high risk of bias (inadequate method) for the domain of blinding of participants and personnel (78.8%, 215/273). Meanwhile, the proportions of trials with inadequate methods (high risk of bias) for the remaining five domains (random sequence generation, allocation concealment, blinding of outcome assessment, incomplete outcome data and selective reporting) were low (≤12.1%). However, of these five domains, the proportions of trials with an unclear risk of bias (poor reporting) for two domains (random sequence generation and allocation concealment) were greater than 30% (38.5%, 105/273% and 34.1%, 93/273, respectively) ([figure 2](#F2){ref-type="fig"}). In the subgroup analysis based on the topics of the trials, ACLS trials had the highest proportion of trials with a low risk of bias for the blinding of participants and personnel (71.9%, 23/32) ([figure 3](#F3){ref-type="fig"}). Larger proportions of trials examining other topics (BLS, postcardiac arrest care, acute coronary syndrome, neonatal and paediatric resuscitation, education, and others) had a high risk of bias for the blinding of participants and personnel (94.7%, 85.7%, 85.0%, 86.0%, 88.9% and 66.7%, respectively) than ACLS trials (25%). ACLS trials represented the highest proportions of trials with an unclear risk of bias for random sequence generation (50%, 16/32) and incomplete outcome data (31.3%, 10/32). Education trials represented the highest proportion of trials with an unclear risk of bias for allocation concealment (58.9%, 53/90). {#F2} {#F3} Regarding the type of intervention, the proportion of drug trials with a high risk of bias for the blinding of participants and personnel (21.5%, 8/37) was less than that of non-drug trials (87.7%, 207/236) ([figure 4](#F4){ref-type="fig"}). Seventy-five per cent of ACSL trials (24/32) were drug trials, among which 91.7% of drug trials on the ACLS topic (22/24) displayed a low risk of bias for the blinding of participants and personnel. {#F4} After stratification according to the three groups of IFs in 2017 \[low group (IF\<5), intermediate group (5--10) and high group (≥10)\], the proportion of trials with a low risk of bias showed an increasing trend as the IF increased, although little difference was observed between the two higher IF groups (intermediate vs high IF group) in the domains of random sequence generation and selective reporting ([figure 5](#F5){ref-type="fig"}). However, the proportions of trials with an unclear risk of bias were still high for the domain of random sequence generation and allocation concealment, even in journals with high IFs (38.5% and 23.1%, respectively). Even after stratification according to the IFs of the year of publication, the proportions of trials with an unclear risk of bias in journals with high IFs were high for the domain of random sequence generation (35.6%) and allocation concealment (22.0%) (online [supplementary appendix 4](#SP1){ref-type="supplementary-material"}). {#F5} After stratification by the year of introduction (1996) and update of the CONSORT statement (2010), the proportion of trials with a low risk of bias also tended to increase after the introduction of the statement in 1996 and after its update in 2010 ([figure 6](#F6){ref-type="fig"}) for four domains (random sequence generation, blinding of outcome assessment, incomplete outcome data and selective reporting). Unclear risk of bias proportions was 65.4% for random sequence generation and 57.7% for allocation concealment before the introduction of CONSORT but decreased to 31.3% and 32.2% after the 2010 CONSORT update, respectively. For allocation concealment, the proportion of trials with a low risk of bias also increased after the introduction of the CONSORT statement in 1996 (30.8%--57.6%); however, the proportion did not increase after the update of the CONSORT statement in 2010 (57.6%--54.8%). For blinding of participants and personnel, the proportion of trials with a low risk of bias showed an opposite trend after the introduction of the CONSORT statement in 1996 and after its update in 2010. Drug-trials were 30.8% (8/26) prior to or in 1995, 14.4% (19/132) from 1996 to 2009, and 8.7% (10/115) in in or after 2010. {#F6} Discussion {#s4} ========== In this study, we assessed the risk of bias in 273 RCTs that were referenced in the 2015 update of the AHA guidelines for CPR and ECC. The largest proportion of trials with a high risk of bias (inadequate methods) was observed for the blinding of participants and personnel (78.8%). This finding might be explained by the fact that only 13.6% (37/273) of the RCTs in this guideline were drug trials that can be sufficiently blinded by placebos. The remaining 86.4% of the RCTs were non-drug trials that did not use placebos. In general, the masking of participants and personnel is difficult in non-drug trials where placebos are not available. In fact, the proportion of drug trials with a high risk of bias for the blinding of participants was less (21.5%) than that in non-drug trials (87.7%) in this study. Furthermore, in some cases, blinding is not performed because of an ethical issue, depending on the outcome of cardiac arrest. In domains other than the blinding of participants and personnel, the proportion of trials with a high risk of bias was less than 12.1%. However, greater than one-third of trials exhibited unclear risks of bias for random sequence generation (38.5%) and allocation concealment (34.1%), indicating that many trials did not sufficiently report the risk of bias. Random sequence generation minimises selection bias and balances baseline characteristics between the study and control groups. Methods such as a computerised random number generator, random number table, or coin tossing should be used to ensure a low risk of bias. Sequences generated by an odd or even date of birth or a non-random process using hospital registration numbers will lead to bias. In this study, 38.5% of RCTs showed unclear risk of bias in random sequence generation, and half of the ACLS trials poorly described random sequence generation, possibly because patients experiencing cardiac arrest were not able to be randomly divided using the above methods to minimise the risk of bias for random sequence generation due to ethical issues. Random sequence generation alone is not sufficient to prevent selection bias, and allocation concealment must be employed to effectively protect unpredictable randomised sequences.[@R2] Central allocation, sequential numbered opaque sealed envelopes (SNOSE) or minimisation should be used to ensure participants and researchers are not aware of assignments in advance. According to Akl *et al*,[@R26] SNOSE leads to a risk of bias when the patient or participants opens the envelope in advance. The authors recommended the use of simple randomization when the sample size is large. Block randomisation is often used if stratification is required; however, a small block size should be avoided because the allocation can be predicted in advance. Allocation by minimisation is much more difficult to predict than block randomisation because minimisation is a dynamic mode of random allocation that depends on the characteristics of the participants who are already enrolled.[@R26] In this study, 12.1% of RCTs had a high risk of bias for allocation concealment. A total of 34.1% of trials showed an unclear risk of bias for allocation concealment, which is less than the percentages reported by a recent study analysing the 20 920 RCTs included in the Cochrane reviews (57.5%)[@R8] or RCTs of simulation-based interventions in emergency medicine (48.5%).[@R9] However, in a paper analysing the RCTs published in four general medical journals (*The BMJ*, *JAMA*, *The Lancet* and *The NEJM*), 26% of trials published in 2002[@R27] and 22% of trials published in 2015[@R26] poorly described information regarding allocation concealment. This finding is similar to our study, as the proportion of RCTs with an unclear risk of bias was 23.1% in the group with an IF≥10. A potential explanation is that journals with high IFs have been included in the 2015 AHA guidelines (28.6% of trials included in this study had an IF greater than 10). If outcome assessors know the assignment of the intervention, detection bias will occur in this assessment.[@R2] The lack of blinding of outcome assessors can lead to different outcomes, and clinicians must decide who will assess the outcome of RCTs at the planning stage. In this study, the proportions of trials with unclear and high risks of bias for the blinding of outcome assessment were relatively low (8.4% and 2.2%, respectively) compared with other domains because we assessed that domain as a low risk when the study end-point was objective or permanent, such as mortality or laboratory findings, which were unlikely to affect the bias even if the outcome assessors were not blinded. Online [supplementary appendix 5](#SP1){ref-type="supplementary-material"} shows the end-point of the trials included in this study. Many of the trials were objective (eg, the presence or absence of return of spontaneous circulation, survival or death, admission or discharge and the results of laboratory findings) and, thus, are unlikely to affect the outcome assessment. We determined that 22.3% of the RCTs showed an unclear risk of bias for incomplete outcome data, which suggests insufficient reporting on attrition and exclusion. Of the RCTs published in the five top general medical journals in 2005--2007, 13% did not discuss loss to follow-up.[@R26] Additionally, 24.7% of the RCTs included in the Cochrane Review in 2011--2014 showed an unclear risk of bias in the domain of incomplete outcome data.[@R8] The number of participants in each intervention group and reasons for attrition and exclusion should be reported to assess this domain. Studies in which participants are excluded by a per-protocol analysis are considered high risk, while an intention-to-treat analysis is often recommended as the least biased method.[@R2] Dechartres *et al* [@R8] reported that the proportions of RCTs with unclear risks of bias in sequence generation, allocation concealment, blinding and incomplete outcome data have decreased from 1986 to 2014. The proportions of RCTs referenced in the 2015 AHA guidelines with an unclear risk of bias also showed decreasing trends after the introduction of the CONSORT statement in 1996 and after its update in 2010 for four domains (random sequence generation, blinding of outcome assessment, incomplete outcome data and selective reporting). However, for random sequence generation and allocation concealment, a larger proportion of RCTs showed an unclear risk of bias than the other four domains in both periods: 1996--2009 and ≥2010. A reporting guideline, such as the CONSORT statement, should be applied to improve reporting.[@R27] When the CONSORT guideline is actively applied, the quality of reporting improves.[@R28] However, only 38% of biomedical journals (168) with a high IF used the CONSORT guideline in 2014,[@R31] and only 14% of surgical journals mentioned using a reporting CONSORT guideline in 2017.[@R32] Journals should suggest to authors that the items of the CONSORT guideline should be reported at the time of submission, and authors should consider the guideline beginning at the planning stage of the study. The proportions of drug trials were decreased after the introduction of the CONSORT statement in 1996 and after its update in 2010, which might be the reason why the proportion of trials with a low risk of bias for the blinding of participants and personnel showed a declining trend. Inadequate or unclear sequence generation, allocation concealment and blinding affect treatment outcomes.[@R33] Clinicians should consider that these biases exaggerate the treatment effect when interpreting the results of RCTs, particularly trials with subjective outcomes, and when applying these data to the management of patients. Additionally, the effect of bias on RCTs should be considered when designing a systematic review and creating a guideline. This study has limitations. The risk of bias assessed using the Cochrane Collaboration tool can be subjective. However, two reviewers evaluated the included studies, and a third party resolved the differences. We also established a detailed protocol (online [supplementary appendix 1](#SP1){ref-type="supplementary-material"}) for evaluation and tried to reduce discrepancies between the reviewers. We calculated kappa values to evaluate inter-rater agreement and observed almost perfect (0.81--1.00) or substantial (0.61--0.80) agreement. The kappa values for the Cochrane tool used in our study were greater than those observed when this tool was used in physical therapy trials[@R36] and similar to those found in quality reporting of the RCTs in five leading neurology journals.[@R5] Second, if we had contacted the authors, some of the trials scored as an unclear risk of bias could have been clarified and changed. According to Jørgensen *et al*, uncertainty can be resolved by contacting trial authors or trial registers.[@R36] When evaluating selective reporting in this study, protocols were searched if the trial was registered, but we did not contact the authors. Third, although the reviewers were blinded to the journal's name during classification to minimise the possibility of classification bias due to IF, the reviewers were able to recognise the journals despite our efforts; thus, the reviewers might have been influenced by the journal's IF. Fourth, we evaluated the journal's IF in 2017, but this value may not accurately reflect the impact of the journal at the time the article was published. Nevertheless, we relied on the most recent IF (2017) because the journal's IF has consistently increased during the study period. Therefore, a direct comparison of the IF values from different years is impossible. Journals with the same IF values in 1980 and 2017 have different impacts (an IF of 1.8 in 1980 is quite different from the same value in 2017). Furthermore, a low IF for a journal does not indicate that the journal is of low quality. IFs may not properly reflect the quality of each journal. For example, many journals are located in the top level in specialist categories, although their IFs are less than five. Conclusions {#s5} =========== In the 2015 AHA guidelines for CPR and ECC, the proportions of RCTs with a high risk of bias were high for the blinding of participants and personnel. In ACLS and drug trials, the proportions of trials with a low risk of bias for the blinding of participants and personnel were high. The proportions of RCTs with an unclear risk of bias were high for random sequence generation and allocation concealment. These proportions were reduced after the introduction of the CONSORT statement but remained at 31.3% for random sequence generation and 33.2% for allocation concealment after the 2010 update of the CONSORT statement. The risk of bias should be considered when interpreting and applying the CPR guidelines in clinical settings. Journals should make an effort to provide authors with information for assessing the risk of bias and should request the use of a reporting guideline, such as the CONSORT guideline, at the submission stage to minimise the risk of bias. In addition, the authors also should plan their research to reduce the risk of bias and clearly report all domains of the Cochrane Collaboration's tool. Supplementary Material ====================== ###### Reviewer comments ###### Author\'s manuscript YC and CK contributed equally. **Contributors:** YC acquired the data, performed the risk of bias assessment and analysis, and drafted the manuscript. CK performed the risk of bias assessment, interpreted the data and critically revised the manuscript. BK contributed to the study design, interpreted the data and critically revised the paper. All authors read and approved the final manuscript. **Funding:** This work was supported by the National Research Foundation of Korea (NRF) grant funded by the Korea government (Ministry of Science, ICT and Future Planning) (No. 2017R1C1B5017218). **Competing interests:** None declared. **Provenance and peer review:** Not commissioned; externally peer reviewed. **Data sharing statement:** There are no additional unpublished data from the study. **Author note:** An abstract of this work was presented at the 2018 American College of Emergency Physicians Scientific Assembly, October 1--4, 2018, San Diego, CA. **Patient consent for publication:** Not required.

The authors confirm that all data underlying the findings are fully available without restriction. All relevant data are within the paper. Introduction {#s1} ============ The prevalence of heart failure including systolic and diastolic functional impairments continues to increase representing a major health problem [@pone.0109164-Murray1], [@pone.0109164-Owan1]. Several diagnostic techniques have been proposed to study systolic and diastolic myocardial performances [@pone.0109164-Karamitsos1], [@pone.0109164-Kowallick1], [@pone.0109164-Pennell1]. Left ventricular (LV) systolic torsion and diastolic recoil describe the myocardial twisting and untwisting motion resulting from apical counter-clockwise and basal clockwise rotation during systole (when viewed from the apex and normalized for LV length). Torsion and recoil rate have shown to be sensitive markers for systolic and diastolic dysfunction [@pone.0109164-Delhaas1], [@pone.0109164-Dong1], [@pone.0109164-Opdahl1]. The phenomenon can be captured by the simple twist angle (difference in apical and basal rotation) [@pone.0109164-Swoboda1], [@pone.0109164-Goffinet1], the circumferential-longitudinal shear angle [@pone.0109164-Russel1], [@pone.0109164-Buchalter1] or torsion (twist per cardiac length) [@pone.0109164-Sorger1], [@pone.0109164-Yoneyama1]. Recent studies showed that torsion might represent an important compensatory mechanism to maintain an adequate systolic function in patients with chronic hypertension as examined in a large community-based population based on cardiovascular magnetic resonance (CMR) myocardial tagging [@pone.0109164-Yoneyama1], [@pone.0109164-Donekal1]. Recoil rate seems particularly appealing to study diastolic function. Its applicability has been demonstrated in isolated diastolic dysfunction [@pone.0109164-Park1], in heart failure with preserved ejection fraction [@pone.0109164-Wang1] and heart disease characterized by both systolic and diastolic dysfunction such as hypertrophic cardiomyopathy (HCM) [@pone.0109164-Abozguia1]. CMR myocardial tagging is considered the reference standard for the evaluation of torsion at the current time. However, this technique has not found widespread implementation into clinical routine since practical obstacles, e.g. the need for additional sequence acquisition and time-consuming post-processing limit its clinical applicability. Recent advances in LV deformation quantification based on CMR feature tracking (CMR-FT) allow straightforward quantification of myocardial physiology from routine steady-state free precession (SSFP) images [@pone.0109164-Hor1], [@pone.0109164-Kowallick2]. The technique has been used to analyse ventricular strain in health and disease [@pone.0109164-Morton1], [@pone.0109164-Schuster1]. However, the feasibility of CMR-FT for quantitative assessment of LV torsion has never previously been demonstrated. We therefore aimed to first develop a model that allows uniform selection of apical and basal slices at standardized LV levels and to evaluate its feasibility and reproducibility to quantify LV systolic torsion and diastolic recoil at rest and during dobutamine stress using CMR-FT. Materials and Methods {#s2} ===================== Ten healthy volunteers underwent CMR on a 1.5 Tesla scanner (Intera R 12.6.1.3, Philips Medical Systems, Best, The Netherlands). The study protocol was approved by the Institutional Review Board at the University of Nebraska Medical Center and complies with the Declaration of Helsinki. Written informed consent was obtained from all participants. CMR Imaging {#s2a} ----------- All CMR measurements were performed in the supine position using a 5-channel cardiac surface coil. LV dimensions and function were assessed with an ECG- gated SSFP cine sequence during brief periods of breath-holding in 12 to 14 equidistant short-axis planes (slice thickness 8 mm; gap 1--2 mm) completely covering the LV. The field of view was 360×480 mm and matrix size 196×172. Dobutamine stress imaging was performed as previously described [@pone.0109164-Nagel1]. Complete short-axis stacks were acquired at rest and with 10 and 20 µg·kg^−1^·min^−1^ of dobutamine. Haemodynamic analysis {#s2b} --------------------- End-diastolic (EDV) and end-systolic volume (ESV), stroke volume (SV), ejection fraction and cardiac index (CI), were calculated using MassK-Mode in QMass Version 7.6 (Medis Medical Systems, Leiden, The Netherland). Ventricular volumes were adjusted to body surface area. All parameters were analysed at rest, 10 and 20 µg·kg^−1^·min^−1^ of dobutamine stress. Furthermore, heart rate and mean blood pressure were measured at rest and with both levels of dobutamine stimulation. Cardiac magnetic resonance myocardial feature tracking {#s2c} ------------------------------------------------------ Myocardial feature tracking was performed using dedicated software (TomTec Imaging Systems, 2D CPA MR, Cardiac Performance Analysis, Version 1.1.2.36, Unterschleissheim, Germany). All short-axis slices were included in the analysis between the most apical slice showing LV cavity at end-systole and the most basal slice including a complete circumference of myocardium at end-systole. LV endocardial and epicardial borders were manually traced in all slices using a point-and-click approach with the initial contour set at end-diastole. Endocardial and epicardial border surfaces were manually delineated and the automated tracking algorithm was applied. The software automatically tracks 48 subendocardial and subepicardial tissue voxels throughout the cardiac cycle. Tracking performance was visually reviewed to ensure accurate tracking of the ventricular myocardium. In case of insufficient border tracking manual adjustments were made to the initial contour and the algorithm was reapplied. Tracking was repeated for three times in each short-axis view and results were based on the average of the three repeated measures. Definition of torsion and recoil rate {#s2d} ------------------------------------- Torsion and peak recoil rate were calculated using MATLAB software (The Math Works, MA, USA). Averaged LV rotation profiles were calculated from the angular displacement of all 48 tissue voxels in all slices. LV rotation was averaged across the three repeated measurements. The rotation of points that would typically be located in between adjacent slices was linearly interpolated resulting in a complete 3D LV model of angular displacement ([**Figure 1**](#pone-0109164-g001){ref-type="fig"}). When viewed from the apex, LV torsion was calculated as the difference in counter-clockwise apical rotation (**φ** ~apex~) and clockwise rotation at the base (**φ** ~base~) divided by the inter-slice distance (D) [@pone.0109164-Sorger1], [@pone.0109164-Yoneyama1]: ![3 D model of LV rotational displacement.\ Rotation between time points (angular difference between red and green contours) is computed in each slice, and it is then linearly interpolated between slices. The 3D ventricular model is automatically fitted to the contours [@pone.0109164-Lamata1] and is used here only for illustration purposes, and not for the definition of the location of the most apical and basal points.](pone.0109164.g001){#pone-0109164-g001} The distance (D) was calculated as the sum of the slice thickness (8 mm) and gap (1--2 mm) of imaging planes that were located in between the basal and apical slices where rotation was measured [@pone.0109164-Yoneyama1] ([**Figure 2**](#pone-0109164-g002){ref-type="fig"} **and** [**3**](#pone-0109164-g003){ref-type="fig"}). Depending on these measurement positions of rotation obtained at different LV locations and the distance between the most apical (defined as 0% of LV distance) and the most basal slice (defined as 100% of LV distance) the following four methods to calculate torsion were examined: Model 1 (difference in rotation between 25% and 75%), model 2 (difference in rotation between 0% and 100%), model 3 (difference in rotation between 25% and 100%) and model 4 (difference in rotation between 0% to 75%) ([**Figure 3**](#pone-0109164-g003){ref-type="fig"}). The analysis included subepicardial and subendocardial torsion as well as LV global torsion (averaged subendocardial and subepicardial torsion). Furthermore, subendocardial, subepicardial and global peak recoil rates were calculated from the maximum slope (−dθ/dt) of the diastolic limb of the torsion-time curve ([**Figure 2**](#pone-0109164-g002){ref-type="fig"}). {#pone-0109164-g002} {#pone-0109164-g003} Statistical Analysis {#s2e} -------------------- Statistical analysis was performed using Microsoft Excel and IBM SPSS Statistics version 22 for Macintosh. Data are expressed as mean (± standard deviation). Differences of torsion and peak recoil rate calculated between models 1--4 were assessed by the Friedman test. Pairwise comparisons between variables at rest and with increasing levels of dobutamine were evaluated using the Wilcoxon test. All statistical tests with p values \<0.05 were considered statistically significant. Two independent observers analysed all cases to assess inter-observer variability (JTK & AS). Intra-observer variability was derived from the repeated analysis by the first observer (JTK) after four weeks. The intra- and inter-observer variability were assessed by intraclass correlation coefficients (ICC) using a model of absolute agreement. Agreement was considered excellent when ICC\>0.74, good when ICC = 0.60--0.74, fair when ICC = 0.40--0.59, and poor when ICC\<0.4 [@pone.0109164-Oppo1]. Furthermore, Bland Altman analysis [@pone.0109164-Bland1] and coefficients of variation (CoV) were calculated. The CoV was defined as the standard deviation of the differences divided by the mean [@pone.0109164-Grothues1]. Results {#s3} ======= The image quality was sufficient to perform CMR-FT in all subjects and for all slices at rest and during dobutamine stress. CMR-FT post-processing took 10 to 12 minutes for a LV short-axis stack on average including adjustments if necessary. Participant demographics are summarised in [**table 1**](#pone-0109164-t001){ref-type="table"}. 10.1371/journal.pone.0109164.t001 ###### Subject characteristics. {#pone-0109164-t001-1} Demographics Healthy volunteers -------------------- -------------------- Study population N = 10 Gender (f/m) 5/5 Age (y) 40.6 (23--51) LV-EDV (ml/m^2^) 51.0 (7.5) LV-ESV (ml/m^2^) 21.7 (5.1) LV-CI (l/min/m^2^) 2.0 (0.4) LV-EF (%) 57.9 (5.6) Continuous variable are expressed as mean (standard deviation), age is expressed as median (range). f: female, m: male, EDV: end-diastolic volume, ESV: end-systolic volume, CI: cardiac index, EF: ejection fraction. Haemodynamic response to dobutamine {#s3a} ----------------------------------- There were no side effects to dobutamine exposure. There was a significant increase of cardiac output, mean blood pressure and ejection fraction between rest and both levels of dobutamine as well as between 10 and 20 µg·kg^−1^·min^−1^ of dobutamine. Stroke volume increased significantly between rest and both levels of dobutamine, whilst no further increase was observed between 10 and 20 µg·kg^−1^·min^−1^ of dobutamine ([**table 2**](#pone-0109164-t002){ref-type="table"}). 10.1371/journal.pone.0109164.t002 ###### Volumetric and hemodynamic response to 10 and 20 µg kg^−1^min^−1^ of dobutamine. {#pone-0109164-t002-2} Parameter Level of dobutamine (µg kg^−1^ min^−1^) P value -------------------- ----------------------------------------- ------------- -------------- ------------ ------------ ------------ LV-CI (l/min/m^2^) 2.0 (0.4) 3.3 (0.8) 4.2 (0.6) **\<0.01** **\<0.01** **\<0.01** LV-EDV (ml/m^2^) 51.0 (7.5) 52.7 (9.1) 43.8 (15.4) 0.33 **\<0.05** **\<0.01** LV-ESV (ml/m^2^) 21.7 (5.1) 14.4 (5.9) 11.4 (4.6) **\<0.01** **\<0.01** **\<0.01** LV-SV (ml/m^2^) 29.4 (4.1) 38.3 (7.4) 37.3 (6.5) **\<0.01** **\<0.01** 0.24 LV-EF (%) 57.9 (5.6) 72.9 (9.5) 77.0 (5.7) **\<0.01** **\<0.01** **\<0.05** Mean BP (mmHg) 91.5 (10.2) 98.6 (10.4) 102.9 (10.7) **\<0.01** **\<0.05** **\<0.05** Heart Rate 69.3 (10.3) 85.2 (16.6) 112.8 (11.9) **\<0.01** **\<0.01** **\<0.01** Bold p values indicate a significance level \<0.05. BP: blood pressure, other abbreviations as in [table 1](#pone-0109164-t001){ref-type="table"}. Torsion and recoil rate at rest {#s3b} ------------------------------- Inter-slice distances used for torsion calculation were 2.9±0.4 cm (model 1), 5.8±0.7 cm (model 2) and 4.4±0.5 cm (model 3 and 4). There was no significant difference of endocardial, epicardial or global torsion and endocardial and global recoil rate between the models 1--4 at rest. Results are displayed in [**table 3**](#pone-0109164-t003){ref-type="table"}. There was a significant difference (p = 0.02) of epicardial recoil rate between model 2 (0--100%) and model 4 (0--75%). 10.1371/journal.pone.0109164.t003 ###### Torsion and recoil rate. {#pone-0109164-t003-3} Model 1 (25--75%) Model 2 (0--100%) Model 3 (25--100%) Model 4 (0--75%) P value --------------------------------- ------------------- ------------------- -------------------- ----------------------------------------------- ----------- **Torsion Endo (°/cm)** 3.0 (1.9) 3.6 (1.6) 3.6 (1.7) 3.4 (2.0) 0.72 **Torsion Epi (°/cm)** 2.3 (1.3) 2.7 (0.9) 2.6 (1.4) 2.6 (0.8) 0.37 **Global Torsion (°/cm)** 2.7 (1.5) 3.2 (1.2) 3.1 (1.5) 3.0 (1.3) 0.56 **Recoil Rate Endo (°/cm/s)** −34.2 (10.3) −46.2 (18.3) −42.2 (9.7) −45.5 (24.7) 0.56 **Recoil Rate Epi (°/cm/s)** −26.1 (14.6) −26.4 (9.3) −24.8 (12.5) −31.4 (10.6)[\*](#nt107){ref-type="table-fn"} **0.048** **Global Recoil Rate (°/cm/s)** −30.1 (11.1) −36.3 (12.7) −33.5 (9.1) −38.5 (16.4) 0.29 Baseline values for torsion and recoil rate in comparison between the four models. **Variables are given as mean (SD)** \***p = 0.022 vs. Model 2** Bold p values indicate a significance level \<0.05. Endo: subendocardial, Epi: subepicardial, Global: average of subendocardial and subepicardial, rest: measurement at rest, 10: level of 10 µg kg^−1^ min^−1^ of dobutamine, 20: level of 20 µg kg^−1^ min^−1^ of dobutamine. Torsion and recoil rate during dobutamine stress {#s3c} ------------------------------------------------ [**Table 4**](#pone-0109164-t004){ref-type="table"} summarises torsion and recoil rate parameters at rest and during dobutamine stress as calculated by the different models. The mean inter-slice distance during dobutamine exposure was slightly lower compared with rest due to increased longitudinal shortening with stress. The inter-slices distances were 2.8±0.3 cm (model 1), 5.6±0.6 cm (model 2) and 4.2±0.4 cm (model 3 and 4) for measurements during 10 and 20 µg·kg^−1^·min^−1^ of dobutamine. 10.1371/journal.pone.0109164.t004 ###### Comparison of torsion and recoil rate between rest and stimulation with dobutamine. {#pone-0109164-t004-4} Level of dobutamine (µg kg^−1^min^−1^) P value ---------------------------- --------------------------------- ---------------------------------------- -------------- -------------- ------------ ------------ ---------- **Model 1 (25**--**75%)** **Torsion Endo (°/cm)** 3.0 (1.9) 4.6 (2.5) 6.0 (2.4) **0.02** **\<0.01** **0.02** **Torsion Epi (°/cm)** 2.3 (1.3) 2.7 (1.5) 4.1 (2.2) 0.33 **0.01** **0.03** **Global Torsion (°/cm)** 2.7 (1.5) 3.6 (2.0) 5.1 (2.2) 0.06 **\<0.01** **0.01** **Recoil Rate Endo (°/cm/s)** −34.2 (10.3) −53.6 (14.8) −79.1 (35.1) **0.01** **\<0.01** **0.04** **Recoil Rate Epi (°/cm/s)** −26.1 (14.6) −40.3 (19.2) −58.6 (32.3) **0.02** **\<0.01** **0.04** **Global Recoil Rate (°/cm/s)** −30.1 (11.1) −46.9 (15.0) −68.9 (32.3) **\<0.01** **\<0.01** **0.03** **Model 2 (0**--**100%)** **Torsion Endo (°/cm)** 3.6 (1.6) 5.2 (2.5) 5.8 (2.5) 0.14 0.09 0.51 **Torsion Epi (°/cm)** 2.7 (0.9) 3.8 (2.0) 3.9 (1.8) 0.29 0.06 0.45 **Global Torsion (°/cm)** 3.2 (1.2) 4.5 (2.2) 4.8 (1.9) 0.17 0.07 0.51 **Recoil Rate Endo (°/cm/s)** −46.2 (18.3) −72.7 (37.9) −82.6 (32.9) 0.17 0.047 0.33 **Recoil Rate Epi (°/cm/s)** −26.4 (9.3) −49.0 (34.0) −55.5 (28.0) 0.047 **\<0.01** 0.39 **Global Recoil Rate (°/cm/s)** −36.3 (12.7) −60.8 (34.7) −69.1 (28.3) 0.11 **0.02** 0.39 **Model 3 (25**--**100%)** **Torsion Endo (°/cm)** 3.6 (1.7) 4.8 (2.6) 6.1 (3.0) 0.07 **\<0.01** 0.09 **Torsion Epi (°/cm)** 2.6 (1.4) 2.9 (1.4) 3.9 (2.1) 0.51 **0.02** **0.03** **Global Torsion (°/cm)** 3.1 (1.5) 3.8 (2.0) 5.0 (2.4) 0.07 **\<0.01** **0.02** **Recoil Rate Endo (°/cm/s)** −42.2 (9.7) −62.9 (29.7) −69.5 (25.0) **0.02** **0.02** 0.45 **Recoil Rate Epi (°/cm/s)** −24.8 (12.5) −37.8 (18.5) −48.3 (25.3) **0.02** **0.01** 0.07 **Global Recoil Rate (°/cm/s)** −33.5 (9.1) −50.4 (23.6) −58.9 (24.6) **0.01** **0.01** 0.33 **Model 4 (0**--**75%)** **Torsion Endo (°/cm)** 3.4 (2.0) 5.4 (2.5) 5.5 (3.0) 0.17 0.11 0.96 **Torsion Epi (°/cm)** 2.6 (0.8) 4.0 (2.3) 4.1 (1.7) 0.24 **0.02** 0.20 **Global Torsion (°/cm)** 3.0 (1.3) 4.7 (2.3) 4.8 (1.8) 0.14 **0.047** 0.88 **Recoil Rate Endo (°/cm/s)** −45.5 (24.7) −69.8 (39.4) −99.5 (45.6) 0.24 **0.01** 0.11 **Recoil Rate Epi (°/cm/s)** −31.4 (10.6) −55.0 (34.7) −70.4 (38.4) 0.06 **\<0.01** 0.14 **Global Recoil Rate (°/cm/s)** −38.5 (16.4) −62.4 (36.4) −84.9 (38.9) 0.09 **0.01** 0.11 Bold p values indicate a significance level \<0.05. Abbreviations as in [table 3](#pone-0109164-t003){ref-type="table"} *.* Model 1 (25--75%) was the only approach that revealed significantly increased torsion and recoil rate (subendocardial, subepicardial and global) between rest and both levels of dobutamine as well as between 10 and 20 µg·kg^−1^·min^−1^ of dobutamine except there was no significant difference of subepicardial torsion (p = 0.33) and global torsion (p = 0.06) between rest and 10 µg·kg^−1^·min^−1^ of dobutamine ([**Figure 4**](#pone-0109164-g004){ref-type="fig"}). The increase of torsion during dobutamine stress was more pronounced on the subendocardial as compared to the subepicarial level. All other models showed only partly significant results or trends towards increased torsion and recoil rates as outlined in [**table 4**](#pone-0109164-t004){ref-type="table"}. {#pone-0109164-g004} Reproducibility {#s3d} --------------- Bland Altman analysis, ICC and CoV are displayed in [**table 5**](#pone-0109164-t005){ref-type="table"}. There was no difference in reproducibility between rest and dobutamine stimulation. ICC for model 1 (25--75%) and model 3 (25--100%) consistently showed values above 0.85. The model 2 (0--100%) and 4 (0--75%) showed slightly higher inter-observer variability for subendocardial torsion with ICC of 0.78 and 0.79, respectively. 10.1371/journal.pone.0109164.t005 ###### Reproducibility for all four models to calculate torsion and diastolic recoil rate. {#pone-0109164-t005-5} Model 1 (25--75%) Model 2 (0--100%) Model 3 (25--100%) Model 4 (0--75%) -------------------- --------------------------------- ------------------- ------------------- -------------------- ------------------ ------------------- ------- --------------- ------------------- ------- --------------- ------------------- ------- **Intra-observer** **Torsion Endo (°/cm)** 0.02 (1.36) 0.92 (0.83--0.96) 30.14 −0.06 (0.79) 0.97 (0.94--0.99) 16.09 −0.16 (0.83) 0.97 (0.95--0.99) 16.91 0.00 (1.04) 0.96 (0.91--0.98) 21.80 **Torsion Epi (°/cm)** −0.02 (0.91) 0.94 (0.87--0.97) 29.81 0.05 (0.42) 0.99 (0.97--0.99) 12.23 0.03 (0.51) 0.98 (0.95--0.99) 16.50 −0.02 (0.61) 0.97 (0.94--0.99) 16.99 **Global Torsion (°/cm)** 0.00 (1.05) 0.93 (0.86--0.97) 27.65 −0.01 (0.50) 0.98 (0.96--0.99) 12.07 −0.06 (0.49) 0.99 (0.97--0.99) 12.14 −0.01 (0.66) 0.97 (0.94--0.99) 15.81 **Recoil Rate Endo (°/cm/s)** 6.48 (15.24) 0.92 (0.82--0.96) 25.88 0.98 (23.49) 0.91 (0.81--0.96) 34.73 −2.82 (10.88) 0.95 (0.90--0.98) 19.17 −0.56 (31.59) 0.88 (0.75--0.95) 44.29 **Recoil Rate Epi (°/cm/s)** −0.44 (9.74) 0.96 (0.92--0.98) 23.50 −0.81 (11.72) 0.95 (0.90--0.98) 27.11 −1.64 (7.51) 0.96 (0.92--0.98) 20.79 −3.80 (16.43) 0.93 (0.85--0.97) 32.63 **Global Recoil Rate (°/cm/s)** 3.02 (9.68) 0.96 (0.92--0.98) 19.30 0.09 (15.77) 0.94 (0.87--0.97) 28.44 −2.23 (7.89) 0.96 (0.92--0.98) 16.97 −2.18 (21.34) 0.91 (0.82--0.96) 35.08 **Inter-observer** **Torsion Endo (°/cm)** 0.07 (1.58) 0.89 (0.77--0.95) 35.30 0.94 (1.33) 0.85 (0.54--0.94) 30.35 0.52 (1.18) 0.94 (0.85--0.97) 25.84 0.62 (1.89) 0.80 (0.59--0.91) 42.31 **Torsion Epi (°/cm)** −0.15 (0.74) 0.96 (0.92--0.98) 23.55 0.35 (0.78) 0.94 (0.85--0.97) 23.73 0.21 (0.50) 0.98 (0.94--0.99) 16.54 0.14 (1.03) 0.93 (0.85--0.97) 29.61 **Global Torsion (°/cm)** −0.04 (1.07) 0.93 (0.86--0.97) 28.04 0.64 (0.97) 0.89 (0.68--0.96) 25.33 0.37 (0.70) 0.96 (0.91--0.99) 18.44 0.38 (1.35) 0.86 (0.71--0.93) 33.78 **Recoil Rate Endo (°/cm/s)** 8.09 (17.07) 0.90 (0.76--0.95) 28.60 −9.31 (17.17) 0.91 (0.77--0.96) 27.47 −0.70 (8.61) 0.97 (0.94--0.99) 14.89 −7.01 (25.30) 0.89 (0.76--0.95) 37.15 **Recoil Rate Epi (°/cm/s)** −0.05 (9.42) 0.96 (0.93--0.98) 22.63 −5.35 (12.64) 0.93 (0.85--0.97) 30.85 −2.35 (7.02) 0.97 (0.93--0.98) 19.61 −6.42 (16.56) 0.92 (0.82--0.96) 33.77 **Global Recoil Rate (°/cm/s)** 4.02 (10.76) 0.95 (0.90--0.98) 21.24 −7.33 (10.86) 0.95 (0.82--0.98) 21.00 −1.52 (6.02) 0.98 (0.96--0.99) 12.86 −6.71 (17.38) 0.92 (0.83--0.96) 29.67 ICC: Intraclass-correlation coefficient; CoV: coefficient of variation; SD: standard deviation; CI: confidence interval. Other abbreviations as in [table 3](#pone-0109164-t003){ref-type="table"}. Discussion {#s4} ========== Myocardial twisting and untwisting represent fundamental mechanisms of LV systolic and diastolic function. It has been suggested that even up to 40% of stroke volume is produced from systolic twisting forces [@pone.0109164-Esch1]. The subsequent fast untwisting during early diastole is a major indicator of the restoring forces that contribute to rapid diastolic filling [@pone.0109164-Burns1], [@pone.0109164-Moon1]. In a clinical context, a simple, quick and robust quantification of LV twisting and untwisting motion is therefore desirable. The present study applied CMR-FT to assess both systolic torsion and diastolic recoil directly from routine SSFP cine images to capture these phenomena and demonstrates several important findings. CMR-FT based quantification of myocardial twisting motion as assessed with torsion and diastolic recoil is feasible and reproducible. Furthermore, we were able to show that CMR-FT allows discrimination of both systolic torsion and diastolic recoil at rest and dobutamine stress when using standardized analyses based on rotation measurements at 25% and 75% LV distance. Feasibility of CMR-FT derived torsion {#s4a} ------------------------------------- The presented approach used a complete LV rotational model to define exact apical and basal slice positions by assessing rotational profiles of all short-axis slices of a short-axis stack with subsequent linear interpolation of inter-slice rotation between adjacent slices. This approach appears suitable since several studies revealed strict linear progression of rotation from base to apex [@pone.0109164-Goffinet1], [@pone.0109164-Buchalter1], [@pone.0109164-Yoneyama1]. Previous reports have shown that the amount of LV rotation is highly dependent on the site of measurements at apical and basal levels. This needs to be taken into account when calculating the simple twist angle [@pone.0109164-Goffinet1]. We therefore applied torsion (twist normalized to inter-slice distance) to quantify the myocardial twisting motion since this technique is supposed to be less influenced by the variation in rotation as compared to the simple twist definition (difference in apical and basal rotation) [@pone.0109164-Young1]. The results of our study indicate, that in fact CMR-FT quantification of torsion at rest is not that dependant on measurement positions since we have not detected any significant differences between the four examined models that used rotation from different standardized LV levels serving for the calculation of torsion. However, differences in recoil rate were more pronounced. Hence, quantification of recoil rate might be more dependent on the exact slice position as compared to quantification of torsion. This is particularly important when quantification of diastolic function is required which underpins the need for a method allowing a uniform slice selection at standardized LV levels. It is important to note that our methodology allows accurate, specific assessments of LV torsion and diastolic recoil at different levels throughout the myocardial wall which could be particularly valuable in the assessment of myocardial diseases that predominantly affect the endocardial layer such as subendocardial myocardial infarction, cardiac amyloid or endomyocardial fibrosis. Reproducibility of CMR-FT derived torsion {#s4b} ----------------------------------------- CMR-FT represents a relatively novel approach to quantify myocardial deformation. Previous CMR-FT studies primarily focused on ventricular strain measurements. The reported amount of reproducibility and repeatability varies between studies with most studies reporting reasonable reproducibility of global strain levels [@pone.0109164-Hor1], [@pone.0109164-Schuster2], [@pone.0109164-Schuster3], [@pone.0109164-Singh1]. Therefore, intra- and inter-observer variability for torsion and recoil rate needs to be taken into special consideration. In order to maximise reproducibility all measurements were repeated three times and corresponding averaged rotational profiles were included in the calculation of torsion. As a result, all torsion and recoil rate parameters were highly reproducible on an intra-observer and inter-observer level. We found slightly lower reproducibility of model 2 (0--100%) and model 4 (0--75%) as compared to model 1 (25--75%) or model 3 (25--100%). Both, model 2 and model 4, involve rotational measurements taken from the most apical slice included in the calculation of torsion. As described in previous reports, reproducibility for CMR-FT derived strain parameters was lower at apical myocardial levels than at mid-ventricular or basal levels [@pone.0109164-Wu1]. Our study shows that this holds true for apical and basal rotation assessment with CMR-FT and emphasises to rather include apical rotation from a slice position at 25% LV distance instead of using the most apical slice of a short-axis stack. Torsion as a function of dobutamine exposure {#s4c} -------------------------------------------- Previous studies investigated the effect of inotropic stimulation with dobutamine on LV torsion in a canine model and found increased torsion paralleled by accelerated recoil rate [@pone.0109164-Buchalter2], [@pone.0109164-Rademakers1]. The findings were confirmed in a human model using echocardiographic speckle tracking [@pone.0109164-Akagawa1]. The authors found that subepicardial torsion remained unchanged during low dose dobutamine infusion and increased with higher doses of dobutamine. In contrast, subendocardial torsion already increased during low dose dobutamine infusion [@pone.0109164-Akagawa1]. In the present study, we showed that model 1 (25--75%) using the shortest distance between apical and basal slices, consistently revealed significant increases of subendocardial and global torsion as well as accelerated recoil rate (subendocardial, subepicardial and global) with increasing levels of dobutamine. Conversely, subepicardial torsion only increased with the application of an intermediate dose of dobutamine, which is in accordance with the published findings by Akagawa et al. [@pone.0109164-Akagawa1]. The other models using either the most apical or basal slice did not reliably differentiate torsion and recoil rate between rest and pharmacological stress and only showed non-significant trends towards increased torsion and recoil rates. This is most likely explained by through-plane motion causing loss of tracked features from the imaging plane. This might be particularly pronounced in the very apical and basal LV slices since longitudinal shortening is increased during dobutamine stress and may explain the need to measure torsion from positions a bit farer away from the apex and the base. CMR-FT in the context of existing methods to calculate torsion {#s4d} -------------------------------------------------------------- Previously described CMR based modalities for the study of myocardial twisting and untwisting motion include myocardial tagging [@pone.0109164-Goffinet1], [@pone.0109164-Donekal2], tissue phase mapping [@pone.0109164-Codreanu1] and displacement encoding with stimulated echoes (DENSE) [@pone.0109164-Haggerty1]. Although accelerated techniques have been reported for myocardial tagging [@pone.0109164-Reyhan1], [@pone.0109164-Reyhan2], all mentioned approaches commonly require additional sequence acquisitions and are usually associated with time-consuming post-processing both limiting their clinical applicability. In contrast, CMR-FT allows direct quantification of myocardial motion from standard SSFP images, which are widely available since they are generally part of every clinical CMR volumetric examination. From a clinical perspective, evaluation of myocardial torsion using CMR-FT has the potential to be integrated into a given clinical CMR protocol. Speckle-tracking echocardiography is another technique that has proved feasible for the quantification of myocardial rotation [@pone.0109164-Opdahl1], [@pone.0109164-Goffinet1], [@pone.0109164-Notomi1]. Speckle tracking echocardiography is widely available and benefits from high temporal resolution but suffers from limited image quality especially occasionally poor acoustic windows [@pone.0109164-Nagel1]. In general echocardiography does not allow a reliable and uniform selection of apical and basal slices in individual subjects, which has proved to be crucial for a robust quantification of myocardial twisting motion as expressed by our data and in previous reports [@pone.0109164-Goffinet1]. Furthermore it is challenging to quantify torsion (twist per length) with speckle-tracking echocardiography, since the assessment of the inter-slice distance between apical and basal slices is generally not possible. Speckle-tracking echocardiography is therefore limited to the simple twist definition, which is even more dependent on the exact slice locations and difficult to reproduce in longitudinal studies [@pone.0109164-Young1]. Notwithstanding these considerations, several echocardiography speckle tracking based studies have shown the feasibility of twisting and untwisting motion quantification in various heart diseases. These involve conditions with a predominant diastolic involvement such as heart failure with preserved ejection fraction [@pone.0109164-Wang1], diseases with predominant systolic involvement such as ischemic cardiomyopathy [@pone.0109164-Bansal1] and also diseases that equally affect systole and diastole such as hypertrophic cardiomyopathy [@pone.0109164-Notomi2] or cardiac amyloidosis [@pone.0109164-Karabay1]. Whether or not such assessments could also be derived from CMR-FT torsion and diastolic recoil measurements need to be investigated with future research studies. Regarding longitudinal assessments, previous studies applied myocardial tagging to investigate the inter-study reproducibility for quantitative analysis of torsion and recoil rate and found good reproducibility for torsion whilst peak recoil rate was only poorly reproducible [@pone.0109164-Donekal2]. In fact, the study of recoil rate may require a more strict definition of measurement positions at apical and basal levels. This may be challenging using myocardial tagging, since this is usually performed at 3 LV levels only (apical, mid-ventricular and basal) and therefore subject to manual planning of slice positions. In contrast, we have now introduced a CMR-FT based technique that allows a uniform and exact definition of apical and basal slice positions independent of different heart sizes. Future work will need to address whether this approach positively impacts on inter-study reproducibility of torsion and of recoil rate in particular. Limitations {#s4e} ----------- The sample size of this feasibility study was relatively small. Notwithstanding this fact the results demonstrate feasibility and good reproducibility for torsion and diastolic recoil quantification with CMR-FT from routine cine images. Ideally we would have compared our results to myocardial tagging representing the reference standard for such acquisitions. However, given the necessity to obtain a whole SSFP short-axis stack at rest and with different levels of dobutamine we felt that the acquisition of an equal number of tagged images would have resulted in scanning times difficult to tolerate during dobutamine stimulation. A general limitation of CMR based quantification of myocardial twisting motion is the current lack of standardized methods. Future studies will need to compare CMR-FT derived torsion and recoil rate with parameters from myocardial tagging using equal computation (twist per length) [@pone.0109164-Donekal2] or alternative computational methods e.g. the circumferential-longitudinal shear angle [@pone.0109164-Russel1], [@pone.0109164-Buchalter1] which has not been addressed in the present study. However, there is evidence to suggest that twisting mechanics as assessed by torsion (twist per length) and shear angle deliver somewhat similar results [@pone.0109164-Donekal1]. The individual merits of both methods remain to be addressed in future investigations. A further limitation of both, conventional myocardial tagging and CMR-FT is the two-dimensional acquisition technique. We developed a pseudo three-dimensional mesh of myocardial torsion based on linear interpolation. However, this method does not compensate for through plane motion. Previous studies compared 2D and 3D derived torsion measured by myocardial tagging [@pone.0109164-Russel2] and found that through plane motion did not significantly impact on 2D torsion since both methods showed reasonable agreement. The authors concluded to rather use the faster and easier 2D method for calculation of LV torsion. If this also holds true for 2D derived torsion from CMR-FT remains to be addressed in future investigations. The clinical applicability of CMR-FT derived torsion and diastolic recoil assessment will depend on the speed of the analysis. At the present time the result files originating from CMR-FT are analysed using an in-house module for MATLAB. Future software developments will need to incorporate both post-processing steps in a single application to further minimize post-processing time. Finally the current work aimed to determine the feasibility and reproducibility to quantify torsion and recoil rate from CMR-FT in a collective of healthy volunteers at rest and during dobutamine stress. Future research needs to prospectively validate this novel technique in pathologies such as coronary artery disease, heart failure or valvular disease. Conclusions {#s5} =========== CMR-FT allows derivation of myocardial torsion directly from standard SSFP cine images. Torsion and recoil rate calculated from apical and basal slices with 25--75% distance consistently discriminated between rest and increasing levels of dobutamine. Reproducibility was good to excellent for all torsion and recoil rate parameters, but was especially high when using an apical level at 25% LV distance instead of the most apical slice. When using CMR-FT to quantify myocardial torsion, it is therefore suggested to select apical and basal slices at standardized 25% and 75% levels. The analysis of myocardial torsion and recoil motion based on CMR-FT may have potential clinical and research applications when exact quantification of systolic and diastolic function is required. Future validation studies in larger cohorts should follow to establish the clinical utility and the prognostic implications of this technique. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JTK PL AS. Performed the experiments: JTK PL STH SK BB AS. Analyzed the data: JTK PL AS. Contributed reagents/materials/analysis tools: MS JMS MF WS CUB JL GH. Wrote the paper: JTK PL AS.

Introduction ============ In 2010, the Carnegie Foundation made recommendations to fundamentally change medical education.[@b1-amep-5-289] It called for curricular reforms that would provide earlier exposure to clinical work, more hands-on experiential learning, and a greater focus on care coordination and working with interprofessional teams. The call for medical education reform parallels the call for health care reform. Medical schools must make certain that efforts toward curricular reform align with the rapidly changing health care delivery system. As the health care delivery landscape changes, medical schools must develop creative strategies for preparing future physicians to provide quality care in this new environment. One such change is the increasing adoption of the patient-centered medical home (PCMH). The PCMH promises to deliver care that is more accessible, effective, efficient, and safe, while simultaneously bending the cost curve.[@b2-amep-5-289] The core tenets of the PCMH are a personal physician, a physician-directed medical practice, a whole-person orientation, integration and coordination of care, quality and safety assurance, enhanced access, and outcome-based payment.[@b3-amep-5-289] As health centers across the country adopt this model of care, future physicians need to be prepared to incorporate these principles in their practices. Despite the growing prominence of the PCMH as effective models for health care delivery, few medical schools have integrated formal education on the PCMH into their curricula. Studies show that few medical students are exposed to this model in their formal education or have opportunities to gain hands-on experience using this model of care.[@b4-amep-5-289]--[@b6-amep-5-289] Only 41% of family medicine departments in one survey reported implementing a PCMH curriculum for students, and teaching modalities largely focused on educational conferences rather than longitudinal exposure.[@b5-amep-5-289] Integrating the tenets of the PCMH into medical school curricula is important to ensure that students have a comprehensive understanding of the different models of health care delivery and can operate effectively as physicians. Student-run free clinics have gained popularity as a unique experiential learning opportunity that provides hands-on experience for students while serving as a safety net for medically underserved and uninsured populations. Approximately 110 student-run free clinics exist in the US.[@b7-amep-5-289] They are operated by health professional students under the supervision of a licensed clinician and provide care to medically uninsured populations.[@b7-amep-5-289],[@b8-amep-5-289] Free clinics place students as early as their first year of medical school on the front lines of problem solving and care coordination for patients who have no or little access to care.[@b7-amep-5-289] The infrastructure of these clinics varies, with students seeing patients after hours in an established care facility or in community-based settings, such as mobile health clinics, shelters, or churches.[@b9-amep-5-289] There is literature to support that patients receive quality care at these student-run clinics and are generally satisfied with the experience.[@b10-amep-5-289] These patients most commonly present to these clinics requesting acute care, chronic disease management, screenings, or social services. However, a notable absence exists in the literature regarding how student-run clinics can be used to expose students to new systems of health care delivery, such as the PCMH. The PCMH model is particularly applicable to the care of patients who are economically and medically underserved. The lack of health insurance and basic necessities often renders this population vulnerable to discontinuous or fragmented care. A PCMH model can help to provide continuous and patient-centric care that addresses social determinants of health through linkages with key community agencies.[@b11-amep-5-289] Therefore, students who care for the underserved in a free clinic setting should be aware of this model. Given the limited exposure to the PCMH within current curricula, new physicians may be ill-prepared to practice within a PCMH model of care.[@b5-amep-5-289] Integrating the PCMH model into the medical school curriculum would allow students to learn how to establish long-term relationships with patients, work in interprofessional teams, integrate methods of quality assurance and patient safety into practice, and develop proficiency in the use of electronic health records (EHRs).[@b12-amep-5-289] The objective of this paper is to provide a detailed description of the processes by which the Weill Cornell Community Clinic (WCCC, New York, NY, USA), a student-run free clinic, adopted elements of the PCMH model into an existing experiential learning program. Methods ======= Description of the WCCC ----------------------- The WCCC was established by medical students in 2003 with the mission of providing care to patients who were uninsured or had insufficient health insurance to cover the cost of care. The activities of the clinic are coordinated by a board of medical students working closely with volunteer physicians and a Faculty Medical Director. Each board member is responsible for specific tasks, such as scheduling patient visits, directing patient care, and coordinating psychosocial screenings and services. At the student level, two Clinical Directors and two Executive Directors oversee the board. Clinical Directors are fourth year medical students who serve a 1-year term and supervise the care of all patients ([Table 1](#t1-amep-5-289){ref-type="table"}). Executive Directors are two MD--PhD students during their graduate training who deal with administrative tasks of the WCCC. The board also operates under the close supervision of a Faculty Medical Director, a physician who provides overall guidance for the program, ensures that patients are provided with quality care, and encourages students to participate as active learners. The Faculty Medical Director was selected by the Dean's office as a faculty member who has practiced in primary care with experience supervising and teaching students. In order to ensure support for this position, the Faculty Medical Director receives 0.1 full time equivalents for her work with the WCCC. The WCCC operates on a \$50,000 annual budget funded by a combination of monetary donations and grants in addition to in-kind donations from the Weill Cornell Physicians Organization, Weill Cornell Internal Medicine Associates, Weill Cornell Medical College, and NewYork-Presbyterian Hospital. Patients are seen every Monday from 5--8 PM in clinical space that is donated by the Weill Cornell Internal Medicine Associates, the adult ambulatory care practice for attendings and Internal Medicine residents. Each patient is seen by a team of students consisting of a Junior Clinician, a first or second year student, and a Senior Clinician, a third or fourth year student, under the supervision of a licensed physician volunteer. The responsibilities of this supervising physician are to discuss and examine each patient with the medical students and develop a plan of action. Supervising physicians volunteer their time at the WCCC and must be a practicing physician and a faculty member at the Medical College. In addition, on a weekly basis, all patients are reviewed with the Faculty Medical Director to discuss patient follow-up and review the laboratory data of patients who were seen. In 2009, the EHR system EpicCare (Epic Systems Corporation, Verona, WI, USA) was introduced into the WCCC. This provided an opportunity to reflect on the clinic practices and adopt different models of health care delivery that would maximize the use of an EHR. The PCMH model was considered because of its potential to provide students with exposure to comprehensive patient care, give them an opportunity to assume different roles as providers, and allow them to participate in interdisciplinary teams. Tenets of the PCMH ------------------ ### Students assume the role of the personal physician In a PCMH, each patient has an ongoing relationship with a personal physician who provides continuous and comprehensive care[@b13-amep-5-289] ([Table 2](#t2-amep-5-289){ref-type="table"}). The physician coordinates communication between members of the health care team and establishes relationships with administrative staff and outside consultants. In an effort to provide students at the WCCC with the opportunity to take on the role of a lead physician, third and fourth year students rotate through the WCCC for 6 weeks as part of their primary care clerkship and take on the responsibilities of the personal physician while directing the patient care team. These Senior Clinicians continue to follow the patients after each clinical session and ensure that all necessary tests and referrals are completed. The Senior Clinician also becomes the patient's advocate, helping to fill out necessary forms for social services and making sure the patient sees necessary service providers, such as referred specialists and social workers. In order to ensure a seamless transition and patient hand-off, each outgoing Senior Clinician provides detailed information about the patient and describes any outstanding tests, referrals, or procedures that are pending to the Clinical Directors. This helps to promote a smooth transition of care when new Senior Clinicians rotate through the WCCC. The Medical Faculty Director also ensures that new students joining the WCCC are aware of pertinent outstanding patient issues. In an attempt to meet the high level of interest among the preclinical students, Junior Clinicians sign up for individual nights rather than blocks of time, as with the Senior Clinicians. Thus, a limitation of the nature of such a student-run free clinic can be a lack of continuity as volunteers come and go. In order to provide the opportunity for preclinical students to be involved in the long-term care of a patient, the Continuity of Care Program was piloted. This pilot project was designed to provide additional experience in communicating with patients and their families and to provide students with first-hand experience of the after visit care process. Each student in the program is paired with a patient and attends all of their appointments throughout the year, serving as Junior Clinician in those sessions. They also are involved in follow-up care -- calling the patient with results and working to schedule any necessary referrals. This gives greater continuity for both student and patient. ### Caring for the whole person In the PCMH model, there is an integrated and whole-person orientation to care. The personal physician attends to all of the patient's health care needs and assumes responsibility for coordinating services with other professionals or organizations. This requires integrating all elements of a complex health system, extending from within the medical practice to various consultants, ancillary services, and community resources.[@b13-amep-5-289] Students in the WCCC learn to coordinate health care services for patients to address both their psychosocial and their medical needs. For example, a review of patient requests identified a need for mental health and gynecological services. In response, the WCCC developed a mental health clinic night for patients requiring psychosocial and mental health care, as well as a women's health clinic. Despite these additions, patients needed additional social services. This posed a challenge for students, as they were not familiar with the breadth of social services that were available to patients. Therefore, the WCCC partnered with the Silberman School of Social Work at Hunter College (New York, NY, USA) to facilitate enrollment into health insurance plans and access to social services. The WCCC thus became a service-learning program for both the WCCC medical students and the Silberman School social work students. In collaboration with the social work students, who are supervised by a licensed social worker, the medical students compiled a list of social services that direct patients to resources such as housing, insurance, access to medication, and nutrition. ### Students learn to develop effective systems of communication Health care that is provided within a PCMH offers enhanced access to care through systems such as open scheduling, expanded hours, and alternative methods of communication.[@b14-amep-5-289] Despite the WCCC's limited weekly hours, the clinic maintains accessibility through electronic and telephonic communication. In order to provide enhanced access to patients and maintain effective communication, the students use an electronic inbox and voicemail system to triage patient concerns, which is monitored at least daily by the Clinical Directors. Patients can use this system to schedule appointments or make requests for referrals. Patients can also request new appointments through the WCCC's website. All telephone interactions with patients are documented in the patient's EHR and discussed with the Faculty Medical Director. ### Students learn to assess quality assurance and risk management In the PCMH model, creating systems for documenting and monitoring patient outcomes and establishing reliable systems of communication are essential components to ensuring quality of care and risk management.[@b15-amep-5-289],[@b16-amep-5-289] Students who volunteer in the WCCC receive training on using EHR to conduct quality assurance checks on patients. Students worked with the information technology team at the Medical College and adopted an existing EHR system -- EpicCare -- to collect and maintain patient data. All students on the WCCC board receive training and secure passwords, granting them access to patient records. This EHR provides a mechanism for collecting patient clinical data, documenting notes that can be reviewed by the supervising attending and the Faculty Medical Director, processing referrals, and documenting the exchange of information with patients. All patient calls are documented in the EHR. This EHR provides a unique opportunity to teach students about proper documentation, how to query their own patient panels, and how to collect data on quality assurance and risk management. The EHR also provides patients secure access to their test results via the Internet. At the WCCC, students have established quality assurance and patient safety initiatives, designing a four-part quality tracking and improvement campaign. In line with the holistic orientation of a PCMH, the WCCC improves preventive care through distribution of a health care maintenance list that includes age-appropriate screenings and vaccinations. Students then do follow-up with patients regarding these measures at every clinic session. Secondly, there is a focus on improving outcomes by tracking chronic disease and preventive measures with a patient dashboard generated by the EHR. As part of a hospital-wide initiative, they also measure health care provider hand-washing compliance through patient surveys completed at the end of the appointment. Finally, students have instituted a nightly checklist that the Senior Clinicians use to ensure that important follow-up issues are not missed when patients leave. Such issues include providing the patient with a follow-up plan, processing referrals, and establishing a means of contacting patients after the visit. As a component of tracking the quality of the experience, the student board of the WCCC received Institutional Review Board approval to perform a chart review and to survey both students and patients. For the year of July 2011 to June 2012, the EHR was interrogated to investigate patient demographics and the most common diagnoses and prescribing information. During the first year of the PCMH adoption, all 174 patients were provided surveys upon arrival at the WCCC, which were submitted before leaving the clinic. The surveys consisted of nine five-point Likert scale questions about the helpfulness of various members of the clinic, including medical students, staff, and physicians; quality of care received; and overall satisfaction with the WCCC, with options ranging from poor to excellent. The survey also included questions about the patient's primary language, total number of visits to the clinic, and time spent during the visit. Descriptive statistics were calculated using SPSS^®^ version 20 (IBM Corporation, Armonk, NY, USA). Student satisfaction was measured in the Continuity of Care pilot project. All ten students who finished the first year in the pilot program completed a 22-question survey regarding their experience. Eight of these questions used a five-point Likert scale to query how the Continuity of Care project supplemented the medical school experience, how it had affected future medical career plans, the student's relationship with the patient, and overall satisfaction with the program. Students also answered open-ended questions regarding positive aspects of their experience, as well as areas for improvement. Descriptive statistics were calculated using SPSS version 20. Results ======= A review of clinic practice over a 1-year period after a PCMH model was adopted showed that 112 students volunteered across all 4 years of medical school, including 70% of the first year class. They were overseen by 19 licensed physician volunteers. All students were enrolled at Weill Cornell Medical College. The ten students who participated in the Continuity of Care pilot project rated their overall satisfaction of the program on a scale from one to five, with five being extremely satisfied, as 4.2 (95% confidence interval 3.4--5.0). The degree to which they felt this experience improved their ability to care for and manage medical conditions was rated as 3.3 (95% confidence interval 2.2--4.5) on a scale from one to five, with five being extremely improved. The degree to which this experience enhanced their medical school education was rated as 3.9 (95% confidence interval 2.8--5.0). Common positive themes from open-ended questions included the opportunity to "follow a patient individually throughout their medical care and watch their symptoms improve", "bridging the gaps in care", and "see(ing) how helpful good medicine can be". Areas for improvement included a need for "more structured orientation", and feeling ill-equipped to provide medical advice. Between July 2011 and June 2012, 174 patients were seen by the WCCC at 337 appointments. A review of this patient series found that 66% were women; 37% fell in the age range of 45--59 years, 26% were aged between 25--34 years, 20% were 60 years and older, and 14% were between 35--44 years. English was the primary language spoken by 87% of the patients, followed by Spanish. A significant portion of patients (87%) reported an annual income of less than \$30,000, and 17% had an income of less than \$10,000. Almost half of the patients had at least two visits in 1 year. The most common diagnoses made during the visits included management of hypertension, well woman exam, dyslipidemia, pain, depression or anxiety, diabetes, and thyroid disease. A total of 189 medications were reimbursed, nearly two-thirds of which were for hypertension, diabetes, and dyslipidemia. Medications for management of psychiatric diagnoses accounted for 6% of the medications; anti-inflammatories and medications for insomnia each accounted for 4% of the medications; and antibiotics, eye drops, and nutritional supplements made up the remainder. A total of 785 laboratory tests were ordered. The most common tests included complete blood count, glucose level, lipid level, sexually transmitted disease screening (including human immunodeficiency virus), and thyroid function test. A total of 87 patients fully completed the satisfaction survey. Incomplete surveys were excluded. Analysis of the patient satisfaction survey found that 87% were very or extremely satisfied with the care they received in the WCCC, and 96% of the patients would recommend the WCCC to others. Patients were asked to describe the interpersonal and technical aspects of care received from poor to excellent. Overall, 80% of patients rated the helpfulness and courtesy of the students, thoroughness and competence, and communication of the treatment plan as excellent or good. The length of time spent at the clinic was rated as excellent or good by 70% of the patients. Discussion ========== Given the popularity of student-run free clinics across medical schools, integrating the PCMH into these practices can address the gap in education about this health care delivery model. By incorporating tenets of the PCMH into the WCCC, 70% of first year medical students at Weill Cornell Medical College gained hands-on exposure to this practice model. Student volunteers at the WCCC participated in a breadth of activities consistent with the guidelines put forth in the *Joint Principles for the Medical Education of Physicians as Preparation for Practice in the Patient-Centered Medical Home* that made recommendations on key aspects of the PCMH to which students should receive exposure, such as a longitudinal experience in continuity of care, work in interdisciplinary teams, experience with the economics of health care, practice with a variety of different types of encounters, including face-to-face and telephonic and electronic messaging, and exposure to the basics of medical informatics.[@b17-amep-5-289] Students involved in the Continuity of Care project reported a high level of satisfaction with their experience, as seen in other studies of student involvement in student-run clinics.[@b18-amep-5-289] A review of patients seen at the WCCC demonstrates that they present with a breadth of medical conditions that are consistent with what is experienced at other free clinics.[@b9-amep-5-289] In general, patients expressed satisfaction with the care they received at the WCCC. Need for improvement was noted in the amount of time they spent waiting to see a care team at the WCCC. Longer wait times may be due in part to the greater time spent by first and second year students conducting the history and physical exam. In addition, after being interviewed by medical students, all patients must be seen and examined by the supervising attending, further lengthening the time of the appointment. In response to receiving this feedback, a new student volunteer position was created to focus specifically on clinic flow and ensure patients are seen in a timely manner. Despite the wait times, over 90% of patients said that they would refer others to the WCCC. Medical schools today must prepare students to navigate the ever-changing landscape of health care. By taking primary responsibility for the day-to-day practice of the WCCC, students put themselves not only on the frontline of clinical care, but also gain exposure to systems-level practice issues. Each barrier to be overcome has served as a learning experience for the students. The most significant obstacle has been ensuring long-term stability within the medical school while maintaining the autonomy of a student-run clinic with the care of a vulnerable population as the focus. Creating stability has involved ensuring a funding stream and advocating for support from the medical school. In order to fund the WCCC, students gained experience with fundraising by working with the alumni association, writing grants, and seeking donations from foundations to support the clinical activities. Students were responsible for allocating the budget. The funding was utilized to support patient medications, laboratory testing, and a subset of referrals for specialty care appointments. To expand the services available to patients while operating in a tight budget, the WCCC has negotiated low-cost referrals with a number of specialty clinics and pays this fee for its patients. The student board has worked closely with the medical school administration to maintain a mutually beneficial relationship. Students give presentations about the WCCC at meetings to promote the role of the WCCC as an important educational and service learning opportunity while also working to raise awareness for the WCCC within the college and hospital. This helps to recruit new physician volunteers. To maintain a stable influx of student involvement during their clinical years, when involvement tends to taper off with the increasing demands of clerkships, the WCCC has become a required component of the primary care/family medicine clerkship. Finally, as a reflection of the college's commitment to the WCCC, the Faculty Medical Director is now supported through a separate budget provided by the Dean's office. Given the low income of patients, students learn to address many of the social determinants of health, such as housing, insurance, and transportation. Students are further exposed to the ethical considerations of having a student-run clinic play such a central role in the care of the indigent patient population. As discussed elsewhere,[@b19-amep-5-289] student-run clinics have a responsibility to put patient care first and to create a culture of professionalism with suitable physician oversight. Patient care is paramount at the WCCC, and all individuals seeking care there are screened for appropriateness with the Faculty Medical Director. Those who are too medically fragile for the WCCC to provide sufficient care are referred elsewhere in the community. Overcoming the challenges of implementing the PCMH model at the WCCC has provided invaluable lessons to all students involved. The WCCC is an evolving practice, with the enduring goal of improving clinic efficiency and patient outcomes. Each class of medical students has the opportunity to contribute to the WCCC and explore novel ways to enhance the PCMH model. Limitations ----------- This paper has a number of limitations. Surveys of students were restricted to those involved in the Continuity of Care pilot project, with a focus on feasibility of the project and the students' perceptions. Thus, while the students self-reported feeling better equipped to treat patients after the experience, this may be the result of simply having more education overall rather than be the direct result of their experience with the WCCC. A future study is needed to compare this experience with those not involved in the project as a control group. Furthermore, there would be benefit from an expanded evaluation of the student experience, including a comparison with other components of the curriculum. Future studies will investigate whether participation with the WCCC impacts a student's career goals. Although all patients were provided with satisfaction surveys, only a subset of these patients completed the survey, which limits generalization. Furthermore, while all patients were informed that their responses were anonymous and participation would not affect their care, these concerns may have impacted the results. There is a need to extend beyond patient satisfaction and study outcomes of care, an area that will be addressed in future studies utilizing the EHR. Conclusion ========== As efforts to implement new models of health care delivery increase, so must the efforts of educators to ensure that students receive proper exposure to these models. The WCCC provides a model for how principles of the PCMH can be taught through experiential learning activities of a student-run free clinic. Integrating core principles of the PCMH into programs such as student-run free clinics can help to better prepare students to respond to health care reform changes in their own practices. The experience of the WCCC may serve as an example for how other programs can incorporate the PCMH models as part of required ambulatory care rotations and other experiential learning initiatives. As medical schools nationwide reconsider their curricula, they must also adapt to changes in health care. The student-run free clinic can provide a model for students to play a significant role and gain hands-on experience in the intricacies of health care delivery. The authors regret to advise of the passing of Dr Anne Kastor prior to publication. This paper is dedicated to the memory of Dr Kastor for her years of service to the WCCC. The authors wish to thank Dr Alvin Mushlin and Dr Olaf Andersen, both of Weill Cornell Medical College, for their insightful editorial suggestions. The authors are grateful to the physicians and students who have volunteered in the WCCC and the patients who allowed us to participate in their care. **Disclosure** The Weill Cornell Community Clinic receives funding from Weill Cornell Physicians Organization, Weill Cornell Internal Medicine Associates, Weill Cornell Medical College, and NewYork-Presbyterian Hospital in addition to private donors. MCR previously received and JBS, and JDS receive funding from NIH MSTP grant T32GM07739 to the Weill Cornell/Rockefeller/Sloan-Kettering Tri-Institutional MD--PhD Program. CBF was supported by Grant Number 1P60MD003421 from the National Institute on Minority Health and Health Disparities. The content of this article is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. JBS is presently director of the WCCC; MCR and JDS were directors of the WCCC. JL and MMR were board members of the WCCC. ASK and CH report no conflicts of interest in this work. This work received approval from the Weill Cornell Medical College Institutional Review Board. ###### Student and faculty positions at the Weill Cornell Community Clinic (New York, NY, USA) Title Qualifications Description Minimum commitment -------------------------------------------------------------------------------------------------------------------------------------------------------- ----------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------- -------------------- Clinical Directors Two third to fourth year medical students • Monitor all patient care through close interactions with Senior Clinicians 1 year • H andle long-term patient follow-up • Meet weekly with Faculty Medical Director to discuss patient care • Conduct board meetings Executive Directors Two MD--PhD students in graduate school • Work with Clinical Directors to set clinic protocols and policies 2--3 years • Manage budgeting and fundraising • Conduct board meetings Faculty Medical Director One faculty physician • Meet regularly with Clinical Directors to discuss patient care Open-ended • Provide guidance on patient issues that arise between clinic sessions Board 20+ medical students -- most are first or second year • Roles include patient scheduling, social work coordination, insurance screening, quality assurance, data management, and information technology 1 year • Attend board meetings to review all patients seen at most recent clinic, address ongoing issues, and resolve clinic matters smoothly and efficiently Senior Clinicians Third or fourth year medical students • Conduct history and physical with Junior Clinician 6 weeks • Work with Junior Clinician to present patient to attending physician, formulate an assessment and plan, and order all necessary tests • Follow-up with patients • Attend board meetings to ensure smooth patient hand-off Junior Clinicians First or second year medical students • Conduct history and physical 1 night • Present findings and work with attending physician and session coordinators to formulate an assessment and plan Attending Physician Faculty volunteer board certified in internal medicine, psychiatry, or gynecology • See all patients on clinic night with student medical team 1 night • Advise Senior Clinicians regarding patient management • Approve all clinical decisions on clinic night ###### Incorporation of the principles of the patient-centered medical home into the Weill Cornell Community Clinic (New York, NY, USA) practice Components of the PCMH Implementation within the WCCC ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------- A personal physician within a physician-directed medical practice • Clinical Directors oversee patient care for a full year • Senior Clinicians act as primary care providers for 6 week blocks • The Continuity of Care Program matches medical students with patients • Board and directors meet weekly with Senior Clinicians to discuss all patients seen during the week and care is reviewed as a team • Policies and practices are discussed as a board Whole-person orientation and integrated care • All new patients receive psychosocial screening and evaluation for insurance eligibility; this is repeated annually • WCCC encourages regular appointments for health care maintenance by scheduling follow-up care at the end of an appointment • In-house women's health and mental health nights occur every 3 weeks • Agreements with various specialists for low-cost referrals paid for by the WCCC Quality and safety • Adoption of EpicCare (Epic Systems Corporation, Verona, WI, USA) for EHR • Health care maintenance lists with screenings and vaccinations are created for each patient at the beginning of a clinic night to encourage appropriate preventative care • Measurement made of health care provider hand-washing compliance • Nightly checklist ensures all patient needs are met before leaving clinic • Clinical operations are updated annually to ensure continuity during annual board transition Enhanced access • Patients can contact the clinic via phone or email, with messages rapidly triaged by the clinical directors to ensure appropriate and timely follow-up **Abbreviations:** EHR, electronic health records; PCMH, patient-centered medical home; WCCC, Weill Cornell Community Clinic. [^1]: \*These authors contributed equally to this work [^2]: †Anne S Kastor passed away on July 5, 2013.

Introduction {#h0.0} ============ The therapeutic benefit of drugs depends on achieving high potency toward the target (an enzyme, cell, bacterium, parasite, virus, etc.) while avoiding damage to the host organism. One approach to killing tumor cells has been to use highly potent bacterial and plant toxins that act catalytically in the cytosol of the targeted cells. Most commonly, specificity has been sought by linking these toxins chemically or genetically to antibodies that bind to cell surface materials enriched on tumor cells. These proteins, initially termed "immunotoxins" and now more generally referred to as "targeted toxins" (TTs), have been under development for decades, but few have reached clinical use ([@B1]--[@B3]). This appears to be due to inadequate specificity for the tumor versus the host (i.e., a low therapeutic index), low efficiency of delivery to the cytosol, and other factors. Several approaches have been explored for improving the therapeutic indices of TTs (reviewed in reference 2) or to increase the uptake of TTs into the cytosol of tumor cells ([@B4]). Our research group has used a different approach to achieve tumor cell specificity. This approach exploits the fact that anthrax toxin from *Bacillus anthracis* activity depends on proteolytic activation of the receptor-bound protective antigen (PA) protein by cell surface proteases ([@B5]--[@B7]). Replacing the site normally cleaved by furin and related proteases with sequences recognized by matrix metalloproteases or urokinase plasminogen activator has yielded potent agents having high specificity and efficacy in mouse tumor models. The protease-activated PA assembles into an oligomeric-protein-conducting channel that efficiently delivers the anthrax toxin catalytic effector proteins to endosomes and then translocates them to the cytosol. The native anthrax effector proteins lethal factor (LF) and edema factor can be replaced with a fusion protein containing the N-terminal 254 amino acids of anthrax toxin lethal factor (lethal factor N terminus \[LFn\]) and the *Pseudomonas aeruginosa* exotoxin A (PE) catalytic domain (PEIII). Once in the cytosol, PEIII will transfer ADP-ribose to eukaryotic elongation factor 2 (eEF2), resulting in protein synthesis inhibition and cell death. This system is highly effective in terms of cytosolic delivery and tumor-specific activation. It has been tested successfully on a number of tumor types ([@B8]) and is expected to be active on nearly all types of solid tumors. One factor that affects the potency of all TTs but that has received limited attention is the issue of the stability of the effector proteins once they have reached the cytosol. It was noted in 1989 that many protein toxins have a strong bias against the presence of lysine residues in their catalytic domains ([@B9]). In retrospect, it is now evident that this feature limits the attachment of ubiquitin and the resulting proteasomal degradation of toxins ([@B10]). The cytosolic stability and resulting potencies of several toxins have been shown to depend on the N-end rule, which specifies that the N-terminal amino acid of a polypeptide determines the efficiency with which side chain lysine residues are ubiquitinated for proteasomal targeting ([@B11]). The N-end rule applies to LF and LFn-based fusion proteins ([@B12], [@B13]), indicating that it will impact the efficacy of anthrax toxin-based TTs. Ubiquitin is a small eukaryotic protein that plays a major role in signal transduction and many other processes in addition to its role in protein degradation. Ubiquitin contains a diglycine motif at its C terminus that is conjugated to the epsilon amine of a lysine within the target protein. According to the N-end rule noted above, ubiquitination occurs on proteins with specific destabilizing N-terminal residues ([@B11]), and ubiquitination may occur on several sites within one protein. Ubiquitinated proteins are targeted for degradation by the 26S proteasome system ([@B14]). Ubiquitin itself may be ubiquitinated after its conjugation to a target protein, leading to creation of polyubiquitin chains. These chains may be built upon any of the seven lysine residues within ubiquitin by ubiquitin ligases, although Lys48 is most often used ([@B15]). Furthermore, polyubiquitination of Lys63 instead of Lys48 seems to result in less protein degradation ([@B16]). Ubiquitination is a balanced process with deubiquitinating enzymes (DUBs) counteracting ubiquitination. The DUBs recognize the C-terminal diglycine motif of ubiquitin and release ubiquitin from labeled proteins. This specific release of cargo from ubiquitin fusions is described by Varshavsky as the ubiquitin fusion technique ([@B17]), which can be used to conditionally stabilize or destabilize a protein. In this study, we examined how the insertion of ubiquitin variants within the TT LFn-PEIII fusion protein altered enzymatic activity, cytotoxicity, and stability of the TTs. We used ubiquitin variants that allowed us to study the accessibility of the TTs to DUBs and polyubiquitination. The results obtained indicate that intracellular release of the catalytic PEIII domain is achievable and that ubiquitination of the TTs controls their persistence in the cytosol and thus determines their potency. RESULTS {#h1} ======= Design of TTs. {#h1.1} -------------- Six different TTs were constructed and analyzed in this study ([Fig. 1A](#fig1){ref-type="fig"}). The five ubiquitin-containing TTs are based on the TT anthrax fusion toxin FP59 that contains LFn at the N terminus and PEIII at its C terminus. An alignment of the amino acid sequences shows the differences in the ubiquitin fusion proteins ([Fig. 1B](#fig1){ref-type="fig"}). The TT designated "Ub" ([Fig. 1A](#fig1){ref-type="fig"}) contains the human wild-type ubiquitin with the C-terminal diglycine motif that is specifically recognized by DUBs. The TT Ub~UN~ has the same sequence but with three glycines (including the diglycine motif) replaced by a sequence of three alanines, which renders the sequence uncleavable by DUBs. The Ub~K48~ and Ub~K63~ mutants have all lysine residues of ubiquitin replaced by arginine residues except for one lysine residue, either Lys48 or Lys63, respectively. Ub~Knull~ contains ubiquitin with all seven lysines replaced by arginine. Ub~K48~, Ub~K63~, and Ub~Knull~ all retain the intact C-terminal diglycine motif. All proteins were successfully purified in yields of at least 1.5 mg per liter of culture ([Fig. 2A](#fig2){ref-type="fig"}). The molecular masses of all six proteins were confirmed by electrospray ionization mass spectrometry. Since the accessibility of PEIII within a larger polypeptide or toxin may limit its activity ([@B18]), we confirmed the catalytic activity of the five TTs containing ubiquitin or variants of ubiquitin in comparison to the fusion protein FP59 ([Fig. 2B](#fig2){ref-type="fig"}). All samples show a band for biotin-containing ADP-ribosylated eEF2 at a molecular mass of 100 kDa and an additional band of the same intensity at around 40 kDa. The latter is a degradation product of eEF2, an artifact of the purification of eEF2 from *Saccharomyces cerevisiae* yeast. This fragment contains the site of ADP-ribosylation of eEF2 and is thus likewise ADP-ribosylated and detected ([@B19]). {#fig1} {#fig2} *In vitro* deubiquitination of TTs. {#h1.2} ----------------------------------- In order to analyze and compare the cleavability of the ubiquitin-containing TTs, the fusion proteins were incubated with HN6 cells (a human head and neck cancer cell line) lysates. The lysates contain active DUBs that are expected to cleave the C-terminal diglycine motif of the ubiquitin ([@B20]). All TTs containing ubiquitin with the natural C-terminal diglycine sequence were cleaved within the 2-h reaction time ([Fig. 3A, B](#fig3){ref-type="fig"}, and C). The cleavage product, LFn-ubiquitin, has an expected molecular mass of 41 kDa and is detected as a band migrating between the 50-kDa and 36-kDa marker bands. FP59, which lacks ubiquitin, showed no cleavage within 2 h ([Fig. 3C](#fig3){ref-type="fig"}). Furthermore, UB~UN~, containing three alanines instead of the diglycine motif, remained intact ([Fig. 3C](#fig3){ref-type="fig"}). All TTs were also exposed to lysates preincubated with the DUB inhibitor ubiquitin aldehyde. Pretreatment of the lysates with the inhibitor prevented cleavage of the fusion proteins, demonstrating the specificity of the cleavage. In order to see the effect of purified DUBs, the TTs were incubated with either Ubiquitin carboxyl-terminal hydrolase isozyme L3 (UCH-L3) or Otubain-1 (Otub1). UCH-L3 incubation resulted in low-efficiency cleavage, while Otub1 did not achieve any detectable cleavage (data not shown). These DUBs are apparently not primarily responsible for the intracellular cleavage of the TTs or require further components in the cytosol to achieve efficient cleavage. {#fig3} Deubiquitination of TTs in cells. {#h1.3} --------------------------------- Studies on deubiquitination of TTs within cells were done using CHO TEM8 T4 cells, Chinese hamster ovary cells stably transfected with tumor endothelial marker 8 (TEM8) that overexpress the TEM8 anthrax toxin receptor and internalize more TT, thereby facilitating its detection in the cytosolic fractions. All full-length TTs were detected in the cytosol after 1-h toxin exposure and Western blotting with anti-LF and anti-PE combined ([Fig. 4A to C](#fig4){ref-type="fig"}, top black arrow). Mutant PAΔFF (PA with Phe~314~ and Phe~315~ deleted) binds and delivers LFn and TTs to endosomes but cannot support their translocation to the cytosol ([@B21]). The presence of small amounts of full-length TTs is probably an indication that the cytosolic extracts are contaminated with some endosomal contents. The bigger cleavage product, LFn-ubiquitin, with an apparent molecular mass of 41 kDa, was most likely detected as a very weak band in all fractions (white arrowhead), even in the PAΔFF samples and in the uncleavable UB~UN~ fractions. However, PEIII, the released catalytic domain of the TTs, is clearly detected in the cytosol of cells incubated with Ub~K48~, Ub~K63~, and Ub~Knull~ ([Fig. 4B and C](#fig4){ref-type="fig"}, bottom black arrow with dashed line at an apparent molecular mass of 24 kDa). The corresponding band for the Ub TT is very weak in comparison, and the Ub~UN~ sample has no band for PEIII ([Fig. 4A](#fig4){ref-type="fig"}). The data demonstrate that DUB processing of the susceptible TT occurs in the cytosol to release the free PEIII domain. The ubiquitination inhibitor Pyrazone-41 (PYR-41) did not increase the amounts of any of the proteins detected, neither the uncleaved TTs nor the cleaved LFn-ubiquitin or PEIII. The FP59 lanes showed only the expected band for full-length FP59. {#fig4} Cytotoxicity of TTs. {#h1.4} -------------------- Cytotoxicity analyses employed HN6 cells. This human cell line was chosen as a model cell line for proof of principle, since this cell line is a suitable model for human head and neck cancer when transplanted into nude mice. A 48-h toxin exposure resulted in dose-dependent cytotoxicity with cytotoxicity in the order Ub~Knull~ \> Ub~K63~ \> FP59 \> Ub~K48~ \>\> Ub \> Ub~UN~ ([Fig. 5A](#fig5){ref-type="fig"}). The 50% survival index (SI~50~) values determined for the TTs are shown in [Table 1](#tab1){ref-type="table"}. The observed SI~50~ values range from 615 pM (Ub~UN~) to 3.7 pM (Ub~Knull~). These values are 0.02-fold and 3-fold changes of the SI~50~ of the ubiquitin-free FP59 (FP59 SI~50~, 11 pM), respectively. A shorter (2-h) period of toxin exposure reduced the observed cytotoxicities for all TTs ([Fig. 5B](#fig5){ref-type="fig"}). However, the reductions in potency were about the same for all TTs, since the relative SI~50~ values (compared to FP59 value) remained similar for all TTs ([Table 1](#tab1){ref-type="table"}). Additionally, FP59-sensitive mouse RAW264.7 cells (murine leukemic monocytes/macrophages) were preincubated with the E1 ubiquitin-activating enzyme inhibitor PYR-41 ([@B22]) and subsequently with the different TTs for a further 18 h before cell survival was measured. RAW264.7 cells were used, since they die within 18 h of continuous toxin exposure, while other cell lines need up to 48 h and are thus affected by PYR-41 toxicity (data not shown). PYR-41 increased the cytotoxicities of Ub~K63~, Ub~Knull~, and FP59 only slightly (see [Fig. S1D to F](#figS1){ref-type="supplementary-material"} in the supplemental material), while increasing the toxicities of Ub~K48~, and especially Ub and UB~UN~, to larger, if still modest extents ([Fig. S1A to C](#figS1){ref-type="supplementary-material"}), consistent with the expectation that the latter TTs are more susceptible to ubiquitination and inactivation. {#fig5} ###### Cytotoxicity of the targeted toxins on HN6 cells^[*a*](#ngtab1.1)^ Targeted toxin 48-h toxin exposure 2-h toxin exposure ---------------- --------------------- -------------------- ------ ---------------------- ----------- ------ Ub~UN~ 615 0.052 0.02 ND[*^d^*](#ngtab1.4) 0.066 ND Ub 100 **0.045** 0.11 773.0 0.053 0.03 Ub~K48~ 014 0.093 0.79 79 **0.012** 0.34 FP59 011 027 Ub~K63~ 006.2 0.051 1.8 018 0.496 1.5 Ub~Knull~ 003.7 **0.038** 3 011 0.114 2.5 SI~50~ values, *P* values, and factors of enhancement for the cytotoxicity of the TTs on HN6 cells obtained from [Fig. 5](#fig5){ref-type="fig"}.  The *P* value was calculated for the FP59 value versus the values of all TTs. Bold *P* values are below 0.05, which is defined to be a significant difference. The factor of enhancement was calculated by comparison of the SI~50~ value of FP59 to those of all other TTs.   ND, not determined. Tumor treatment. {#h1.5} ---------------- The Ub and Ub~Knull~ fusion proteins as well as the control FP59 protein were administered with the matrix metalloprotease-activated PA-L1 to mice with implanted LL3 mouse melanoma cell tumors. Treatment with PA-L1 plus Ub or Ub~Knull~ resulted in strong tumor growth inhibition ([Fig. 6A](#fig6){ref-type="fig"}). Ub~Knull~ significantly reduced tumor growth compared to the group treated with phosphate-buffered saline (PBS group) (the control group) on day 13 (*P* = 0.05). Due to rapid tumor growth and ulceration in the PBS group, several mice were euthanized before the end of the experiment. Therefore, only the Ub and Ub~Knull~ groups were compared at days 15 and 17; Ub~Knull~ treatment was significantly better than Ub treatment (*P* = 0.04 and *P* = 0.01, respectively). Ub~Knull~ treatment resulted in the development of inflammation and edema after the third injection, while Ub treatment did not result in any observable side effects. More aggressive tumor growth in the PBS and Ub groups resulted in a higher number of mice being euthanized before the end of the experiment ([Fig. 6B](#fig6){ref-type="fig"}). Individual tumor masses at day 13 and day 17 demonstrate a very good antitumor effect of the Ub~Knull~ treatment ([Fig. 6C and D](#fig6){ref-type="fig"}, respectively). Eight of 10 mice showed a strong response to the treatment, and one mouse was tumor-free. Treatment with PA-L1 plus FP59 resulted in severe inflammation and edema after the first injection, and no further injections were possible on day 8 or 10. One mouse was euthanized due to lethargy, most likely related to the FP59 treatment. Even at the end of the experiment, 7 of 10 mice showed symptoms like those following the first treatment. Due to the toxic effects of this treatment, the FP59 group was not further analyzed for tumor growth-inhibitory effects. {#fig6} DISCUSSION {#h2} ========== TTs containing bacterial toxin catalytic domains possess high potency, at least in theory, but this high potency requires that the TT be targeted with high specificity to avoid damage to nontarget cells. A high inherent potency allows lower doses to be used, which can help to limit the inevitable damage that comes from clearance of administered proteins to, and accumulation in, the liver and kidneys ([@B23]). However, for low protein doses to be effective, the protein must be efficiently bound and internalized and translocated to the cytosol, and once there, it must have sufficient stability to inactivate its target. The studies described here focus on the latter steps in TT action. The modified anthrax toxin used in our study utilizes a highly efficient delivery system which can deliver a number of different "payloads" in addition to the natural toxin lethal factor, as it was shown that the system is able to deliver PEIII ([@B24]), diphtheria toxin A chain, Shiga toxin catalytic domain ([@B25]), Bcl-XL protein ([@B26]), as well as the reporter beta-lactamase ([@B27], [@B28]). A key feature of the TTs described here is the insertion of a ubiquitin domain between the targeting domain (LFn) and the catalytic payload (PEIII). A prerequisite for success of this design is that the ubiquitin domain not limit the efficiency of translocation of the enlarged, 3-domain polypeptides to the cytosol. Previous studies of the translocation process show that polypeptides must completely unfold to pass through the narrow lumen of the oligomeric PA channel ([@B13], [@B29]). Proteins that are tightly folded, either naturally or due to ligand binding, will not be translocated, although it is possible that chaperones may facilitate translocation of certain proteins ([@B30]). Ubiquitin is a tightly folded protein ([@B31]), and it was not certain that it would readily translocate. However, the fact that several of the proteins characterized here had potencies very similar to that of the FP59 protein indicate that ubiquitin unfolding occurred readily, at least in the context of these LFn fusion proteins and in the HN6 cells. A second prerequisite for success of the strategy described here is that cytosolic DUBs be able to cleave the fusion proteins at an adequate rate. Here, the strategy is aided by the great diversity of DUBs present in cells ([@B32]). It was evident from the cleavages we observed ([Fig. 3 and 4](#fig3 fig4){ref-type="fig"}) and the pattern of toxicities ([Fig. 5](#fig5){ref-type="fig"} and [Table 1](#tab1){ref-type="table"}) that at least a few of the DUBs recognized the ubiquitin and the modified ubiquitins in the context of the 3-domain fusions. Experiments with purified DUBs showed that UCH-L3 could be one of the DUBs involved to some degree in cleavage of the TTs, while Otub1 does not recognize the ubiquitin-containing TTs as the substrates. Even replacement of all lysines with arginine (in the Ub~Knull~ protein) did not greatly diminish the ability of the DUBs to cleave the fusion proteins. The release of a free catalytic domain (e.g., PEIII) in the cytosol has a number of potential and real advantages, including the opportunity to increase its stability and potency. Typically, it can be expected that any polypeptide used in the delivery of the catalytic domain will contain a number of lysines that provide targets for ubiquitination. This is clearly the case with LFn fusion proteins, since 32 of the 254 residues of LFn are lysine. Thus, an LFn-PEIII fusion protein (e.g., FP59) might be expected to be rapidly degraded. In fact, we and others have shown that LF and LFn are subject to the N-end rule ([@B12], [@B13]) and therefore that the side chain lysines are ubiquitinated, provided the N-terminal region has a destabilizing residue under the N-end rule. In retrospect, it is rather surprising that LF and LFn fusion proteins possess high potency. Thus, the introduction of a ubiquitin domain into a TT provides a way to cause release of the catalytic domain, freeing it from other domains (usually in the N-terminal region) that would promote its degradation. In the case of the PEIII fusions described here, the catalytic domain contains only two lysines, and therefore is inherently resistant to proteasomal degradation. Another advantage of the strategy described here is the ability to release the TT catalytic domain having a specific, desirable N-terminal residue. This objective was one impetus to the original development of the ubiquitin fusion method ([@B17]). In the case of TTs, it is often desirable to generate a free catalytic domain having an N-terminal residue that is an N-end rule stabilizing residue. Introducing ubiquitin as a linker between the two domains of FP59 to produce the TT designated "Ub" decreased its potency nearly 10-fold ([Table 1](#tab1){ref-type="table"}). The introduction of ubiquitin targets the fusion protein to the proteasome after polyubiquitination ([@B33]) and prevented efficient accumulation of PEIII in the cytosol of cells ([Fig. 4](#fig4){ref-type="fig"}). Even though *in vitro* cleavage was as efficient as for the other cleavable TTs ([Fig. 3](#fig3){ref-type="fig"}), release of PEIII may be too slow to save most of the fusion protein from degradation. The uncleavable Ub~UN~ showed no cleavage and was apparently degraded too fast for the PEIII to efficiently inhibit protein synthesis ([Fig. 4](#fig4){ref-type="fig"}). Only the DUB-cleavable TTs having a reduced number of Lys residues demonstrated successful accumulation of PEIII in the cytosol and thus much higher cytotoxicities. The release of PEIII from the rest of the molecule helps PEIII to persist longer in the cytosol, which is a requisite for high cytotoxicity ([@B34]). Ub~Knull~, which is resistant to ubiquitination of the inserted ubiquitin domain ([@B35]), is even more toxic than FP59 and presents the lowest SI~50~ of all TTs studied here ([Table 1](#tab1){ref-type="table"}). Its high cytotoxicity is apparently due to the release of PEIII at rates well above those at which ubiquitination of the lysines on the LFn domain targets the intact fusion protein to the proteasome. Due to severe side effects of the FP59 treatment, Ub~Knull~ could not be compared directly to FP59 in the tumor model. However, lower side effects combined with a strong antitumor effect indicate the advantageous effect of the Ub~Knull~ insertion in FP59. The comparison to Ub in the tumor experiment clearly shows the importance of mutating the lysines of ubiquitin for increasing the protein's stability ([Fig. 6A](#fig6){ref-type="fig"}). In cell culture experiments, the Ub~K63~ mutant was slightly more cytotoxic than the Ub~K48~ mutant ([Table 1](#tab1){ref-type="table"}), consistent with prior evidence that ubiquitination of these two lysines leads to different outcomes. Thus, Jacobson et al. described differences for the proteasomal processing of polyubiquitin chains built on lysine 48 versus lysine 63, with faster deubiquitination of lysine 63-linked polyubiquitin chains and less proteasomal accessibility ([@B16]). This would explain the lower SI~50~ value for Ub~K63~ compared to Ub~K48~. Ub~K63~ would persist longer in the cytosol before it is degraded and counteract the lower release of PEIII detected ([Fig. 3C](#fig3){ref-type="fig"}). Thus, the different ubiquitin variants allow the modulation of PEIII release and protein stability so as to achieve different levels of cytotoxicity. The effect of the ubiquitination inhibitor PYR-41 was rather weak in the cytotoxicity studies (see [Fig. S1](#figS1){ref-type="supplementary-material"} in the supplemental material). Higher concentrations are typically needed to obtain sufficient inhibition of ubiquitin activation ([@B22]). However, these concentrations were toxic for the cell lines used here. Furthermore, there was no increased stability of the TTs observed due to preincubation of PYR-41 ([Fig. 4](#fig4){ref-type="fig"}). As [Fig. S1](#figS1){ref-type="supplementary-material"} shows, the effects of PYR-41 are rather weak, probably due to the instability of PYR-41. Cleavable sequences have been introduced in other TTs for separation of the different moieties. Thus, Heisler et al. introduced furin- and cytosolic-protease-cleavable peptide sequences in a fusion of epidermal growth factor and the plant toxin saporin to release the toxin in the cytosol ([@B36]). Furin cleavage sites were also introduced into TTs by other groups to achieve increased cytotoxicity ([@B37], [@B38]). Fusions to ubiquitin have more commonly been used to decrease the stability of proteins, including green fluorescent protein ([@B39]) and beta-lactamase ([@B33]). In an approach to enhance TT potency more like that described here, Tcherniuk et al. fused ubiquitin to saporin, a plant protein toxin. In that study ([@B40]), the DUB cleavage sequence of ubiquitin was exchanged with a prostate-specific antigen cleavage site. Thus, the authors intended to separate ubiquitin and saporin only in the vicinity of tumor cells expressing prostate-specific antigen, following the same idea used for the activation of anthrax toxin PA by urokinase plasminogen activator or matrix metalloproteases ([@B6]). In contrast to PA, where the proteolytic cleavage results in activation of PA, prostate-specific antigen cleavage would separate ubiquitin and saporin with the goal of enrichment of ubiquitin-free saporin in the vicinity of tumor cells without including a strategy for achieving binding and delivery of the payload to the cytosol, and no gain in potency was achieved. The data presented here demonstrate the value of introducing ubiquitin to the anthrax toxin delivery system (PA and LFn) in order to efficiently deliver enzymes to the cytosol. Variants with the lysines of ubiquitin replaced by arginines have improved stability and significant antitumor effects. Thus, the ubiquitin system for intracellular release provides an effective system for the fine-tuning of toxin uptake, half-life, nonspecific toxicity, and overall efficacy. Future studies on this system will hopefully provide further drugs capable of the safe and efficient elimination of tumor cells and, eventually, may warrant clinical trials in humans. MATERIALS AND METHODS {#h3} ===================== Cloning of ubiquitin-containing targeted toxins. {#h3.1} ------------------------------------------------ All TTs used for this study are based on the anthrax fusion toxin FP59, consisting of LFn-PEIII ([@B5]) and were modified by inserting different ubiquitin variants between LFn and PEIII. LFn consists of the N-terminal 254 residues of anthrax toxin lethal factor ([Fig. 1](#fig1){ref-type="fig"}) and has the sequence AGG...QEINL (but in FP59 includes an additional N-terminal HM added due to cloning). PEIII consists of the C-terminal 216-residue catalytic domain of PE and has the sequence AEFL...REDLK. Details on cloning, expression, and purification of the ubiquitin-containing targeted toxins are described in the supplemental information (see [Text S1](#supplS1){ref-type="supplementary-material"} in the supplemental material). PA and the mutant PAΔFF (PA with Phe~314~-Phe~315~ deleted) were expressed as described earlier ([@B21], [@B41]). Enzymatic activity of TTs. {#h3.2} -------------------------- The enzymatic activities of PEIII within all TTs were measured as described previously ([@B42]). Incubation of the TTs with purified eEF2 and biotinylated NAD^+^ allows the detection of ADP-ribosylated eEF2 by using streptavidin to detect biotin on eEF2 after SDS-PAGE and Western blotting. Cell culture. {#h3.3} ------------- Cell culture experiments were performed on HN6 cells (a human head and neck cancer cell line) ([@B43]), CHO TEM8 T4 cells (Chinese hamster ovary cells stably transfected with anthrax receptor tumor endothelial marker 8) ([@B41], [@B44]), and RAW264.7 cells (murine leukemic monocytes/macrophages). See Text S1 in the supplemental material for details on cell culture and cytotoxicity assays. Cleavage by deubiquitinating enzymes. {#h3.4} ------------------------------------- Lysates of HN6 cells were obtained after growing 6 × 10^6^ HN6 cells overnight, washing them twice with phosphate-buffered saline (PBS) (150 mM NaCl, 8.3 mM Na~2~HPO~4~, 1.7 mM KH~2~PO~4~ \[pH 7.4\]), and incubating them with 300 µl of PBS supplemented with 1% Triton X-100. After 30 min at 4°C on a rotary shaker, cells were resuspended and centrifuged (30 min, 4°C, 16,000 × g). The supernatant (cell lysates) were used for *in vitro* cleavage of ubiquitin fusion toxins by incubation of 75 ng of the TTs for up to 120 min at 37°C with 8 µl of the cell lysate supplemented with 2 mM dithiothreitol (DTT) in a total volume of 9 µl. Additional experiments were performed with ubiquitin aldehyde ([@B16]) to inhibit deubiquitinating activity. Ubiquitin aldehyde (0.5 µl; final concentration, 2.5 µM) was incubated with the lysate 5 min at 37°C prior to TT addition. Following incubation with the lysate, samples were separated by SDS-PAGE and Western blotted using the iBlot system (Invitrogen, Life Technologies, Grand Island, NY). TTs were detected by a polyclonal rabbit anti-LF serum and infrared dye-conjugated secondary antibodies on the Odyssey imager infrared detection system (LI-COR, Lincoln, NE). Samples for detection of intracellular cleavage were separated by SDS-PAGE and Western blotted using the iBlot system and the Western blot signal enhancer kit (Thermo, Waltham, MA) for signal enhancement. TTs were detected by polyclonal rabbit anti-LF serum, polyclonal rabbit anti-PE (anti-*Pseudomonas* exotoxin A) serum (Sigma-Aldrich, St. Louis, MO), and infrared dye-conjugated secondary antibodies (Rockland Immunochemicals, Gilbertsville, PA) on the Odyssey imager infrared detection system (LI-COR, Lincoln, NE). Detection of intracellular cleavage was performed on CHO TEM8 T4 cells (1 × 10^6^ cells overnight in 12-well plates). The cells were incubated with 1 µg/ml PA or PAΔFF (PA with Phe314 and Phe315 deleted, a mutant that fails to deliver LF to the cytosol \[21\]), and 1 µg/ml of the TTs in 0.5 ml medium for 1 h or washed off immediately (0 h). Further samples were preincubated with 50 µM PYR-41 ([@B22]) for 1 h at 37°C before the addition of PA and the TTs at the indicated concentrations for 1 h. All cells were washed twice with PBS and incubated with trypsin-EDTA at 37°C until all cells could be transferred into new tubes for centrifugation (5 min, 4°C, 1,000 × g). For cytosol isolation, the cells were resuspended in 125 µg/ml saponin (Sigma-Aldrich, St. Louis, MO) in PBS, supplemented with protease inhibitor cocktail (Roche), and incubated for 10 min on ice ([@B45]). The complete supernatants after centrifugation (30 min, 4°C, 16,000 × g) were analyzed by SDS-PAGE and Western blotting. Animal experiments. {#h3.5} ------------------- All animal experiments were performed under protocols approved by the NIAID Animal Care and Use Committee. Female C57BL/6 mice (Jackson Labs, Bar Harbor, ME) were injected with 0.8 × 10^6^ LL3 mouse melanoma cells intradermally in the neck on day 0. After 6 days (all tumors had a width of at least 4 mm), mice were randomly assigned to four groups of 10 mice each and injected with 50 µl sterile PBS with different drug combinations intratumorally every other day (day 6, day 8, and day 10). Mice were treated with PBS, 5 µg PA-L1 (PA with a mutated furin cleavage site to achieve tumor-selective cleavage and activation of PA by matrix metalloproteinase 2 \[5\]) plus 1 µg FP59 in PBS, 5 µg PA-L1 plus 1.15 µg LFnUbPEIII in PBS, or 5 µg PA-L1 plus 1.15 µg LFnUb~Knull~PEIII in PBS). Tumors were measured every other day with a caliper, and tumor mass was calculated \[tumor mass = (width × depth × height)/2, with tumor mass measured in milligrams and width, depth, and height all measured in millimeters\] with a final measurement on day 17. Mice with tumors with one diameter exceeding 20 mm or ulceration were euthanized. SUPPLEMENTAL MATERIAL {#h4} ===================== ###### Additional detailed Materials and Methods. Download ###### Text S1, DOCX file, 0.1 MB ###### Cytotoxicity of TTs on RAW264.7 cells in the presence of the E1 ubiquitin-activating enzyme inhibitor PYR-41. RAW264.7 cells (15,000 cells/well) were preincubated with 21 µM PYR-41 or without PYR-41 for 1 h and further exposed to the TTs for 18 h. TTs were added to the cells in various concentrations with a fixed concentration of 250 ng/ml PA. Relative survival was calculated as described in the legend to [Fig. 5](#fig5){ref-type="fig"}. Error bars indicate SEM of two independent experiments performed in triplicate. Each panel presents the relative survival data for the indicated TT. Download ###### Figure S1, TIF file, 0.7 MB **Citation** Bachran C, Morley T, Abdelazim S, Fattah RJ, Liu S, Leppla SH. 2013. Anthrax toxin-mediated delivery of the *Pseudomonas* exotoxin A enzymatic domain to the cytosol of tumor cells via cleavable ubiquitin fusions. mBio 4(3):e00201-13. doi:10.1128/mBio.00201-13 This work was supported by the Intramural Research Program of the National Institute of Allergy and Infectious Diseases (NIAID), Bethesda, MD. We thank J. Eric Anderson for mass spectrometric analysis of the proteins described here. [^1]: **Editor** R. John Collier, Harvard Medical School

{#sp1 .39}

Dear Editor, We reviewed the manuscript recently published by Cumhur Cure M, et al. \[[@CR1]\], regarding colchicine and COVID-19. The authors discussed the effects of colchicine in intracellular and extracellular pH conditions and argued that low pH levels secondary to colchicine may increase the viral load of SARS-CoV-2 and, therefore, cytokine storms will be more severe. In contrast, we consider that colchicine can be a good therapeutic option because of several effects in the immunology system involved in SARS-CoV-2 infection and in the acute respiratory distress syndrome (ARDS). Colchicine has effects on the chemotaxis of inflammatory cells such as neutrophils and monocytes and on the intracellular transportation of vesicles such as endosomes and exosomes. Colchicine also inhibits the expression of E-selectin, an adhesion molecule important for binding leukocytes to endothelial cells, and the recruitment of monocytes and neutrophils to inflamed tissue. Finally, colchicine reduces neutrophil production of free radicals like superoxide \[[@CR2]\]. The recruitment of neutrophils in COVID-19 is a key factor in the severity of cases \[[@CR3]\]. Moreover, colchicine has also been associated with disrupting inflammasome activation, thereby suppressing caspase-1 activation and the subsequent release of IL-1β and IL-18 \[[@CR4]\]. In SARS-CoV-1, the inflammasome activation has been associated with this disruption \[[@CR5]\]. The authors also cited an article published in 1986 by Maurizi M, et al. \[[@CR6]\] regarding two patients who developed ARDS after treatment with toxic doses of colchicine. The first patient received approximately 80 mg of colchicine or 1.6 mg/kg, while the other patient received 15 to 20 mg or 0.25--0.3 mg/kg of colchicine for acute gout. Both patients had ARDS between 24 and 72 h after the colchicine doses \[[@CR1]\]. Both articles \[[@CR1], [@CR6]\] attributed a direct toxic action of colchicine on pneumocyte microtubules, and the inhibition of surfactant production was deemed the probable cause of death of those patients. The dose used in those cases is not recommended due the toxicity. The usual adult oral dose in acute gout is 1.2 mg/day or as a prophylactic, 0.5--1 mg/day; however, dose must be adjusted in patients with renal impairment. The high fatality rate was reported after acute ingestions exceeding 0.5 mg/kg \[[@CR7]\], and therefore, there is no support for that outcome in therapeutic doses. Classically colchicine is commonly used to treat different inflammatory diseases such as gout, as well as some autoinflammatory diseases, and cardiac conditions including viral pericardial syndromes. Recently, a patient with cardiac tamponade secondary to COVID-19 was treated with colchicine in addition to corticosteroids and antimalarials with positive clinical response \[[@CR9]\]. Other research demonstrates the positive effect of colchicine in respiratory syncytial virus replication and suppression of secondary airway inflammation given the promoted expression of IFN-α and IFN-β1 of colchicine and regulation of anti-oxidative factor production \[[@CR10], [@CR11]\]. In addition, there is a wide range of preclinical and clinical literature on the effect of colchicine inhibiting viral disease like adenoviral and adeno-associated viral \[[@CR12]\], herpes simplex virus type 1 \[[@CR13]\], Epstein-Barr virus \[[@CR14]\], and hepatitis virus \[[@CR15], [@CR16]\], among others. On the other hand, Cumhur Cure M, et al. \[[@CR1]\] discussed the interactions of colchicine with other drugs. We would agree that colchicine may indeed have interactions with other drugs; therefore, we recommend adjusting the dose in close consideration of the interactions with inhibitors of CYP3A4 as antibiotics and antivirals even though some of them are used in COVID-19 as lopinavir/ritonavir. In our experience, we reported 5 patients (age 38--61 years) with comorbidities (arterial hypertension, type 2 diabetes, among others) in treatment with colchicine for iatrogenic allogenosis 1 to 3 weeks before COVID-19 test positive. They developed mild symptoms such as headache, cough without dyspnea, and arthralgias. It should be noted that some close contacts presented severe symptoms and three of them died \[[@CR8]\]. According to a potential benefit of colchicine in COVID-19 patients, since March 26, 2020, a total of twelve studies have been registered in [www.clinicaltrials.gov](http://www.clinicaltrials.gov) and the European Union Clinical Trials Register considering the clinical utility of this well-known medication. Most of these studies correspond to randomized, open-label, phase 2 clinical trials. It is aspired to include more than 11,000 patients on three continents. The experimental arm in most studies includes an oral colchicine regimen with loading and maintenance doses plus the standard of care for COVID-19. Primary outcomes to evaluate include change in clinical condition (according to the semiquantitative ordinal scale suggested by WHO), requirement for invasive mechanical ventilation/intensive care unit, delta in the score for the Sequential Organ Failure Assessment, length of hospital stay, all-cause mortality, and change in prognostic biomarkers. The evaluation of these outcomes will be carried out in a time frame up to 30 days, and hopefully, their results will be available soon. **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. None. Written informed consent for publication of their clinical details and clinical images was obtained from the patient.

Introduction {#Sec1} ============ In Europe, on average 70% of total expenditure on dental care is private expenditure and 16% of this private expenditure is covered by complementary dental insurance (CDI) \[[@CR1]\]. However, most CDI products currently on the market in Belgium, France, Germany, and the Netherlands are not optimal because (1) they provide little protection against financial risk and do not improve access to otherwise unaffordable dental treatment (e.g., implants and crowns) and (2) moral hazard and adverse selection are not sufficiently counteracted. The suboptimal character of CDI can be explained by supply-side aspects (the limits of insurability) and demand-side aspects (behavioral economics). On the basis of these potential explanations, strategies will be drawn to optimize dental insurance. We begin by presenting a framework for optimal insurance design, as well as the current situation of CDI in Belgium, France, Germany, and the Netherlands. Optimal health insurance design {#Sec2} =============================== Health insurance has two important advantages for the consumer, but also two disadvantages (see Table [1](#Tab1){ref-type="table"} below). On the one hand, health insurance reduces the financial risk for the insured and provides access to health care that would otherwise be unaffordable \[[@CR2]\]. On the other hand, insurance increases costs due to loading costs---the administrative and other expenses of the insurer---and moral hazard. In relation to dental care, perhaps more so than in relation to other types of care, a choice is possible between cheaper basic treatments and more expensive 'luxury' treatments (e.g., placing a metal crown versus a porcelain crown), which may result in substantial moral hazard (both consumer- and supplier-induced moral hazard).Table 1Advantages and disadvantages of health insuranceAdvantagesDisadvantagesReduction of financial risk for the insuredLoading costsAccess to health care that would otherwise be unaffordableMoral hazard Optimal health insurance should maintain the advantages and reduce the disadvantages as much as possible. First, it should reduce the insured's financial risk as far as possible. Because trivial risks lead to losses that can be borne by the insured without any noticeable burden, optimal insurance should not provide coverage for trivial risks. This avoids relatively high administrative expenses and loss settlement costs (loading costs) that are very high for these risks compared with the pure risk premium. Secondly, optimal insurance should provide 'access to health care that would otherwise be unaffordable'. This implies that cover limits should be avoided and expensive care should be covered as much as possible. Third, the optimality of CDI can be increased by restricting the loading costs of dental insurance by increasing insurers' efficiency. Fourth, to reduce moral hazard, optimal insurance should involve cost-sharing arrangements such as deductibles and co-insurance, and apply managed care. A deductible makes the enrollee responsible for all costs up to a defined threshold. A co-insurance rate makes the enrollee responsible for a percentage of costs. Optimal insurance contracts should also have a stop-loss, a limit on out-of-pocket expenses \[[@CR3]\]. Fifth, optimal insurance should also counteract adverse selection as much as possible. Adverse selection occurs when the insured knows more information about his expected losses than the insurer knows or uses in his premium setting and underwriting process. Adverse selection can be counteracted by selective underwriting (using medical questionnaires), risk rating (setting higher premiums for groups presenting high risk, e.g., age-related) and product differentiation (designing benefits so to attract lower risks). The features of optimal health insurance are summarized in the first column of Table [4](#Tab4){ref-type="table"}. Complementary dental insurance in Belgium, France, Germany, and the Netherlands {#Sec3} =============================================================================== Complementary dental insurance (CDI) provides coverage for care that is either not covered or not fully covered by the mandatory basic insurance (MBI). In the four countries studied, private expenditure on dental care (i.e., not covered by MBI) expressed as a percentage of total expenditure on dental care ranges between 30% in Germany and 74% in the Netherlands (Table [2](#Tab2){ref-type="table"}). CDI represents 3% (Belgium) to 55% (Netherlands) of total expenditure on dental care. Out-of-pocket expenditure on dental care is the highest in Belgium (42%) and the lowest in the Netherlands (19%).Table 2Dental care: expenditure and insuranceBelgiumFranceGermanyNetherlandsAverage expenditure per person on dental care€150€160€314€179Private expenditure (i.e., not covered by MBI)^a^45%65%30%74%Complementary dental insurance (CDI)^a^3%^b^43%5.4%55%Percentage of population with CDI5%95%18%62%Source: \[[@CR1]\]^a^As share of total expenditure on dental care^b^Authors' estimate The coverage provided by CDI is complementary to that provided by MBI, which covers 26% (the Netherlands) to 70% (Germany) of total expenditure on dental care. Table [3](#Tab3){ref-type="table"} provides an overview of the dental care covered by MBI in the four countries.Table 3Dental care covered by mandatory basic health insuranceBelgiumFranceGermanyNetherlandsConservative care (e.g., fillings)+++^a^+++^a^++++^b^Orthodontics (e.g., braces)++++−Prosthetics (e.g., implants, bridges, crowns)−++−Periodontics (gum disease treatment)++++−+++: good coverage++: medium coverage+: low coverage−: no coverage^a^Extra billing is possible^b^Conservative dental care is covered for children only (under age 18) No optimal dental insurance {#Sec4} --------------------------- Currently, CDI offered in Belgium, France, Germany, and the Netherlands cannot be described as 'optimal' (Table [4](#Tab4){ref-type="table"}). German CDI responds best to the criteria of optimal health insurance.Table 4Complementary dental insurance: presence of features of optimal health insurance designBelgiumFranceGermanyNetherlandsNo upper limit on coverage−−+−No coverage of trivial risks−−−−Deductible−−−−Co-insurance+−++Cap on out-of-pocket expenses−−−−Selective underwriting−^a^−^a^+−^a^Risk rating++++Product differentiation++++Managed care−+++−: The feature is not present in CDI products offered+: The feature is present in CDI products offered^a^Selective underwriting is applied only for a limited number of contracts, i.e., those with the highest upper limits on coverage Only in Germany are there no upper limits on coverage for dental care (after an initial period). CDI in the other three countries does not provide access to otherwise unaffordable health services and protection against unpredictable high financial risks. For example, replacing four teeth by implants and crowns can cost about €10,000. Upper limits of only €250 (the Netherlands) or €1000 (Belgium) do not provide protection against high financial risks nor do they make this kind of dental care more accessible. In France, even the most extensive complementary covers provide a maximum amount of only €750 per year for implants. With total costs for an implant easily amounting to about €2500, €1750 still needs to be paid for out-of-pocket after complementary insurance has kicked in. In all four countries, CDI provides coverage of trivial risks. In all four countries, cost sharing is applied, but only in the form of co-insurance. Co-insurance rates vary between 0 and 50% in Belgium, 0 and 55% in Germany, and 0 and 25% in the Netherlands. In France, co-insurance is generally not used. In Germany and the Netherlands, many products are offered with 0% co-insurance. Deductibles are not used. Caps on out-of-pocket expenses, which protect the consumer against high financial risk, are not applied in any of the four countries studied. Selective underwriting is primarily used in Germany, and to a lesser extent in the other three countries. In Belgium, selective underwriting is used by only one insurer offering CDI. In France, where a 7% tax has to be paid for contracts that apply selective underwriting, most CDI contracts abstain from selective underwriting. In Germany, a medical questionnaire needs to be filled out for most CDI products. The insurer can decline to cover the candidate or charge an additional premium or exclude missing teeth or the use of certain techniques from the scope of coverage. Insurance products without selective underwriting usually have contractual clauses excluding reimbursement for problems that existed well before the start of the contract and for treatments running at the moment of the conclusion of the contract ('pre-existing conditions'). In the Netherlands, selective underwriting is rarely applied (only for the high-end dental coverage). In all four countries, risk rating and product differentiation are used to a certain extent (e.g., age-related premiums). Products are designed and marketed to attract certain market segments. In all countries except Belgium, preferred provider networks are used as an element of managed care for CDI. In all countries, waiting times are used as a means to contain costs and to counteract adverse selection. Why is dental insurance suboptimal? {#Sec5} =================================== Both supply-side aspects (the limits of insurability) and demand-side aspects (behavioral economics) may explain why dental insurance is not optimal. Limits of insurability {#Sec6} ---------------------- A potential explanation of why insurers offer suboptimal CDI is that optimal CDI exceeds the limits of insurability. According to Berliner \[[@CR4]\], risks properly belong in the area of insurability where: (1) losses occur with a high degree of randomness; (2) the maximum possible loss for the insurer is limited; (3) the average loss amount upon loss occurrence is small; (4) the average time interval between two loss occurrences is small (i.e., losses occur frequently); (5) the insurance premium is sufficiently high; (6) there is virtually no possibility of moral hazard; (7) coverage of the risk is consistent with public policy; and (8) the law permits the cover. Public policy (7), moral hazard (6), the degree of randomness of losses (1) and the maximum possible loss (2) are important issues as far as the suboptimality of CDI is concerned. Public policy plays an important role in Belgium, France, and the Netherlands, where the 'solidarity' principle is paramount in health care financing. Equal access to health insurance is at the heart of the values of the mutual insurers ('mutuelles') established in those countries. Their goal is to organize solidarity between their members for the reimbursement of health care costs. According to the French 'Code de la mutualité', medical questionnaires may not be used by mutuals and thus selective underwriting cannot be applied. The solidarity idea is not restricted to mutuals. In France, for instance, a 7% tax is due when selective underwriting is applied. So, commercial insurance companies are encouraged by the French government not to apply selective underwriting. In Belgium as well, the government intervenes in the organization of CDI. Premium increases for existing clients are strictly regulated. Premiums can only be adjusted in line with the consumer price index or a specific 'medical index' for dental care, which is calculated annually by the Ministry of Economic Affairs. Fear of reputational damage and the wish to avoid (further) restrictive regulation being adopted may help to explain why (commercial) insurance companies in Belgium, France, and the Netherlands refrain from applying 'hard insurance logic' such as selective underwriting. By offering limited coverage, which is equally accessible for all citizens, insurers willingly refrain from applying private insurance logic. Rather, they apply 'social security mechanisms' (open enrolment, no selective underwriting, community rating). Insurance companies may be concerned that the unfettered application of insurance logic could provoke a reaction from the regulator. In a market where it is impossible or very difficult for mutuals to engage in selective underwriting, commercial insurance companies applying selective underwriting could easily be accused of 'cherry-picking'. By sticking to social security-type mechanisms, insurance companies err on the safe side. However, insurers' reluctance to apply private insurance logic (such as selective underwriting) and reliance on other methods (such as setting upper limits on coverage) lead to the development of suboptimal CDI products. Moral hazard is an important problem due to the very nature of dental care. Choices among different treatment options are strongly influenced by individual consumer preferences, where aesthetic aspects often play a role. Dentists also have their preferences, which can be influenced by the consumer's and insurer's 'willingness to pay'. Dental insurance can thus provide fertile ground for both consumer- and provider-induced moral hazard. Therefore, in an optimally designed scheme, insurers ought to fully invest in countermeasures. However, classic private insurance measures such as deductibles and co-insurance are not or are not fully implemented due to public policy concerns. Rather, insurers prefer to use other measures such as offering restricted coverage, setting upper limits on coverage, and not applying a cap on out-of-pocket expenditure. However, this contributes to the development of suboptimal CDI products. The degree of randomness of losses varies between total randomness and absolute predictability. Pre-existing conditions come close to being absolutely predictable. In Belgium, France, and the Netherlands, CDI covers pre-existing conditions. Consequently, adverse selection is inadequately counteracted and a vicious circle arises whereby the insurer needs to repeatedly increase premiums to be able to continue coverage for the high risks that have subscribed the insurance policy. This leads to the development of suboptimal CDI products \[[@CR5]\]. By comparison, in Germany, CDI is effectively limited to future, unforeseen events. It may not be a coincidence that CDI in Germany generally provides unlimited coverage, in contrast to the situation in Belgium, France, and the Netherlands. The 'maximum possible loss' is also a potential explanation for the sub-optimality of CDI. Certain dental treatments, i.e., prosthetic treatments such as the replacement of multiple teeth, constitute a risk with a relatively large loss amount (more than €10,000) and a low loss frequency. Such risks can only be made insurable if the insurer is given the opportunity to build up long-term loss reserves from its premium income. However, CDI contracts can be cancelled by the insured every year. Uncertainty about the duration of the contract together with moral hazard and adverse selection may lead insurers to limit coverage, resulting in suboptimal dental insurance. Behavioral economics {#Sec7} -------------------- Van Winssen et al. \[[@CR6]\] explored potential explanations of why individuals choose suboptimal complementary health insurance. Based on key insights from behavioral economics, they discuss several factors that can have an impact on the high uptake of suboptimal insurance by consumers. For CDI, the following factors are relevant. Factors such as liquidity constraints and debt aversion may help to explain why people buy dental insurance products that provide only limited coverage. Liquidity constraints imply that individuals do not have the means to free up (substantial) funds at a given point in time. Debt aversion stems from mental accounting theory \[[@CR7]\] and is illustrated by individuals' preference to prepay for consumption and to get paid for work after completion. Ignorance and social comparison may also affect individuals' willingness to purchase suboptimal insurance. People often do not know exactly what they are insuring themselves against by taking out complementary dental insurance (see, e.g., \[[@CR8]\]) and they often do not know the costs of dental care that is (not) covered by their insurance. They often rely on what their peers decide \[[@CR9]\]. Strategies for optimizing complementary dental insurance {#Sec8} ======================================================== In many countries, MBI does not cover certain types of dental care such as prosthetic treatments (e.g., crowns, implants, and bridges) or provides only limited coverage (see, e.g., Table [3](#Tab3){ref-type="table"}). With 70% of total expenditure on dental care being privately financed in Europe, CDI can play an essential role in the affordability and accessibility of dental care. Therefore, it is important for CDI products to respond as closely as possible to the features of optimal insurance design. However, currently, many CDI products on the market are suboptimal. The gap with optimal insurance design can be explained by both supply-side aspects and demand-side aspects. From these potential explanations, the following strategies to optimize CDI can be drawn. First, public policy would like voluntary CDI products to provide both optimal insurance coverage and equal access to insurance. However, this is not possible because optimal insurance requires selective underwriting and risk rating (to counteract adverse selection to protect existing clients against free riders who abuse the insurance system), which is inconsistent with the principle of equal access. Therefore, policymakers should carefully decide which types of dental care are essential and ought to be covered by MBI. Dental care that is considered non-essential by policymakers and that is therefore not covered by MBI should be subject to private insurance logic. If, because of budgetary constraints, essential dental care cannot be covered by MBI, subsidization of private insurance for persons with low incomes might be an alternative to full public provision. Second, moral hazard could be counteracted by the systematic use of deductibles and co-insurance. Standard lists of usual market prices could be compiled (as, e.g., in the Netherlands and France) and provider networks adhering to a price list could be created. Insurers should not shy away from legal action in case of excessive amounts being claimed (i.e., excessive extra billing). Third, selective underwriting and risk rating could be used to counteract adverse selection and to protect existing clients against free riders who abuse the insurance system. Providing insurance for pre-existing conditions is incompatible with the insurance principle that only future, unforeseen risks can be covered: a burning house cannot be insured. Fourth, applying waiting times for expensive treatments such as prosthetics and providing only limited coverage during the initial years of the contract constitute alternatives to a general limitation of coverage. For instance, in Germany, limited coverage typically applies during the first 4 years of the insurance contract. Fifth, behavioral economics aspects such as liquidity constraints and debt aversion could be dealt with by offering a combination of optimal dental insurance in combination with a health (dental) savings account. The dental savings account could be used to finance trivial costs. In this way, CDI could be optimized and would not be tainted by attempts to also cover trivial risks. Ignorance and social comparison can be taken care of by improving the transparency of CDI products. Consumer organizations can play an important role in clarifying the market offer for the consumer. The authors wish to thank Richard Burton and Seraf Van Den Bosch for their corrections and suggestions.

Background ========== The post thrombotic syndrome ---------------------------- The post thrombotic syndrome (PTS) (sometimes called \'post-phlebitic syndrome\') is a chronic condition that develops in 20--40% of patients within 1--2 years after symptomatic deep venous thrombosis (DVT). A severe form, which can include venous ulcers, affects 1/4 to 1/3 of patients with PTS \[[@B1]\]. In contrast to research gains in the areas of diagnosis, prevention and treatment of acute venous thromboembolism (VTE), PTS has been understudied and as a result, treatment options are limited. Patients with PTS experience pain, heaviness, swelling, or other symptoms in the affected limb, which are typically aggravated by standing or walking and improve with rest and recumbency. Edema, venous ectasia, hyperpigmentation, eczema, and varicose collateral veins may be apparent. In severe cases, ulceration can occur \[[@B2]\]. Since PTS is a direct consequence of DVT, its prevalence is influenced by the incidence of DVT. Despite advances in VTE prevention and treatment, the annual incidence of VTE has not decreased over time, and remains at 1.0--1.6 per 1000 persons per year, with a per-person lifetime incidence of 2--5% \[[@B3],[@B4]\]. Because of its prevalence and chronicity, PTS is expensive, both in terms of direct medical costs and indirect costs such as loss of productivity and thus it is burdensome to patients and society \[[@B5]\]. Risk factors for the development of PTS --------------------------------------- While hereditary and acquired risk factors that predispose to the development of VTE are widely known \[[@B6]\], factors that influence the development of PTS after DVT have not been well elucidated. The only clearly identified clinical risk factor for PTS is recurrent, ipsilateral DVT \[[@B7],[@B8]\]. The site (proximal vs. distal), size and occlusiveness of the initial DVT and the intensity (i.e. target INR) of long term anticoagulation for DVT do not appear to reliably predict PTS \[[@B1],[@B7]-[@B10]\]. It is not clear why some patients develop PTS while others recover from their DVT. As a result, physicians are unable to provide their DVT patients with reliable, individualized prognostic information. Potential role of inflammation, d-Dimer and thrombophilia in the post-thrombotic syndrome ----------------------------------------------------------------------------------------- While the pathophysiology of PTS is incompletely understood, it is likely that the acute thrombus itself, associated mediators of inflammation, and the process of vein recanalization in the weeks following DVT induce damage to venous valves, leading to valvular incompetence (reflux). Valvular incompetence and/or persistent venous obstruction by thrombus cause venous hypertension, which promotes capillary leakage of plasma proteins, erythrocytes and leukocytes and the development of venous ectasia and varicosities. The result is edema, tissue hypoxia, and ultimately, in some cases, skin ulceration \[[@B11]-[@B14]\]. Inflammation and thrombosis are closely interrelated \[[@B15]-[@B17]\]. It has been appreciated over the last few years that arterial atherothrombosis is, at least in part, an inflammatory disease, and that elevations in markers of inflammation, such as C-reactive protein (CRP), increase the risk of future clinical cardiovascular events \[[@B18]\]. Clinically, patients with DVT exhibit cardinal signs of inflammation such as redness, warmth, swelling, pain and fever. Several clinical studies have examined the association between levels of inflammatory markers and venous thrombosis, and found that a two- to six-fold increase in the risk of DVT associated with elevations in plasma levels of CRP, interleukin (IL)-6, IL-8, monocyte chemotactic protein (MCP)-1 or tumor necrosis factor (TNF)-alpha \[[@B19]\]. A recent paper showed that CRP is elevated in patients with acute DVT (median 37.5 mg/L) compared with controls (5.0 pg/L), and that levels decline during the first 5 days of DVT treatment. Similar trends were noted for IL-6 \[[@B20]\], leading the authors to conclude that the thrombotic process produces a systemic inflammatory response and to speculate whether the observed decrease in levels was partly caused by treatment with heparin, which is known to have anti-inflammatory properties distinct from its anticoagulant properties \[[@B21]\]. Whether levels of markers of inflammation are predictive of PTS has not previously been examined. D-dimer is a degradation product of cross-linked fibrin that reflects fibrinolysis and is an indirect marker of coagulation activation. Recent work performed by our group and by others suggests that D-dimer levels appear to be a significant predictor of first VTE \[[@B22]\] and of recurrent VTE \[[@B23]-[@B28]\], whether measured during or after anticoagulation. Independent of VTE recurrence, persistent coagulation activation, as reflected by elevated D-dimer levels, may also predict a higher risk of developing PTS after DVT \[[@B29]\]. Thrombophilia refers to an inherited or acquired predisposition to VTE. Over 10% of the general population is affected by one or more identifiable inherited thrombophilias which have been shown to underlie at least 1/3 of cases of VTE \[[@B30]\]. The most common inherited thrombophilias are the single nucleotide polymorphisms Factor V Leiden, which renders coagulation factor V resistant to the anticoagulant effects of activated protein C \[[@B31]\], and Prothrombin G20210A, which leads to elevated plasma prothrombin levels \[[@B32],[@B33]\]. Recent studies have also established that elevation of Factor VIII level is an inherited risk factor both for first and for recurrent VTE \[[@B34]-[@B36]\], independent of age, sex, Factor V Leiden, Prothrombin G20210A or markers of acute phase activation such as CRP \[[@B35],[@B37],[@B38]\]. While results of a few studies to date suggest that thrombophilia does not increase the risk of developing PTS \[[@B7],[@B8],[@B10]\], not all thrombophilias have been examined and interactions among disorders have not been studied. Further investigation of the potential link between thrombophilia and PTS is warranted. Management of PTS ----------------- PTS could be averted by primary prevention of the initial DVT with the judicious use of thromboprophylaxis \[[@B39]\], and by preventing recurrent ipsilateral DVT by prescribing adequate anticoagulation for the initial DVT \[[@B40]\]. However, at least 1/2 of all cases of DVT occur unpredictably, hence are not preventable \[[@B41],[@B42]\]. There is no definitive evidence that using thrombolysis to treat DVT reduces the incidence of PTS \[[@B43]\]. The treatment of established PTS is limited and frustrating for patients. Severe PTS can be managed with long-term use of an intermittent compression extremity pump \[[@B44],[@B45]\]. Management of venous ulcers is labor intensive and protracted, and involves compression therapy, leg elevation, topical dressings, and sometimes surgery \[[@B2],[@B46]\]. Ulcers are often recalcitrant to treatment, and tend to recur \[[@B47]\]. A significant reduction in the overall burden of PTS is unlikely to be achieved by attempts to prevent the initial DVT or by treatment of established PTS. Rather, strategies that focus on preventing the development of PTS in patients with DVT are more likely to be effective and cost-effective in reducing the patient and societal impact of PTS. Elastic compression stockings for the prevention of PTS ------------------------------------------------------- Graduated elastic compression stockings (ECS) work by providing graded compression to the leg that is highest at the ankle, which assists the calf muscle pump, reduces venous hypertension and valvular reflux, and consequently reduces edema, improves tissue microcirculation, and prevents skin breakdown \[[@B48],[@B49]\]. Both knee-length and thigh-length stockings appear to have equal physiological effects, but knee-length ECS are easier to apply and are more comfortable \[[@B50]\]. The effectiveness of daily use of ECS to prevent PTS is supported by two studies \[[@B8],[@B51]\] but challenged by another \[[@B52]\]. All three studies have limitations that could affect their validity and generalizability. The first trial, Brandjes\' study of 194 patients with symptomatic proximal DVT, provides evidence supporting the effectiveness of ECS. Patients were randomized to daily use of custom-made, knee-length ECS applied within 2--3 weeks of diagnosis for at least 2 years (class II compression, i.e. 30--40 mm Hg pressure at the ankle), or no stocking. Use of ECS resulted in a 50% reduction in the incidence of PTS, diagnosed using a modification of Villalta\'s clinical PTS scale \[[@B53]\], or an absolute decrease from 47% to 20% of mild/moderate PTS and from 23% to 11% of severe PTS \[[@B51]\]. In contrast, a randomized trial conducted by Ginsberg suggested that ECS were not of benefit in preventing PTS \[[@B52]\]. A strength of Ginsberg\'s study was the use of a control comparison group that wore sham stockings (i.e. stockings that were 1--2 sizes too big to be effective). However, because of the small number of patients with PTS, benefit or harm of up to 30% could not be definitively excluded. Recently, Prandoni published the results of a trial performed at a single center in Italy to evaluate the effectiveness of ECS to prevent PTS \[[@B8]\]. Among 180 patients with proximal DVT, those randomized to wear daily ECS had a 50% reduction in the rate of PTS after a 2 year period, compared with controls. This study, like Brandjes\', lacked a placebo control, an important limitation due to the subjective nature of many of the components of the standardized scale that was used to diagnose PTS \[[@B54]\]. As the Prandoni study was the first, single positive study of \"off-the-rack\" elastic compression stockings, it requires replication. Also, as the study was conducted entirely at a single center in Italy, further evaluation of the generalizability of this data to North American clinical practice in a multi-center study is needed. In light of the above, we believe that a large scale, randomized placebo-controlled trial of ECS to prevent PTS is needed to provide definitive evidence of effectiveness, or lack of effectiveness, of ECS. This will allow physicians to make informed, evidence-based decisions regarding their use in DVT patients. Furthermore, such a trial will permit prospective evaluation of the predictive role of markers of inflammation, d-Dimer and thrombophilia in PTS. Methods/Design ============== The SOX Trial is a Canadian, multicenter, randomized double-blind controlled trial in patients with a first episode of proximal DVT. Aims of the study ----------------- The primary aim of the study is to evaluate whether graduated elastic compression stockings compared to inactive (placebo) stockings, worn daily for 2 years after DVT is diagnosed, decrease the incidence of PTS. Secondary aims are 1) to evaluate whether active compared with inactive stockings reduce the severity of PTS, 2) to compare mean quality of life (QoL) scores during follow-up in the active intervention vs. control groups, 3) to describe the rates of recurrent DVT, death from VTE, venous ulcers, and major bleeding during the 2-year follow-up in the active intervention vs. control groups, 4) to evaluate the cost-effectiveness of ECS for the prevention of PTS, and 5) to determine whether markers of inflammation, D-dimer and thrombophilia influence the development and severity of PTS (Bio-SOX biomarker substudy). Trial design and description of the intervention ------------------------------------------------ ### Original trial design The study was initially designed as a factorial design randomized clinical trial whose primary aims were to determine whether (1) elastic compression stockings used for 2 years compared to inactive (placebo) stockings, and (2) celecoxib, a COX-II inhibitor, used for 30 days compared to placebo, were effective in preventing PTS in patients with symptomatic proximal DVT. The rationale for the use of a COX-II inhibitor was to assess whether an anti-inflammatory drug used acutely could limit the extent of venous valvular damage and reduce the frequency of subsequent PTS. However, in December 2004, The National Institutes of Health announced suspension of the use of celecoxib for all participants in a large colorectal cancer prevention clinical trial conducted by the National Cancer Institute, because analysis by an independent Data Safety and Monitoring Board (DSMB) showed a 2.5-fold increased risk of major fatal and non-fatal cardiovascular events for participants taking the drug compared to those on a placebo. As a result of this announcement, the SOX Trial Steering Committee decided on December 21, 2004 to discontinue the Celecoxib intervention of the SOX Trial \[[@B55]\]. This decision was made after independent meetings of the Steering Committee and the SOX Trial DSMB and was based on patient safety and also on study feasibility. The SOX Trial thus resumed in February 2005 as a parallel group randomized trial, i.e. without the celecoxib/placebo intervention. ### Current trial design The study is a randomized, allocation concealed, double-blind multicenter clinical trial with an intervention allocation ratio of 1:1 (figure [1](#F1){ref-type="fig"} summarizes the study design). Once a patient is deemed eligible and consents to the study, the research nurse at each site logs onto a web-based interface which allocates the patient to one of 2 treatment groups: active stockings or inactive (placebo) stockings. Treatment group is assigned randomly using permuted blocks of randomly varying sizes at each study site to maintain close balance of the numbers in each treatment group, at any time during the trial, and to ensure allocation concealment. The active stockings intervention consists of knee-length, 30--40 mm Hg (Class II), graduated ECS worn on the DVT-affected leg daily, applied upon waking and removed upon retiring, beginning as early as possible within 14 days after DVT diagnosis and continued for 2 years. The inactive stockings intervention (the control intervention) consists of knee-length, inactive stocking (5 mm Hg compression at ankle; similar to store bought trouser sock), identical in appearance to the active stockings, worn on the DVT-affected leg daily, applied upon waking and removed upon retiring, beginning as early as possible within 14 days weeks after DVT diagnosis, and continued for 2 years. Active and inactive stockings are manufactured by SIGVARIS Corp. {#F1} ### Patient eligibility criteria Patients presenting with a first, symptomatic, objectively confirmed proximal DVT diagnosed within the last 10 days (with or without concurrent distal DVT or pulmonary embolism), who have no contraindications to standard anticoagulant therapy and who provide informed consent are eligible for the study. Patients are excluded if they 1) have a contraindication to compression stockings e.g. previously documented moderate to severe peripheral arterial disease of the lower extremities, absence of palpable pedal pulses or arterial compromise due to massive venous obstruction; 2) have a limited lifespan (estimated \< 6 months); 3) have geographic inaccessibility preventing their return for follow-up visits; 4) demonstrate inability to apply stockings daily (e.g. severe arthritis, arm paralysis, and unavailability of a caregiver to apply stockings daily); or 5) were prescribed lytic therapy to treat their acute DVT. ### Study procedures Once the patient is recruited and randomized via the SOX Trial\'s web-based interface, the computer generates a unique treatment code for each patient, according to treatment group and center. The study nurse faxes the patient\'s leg measurements to the stocking supplier, who then ships the assigned pair of ECS to the patient (i.e. correct size, according to patient\'s measurements; active or inactive stocking, according to patient\'s treatment code). ECS are pre-packaged by the manufacturer in a plain box, labeled with the unique stocking code (i.e. patient and health personnel blinded). The same procedure is used to replace the patient\'s ECS every 6 months during the trial. Several strategies are used to protect against bias, including: randomization with allocation concealment, enrolling consecutive patients, blinding subjects to treatment assignment, blinding investigators, study nurses and outcome assessors (e.g. vascular technicians, radiologists) to treatment assignment, strict inclusion and exclusion criteria, and use of validated measures to diagnose PTS, recurrent VTE and to measure QoL. To prevent unblinding of study nurses, patients are instructed not to wear ECS on the day of their study visits. While it is possible that patients may be able to distinguish active from inactive ECS, this is minimized by (1) using inactive ECS, developed by the manufacturer, that appear identical to active ECS, and (2) recruiting patients with a first DVT, i.e. likely to be \"ECS-naïve\". In addition to such pre-trial design features, we will assess the blinding procedures by asking patients, investigators and research nurses at the end of the trial (i.e. at 2 years or at time of study termination) if they were aware of the treatment assignment. Study patients are followed for 2 years after enrollment. The procedures at each visit are shown in Table [1](#T1){ref-type="table"}. At any time during the study, if a patient has symptoms or signs suggestive of recurrent VTE, he contacts the study team for medical assessment, at which time recurrent VTE will be ruled in or ruled out. Similarly, if an adverse event occurs, the patient contacts the study team for evaluation. If a patient withdraws from the study, a Study Termination form is completed to account for withdrawal, which will document the reason(s) for withdrawal. If a patient dies, a Death Form is completed which documents the cause(s) of death, further classified as VTE likely or unlikely. Throughout study follow-up, co-interventions are recorded (e.g. diuretics, analgesics, exercise). While not encouraged, any temporary use of elastic bandages/wraps or active stockings during the acute phase of DVT is systematically tracked. ###### Details of study visits and telephone calls -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ***Enrollment visit*** • The nurse completes the baseline form (inclusion/exclusion criteria, demographic and clinical variables, results of objective tests for DVT), PTS scale and pain scale • The patient self-completes the QOL form • The nurse performs the leg measurements to size the stockings • The nurse teaches patients how to apply (using a plastic leg model) and care for stockings (handling, washing), and reinforces the importance of daily use • Study stockings are supplied to the patient (within a few days after enrollment) • A venous blood sample is taken for markers of inflammation **2-week telephone call** • The nurse reinforces the importance of wearing stockings daily • Bleeding/Other Adverse Event form and pain scale are completed For the following visits, the patient does not wear study stocking to prevent unblinding of study nurse: **1-month visit** • The nurse completes the follow-up form (includes data on symptoms and signs of PTS, suspected recurrent VTE, compliance with stockings, pain scale) • The patient self-completes the QOL form • Bleeding/Other Serious Adverse Event form is completed • A venous blood sample is taken for markers of inflammation and genetic thrombophilia • The nurse reinforces the importance of wearing stockings daily **2-month telephone call** • Same as the 2-week telephone call **6-month visit** • The nurse completes the follow-up form • The patient self-completes the QOL form • Bleeding/Other Serious Adverse Event form is completed • Leg measurements are taken, and a new pair of stockings is supplied • A venous blood sample is taken for markers of inflammation, d-dimer and coagulation-based assays • The nurse reinforces the importance of wearing stockings daily **1-year visit** • Same as 6-month visit except no blood sampling • A venous ultrasound is performed to assess reflux **18-month visit** • Same as 6-month visit except no blood sampling **2-year visit** • The nurse completes the follow-up form • The patient self-completes the QOL form • Bleeding/Other Serious Adverse Event form is completed • The success of blinding (patient, nurse, and investigator) is assessed (or at Study Termination if earlier than 2 years). -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Primary and secondary outcome measures -------------------------------------- The primary outcome measure is the incidence of PTS during 2 year follow up. PTS will be diagnosed at the 6 month visit or later if the patient reports pain and swelling of ≥ 1 month\'s duration that is typical in character (worse at the end of the day or with prolonged sitting or standing, and better after a night\'s rest and leg elevation), and that is present ≥ 6 months after the acute DVT \[[@B52]\]. Patients diagnosed with PTS will be considered to have reached the study endpoint. Severity of PTS will be graded according to the Villalta scale \[[@B53]\], a clinical post-thrombotic syndrome measure that rates the severity of five patient-rated symptoms and six clinician-rated clinical signs \[[@B53]\]. A Villalta score of 5--9 represents mild post-thrombotic syndrome; 10--14, moderate post-thrombotic syndrome; and \> 15 or presence of a venous ulcer, severe post-thrombotic syndrome \[[@B53]\]. Secondary outcomes measures include: • Reflux: Reflux will be assessed and quantified in a standardized fashion by venous ultrasound \[[@B52],[@B56]\]. • Recurrent VTE: Objective tests and validated algorithms will be used to diagnose recurrent DVT and PE, as described previously \[[@B57]-[@B59]\], and all events and deaths will be independently and blindly adjudicated. Death from VTE will be assessed as \'likely\' or \'unlikely\' after review of the death certificate, patient chart, autopsy report (where available) and by contacting the treating physician. • Bleeding: Clinically suspected bleeding events will be documented by recording the clinical event and results of any objective diagnostic testing performed. Bleeding will be defined as major if it is clinically overt and associated with a fall in hemoglobin of 20 g/L or a need for transfusion of ≥ 2 units of red blood cells; if it was intracranial or retroperitoneal; or if it warranted the permanent discontinuation of anticoagulation. Less severe clinically overt bleeding which is considered abnormal will be classified as minor. • Markers of inflammation: C-reactive protein will be assayed using a nephelometric assay. IL-6, IL-8 and MCP-1 will be measured using commercial ELISA kits which include plasma controls (Biosource Europe SA, Nivelles, Belgium). At the time of blood draw for markers of inflammation, data will be recorded on time of day, type and duration of heparin use, concomitant medications and medical conditions. • DNA polymorphisms associated with thrombophilia: QIAGEN kits and standard protocols will be used to extract DNA from peripheral blood leukocytes. Samples will be run on the ABI Prism 3100 Genetic Analyzer to analyze Factor V Leiden, Prothrombin G20210A, PAI-1 4G/4G, TAFI T1053C, TFPI C536T, TFPI T33C and FXIII Val34Leu. • Coagulation and ELISA-based assays: Factor VIII assays will be performed using the PTT based Diagnostica Stago (Abbott Diagnostics Canada, Mississauga). D-dimer testing will be performed with the IL-Test Latex Agglutination method (Instrumentation Laboratory, Lexington, USA). Lupus anticoagulant testing will be performed on an ACL9000 (Instrumentation Laboratory, Lexington, USA). Standard aPTTs are performed using Alexin HS (Trinity Biotech USA, St. Louis, USA) and HemosIL APTT-SP (Instrumentation Laboratory, Lexington, USA) as a screening test. Abnormal results are followed by a 50:50 mix using Cryocheck Normal (Precision Biologic, Dartmouth, NS), followed by Platelet Neutralization Procedure. Additionally, a Dilute Russell Viper Venom test is performed using IL test LAC Screen followed by Confirm (Instrumentation Laboratory, Lexington, MA, USA). IgG and IgM antiphospholipid antibodies are tested usingapHL-HRP ELISA Kit (Louisville APL Diagnostics, Doraville, USA). • Quality of life measures: Quality of life will be measured using the SF-36 questionnaire for generic QOL \[[@B60],[@B61]\]., and the VEINES-QOL/Sym questionnaire for venous disease-specific quality of life, which our group developed and validated \[[@B62],[@B63]\] • Cost Effectiveness measures: A decision tree-based cost effectiveness analysis will be used to assess the relative cost-effectiveness of the active stockings in reducing the incidence of PTS, as compared to a strategy of inactive stockings \[[@B64],[@B65]\]. Tree structure will schematically reflect the clinical trajectory for each strategy \[[@B66]\]. Sensitivity analyses will be conducted to verify the robustness of the qualitative conclusions to variations in the data \[[@B67]\]. Sample size and power calculations ---------------------------------- We hypothesized that event rates (i.e. PTS at 2 years) will be 30% in the inactive stocking vs. 20% in the active stocking group, i.e. a risk reduction of 33%. Our baseline event rate and effect size for the stocking group were based on a systematic review of available published data \[[@B68]\]. The total sample size required to detect this difference in event rates with a two-tailed α of 0.05 and 80% power is 600 \[[@B69]\]. We increased our sample size to 800 to take into account a projected 25% loss-to-follow up by 2 years. For the biomarker analyses, our sample size provides \> 80% power to detect ORs of 0.5 or lower, or 2.0 or higher. Larger alternatives will be associated with increased power. With regard to inflammatory markers, the analysis will be largely exploratory, as this study will be among the first to provide prospective data on levels of markers of inflammation in patients with DVT and PTS. However, the study design provides power of at least 80% to detect differences on the order of 1/3 of a standard deviation or larger between initial values of inflammatory markers in the two allocation groups. Statistical analysis -------------------- An intention-to-treat analytical approach will be used for all outcomes. Descriptive statistics for baseline variables will be calculated to describe the baseline status of the treatment groups. For the primary outcome variable PTS, a logistic regression analysis initially adjusted only for center will be used to compare the incidence rates between the active vs. inactive stockings groups. In secondary analyses, other covariates that will be adjusted for are duration of anticoagulation with warfarin, recurrent ipsilateral DVT during follow-up, temporary use of active stockings, initial type and duration of heparin anticoagulation, co-morbid conditions that could lead to leg symptoms/signs and differences in pertinent baseline variables. The number needed to treat for benefit will be calculated. As an adjunct to the primary analysis, a secondary explanatory (per-protocol) analysis will be performed as well. Subgroup analyses by age, sex, level of compliance, duration of warfarin anticoagulation and recurrent VTE during study follow-up are also planned. For the secondary outcomes valvular reflux, recurrent VTE, death from VTE, venous ulcer, major bleeding, and severity of PTS, rates in the two intervention groups will be described, and between-group differences explored in similar regression procedures. For differences in mean quality of life scores between intervention groups, analysis of variance incorporating treatment and center will be used. A difference of 4--5 points between groups is considered to be clinically meaningful. Analysis of covariance with adjustment for age, sex and comorbidity will also be used to compare quality of life between groups. For the biomarker variables, univariate analysis will be used to determine the strength of association between each biomarker variable and the occurrence of PTS. For genetic markers, heterozygotes and homozygotes will be analyzed together, given the rarity of homozygotes. IgG, IgM and lupus anticoagulant results will be dichotomized as normal or abnormal. Factor VIII, D-dimer and markers of inflammation will be analyzed both as continuous data and as proportion of values exceeding the upper 90^th^percentile. For binary and nominal data, the appropriate chi-square tests will be used to examine the respective crude associations, and for continuous variables, the unpaired 2-tailed t-test. To examine adjusted associations, multiple logistic regression analyses will be performed in which the dependent variable is the presence of PTS. The independent variables to be tested are markers of thrombophilia, D-dimer, and levels of CRP, IL-6, IL-8 and MCP-1. Other variables of interest (for interaction and/or confounding) are age, sex, ethnicity, comorbid illness, allocation to active ECS, type and duration of heparin and warfarin use, type of DVT (unprovoked vs. secondary; anatomic extent of DVT), and recurrent DVT during follow-up. The analyses will aim to (1) identify the strength of the association between relevant biomarkers, other relevant explanatory variables and PTS; and (2) explore relevant interactions. Similarly, multiple linear regression will be performed in which the dependent variable is the severity of PTS, as measured by the Villalta scale. An interim analysis for efficacy will be performed when half of the subjects have completed 2 years of follow-up. The analysis plans call for a Lan-DeMets alpha spending approach with an O\'Brien Fleming boundary \[[@B70]\]. Data management --------------- All data are collected using standardized case report forms. Data is entered on-line at study sites using a customized web-based data entry tool, and data quality is maintained via use of validation checks at the time of data entry. Data will be reviewed and cleaned by the database coordinator on an on-going basis by initiating and following up on queries to the sites. Data management is being overseen by TrialStat^®^(Ottawa, Canada). Analysis of the cleaned database will be carried out at the coordinating center under the supervision of study principal investigators. Trial committees ---------------- The Steering Committee is chaired by the study principal investigator and includes the trial coordinator and all SOX Trial grant co-applicants. The Steering Committee meets regularly and on an as-needed basis to monitor trial progress, assess the need for changes in procedures, and address issues that could affect the integrity or projected timeframe of the trial. The Expert Adjudication Committee for all outcome events is comprised of two thrombosis experts who are not co-investigators or collaborators. Clinical details and diagnostic studies relating to suspected outcome events and deaths are reviewed by the committee, who are blinded to treatment assignment. The Independent Data Safety and Monitoring Committee consists of an experienced thrombosis physician, a biostatistician and an experienced clinical trialist. Episodes of symptomatic recurrent venous thromboembolism, the post-thrombotic syndrome, and death from all causes are reported to the Chair of the Committee on a quarterly basis. Ethical considerations ---------------------- Each patient is provided with written information about the trial and written informed consent is obtained prior to study inclusion. The study protocol has been approved by Health Canada and by the local Research Ethics Committees of the participating centres. Discussion ========== In this report, we describe the protocol of a multicenter randomized placebo controlled trial of elastic compression stockings for the prevention of post-thrombotic syndrome (The SOX Trial). This large, methodologically rigorous study is evaluating the effectiveness and cost-effectiveness of ECS to prevent PTS and has been designed to provide definitive data on the role of ECS in the prevention of PTS. As such, our results will impact directly on the care of patients with DVT. This will also be the first study to prospectively evaluate the predictive role of biomarkers that are reflective of putative underlying pathophysiological mechanisms in the development of clinical PTS (Bio-SOX biomarker substudy). Our findings will increase understanding of PTS and could result in the development of novel therapies aimed at candidate biomarkers. Competing interests =================== The author(s) declare that they have no competing interests. Authors\' contributions ======================= SRK wrote the initial protocol and designed this study. HS is the trial coordinator and co-authored the manuscript. All other co-authors contributed to study design and/or manuscript revisions. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2261/7/21/prepub> Acknowledgements ================ The SOX Trial is funded by the Canadian Institutes of Health Research (MCT\# 63142) and the BioSOX Study is funded by the Heart and Stroke Foundations of Quebec and Ontario (HSFO\#5629). SIGVARIS corp. manufactures and supplies all active and placebo stockings as in-kind support. We acknowledge the contribution of the SOX Trial site coordinators and investigators, and of Ms. Tatiana Vydykhan MSc, SOX Trial Coordinator from 2003--2006. The SOX Trial investigators and centers Quebec: Susan R. Kahn (Principal Investigator) Hadia Shbaklo (SOX Trial Coordinator) Vicky Tagalakis Stan Shapiro SMBD -- Jewish General Hospital (Trial Coordinating Center) Mira Johri Université de Montréal Isabelle Chagnon Hôpital du Sacré-Coeur de Montréal Christine Demers CHA Hôpital de l\'Enfant-Jésus Sylvie Desmarais Centre Hospitalier Pierre-Boucher Jeannine Kassis Hôpital Maisonneuve-Rosemont Marie-José Miron CHUM Hôpital Notre-Dame Lucie Opatrny Jacques Genest (Biomarker analyses) Royal Victoria Hospital Susan Solymoss Montreal General Hospital St. Mary\'s Hospital Center Ontario: Jeffrey S. Ginsberg McMaster University Medical Centre HHS -- Chedoke Division Rajendar Hanmiah St. Joseph\'s Healthcare Clive Kearon Henderson General Hospital Michael J. Kovacs London Health Sciences Centre Sam Schulman HHS General Hospital Rita Selby Sunnybrook & Women\'s College Health Sciences Centre Philip S. Wells Marc A Rodger The Ottawa Hospital -- Civic Campus Erik Yeo UHN -- Toronto General Hospital British Columbia: Reginald Smith Victoria Heart Institute Foundation Manitoba: Turnly Wong St. Boniface General Hospital Nova Scotia: David Anderson QEII Health Sciences Centre

1. Introduction {#sec1} =============== Overweight and obesity correspond to an abnormal or excessive fat accumulation, defined as a Body Mass Index (BMI) ≥ 25.0 kg/m^2^ and ≥30 kg/m^2^, respectively \[[@B1]\]. These chronic situations described as epidemic should be considered as pandemic with a prevalence that has more than doubled since 1980 and which continues to increase dramatically to reach about 1.5 billion overweight or obese people worldwide \[[@B1], [@B2]\]. Overweight and obesity represent a major risk factor of ischemic heart diseases, ischemic stroke, type 2 diabetes, osteoarthritis, some cancers (endometrial, breast, and colon) \[[@B1], [@B2]\], and some of the leading diseases in terms of frequency, quality of life impairment, morbidity, mortality, and heath expenses \[[@B3]\]. Public health campaigns and industry-supported changes in our food supply have obviously failed to control the pandemic to date. Many individuals appear to conclude that the benefits of weight management strategies are not worth the cost (i.e., time, money, and continued unrewarding efforts) \[[@B4], [@B5]\]. This underlines the critical need to implement new, practical, and affordable strategies to complete the armamentarium to fight excess weight. Balneotherapy (BT), or SPA therapy, is defined as the treatment of disease by bathing, usually at a SPA resort, using hot or cold water rich in minerals, and including also drinking, inhalation, massage through moving water, mud-baths, relaxation, or stimulation \[[@B6]\]. The medical community, mainly in Europe and Asia, has long used BT to improve the symptoms of several chronic diseases, such as osteoarthritis for which recent studies show evidence-based benefit \[[@B7]--[@B9]\]. Obesity is another common indication of BT, with preliminary positive results \[[@B10]--[@B13]\]. In France, BT has been subsidized by national health insurance for more than 50 years, but scientific evidence to support long-term BT benefit on weight loss has been requested by national health authorities. Long-term weight control may be facilitated by an appropriate weight loss maintenance strategy such as motivational interviewing, with a weak but significant difference reduction in body mass compared to the control group \[[@B14]\]. Numerous reports have concluded that this modest weight loss contributes to important health benefits \[[@B15]--[@B17]\]. The objective of the present study was to assess the one-year effectiveness of a 3-week program of BT for overweight or obese patients and whether a monthly phone dietary motivational interview (DMI) during the follow-up could improve weight loss. 2. Methods {#sec2} ========== The study was a Zelen double consent randomised controlled trial \[[@B18], [@B19]\] to assess the effectiveness of BT compared to usual care (UC) with or without a monthly phone DMI during follow-up, using a 2 × 2 factorial design. After inclusion by their GP, subjects were equally allocated by a centralized randomisation to one of the following groups: BT alone, BT and DMI, UC alone, and UC and DMI. Zelen double consent was applied to BT randomisation but not to DMI intervention. BT should be done in the 2 months following inclusion. All patients were followed up by the same GP at baseline, 7 months and 14 months after inclusion (i.e., about one year after BT). 2.1. Participants {#sec2.1} ----------------- Inclusion criteria were generally healthy adults between the ages of 20 and 70 years consulting their GP for excess weight, with a BMI \> 27 kg/m^2^ and ≤35 kg/m^2^. Exclusion criteria were BT contraindication (severe general weakness, inflammatory bowel disease, cirrhosis, severe disability, psychosis and dementia, or immunodeficiency), major eating disorders (bulimia nervosa, compulsive overeating), pregnancy, previous BT for weight problems, poor proficiency, and involvement in another clinical trial. All participants signed an informed consent to participate. 2.2. Interventions {#sec2.2} ------------------ BT was a 3-week program in Brides les Bains, Capvern les Bains, Vals les Bains, Vichy, or Vittel BT resorts. The core of the program included 18 daily sessions of (i) individual mineral water bubble bathing at 37°C during 10 minutes; (ii) mineral water manual massages during 10 minutes; (iii) mud body wrap applied at 42°C during 10 minutes; (iv) mineral water pool supervised callisthenic exercises (34°C, 15 minutes); (v) daily dinking of resort mineral water. Mineral waters were mainly sulfate (Brides les bains, Capvern les bains, Vittel) and bicarbonate waters (Vals les bains, Vichy). The specific chemical composition of the five resorts\' water is reported in [Table 1](#tab1){ref-type="table"}. Dieticians and personal trainers provided nutrition and physical activity counselling. However, no particular caloric restriction or physical training was mandatory during the patient\'s stay. Meals were taken freely either at the resort restaurants or anywhere in the town, offering low-fat/low-calorie menu. This program was the usual one for overweight or obese patients, and the staff was not informed which patients were taking part in the clinical trial. Patients in the UC group received usual weight management advice from the GP consisting of verbal and/or written advice based on the French national guideline, and a brochure "Health comes when eating" \[[@B20]\] was given to the patients at inclusion. Patients were advised to reduce calories, fat, and alcohol, to increase fruit, vegetable, and whole cereal intake, and to incorporate low-intensity, long-duration physical activity into their lifestyle. DMI was a monthly half-an-hour phone interview with a dietician, consisting in an overview of the current situation, to examine the last 24 hours intake and physical activity, followed by meal composition and food behaviour advice as well as to define 3 or 4 objectives for the next month including at least one about physical activity. DMI started one month after the end of BT for patients attending BT and 3 months after inclusion for UC patients. 2.3. Data Collection {#sec2.3} -------------------- Data collected at inclusion were gender, height, birth date, marital status, professional occupation, overweight family history, medical history, weight loss objective, previous drug and nondrug overweight treatment, and drug, examination, dietary, and physical activity prescribed for overweight. Other data collected at inclusion and at the two follow-up visits were the date of visit, weight, body fat mass, waistline measurements, blood pressure and heart rate, physical activity, and tobacco and alcohol consumption. Patient quality of life was assessed using SF12 at inclusion and at the two follow-up visits. Weight was measured by GPs, as well as by a dietician at the end of BT, using the same scale provided for the study. 2.4. Outcome {#sec2.4} ------------ The primary outcome was body mass index (BMI) loss at 14 months of follow-up. Secondary outcomes were weight loss, a weight loss ≥ 5% at 14 months of follow-up and intervention tolerance. 2.5. Sample Size {#sec2.5} ---------------- A sample size of at least 139 subjects per group was required to demonstrate a 25% difference between BT and UC groups, with a two-tailed alpha risk of 5%, a power of 90%, and a BMI loss of 1.4 kg/m^2^ (±0.9 kg/m^2^) one year after BT therapy, according to the results of an open pilot study. 2.6. Statistics {#sec2.6} --------------- According to the Zelen double consent design, a patient could decline the BT or UC randomised allocation to switch to the alternative treatment \[[@B18], [@B19]\]. This occurred for about half of the patients ([Figure 1](#fig1){ref-type="fig"}). With such a complete dilution of the randomisation, it was not relevant to perform intent-to-treat (ITT) analysis. Therefore, in order to reduce a treatment-selection and potential confounding factors as in an observational study, a propensity score to attend BT or to be managed by UC was used. The main analysis compared patients attending BT to optimally matched patients managed by UC. Patients were matched by sex, overweight or obese status, DMI randomisation, and propensity score nearest neighbour with a threshold of 0.1 \[[@B21]\]. A secondary analysis compared all BT and UC patients with DMI stratification and propensity score adjustment. The baseline comparison between 2 groups used the Student\'s *t*-test and Chi-square test according to the variables. Propensity scores were estimated using a logistic model with all patient inclusion characteristics. The effect of BT versus UC, DMI versus no DMI, and the interaction between both interventions were assessed using generalized linear model for BMI and weight loss and using logistic model for weight loss ≥ 5% frequency. Adjusted mean and 95% confidence intervals \[95%CI\] were estimated using Least Squares Mean. A two-tailed *P* value \< 0.05 was considered statistically significant. Statistical analysis was performed using SAS software. The study protocol was ethically approved by the Persons Protection Committee (Paris---Eudract number 2006-A00309-42) and registered to Clinicaltrials.gov number NCT01258114. 3. Results {#sec3} ========== From March 2007 to July 2008, 71 GPs assessed 298 patients, 74 randomised to BT alone, 75 to BT with DMI, 74 to UC alone, and 75 to UC with DMI, of whom 10 were excluded for major exclusion criteria (one with a severe depression, one non-French speaking, and 8 for BMI ≥ 37.5 kg/m^2^). For the 288 remaining patients, 126 (43.8%) declined the randomised allocation to switch to the alternative treatment; 31 dropped out of the study for withdrawal of consent, pregnancy, cancer, major surgery, or other BT during the study or were lost to follow-up ([Figure 1](#fig1){ref-type="fig"}). Finally, 257 patients were followed up, 120 patients attending BT (63 randomised DMI), 137 managed by UC (67 randomised DMI, and 70 patients (36 randomised DMI) of each group could be matched (54.5%). Patients who completed the study had a mean age of 50.9 (SD 11.2) years, 79% were women, 61.1% were obese (≥30 kg/m^2^), 48.6% had a fat mass ≥ 35%, 7.8% were diabetics, 12.8% were smokers, 24.5% were alcohol consumers, 23.3% lived in a large town (≥100 000 inhabitants), 19.8% declared having regular sport activities, and 16.3% had previous treatment for obesity. There were some differences between the 120 patients who attended BT and the 137 managed by UC, whereas the 70 BT and 70 UC matched patients were remarkably similar ([Table 2](#tab2){ref-type="table"}). All patients attending BT completed the 3-week program with a significant mean weight loss of 2.98 kg (\[95%CI: 3.44; 2.52\], *P* \< 0.001) at the end of BT. Then, the benefit continued to improve to 1.47 kg (\[0.64; 2.31\], *P* = 0.001) from the end of BT to 14 months of follow-up ([Figure 2](#fig2){ref-type="fig"}). One-third (32.2%) of the patients reached a weight loss ≥ 5% at the end of BT and half (50.8%) at the end of the study. Adjusted mean BMI loss at 14 months of follow-up was 1.91 kg/m^2^ \[1.46; 2.35\] for the 70 BT patients and 0.20 kg/m^2^ \[−0.24; 0.64\] for the 70 matched UC patients (*P* \< 0.001), corresponding to a 1.71 kg/m^2^  \[1.08; 2.33\] BT benefit. The adjusted mean weight loss was 5.17 kg \[3.95; 6.39\] and 0.54 kg \[−0.68; 1.76\], respectively (*P* \< 0.001), with a BT benefit of 4.63 kg \[2.91; 6.35\] ([Figure 2](#fig2){ref-type="fig"}). A weight loss ≥ 5% was reached by 40 patients of the BT group (57.1%) and 13 of the UC group (18.6%), with odds ratio of 5.9 (\[2.7; 12.8\], *P* \< 0.001). For the whole population, BMI and weight adjusted mean loss at 14 months of follow-up were 1.71 kg/m^2^  \[1.33; 2.09\] and 4.64 kg \[3.60; 5.68\] for the 120 BT patients compared to 0.50 kg/m^2^  \[0.15; 0.85\] and 1.32 kg \[0.36; 2.28\] for the 137 UC patients (*P* \< 0.001 for both comparisons). The adjusted mean BT benefit was 1.21 kg/m^2^ \[0.65; 1.76\] and 3.32 kg \[1.80; 4.84\]. Results remain significant with the 288 included patients without major exclusion criteria and a Last Observation Carried Forward (LOCF) for the dropout patients (data not shown). For all analyses, DMI had no significant effect on weight or BMI. There was no significant interaction between the two interventions. No adverse reaction was reported for patients attending BT. 4. Discussion {#sec4} ============= This is the first multicentre clinical intervention trial to examine the one-year effectiveness of BT on weight loss. It shows that a 3-week BT program was associated with significant weight loss at the end of BT and that the benefit continues to improve one year later. The one-year BT benefit, compared to matched UC patients, was 4.6 kg and 1.7 kg/m^2^, with three times more patients reaching a loss of at least 5% of their initial weight. Results were the same when all patients were considered. A monthly phone dietary motivational interviewing (DMI) did not improve this benefit. The main limit of the study is in relation to the Zelen double consent design. In the review of 58 trials using Zelen design, it was shown that the median of the crossover from one group to the other was 8.9% and interquartile range 2.6% to 15%, which is considered within acceptable limit for an ITT analysis \[[@B22]\]. With about half of patients who crossed over, this study is far from the acceptable limit and overtakes the limit of a Zelen design with a complete dilution of the randomisation (about half of the patients receiving each treatment in both randomized groups). Consequently, this study looks like a prospective nonrandomised clinical intervention trial and it can only be concluded that the weight loss result is more positive for patients who made the choice to go to BT than for those who did not. Nevertheless, this situation may not be really different from most of the randomised weight management trials where the dropout rate usually reaches 30% to 50% of the patients, generally with different dropout rates between groups \[[@B4], [@B23]--[@B26]\]. The aim of randomisation is to balance the distribution of known and unknown parameters in patients who can receive equally either of the treatments (ambivalence). Double blind maintains the comparability over time. With an open randomised trial and one-third to half of patients dropping out during the follow-up with different dropout rates between groups, can we really assume that the comparability be maintained at the end of the study? Can we also really assume that dropouts keep their last or baseline weight evaluation and do not increase their weight thereafter, as is done with Last Observation Carried Forward (LOCF)? This is the hypothesis underlying ITT analysis. In such a situation, a per-protocol analysis taking into account patient initial differences, as in this study analysis, would probably be more suitable. In any case, the methodological issue is the validity of the comparison with the control group to define the magnitude of the BT benefit and not the weight loss at the end of BT and one year later. Most primary care physicians believe that weight loss advice and counselling is not a worthwhile activity in clinical practice \[[@B2], [@B27]\] and they often limit their usual care to verbal or written advice based on national guidelines for healthy lifestyle, without little active follow-up. There is substantial evidence to show that a minimal intervention attitude is ineffective \[[@B24], [@B26], [@B28]--[@B31]\] and corresponds to the results observed for the weight and BMI variation at one year in our control group managed by usual weight advice from their GP, as well as in the placebo group of most of the randomized trials. Lifestyle modification programs now represent the cornerstone of treatment for overweight or obesity \[[@B30]\]. The results of our study are compared favourably with those of two reviews on lifestyle modification programs. The first review compared the effect of various diet and activity interventions on weight loss and concluded that lifestyle modification programs induce a weight loss of 4.5% to 6.5% of baseline weight during the year following treatment (16 to 26 weeks of active phase followed by a maintenance phase) \[[@B30]\]. The second review included 80 studies assessing the effect of 8 types of intervention on weight loss (diet alone, diet plus exercise, exercise alone, meal replacement, very low-calorie diet, orlistat, sibutramine, and advice alone) and provided evidence that meal-planning strategies resulted in a mean weight loss of 4.8% to 8% at 12 months \[[@B23]\]. BT benefit is also competitive with commercial weight management programs, for which two recent studies demonstrated a one-year benefit. The first one included 772 overweight and obese patients in Australia, Germany, and UK to compare a commercial program to standard care. For the 61% of patients who completed the 12-month assessment, the weight loss was 6.65 kg in the commercial program group with 3.16 kg benefit from the standard care group \[[@B24]\]. The second one included 740 overweight and obese patients in primary care in the UK to compare 8 weight loss programs of 12 weeks. One year later, 70.5% of the patients were followed up and weight loss ranged from 4.43 kg (2.7; 6.1) to 3.27 kg (1.4; 5.1) for the 3 commercial weight loss program, and 1.26 (−0.6; 3.1) for UC from general practice \[[@B26]\]. Our results also suggest that BT can challenge management of obesity by pharmacological compounds. Indeed, they are surprisingly very similar to those of the latest controlled trial on lorcaserin, a selective serotonin 2C receptor agonist developed for weight loss and recently approved by FDA \[[@B32]\]. This double-blind clinical trial randomly assigned 3,182 obese or overweight adults to lorcaserin, diet, and exercise versus diet and exercise alone; after 1 year, 47.5% of patients in the lorcaserin group lost 5% or more of their body weight compared to 20.3% of patients receiving placebo. Furthermore, BT appears to be a safe intervention in comparison with drugs. Indeed, most drugs developed for weight loss have been removed from the market for severe adverse effects (i.e., fenfluramine, dexfenfluramine, rimonabant, and sibutramine), put under surveillance for severe adverse effects (i.e., orlistat), or discussed for a long time for safety concerns before recent FDA approval (i.e., lorcaserin) \[[@B33]\]. These different data show that the effectiveness of BT on weight loss is similar to the results obtained by drug treatment, commercial weight management programs, and comprehensive lifestyle modification programs. However, a key advantage of a 3-week BT program is the shorter duration of the active phase, which, on average, is 5 to 9 times shorter than other life style modification programs. Shorter programs are more readily acceptable and better tolerated \[[@B4]\], with a reduction of the economic burden that appears to be correlated with the duration of the program \[[@B34]\]. Attrition rates attributable to lifestyle modification programs are above 35% in one-third of the clinical trials \[[@B31]\] and appear to be strongly correlated with the treatment duration \[[@B4]\]. Conversely, all patients who started BT completed this intervention. This study does not allow the individualization of the effectiveness of different compounds of the 3-week program: water care, water drinking, lifestyle modification from nutrition or physical activity counselling, and 21 days of peace and quiet outside the usual home. However, a study involving 30 patients showed that the two weeks BT have an impact on the plasma level of the two adipocytokines leptin and adiponectin: a slight but not significant increase of leptin which regulates the appetite and energy at the hypothalamic level and a significant decrease of adiponectin, whose main action is to improve insulin sensitivity \[[@B35], [@B36]\]. The relationship between weight loss and adipocytokine variations after BT needs further investigation. 5. Conclusion {#sec5} ============= BT is an efficient lifestyle modification program, which provides significant weight loss and long-term maintenance with only 3 weeks of intervention. BT appears to be an effective and safe program that can be used by primary care physicians as a first-line treatment option for overweight and obese patients. The authors wish to thank all participating patients, their referring GPs and practice team, and the partners and staff of Spa Care Facilities of Brides, Capvern-les-Bains, Vals-les-Bains, Vichy, and Vittel for their support and overall assistance throughout the study. All authors declare no conflict of interests. No other potential conflict of interests relevant to this paper was reported. Investigator GPs had no financial interest in BT resorts. This study was funded by the French Society for Balneotherapy Research (Association Française pour la Recherche Thermale, AFRETH), a nonprofit organisation, after being submitted to and approved by an independent scientific committee. The sponsors had no role in the design and conduct, data collection and datamanagement, analysis, interpretation of the data, preparation, or review of this paper. They were given the opportunity to read and comment on this paper before submission, but the content of the paper was entirely under the control and responsibility of the authors. P. Blin and T. Hanh had full access to all of the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. {#fig1} {#fig2} ###### Chemical composition of the water of five resorts. --------------------------------------------------------------------------------------------------------------------------- Balneotherapy\ Main ions (mg/L) resorts ------------------- ------------------ ------ ------ ------ ------ ------ ------ ------ ------ ------ --------------- ----- Brides les bains 634 104 1365 89.8 2720 1413 427 41.8 634 104 F, Li, Sr No Capvern les bains 335 64.7 5.0 1.4 984 5.0 107 8.7 335 64.7 Sr No Vals les bains 60.0 46.5 2275 --- 20.3 126 6196 --- 60.0 46.5 --- yes Vichy 113 12.4 1909 89.7 178 369 4709 63 113 12.4 F, Li, Fe, As yes Vittel 575 118 12.5 4.7 1584 7.0 377 7.6 575 118 F, Li, Sr No --------------------------------------------------------------------------------------------------------------------------- ###### Patient characteristics at baseline.   All patients Matched patients ---------------------------------------------- -------------- ------------------ -------- ------------- ------------- ------ Gender, *n* (%)     0.54     1.0  Male 23 (19.2%) 31 (22.6%)   11 (15.7%) 11 (15.7%)    Female 97 (80.8%) 106 (77.4%)   59 (84.3%) 59 (84.3%)   Age---years, mean ± SD 51.3 ± 11.5 50.5 ± 10.9 0.54 52.1 ± 11.3 51.9 ± 10.8 0.90 Married or living with a partner, *n* (%) 78 (65.0%) 99 (72.3%) 0.23 42 (60.0%) 48 (68.6%) 0.38 Number of children, *n* (%)     0.83     0.84  None 22 (18.3%) 28 (20.4%)   16 (22.8%) 13 (18.6%)    One 23 (19.2%) 22 (16.1%)   13 (18.6%) 13 (18.6%)    ≥2 75 (62.5%) 87 (63.5%)   41 (58.6%) 44 (62.8%)   Living in town ≥ 100.000 population, *n* (%) 23 (19.2%) 37 (27.0%) 0.14 13 (18.6%) 17 (24.3%) 0.54 BMI---kg/m^2^, mean ± SD 31.3 ± 2.7 31.2 ± 2.5 0.76 31.4 ± 2.8 31.3 ± 2.6 0.75 Weight status, *n* (%)     0.70     1.0  Overweight 75 (62.5%) 82 (59.9%)   42 (60.0%) 42 (60.0%)    Obese 45 (37.5%) 55 (40.1%)   28 (40.0%) 28 (40.0%)   Fat mass \> 35%, *n* (%) 64 (53.3%) 61 (44.5%) 0.17 38 (54.3%) 40 (57.1%) 0.87 Waistline measurement \>100 cm, *n* (%) 73 (60.8%) 86 (62.8%) 0.80 46 (65.7%) 45 (64.3%) 1.0 Paternal history of obesity, *n* (%) 46 (38.3%) 44 (32.1%) 0.36 24 (34.3%) 28 (40.0%) 0.60 Maternal history of obesity, *n* (%) 59 (49.2%) 58 (42.3%) 0.32 34 (48.6%) 32 (45.7%) 0.87 Sibship history obesity, *n* (%) 40 (33.3%) 45 (32.9%) 1.0 25 (35.7%) 25 (35.7%) 1.0 Diabetic, *n* (%) 14 (10.2%) 6 (5.0%) 0.16 8 (11.4%) 4 (5.7%) 0.37 Current smoker, *n* (%) 12 (11.7%) 21 (15.3%) 0.26 7 (10.0%) 9 (12.9%) 0.79 Regular alcohol consumption, *n* (%) 29 (24.2%) 34 (24.8%) 1.0 17 (24.3%) 11 (15.7%) 0.29 Regular physical activities, *n* (%) 23 (19.2%) 28 (20.4%) 0.88 14 (20.0%) 11 (15.7%) 0.66 SBP ≥ 130 mmHg, *n* (%) 81 (67.5%) 70 (51.1%) \<0.01 44 (62.9%) 44 (62.9%) 1.0 DBP ≥ 80 mmHg, *n* (%) 75 (62.5%) 70 (51.1%) 0.08 44 (62.9%) 41 (58.6%) 0.73 Pulse ≥ 70 beats per minute 65 (54.2%) 53 (38.7%) 0.02 36 (51.4%) 32 (45.7%) 0.61 SF12 mental score, mean ± SD 59.3 ± 1.1 59.4 ± 1.1 0.44 59.4 ± 1.1 59.2 ± 1.1 0.26 SF12 physical score, mean ± SD 44.6 ± 9.8 46.5 ± 9.2 0.16 44.0 ± 10.4 46.5 ± 9.8 0.18 [^1]: Academic Editor: MyeongSoo Lee

1. Introduction =============== Fish, marine oils, and their concentrates all serve as sources of the two marine omega-3 fatty acids eicosapentaenoic acid (EPA) and docosahexaenoic acid (DHA), as do some products from algae. To demonstrate an effect of EPA + DHA on heart health, a number of randomized, controlled intervention studies with clinical endpoints like overall mortality or a combination of adverse cardiac events were conducted in populations with elevated cardiovascular risk. One early intervention study with oily fish, rich in EPA + DHA, and some early studies with fish oil or fish oil concentrate or even purified EPA at doses ranging between 0.9 and 1.8 g/day indeed demonstrated effects in terms of fewer sudden cardiac deaths, fatal or non-fatal myocardial infarctions, or a combination of adverse cardiac events \[[@B1-nutrients-06-00799],[@B2-nutrients-06-00799],[@B3-nutrients-06-00799],[@B4-nutrients-06-00799],[@B5-nutrients-06-00799],[@B6-nutrients-06-00799]\]. More recent trials did not demonstrate such effects \[[@B7-nutrients-06-00799],[@B8-nutrients-06-00799],[@B9-nutrients-06-00799],[@B10-nutrients-06-00799],[@B11-nutrients-06-00799],[@B12-nutrients-06-00799]\]. Recent meta-analyses found no significant benefits on total mortality, cardiovascular mortality, and other adverse cardiac or cardiovascular events \[[@B13-nutrients-06-00799],[@B14-nutrients-06-00799],[@B15-nutrients-06-00799],[@B16-nutrients-06-00799],[@B17-nutrients-06-00799],[@B18-nutrients-06-00799]\]. This is in contrast to findings in epidemiologic studies, where intake of EPA + DHA had been found to correlate generally with an up to 50% lower incidence of adverse cardiac events \[[@B18-nutrients-06-00799],[@B19-nutrients-06-00799]\], and in even sharper contrast to epidemiologic studies based on levels of EPA + DHA, demonstrating e.g., a 10-fold lower incidence of sudden cardiac death associated with high levels of the fatty acids, as compared to low levels \[[@B20-nutrients-06-00799],[@B21-nutrients-06-00799]\]. This seemingly contradictory evidence has led the American Heart Association to recommend "omega-3 fatty acids from fish or fish oil capsules (1 g/day) for cardiovascular disease risk reduction" for secondary prevention, whereas the European Society for Cardiology recommends "Fish at least twice a week, one of which to be oily fish", but no supplements for cardiovascular prevention \[[@B22-nutrients-06-00799],[@B23-nutrients-06-00799]\]. The more recent guidelines on treating patients with stable ischemic heart disease or patients after a myocardial infarction, targeting similar patient populations, do not recommend EPA + DHA \[[@B24-nutrients-06-00799],[@B25-nutrients-06-00799]\]. At least in Europe, cardiologists do not routinely use EPA + DHA to reduce cardiovascular risk. A similar picture emerges for atrial fibrillation: In epidemiologic studies, consumption of EPA + DHA or higher levels of EPA + DHA were associated with lower risk for developing atrial fibrillation, while intervention studies found no effect \[[@B26-nutrients-06-00799],[@B27-nutrients-06-00799],[@B28-nutrients-06-00799]\]. Pertinent guidelines do not mention EPA + DHA \[[@B29-nutrients-06-00799]\]. A similar picture also emerges for severe ventricular rhythm disturbances \[[@B20-nutrients-06-00799],[@B21-nutrients-06-00799],[@B30-nutrients-06-00799],[@B31-nutrients-06-00799]\]. Why is it that trial results are at odds with results from epidemiology? What needs to be done to better translate the epidemiologic findings into trial results? The current review will try to shed some light on this issue, with a special consideration of the Omega-3 Index. 2. The Omega-3 Index as a Cardiovascular Risk Factor ==================================================== At least some nutritional surveys do not provide valid data \[[@B32-nutrients-06-00799]\]. This may explain, why the relation of EPA + DHA in the diet to clinical events has been found to be looser than the relation of levels of EPA + DHA measured in blood to clinical events (e.g., \[[@B20-nutrients-06-00799],[@B33-nutrients-06-00799]\]). A detailed discussion of the pros and cons of the various fatty acid compartments in which levels of omega-3 fatty acids (whole blood, whole plasma, plasma phospholipids, and others) should be measured is outside the scope of this review and can be found elsewhere \[[@B34-nutrients-06-00799]\]. The following points argue for the use of erythrocytes: erythrocyte fatty acid composition has a low biological variability, erythrocyte fat consists almost exclusively of phospholipids, erythrocyte fatty acid composition reflects tissue fatty acid composition, pre-analytical stability, and other points \[[@B34-nutrients-06-00799],[@B35-nutrients-06-00799],[@B36-nutrients-06-00799],[@B37-nutrients-06-00799],[@B38-nutrients-06-00799]\]. In 2004, EPA + DHA in erythrocyte fatty acids were defined as the Omega-3 Index and suggested as a risk factor for sudden cardiac death \[[@B39-nutrients-06-00799]\]. Integral to the definition was a specific and standardized analytical procedure, conforming the quality management routinely implemented in the field of clinical chemistry \[[@B39-nutrients-06-00799]\] In fatty acid analysis, methods have a large impact on results: when one sample was sent to five different laboratories offering determination of an Omega-3 Index, results differed by a factor of 3.5 \[[@B34-nutrients-06-00799]\]. While results may be internally valid in one laboratory, a difference by a factor of 3.5 makes it impossible to compare results among laboratories. Therefore, the Omega-3 Index was renamed HS-Omega-3 Index^®^. In contrast, the laboratories adhering to the HS-Omega-3 Index methodology perform regular proficiency testing, as mandated in routine Clinical Chemistry labs \[[@B34-nutrients-06-00799]\]. So far, the HS-Omega-3 Index is the only analytical procedure used in several laboratories. A standardized analytical procedure is a prerequisite to generate the data base necessary to transport a laboratory parameter from research into clinical routine. Moreover, standardization of the analytical procedure is the first important criterion for establishing a new biomarker for cardiovascular risk set forth by the American Heart Association and the US Preventive Services Task Force \[[@B40-nutrients-06-00799],[@B41-nutrients-06-00799]\]. As exemplified by [Table 1](#nutrients-06-00799-t001){ref-type="table"}, the HS-Omega-3 Index has been measured in many populations. Of note, a lower HS-Omega-3 Index was always associated with a poorer clinical condition ([Table 1](#nutrients-06-00799-t001){ref-type="table"}). nutrients-06-00799-t001_Table 1 ###### Mean HS-Omega-3 Index values in various populations, Mean (±standard deviation (SD)). Please note that in every population studied, a lower value was found to be associated with a worse condition than a higher value. References are given, if not, unpublished, *n =* number of individuals measured. Population HS-Omega-3 Index --------------------------------------------------------------------------------------------------------------------------------- ------------------ **Western countries (high incidence of coronary heart disease)** *Germany* Unselected Individuals (*n* = 5000) 7.15 (±2.19)% Patients with atherosclerosis \[[@B42-nutrients-06-00799]\], (*n* = 190) 5.94 (±1.41)% Patients with hyperlipidemia \[[@B43-nutrients-06-00799]\], (*n* = 47) 7.00 (±1.90)% Pregnant women, week 24 (*n* = 103) 7.66 (±1.83)% Patients with congestive heart failure (*n* = 895) 3.47 (±1.20)% Patients with major depression \[[@B44-nutrients-06-00799]\], (*n* = 90) 3.93 (±1.50)% *Spain* Individuals with high risk for, but without cardiovascular disease \[[@B45-nutrients-06-00799]\], (*n* = 198) (SD not reported) 7.10% *Norway* Patients with myocardial infarction \[[@B46-nutrients-06-00799]\] (SD not reported) With ventricular fibrillation (*n* = 10) 4.88% Without ventricular fibrillation (*n* = 185) 6.08% *Europe* Unselected data from routine determinations, *n* = 10,000 6.96 (+2.15)% *USA* Healthy in Kansas City \[[@B47-nutrients-06-00799]\], (*n* = 163) 4.90 (±2.10)% Framingham-Offspring \[[@B48-nutrients-06-00799]\], (*n* = 3196) 5.60 (±1.70)% Patients with stable coronary heart disease \[[@B49-nutrients-06-00799]\], (*n* = 956) (SD not reported) 4.60% Patients with major depression \[[@B50-nutrients-06-00799]\], (*n* = 118) 2.90 (±1.50)% Adolescents with major depression \[[@B51-nutrients-06-00799]\], (*n* =150) (SD not reported) 3.46% Patients with severe obstructive sleep apnea \[[@B52-nutrients-06-00799]\], (*n* = 52) (SD not reported) 4.00% *Saudi Arabia* Individuals, most with diabetes (*n =* 69) 3.47 (±1.20) % **Asian countries (low incidence of coronary heart disease)** *Korea* Healthy controls \[[@B53-nutrients-06-00799]\], (*n* = 50) (SD not reported) 11.81% Healthy control \[[@B54-nutrients-06-00799]\], (*n* = 40) 10.55 (±0.48)% Patients with myocardial infarction \[[@B53-nutrients-06-00799]\], (*n* = 50), (SD not reported) 9.57% Patients with hemorrhagic brain infarction \[[@B54-nutrients-06-00799]\], (*n* = 40) 8.55 (±0.41)% Patients with ischemic brain infarction \[[@B54-nutrients-06-00799]\], (*n* = 40) 8.19 (±0.64)% Hemodialysis-patients without calcification on plain chest radiograph \[[@B55-nutrients-06-00799]\], (*n* = 11) 9.82 (±2.37)% Hemodialysis-patients with calcification on plain chest radiograph \[[@B55-nutrients-06-00799]\], (*n* = 20) 9.23 (±2.34)% Peritoneal Dialysis Patients \[[@B56-nutrients-06-00799]\], (*n* = 14) 12.83 (±3.30)% Patients with a kidney transplant \[[@B57-nutrients-06-00799]\], (*n* = 49) 9.70 (±1.85)% *Japan* Unselected men (*n* = 262), (SD not reported) 9.58% All levels of fatty acids are determined by the balance of substance entering the body and those leaving the body. Neither a recent meal, even if rich in EPA + DHA, nor severe cardiac events altered the HS-Omega-3 Index \[[@B38-nutrients-06-00799],[@B58-nutrients-06-00799],[@B59-nutrients-06-00799],[@B60-nutrients-06-00799],[@B61-nutrients-06-00799]\]. However, while long-term intake of EPA + DHA, e.g., as assessed with food questionnaires, was the main predictor of the HS-Omega-3 Index, long-term intake explained only 12%--25% of its variability \[[@B46-nutrients-06-00799],[@B62-nutrients-06-00799],[@B63-nutrients-06-00799]\]. A hereditary component of 24% exists \[[@B64-nutrients-06-00799]\]. A number of other factors correlated positively (+) or negatively (−), like age (+), body mass index (−), socioeconomic status (+), smoking (−), but no other conventional cardiac risk factors \[[@B47-nutrients-06-00799],[@B64-nutrients-06-00799],[@B65-nutrients-06-00799],[@B66-nutrients-06-00799],[@B67-nutrients-06-00799],[@B68-nutrients-06-00799],[@B69-nutrients-06-00799],[@B70-nutrients-06-00799],[@B71-nutrients-06-00799]\]. More factors determining the level of the HS-Omega-3 Index, especially regarding efflux remain to be defined. Therefore, it is impossible to predict the HS-Omega-3 Index in an individual, as it is impossible to predict the increase in the HS-Omega-3 Index in an individual in response to a given dose of EPA + DHA \[[@B42-nutrients-06-00799],[@B46-nutrients-06-00799],[@B62-nutrients-06-00799],[@B63-nutrients-06-00799]\]. In [Table 2](#nutrients-06-00799-t002){ref-type="table"}, current evidence is presented on the relation of the HS-Omega-3 Index to cardiovascular events. This evidence is supported by measurements of EPA + DHA in other fatty acid compartments, as discussed in more detail elsewhere \[[@B72-nutrients-06-00799],[@B73-nutrients-06-00799]\]. Within the framework of "Heart and Soul" and "Triumph", it was investigated whether determination of the HS-Omega-3 Index added to the information obtained by assessing cardiovascular risk with a conventional scoring system, like the Framingham or GRACE scores for predicting fatal events. The HS-Omega-3 Index provided additional information, as demonstrated by larger areas under the curves in various c-statistics for fatal \[[@B74-nutrients-06-00799]\] and non-fatal events \[[@B53-nutrients-06-00799],[@B75-nutrients-06-00799]\]. Taken together, the HS-Omega-3 Index predicts risk, appears largely independent of conventional risk factors, and adds to the information obtained by conventional risk scoring, thus fulfilling the second criterion for establishing a new biomarker for cardiovascular risk set forth by the American Heart Association and the US Preventive Services Task Force \[[@B40-nutrients-06-00799],[@B41-nutrients-06-00799]\]. nutrients-06-00799-t002_Table 2 ###### Summary of epidemiologic studies relating the Omega-3 Index to cardiovascular events. Acronym \[reference\] Design Disease *n* Years Criterion Comparison Result -------------------------------------------- -------------- ------------ ------------------------ ------- ----------------- ------------------------------------ --------------------------------------------------------------------------------------------------------------- *Total mortality* Heart & Soul \[[@B49-nutrients-06-00799]\] cohort stable CAD 956 5.9 total mortality HS-Omega-3 Index HR 0.73; 95% CI, 0.56--0.94 Triumph \[[@B74-nutrients-06-00799]\] cohort recent MI 1144 2 total mortality EPA in red cells tertiles EPA \< 0.25% total mortality 26%, 0.25 \< EPA \< 0.8% total mortality 13%, EPA \> 0.80% total mortality 7% Triumph \[[@B76-nutrients-06-00799]\] cohort recent MI 1424 1 total mortality HS-Omega-3 Index \< 4%*vs*. \>4.0% HR 2.0; 95% CI 1.2--3.3 Racs \* \[[@B77-nutrients-06-00799]\] cohort recent ACS 460 2 total mortality HS-Omega-3 Index in quartiles not significant. *Sudden cardiac death* \[[@B20-nutrients-06-00799]\] case-control SCD 82/108 cases/controls SCD red cell EPA + DHA in quartiles OR 1.0--0.1 (95% CI 0.1--0.4) Phys Health \[[@B21-nutrients-06-00799]\] case-control SCD 84/182 cases/controls SCD whole blood EPA + DHA in quartiles OR 1.0--0.1 (95% CI 0.02--0.48) across quartiles Cardiac morbidity \[[@B78-nutrients-06-00799]\] case-control ACS 94/94 cases/controls ACS whole blood EPA + DHA in quintiles OR 1.0--0.2 (95% CI not reported), OR 0.67 (95% CI 0.46 to 0.98) per, 1 standard deviation increase EPA + DHA \[[@B79-nutrients-06-00799]\] case-control ACS 768/768 cases/controls ACS HS-Omega-3 Index in tertiles OR 1.0--0.31 (95% CI 0.14--0.67) across tertiles \[[@B53-nutrients-06-00799]\] case-control ACS 50/50 cases/controls ACS HS-Omega-3 Index in tertiles OR 1.0--0.08 (95% CI 0.02--0.38) across tertiles no acronym \[[@B80-nutrients-06-00799]\] case-control ACS 24/68 cases/controls STEMI HS-Omega-3 Index in tertiles OR 6.38 (95% CI 1.02--39.85)--1.0 across tertiles Abbreviations: *n*: number of individuals studied; Coronary artery disease: CAD; HR: hazard ratio; MI: myocardial infarction; EPA: eicosapentaenoic acid; ACS: acute coronary syndrome; SCD: sudden cardiac death; DHA: docosahexaenoic acid; OR: odds ratio; STEMI: ST-elevation myocardial infarction. \* No case estimate was reported in Racs. Therefore, by definition, it is unclear, whether the discriminatory power of the HS-Omega-3 Index was too small, or the study was too small to detect the discriminatory power. Moreover, the HS-Omega-3 Index has made it possible to reclassify individuals from intermediate cardiovascular risk into the respective high risk and low risk strata \[[@B74-nutrients-06-00799],[@B75-nutrients-06-00799]\], the third criterion for establishing a new biomarker for cardiovascular risk \[[@B40-nutrients-06-00799],[@B41-nutrients-06-00799]\]. Increasing the HS-Omega-3 Index by increased intake of EPA + DHA in randomized controlled trials improved a number of surrogate parameters for cardiovascular risk: heart rate was reduced, heart rate variability was increased, blood pressure was reduced, platelet reactivity was reduced, triglycerides were reduced, large buoyant low-density lipoprotein (LDL)-particles were increased and small dense LDL-particles were reduced, large buoyant high-density lipoproteins (HDL)2 were increased, very low-density lipoprotein (VLDL1) + 2 was reduced, pro-inflammatory cytokines (e.g., tumor necrosis factor alpha, interleukin-1β, interleukins-6,8,10 and monocyte chemoattractant protein-1) were reduced, anti-inflammatory oxylipins were increased \[[@B43-nutrients-06-00799],[@B81-nutrients-06-00799],[@B82-nutrients-06-00799],[@B83-nutrients-06-00799],[@B84-nutrients-06-00799],[@B85-nutrients-06-00799],[@B86-nutrients-06-00799],[@B87-nutrients-06-00799],[@B88-nutrients-06-00799],[@B89-nutrients-06-00799],[@B90-nutrients-06-00799],[@B91-nutrients-06-00799],[@B92-nutrients-06-00799],[@B93-nutrients-06-00799],[@B94-nutrients-06-00799]\]. Importantly, in a two-year randomized double-blind angiographic intervention trial, increased erythrocyte EPA + DHA reduced progression and increased regression of coronary lesions, an intermediate parameter \[[@B95-nutrients-06-00799]\]. Taken together, increasing the HS-Omega-3 Index improved surrogate and intermediate parameters for cardiovascular events. A large intervention trial with clinical endpoints based on the HS-Omega-3 Index remains to be conducted. Therefore, the fourth criterion, proof of therapeutic consequence of determining the HS-Omega-3 Index, is only partially fulfilled \[[@B40-nutrients-06-00799],[@B41-nutrients-06-00799]\]. 3. Discussion of Neutral Results of Large Intervention Trials ============================================================= Why is it that a low HS-Omega-3 Index can be a cardiovascular risk factor, and yet the results of the large trials testing EPA + DHA on clinical endpoints were neutral? 3.1. Bioavailability Issues --------------------------- According to personal information from the respective first authors, participants of recent large intervention trials were advised to take their supplements, frequently an encapsulated EPA + DHA ethyl-ester with breakfast---in many countries a low-fat meal \[[@B7-nutrients-06-00799],[@B8-nutrients-06-00799],[@B9-nutrients-06-00799],[@B10-nutrients-06-00799],[@B11-nutrients-06-00799]\]. As discussed in more detail in a recent review, bioavailability of EPA + DHA depends on the chemical form in which they are bound (phospholipids \> recombined triglycerides \> triglycerides \> free fatty acids \> ethyl-esters) \[[@B96-nutrients-06-00799],[@B97-nutrients-06-00799]\], on matrix effects (capsule ingestion with concomitant intake of food, fat content in food) or galenic form (*i*.*e*., microencapsulation, emulsification). The chemical binding form impacts on bioavailability roughly with a factor of two, whereas matrix effects can impact bioavailability up to a factor of 13, and the galenic form up to a factor of 21 \[[@B96-nutrients-06-00799],[@B97-nutrients-06-00799],[@B98-nutrients-06-00799],[@B99-nutrients-06-00799]\]. When the large trials mentioned here were designed, the bioavailability issues just mentioned were unknown. Thus, involuntarily, the combination used in many of the large trials---An unemulsified ethyl-ester or triglyceride with a low fat meal---guaranteed a very low bioavailability of EPA + DHA. 3.2. Issues in Trial Design --------------------------- In all large intervention trials conducted so far, study participants were recruited based on clinical conditions, but irrespective of their baseline omega-3 fatty acid status \[[@B1-nutrients-06-00799],[@B2-nutrients-06-00799],[@B3-nutrients-06-00799],[@B4-nutrients-06-00799],[@B5-nutrients-06-00799],[@B6-nutrients-06-00799],[@B7-nutrients-06-00799],[@B8-nutrients-06-00799],[@B9-nutrients-06-00799],[@B10-nutrients-06-00799],[@B11-nutrients-06-00799],[@B12-nutrients-06-00799]\]. In all populations studied so far, the HS-Omega-3 Index had a statistically normal distribution ([Table 1](#nutrients-06-00799-t001){ref-type="table"}). Thus, the proportion of the study population with high levels was not prone to the effects of EPA + DHA, if any. In order to recruit a study population, in which an effect of EPA + DHA can be demonstrated, recruiting study participants with a low HS-Omega-3 Index is a logical choice. In all large intervention trials conducted so far, study participants were exposed to a trial-specific, but fixed dose of EPA + DHA or placebo \[[@B1-nutrients-06-00799],[@B2-nutrients-06-00799],[@B3-nutrients-06-00799],[@B4-nutrients-06-00799],[@B5-nutrients-06-00799],[@B6-nutrients-06-00799],[@B7-nutrients-06-00799],[@B8-nutrients-06-00799],[@B9-nutrients-06-00799],[@B10-nutrients-06-00799],[@B11-nutrients-06-00799],[@B12-nutrients-06-00799]\]. The inter-individual variability in response to a fixed dose of EPA + DHA has been found to be large, *i*.*e*., vary up to a factor of 13 \[[@B42-nutrients-06-00799],[@B61-nutrients-06-00799]\]. This fact alone suggests individualizing the dose given in a trial, in order to reach a predefined target range of the HS-Omega-3 Index, e.g., 8%--11%. The statistically normal distribution of the baseline HS-Omega-3 Index further complicates this problem: A large overlap of omega-3 levels in the EPA + DHA group and placebo or control group can be expected, and has been seen in at least one large trial (Mühlhäusler, B., personal communication) \[[@B100-nutrients-06-00799]\]. With levels of omega-3 fatty acids not differing between intervention and placebo or control groups, a difference in study outcome cannot be expected, even if the condition studied would be susceptible to treatment with EPA + DHA. It is worth noting that when a neutral intervention trial was analyzed in a cross-sectional way, EPA + DHA levels directly related to study outcome and less to treatment allocation \[[@B101-nutrients-06-00799]\]. Conversely, if a trial reports a positive result, it is likely to have been conducted in a study population with low baseline levels of EPA and DHA, like congestive heart failure: a positive result of a large trial was reported \[[@B6-nutrients-06-00799]\], and we found a low mean HS-Omega-3 Index in patients with congestive heart failure (unpublished data, [Table 1](#nutrients-06-00799-t001){ref-type="table"}). A similar case can be made for major depression ([Table 1](#nutrients-06-00799-t001){ref-type="table"}, references \[[@B44-nutrients-06-00799],[@B50-nutrients-06-00799],[@B51-nutrients-06-00799],[@B84-nutrients-06-00799]\]). In the future, recruiting study participants with a low baseline HS-Omega-3 Index and treating them within a predefined target range will allow clearer trial results to be a distinct possibility. Dose adjustments will need to be performed in the placebo group. Since a larger treatment effect can be assumed in the study size estimation, it can be expected that study sizes will be smaller and thus studies less expensive. Clearly, these thoughts are not restricted to trials with patients with cardiovascular risk, atrial fibrillation or ventricular arrhythmia, but can be extended to all areas of omega-3 fatty acid research. This will facilitate scientific progress and lead to a faster recognition of the effects of EPA + DHA. 4. Conclusions ============== In an inconsistent manner, EPA and DHA are either recommended or not included in guidelines of cardiac scientific societies. The use of EPA and DHA is not supported by results of recent intervention trials or their meta-analyses. However, epidemiologic data based on assessments of diet and, even more so, data based on levels of EPA + DHA measured in humans, clearly demonstrate that EPA + DHA are associated with a low risk for total mortality, sudden cardiac arrest, and fatal and non-fatal myocardial infarctions. For a number of reasons, like a standardized analytical procedure and a large data base, levels of EPA + DHA are best assessed with the HS-Omega-3 Index. According to current criteria of the American Heart Association and others, the HS-Omega-3 Index is a novel cardiovascular risk factor. Moreover, the HS-Omega-3 Index has led to a fresh look at the field of omega-3 fatty acids and has made it possible to identify issues of bioavailability and study design, explaining at least in part the neutral results of previous intervention trials. In the future, more efficient intervention studies can be conducted based on the HS-Omega-3 Index, thus providing a clearer picture of the effects of EPA + DHA. Research grants were provided by several government agencies, AkerBiomarine, Neptune, and Fresenius Kabi. The cost of travel to the AAOCS meeting was provided by Sanofi Consumer Health Care. CvS operates Omegametrix, a laboratory for fatty acid analyses. Speaker honoraria were received from Reckitt-Benckiser and the Portuguese National Fisheries.

DOI: [10.1002/bjs5.50177](https://doi.org/10.1002/bjs5.50177) *The Editor‐in‐Chief welcomes topical correspondence from readers relating to articles published in BJS Open. Letters should be submitted via ScholarOne Manuscripts and, if accepted, will be published online*. We read with interest the work of Charehbili and colleagues[1](#bjs550285-bib-0001){ref-type="ref"}, which 'aimed to investigate whether there is a superiority of chlorhexidine--alcohol over iodine--alcohol for preventing SSI'. This cluster‐randomized crossover trial was conducted in five hospitals and 3665 patients were included. The authors found that the incidence of surgical‐site infection (SSI) was not different between the groups: 3·8 per cent among patients in the chlorhexidine--alcohol group *versus* 4·0 per cent in those in the iodine--alcohol group (odds ratio 0·96, 95 per cent c.i. 0·69 to 1·35). We commend the authors for performing this interesting study, as these results are useful for the choice of the most appropriate preoperative antiseptic. However, we have several statistical suggestions and queries that we would like to communicate to the authors. The authors concluded that 'Preoperative skin disinfection with chlorhexidine--alcohol is similar to that for iodine--alcohol with respect to reducing the risk of developing an SSI'. This may be due to an underpowered study. In fact, sample size was estimated by simulation. Although this approach is efficient, the authors do not provide enough details on the parameters they used. Thus it is not easy to replicate calculations. As mentioned by the CONSORT statement[2](#bjs550285-bib-0002){ref-type="ref"}: 'the reports of cluster randomized trials should state the assumptions used when calculating the number of clusters and the cluster sample size'. The authors mention R software for the simulations that led to the final estimation of sample size. But several R packages are available and one may suppose that a package such as clusterPower was used[3](#bjs550285-bib-0003){ref-type="ref"}. Not knowing which package was used does not permit the analysis to be replicated. Moreover, algorithms used may vary between packages and lead to different estimations of sample size. In a cluster‐randomized crossover trial, a sequence of interventions is assigned to a cluster (group) of individuals. Each cluster receives each intervention in a separate period of time and this leads to 'cluster periods\'[4](#bjs550285-bib-0004){ref-type="ref"}. There is usually a correlation between patients in the same cluster. In addition, within a cluster, patients within the same period may be more similar to one another than to patients in other periods[5](#bjs550285-bib-0005){ref-type="ref"}. In a cluster‐randomized crossover trial, the sample size estimated by not taking into account the above‐mentioned features must be multiplied by a defined inflation factor. The latter can be approximated by (1 + (*n* − 1)ρ) − η[6](#bjs550285-bib-0006){ref-type="ref"}, [7](#bjs550285-bib-0007){ref-type="ref"}, where *n* is the average number of patients in a cluster during one of the periods, ρ is the intraclass correlation (ICC), and η the interperiod correlation. See, for example, Turner *et al*.[8](#bjs550285-bib-0008){ref-type="ref"} or Moerbeek and Teerenstra[9](#bjs550285-bib-0009){ref-type="ref"} (p. 94) for other approaches to estimate sample size in this context. Parameters ρ and η can be retrieved from literature or estimated using assumptions or approximations[10](#bjs550285-bib-0010){ref-type="ref"} (p. 203), for instance: the logarithm of the ICC can be approximated by the logarithm of the prevalence of disease (here, the SSI rate)[11](#bjs550285-bib-0011){ref-type="ref"}; the interclass correlation is intrinsically lower than the ICC[12](#bjs550285-bib-0012){ref-type="ref"}. Data were analysed using a multilevel model, which is appropriate. Treatment period was considered as a fixed effect and hospitals as random effect. Treatment period could also be considered as a random effect. In their simulations, Morgan *et al*.[5](#bjs550285-bib-0005){ref-type="ref"} actually demonstrated that 'hierarchical models without random effects for period‐within‐cluster, which do not account for any extra within‐period correlation, performed poorly with greatly inflated Type I errors in many scenarios'. The authors did not report variance components of outcomes: within‐ and between‐participant variance, the ICC, as recommended by some authors[13](#bjs550285-bib-0013){ref-type="ref"}. In a cluster‐randomized crossover trial, three components of variation are available: variation in cluster mean response; variation in the cluster period mean response; and variation between individual responses within a cluster period[4](#bjs550285-bib-0004){ref-type="ref"}. Small changes in the specification of the within‐cluster--within‐period correlation, or the within‐cluster--between‐period correlation, can increase the required number of clusters[4](#bjs550285-bib-0004){ref-type="ref"}. Thus, as the above‐mentioned correlation parameters were not reported by Charehbili *et al*.[1](#bjs550285-bib-0001){ref-type="ref"}, the number of clusters required may be larger than that used in the study. A simulation study showed an association between an increase in cluster size variability and a decrease in statistical power[14](#bjs550285-bib-0014){ref-type="ref"}. The authors did not address this point. In summary, the results of this study are interesting, but readers should interpret them with caution, according to the statistical methods used for design and analysis of cluster‐randomized crossover trials. Disclosure {#bjs550285-sec-0002} ========== The authors declare no conflict of interest.

Debates over fundamental concepts in quantum theory such as reality, locality and quantum entanglement not only deepen our understanding of quantum mechanics but also promote the development of quantum information technology. One of such debates, which has existed since the beginning of quantum mechanics, is whether the quantum state corresponds to a real physical state (i.e., it is ontic) or whether it merely represents an observer's knowledge about the underlying reality (i.e., it is epistemic)[@b1][@b2][@b3][@b4]. A major reason for doubting the reality of the quantum state is that an epistemic interpretation of the quantum state could provide an intuitive explanation for many counterintuitive but fundamental quantum phenomena, such as the measurement postulate and wave function collapse[@b3][@b4]. Recently Harrigan and Spekkens formulated a theory regarding the ontic and epistemic concepts of the quantum states[@b2]. Their model is based on a reasonable assumption that every quantum system possesses a real physical state denoted *λ*, which is called the ontic state. Every ontic state corresponds to a different ontic *λ*, on which the possibility of measurement outcomes depends. Under this assumption, quantum states are supposed to be distinguishable via two models. A state is said to be of the *ψ*-ontic model if distinct pure quantum states always match distinct real states. On the other hand, a state is of the *ψ*-epistemic model if distinct states may result in the same ontic *λ*. Pusey, Barrett, and Rudolph (PBR) further develop the argument for the *ψ*-ontic and *ψ*-epistemic models through a no-go theorem. The theorem states that *ψ*-epistemic model cannot reproduce the predictions of quantum theory if the preparation independence assumption is taken[@b3]. Patra *et al.*[@b4] conclude on a similar result but with a different assumption of continuity for a single quantum system (Hardy also obtains a similar result independently[@b5]). Further theoretical works focusing on the no-go theorems for *ψ*-epistemic models have been presented in refs [@b6], [@b7], [@b8], [@b9], [@b10], [@b11], [@b12], [@b13], [@b14]. Given these growing theoretical works, experimental tests of the *ψ*-ontic and *ψ*-epistemic models are needed. The first experiment testing of the existence of *ψ*-epistemic models was implemented by Nigg *et al.*[@b15] with two atoms in an ion trap following the PBR theorem[@b3]. Very recently, Ringbauer *et al.* reported an experiment to rule out the maximal *ψ*-epistemic model with single photons in 3 and 4 dimensions[@b16]. The no-go theory of continuous *ψ*-epistemic models[@b4] was also experimentally tested using attenuated coherent states of light to simulate the high-dimensional single photon quantum states by Patra *et al.*[@b17]. However, due to the nonideal state preparation (the phase fluctuations of the coherent states) in the experiment[@b17], one additional assumption that the ontic state depends on control measurements should be included in the explanation of the experiment. Furthermore, the no-go theorem is proposed for a state of single particles whereas a coherent state with pretty large possibility of multi-photons was used in the experiment[@b17]. Therefore, further experiments to test the epistemic models with a state of single particles and with fewer assumptions are needed. In this paper, we report an experimental test of the existence of the continuous *ψ*-epistemic models with high-dimensional single photon quantum states. We produce the heralded narrowband single photon quantum states in dimensions 3, 7, 16, 25 and 32 to test the no-go theorem and our results support the no-go theorem for continuous *ψ*-epistemic models. In comparison with Patra *et al.*'s experiment[@b17], our experiment benefits from the ideal single photon source and avoids the nonideal state preparation loophole. Thus the no-go theory of continuous *ψ*-epistemic models is tested without additional assumptions except for the fair-sampling assumption for the single photon detection loophole made in all of the above experiments[@b15][@b16][@b17] and often made in nonlocality experiments[@b18]. Results ======= The no-go theorem and the experimental setup -------------------------------------------- The difference between the *ψ*-ontic models and the continuous *ψ*-epistemic models (also named *δ*-continuous *ψ*-epistemic models) is shown in [Fig. 1](#f1){ref-type="fig"}. The *δ*-continuous *ψ*-epistemic models assume that there are ontic states *λ* in the initial ensemble (e.g. *ψ*~1~) that will remain part of the slightly perturbed ensemble (e.g. *ψ*~2~) (see Method for a detail definition of the *δ*-continuity). There are two parameters to characterize the *δ*-continuous *ψ*-epistemic models: the parameter *δ* characterizes how continuous the model is, and characterizes how epistemic it is. As proven in ref. [@b4], there is a fundamental constraint (the no-go theorem) on *δ*-continuous models for single systems: there are no *δ*-continuous *ψ*-epistemic models with reproducing the measurement statistics of quantum states in a Hilbert space of dimension *d*. Mathematically, the *δ*-continuous *ψ*-epistemic models predict the following (with preparations *Q*~*k*~ corresponding to distinct quantum states \|*ψ*~*k*~〉 all contained in a ball of radius *δ*): Here *P*(*k*\|*M*, *Q*~*k*~) stands for the probability of outcome *k* with a measurement *M* carried out on a system in ontic state *λ* which is prepared with procedure *Q*~*k*~ and associated with a probability distribution *P*(*λ*\|*Q*~*k*~). The parameter can be viewed as a measure of the extent to which distributions over real states overlap in the neighborhood of a given quantum state[@b4], or it can be seen as related to the variational distance between the distributions *P*(*λ*\|*Q*~*k*~)[@b3]. According to quantum theory, however, there invariably exist some states for which the left-hand side of [Eq. (1)](#eq19){ref-type="disp-formula"} is equal to 0. As a typical example, we consider *d* distinct states (*j*, *k* = 1, ...*d*), which are all at mutual distance from the state . They have *δ*-continuity property since \|*ψ*~*k*~〉 contain in a ball of radius *δ* centered on \|*ψ*〉. If the basis of the measurement *M* is chosen to be \| *j*〉, then *P*(*k*\|*M*, *Q*~*k*~) = 0 for all *k* = 1, ...*d* and thus the left hand side of [Eq. (1)](#eq19){ref-type="disp-formula"} is equal to 0. Hence, the *δ*-continuous *ψ*-epistemic models' existence can be experimentally tested. We use the time bins of the narrowband single photon (labeled by the time *jτ* when it is detected) as the basis states \| *j*〉 to realize the quantum state \|*ψ*~*k*~〉 in the experiment. As shown in [Fig. 2](#f2){ref-type="fig"}, narrowband paired photons are generated through spontaneous four-wave mixing (SFWM) in a two-dimensional ^85^Rb magneto-optical trap (MOT) (see Methods for detail)[@b19][@b20]. A detection (the single photon counter: PerkinElmer SPCM-AQRH, *C*~*dk*~ = 500 ± 100 *c*/*s*, *D*~0~, *D*~1~/*D*~2~) of the Stokes photon heralds the presence of its paired anti-Stokes photon, which also determines the start point of the experiment. After collection of the photons with single mode fibre (SMF), the anti-Stokes photon passes through an electro-optical amplitude modulator (EOAM, fiber-based, 10 GHz, Eospace), which is used to shape the waveform of the photons[@b21][@b22]. With the EOAM and an arbitrary waveform generator (AWG, Agilent 81150A), the anti-Stokes photon can be prepared with the waveform consisting of a train of *d* pulses with one missing. To perform under a stable time base, the AWG's master clock is phase-locked to an external 10 MHz reference (SRS FS725) with exceptionally low phase noise (−130 dBc/Hz at 10 Hz offset) and one second Allan variance (\<2 × 10^−11^). The modulated anti-Stokes photons then pass through a 180 *m* fiber spool to end the preparation process, which is long enough to store the entire photon wave package. The coincidence counts are implemented by a time-to-digital converter (Fast Comtec P7888) with a 2 ns bin width. The verification of the single-photon nature of the heralded anti-Stokes photons is carried out with standard Hanbury-Brown-Twiss interferometer (a fiber-based 50/50 beam splitter and two detectors *D*~1~ and *D*~2~)[@b23]. The high-dimensional single photon quantum states ------------------------------------------------- We set the pump and coupling light powers to 40 *μW* and 1.4 *mW*, respectively. With the EOAM working at its maximum transmission, the coincidence counts of the unmodulated photon pairs for 240 s are illustrated as the red curve in [Fig. 3(a)](#f3){ref-type="fig"}. The heralded anti-Stokes photon has a coherence time of about 600 ns. Controlling the EOAM with the desired square waveform allows the single photon to be modulated into a train of *d* pulses with one missing. For instance, *d* = 16 is illustrated as the blue curve in [Fig. 3(a)](#f3){ref-type="fig"}. The full width at half maximum of each pulse is about 5 ns and the time interval between neighbour pulses is 12 ns, which are limited by the bandwidth (240 MHz) of the AWG (the square waveform has a 2.5 ns rising/falling time). To minimize the defects in the rising and falling regimes of the pulse, we keep the single-photon counts only at the middle of each pulse within the time window *T*~*p*~ = 2 *ns*. The extinction ratio of the EOAM is estimated to be *R*~*ext*~ = 20 ± 2 *dB*. We confirm the quality of the heralded single-photon source by measuring its conditional autocorrelation function , where *N*~0~ denotes the Stokes counts at *D*~0~, *N*~01~ (*N*~02~) denote the twofold coincidence counts at two detectors *D*~0~ and *D*~1~ (*D*~0~ and *D*~2~), and *N*~012~ is the threefold coincidence counts among three detectors *D*~0~, *D*~1~ and *D*~2~. An attenuated coherent light source has , while a two-photon source gives . As shown in [Fig. 3(a)](#f3){ref-type="fig"}, holds well within the overall waveform and it suggests the good quality of our single-photon source. Compared with the attenuated coherent light source[@b17], the single-photon source reduces the multi-photon probability at least by a factor of with the same generation rate. With all the losses accounted for (the fiber coupling efficiency 70%, filter transmission 70%, EOAM transmissions 55%, detector quantum efficiency 50% and fiber connection efficiency 90%), the efficiency of our heralded narrowband single photon source is about 4.8%. In the *d*-pulse modulation case, the utilization efficiency defined as the ratio of modulated photon counts to the unmodulated photon counts, is about (*d* − 1) × 1.4%. Here 1.4% is the ratio of the counts for one pulse to the entire unmodulated photon counts. Therefore, the overall possibility of detecting one heralded anti-Stokes photon in *d*-pulse train is with the prefactor *η*′ = 1.4% × 4.8%. Therefore, this preparation yields the heralded single photon state as where \|0〉 denotes the vacuum state, and \| *j*〉 represents the single photon at time bin *jτ*. To deal with the vacuum component that never gives rise to a click, we consider its complementary set of ontic states Λ~*clk*~ = Λ\\Λ~0~ (here, Λ~*clk*~ and Λ~0~ stand for the ontic states that give rise to a click with positive probability and states that give rise to no-click, respectively)[@b17]. The distance should thus be redefined as , where is the projection of \|*ψ*~*k*~〉 onto the space orthogonal to the vacuum, and is similarly defined. Since in our experiment, the above no-go theorem holds as [Eq. (1)](#eq19){ref-type="disp-formula"} with a slight modification: for all choices of *Q*~*k*~, the *δ*~0~-continuous *ψ*-epistemic models have where *clk* is the event that the detector clicks in one of the time bins. Hence, the epistemic models would predict a nonzero count rate in the unregistered bins which contradicts the quantum prediction. In addition, we would like to point out that the *δ*~0~-continuous *ψ*-epistemic models will still predict a nonzero counts as shown in [Eq. (3)](#eq19){ref-type="disp-formula"} even in the presence of inefficient detectors under the fair-sampling assumption[@b17][@b24]. This class of *ψ*-epistemic models are what we test in the experiment. Experimental results -------------------- In our experiment, we produce the state *ψ*~*k*~ () for several dimensions *d* = 3, 7, 16, 25, 32. As an example, the measured results for \|*ψ*~2~〉, \|*ψ*~5~〉, and \|*ψ*~7~〉 at *d* = 7 are shown in [Fig. 3](#f3){ref-type="fig"}. The measured fraction of counts in the unregistered time bins (where *N*(*j*, *Q*~*k*~) represents the counts registered in bin *j* when we prepare state \|*ψ*~*k*~〉) and their statistical errors for different dimensions *d* are shown in [Fig. 4](#f4){ref-type="fig"} (open red squares and red bars). As expected, the count rate in the empty bins is not strictly zero due to the imperfect optical components. The expected quantity *ε*~*expt*~ can be obtained through the estimation from the finite extinction ratio *R*~*ext*~ of the EOAM and the detector dark counts *C*~*dk*~*T*~*p*~ during a time bin, i.e., The dependency of on *d* calculated by [Eq. (4)](#eq22){ref-type="disp-formula"} is plotted in [Fig. 4(a)](#f4){ref-type="fig"} (black line). In addition, the uncertainty of are shown in the gray region. Any positive deviation from this gray region could support an epistemic model. The fact that the measured values of *ε*~*expt*~ reveal no positive deviation excludes a large class of *ψ*-epistemic models. Since the quality of the single-photon source is crucial, we also measure the conditional autocorrelation function for each dimensions *d* and the results are shown in [Fig. 4(b)](#f4){ref-type="fig"}. It shows that the measured is kept within the whole coincidence window of the modulated anti-Stokes photon. Furthermore, in order to shown which kinds of continue *ψ*-epistemic models are excluded in our experiment, we plot the experimental bound in the *δ*~0~, plane in [Fig. 4(c)](#f4){ref-type="fig"}. As shown in [Fig. 4(c)](#f4){ref-type="fig"}, the models with are ruled out in the experiment. Discussions and Conclusion ========================== Our experiment is based on the theoretical work of ref. [@b4]. Comparing with the previous experiment of ref. [@b17], our experiment utilizes a heralded narrowband single photon source, which has several merits that advance the test of the no-go theory of continuous *ψ*-epistemic models. First, the state we use is a really pure quantum state rather than an equivalent pure state obtained by weakening a laser beam. Second, in our experiment, the nonideal state preparation loophole is plugged up for the utilization of the single photon Fock state that is immune to phase fluctuation[@b25][@b26]. This greatly simplifies the modification of the tested *ψ*-epistemic models and also avoids the need to introduce additional assumptions. Furthermore, the overall efficiency *η* in Eq. (6) is not constant but depends on *d*, which leads to two significantly new results when compared with ref. [@b17]. First, the *δ*-continuous *ψ*-epistemic models are characterized by two parameters *δ* and . Logically, when distinctive quantum states become closer to each other, they should share more common ontic states *λ* in the ontic space Λ. In other words, *δ*-continuous *ψ*-epistemic models allow the parameter to increas when the distance *δ* decreases. However, as shown in [Fig. 4(c)](#f4){ref-type="fig"}, the experimentally measured quantity clearly decreases when the distance *δ*~0~ decreases, which is counterintuitive in the view of the epistemic models. Second, as shown in [Fig. 4(a)](#f4){ref-type="fig"}, the experimentally measured quantity decreases when the dimension *d* increases. Therefore if we use a better single photon detector with fewer dark counts (e.g., a superconducting single-photon detector) and a better modulator with higher extinction ratio to reduce the undesired counts, the measured quantity can possibly be arbitrarily small. It implies that, for any given *δ*~0~ and , one can always find an appropriate dimension *d* and detecting apparatus so that , which suggests a reliable method to exclude almost all kinds of continuous *ψ*-epistemic models. In summary, we have reported an experiment to test the no-go theorem for *δ*-continuous *ψ*-epistemic models with high-dimensional single photon quantum states. The tested no-go theorem is based on a natural assumption of continuity. However, as proposed in ref. [@b27], the *δ*-continuity does not hold well for all preparation procedures, one of such example is the purely classical distributions of the form . So there are debates[@b27] on the validity of the no-go theorem in ref. [@b4]. Nevertheless, the continuity may hold for those associated with pure quantum states. Here, the heralded single photon has a purity of *γ* \> 0.98 (see Methods for detail analysis). Hence, the assumption of continuity is reasonable in our experiment. Last, we also would like to point out that there still exist some *ψ*-epistemic models which are not *δ*-continuous, such as pairwise continuous *ψ*-epistemic models[@b11]. Thus, whether a no-go theory exists for all the *ψ*-epistemic models is still an open question deserving further exploration. Methods ======= Definition of the *δ*-continuity -------------------------------- Let *δ* \> 0 and let be the ball of radius *δ* centered on \|*ψ*〉, i.e., is the set of states \|*ϕ*〉 such that . If for any preparation *Q*, there exists an ontic state *λ* (which can depend on *Q*) for all preparations *Q*′ corresponding to quantum states in the ball centered on the state \|*ϕ*~*Q*~〉, we have *P*(*λ*\|*Q*′) \> 0. Then we say this model is *δ* continuous. The narrowband paired photon source ----------------------------------- The narrowband photon pairs are generated through spontaneous four-wave mixing (SFWM) in a two-dimensional ^85^Rb magneto-optical trap (MOT) with a longitudinal length of *L* = 1.7 *cm* as addressed in our previous work[@b20][@b28]. The atoms are prepared on the ground level \|1〉 and have an optical depth of about 50 in the \|1〉 → \|3〉 transition, as shown in [Fig. 2](#f2){ref-type="fig"}. The pump laser (780 nm, *ω*~*p*~) is 80 MHz blue detuned from the transition \|1〉 → \|4〉, and the coupling laser (795 nm, *ω*~*c*~) is on resonance with the transition \|2〉 → \|3〉. The Stokes and anti-Stokes beams, focused at the MOT center with a 1/*e*^2^ diameter of 0.3 mm, are aligned at a 2.5° angle with respect to the pump-coupling axis. In the presence of two counter propagating pump and coupling lasers, phase-matched Stokes (*ω*~*s*~) and anti-Stokes (*ω*~*as*~) photon pairs can be generated. After the quarter wave plates (QWP) and filters (F1/F2), the photon pairs are collected by single-mode fibers (SMF) and finally detected with single-photon counters (PerkinElmer SPCM-AQRH, *C*~*dk*~ = 500 ± 100 *c*/*s*, *D*~0~, *D*~1~/*D*~2~). The source is run periodically with a magneto-optical trap for a trapping time of 4.5 ms and a biphoton generation time of 0.5 ms. The purity of the heralded single photon ---------------------------------------- The narrowband time-frequency entangled paired photons generated through SFWM are used to produce the heralded single photon. The time-frequency quantum-state purity of the heralded single photon depends on the response time uncertainty of the trigger photon detector[@b25]: , where is the density operator of the heralded single photon, and Δ*τ* is the response time of the detector. With a typical time resolution of Δ*τ* = 2 *ns* and the power spectrum bandwidth of the source being smaller than 5 *MHz*, the heralded single photon has a purity of *γ* \> 0.98. Additional Information ====================== **How to cite this article**: Liao, K.-Y. *et al.* Experimental test of the no-go theorem for continuous **ψ**-epistemic models. *Sci. Rep.* **6**, 26519; doi: 10.1038/srep26519 (2016). We thank S. Massar for helpful discussions. This work was supported by the NSF of China (Grants Nos 11474107 and 11474153), the GNSFDYS (Grant No. 2014A030306012), the FOYTHEG (Grant No. Yq2013050), the PRNPGZ (Grant No. 2014010), and the PCSIRT (Grant No. IRT1243). K.L. was also supported by the SRFGS of SCNU. The authors declare no competing financial interests. **Author Contributions** K.-Y.L., X.-D.Z. and H.Y. designed and carried out the experiments. S.-L.Z., B.-Q.A. and G.-Z.G. provided theoretical assistance. Y.H. and S.-L.Z. wrote the paper and all authors discussed the contents. B.-Q.A., H.Y. and S.-L.Z. supervised the whole project. {#f1} {#f2} {#f3} {#f4} [^1]: These authors contributed equally to this work.

Background {#Sec1} ========== Our understanding of the genetic basis of Parkinson's disease (PD) has increased tremendously over the past 20 years. Mutations in the gene encoding alpha-synuclein (α-syn) were the first to be associated with genetic PD. Another monogenic causative factor in PD patients is *leucine-rich repeat kinase 2* (*LRRK2*), of which more than 100 variants have been identified \[[@CR1]\]. Asymptomatic carriers of *LRRK2* mutations constitute an ideal population for identifying predictive biomarkers of PD for several reasons: 1) a high risk of conversion to PD, 2) dopaminergic neuronal loss demonstrated by positron emission tomography (PET) scanning, and 3) similarity of the clinical phenotype of LRRK2-associated PD to that of patients with sporadic PD (sPD). While the exact involvement of LRRK2 in PD pathogenesis remains only partially understood, converging evidence suggests a role for LRRK2 in modulating inflammation \[[@CR2], [@CR3]\]. As PD has been proposed to start as an inflammatory disease \[[@CR4], [@CR5]\], it is plausible to suggest that there may be a link between *LRRK2* mutations and inflammation. Several research groups, including ours, have explored the potential of CSF alpha-synuclein (α-syn) forms as diagnostic or progression biomarkers for PD. Total α-syn (t-α-syn) levels were reported to be lower in PD, whereas oligomeric (o-α-syn) and phosphorylated Ser129-α-syn (pS129-α-syn) appear to be elevated \[[@CR6]--[@CR9]\]. CSF core biomarkers of Alzheimer's disease (AD) pathology have also been widely explored in PD cases. While a drop in CSF Amyloid-beta (Aβ-42) levels have been reported in PD \[[@CR10]\], the biomarker profile of total tau (tTau), and phosphorylated threonine 181 tau (pTau) were variable \[[@CR11], [@CR12]\]. More importantly, the potential of the aforementioned proteins as markers for PD at preclinical stage remains largely unexplored. Carriers of *LRRK2* mutations have an elevated risk of developing PD and they therefore represent a useful population in which to identify biomarkers of prodromal PD \[[@CR13]\]. However, there is a paucity of data on different forms of α-syn, AD-related proteins and inflammatory biomarkers in *LRRK2* mutation carriers \[[@CR14]--[@CR16]\]. In the present study, our primary objective was to identify a panel of CSF biomarkers for the early detection of PD, preferably at the presymptomatic stage. A secondary objective was to study whether CSF levels of particular biomarkers were associated with severity of clinical symptoms of PD. Towards that end, we measured the levels of different α-syn species, AD-related proteins and 40 different inflammatory markers in CSF samples from a well-characterized Norwegian cohort of 74 subjects with *LRRK2* mutations: 23 symptomatic individuals and 51 asymptomatic mutation carriers. In parallel, we included 60 patients with sporadic (i.e. idiopathic) PD (sPD) and 43 healthy control subjects (first-degree relatives of *LRRK2* mutation carriers (Ctrl)). Methods {#Sec2} ======= Patient selection and CSF sampling {#Sec3} ---------------------------------- Patient selection criteria and the method of CSF collection were as described in previous publications \[[@CR16], [@CR17]\]. In total, 74 Norwegian individuals from 12 different families with *LRRK2* mutations were assessed in the current study. Twenty-three patients were clinically diagnosed with PD, whereas 51 patients were healthy, asymptomatic *LRRK2* mutation carriers when enrolled in the study. These families have been extensively described in previous reports \[[@CR17]--[@CR19]\]. In addition, 60 patients with sPD and 43 age-matched controls were recruited for this study from St. Olav's Hospital at the University Hospital of Trondheim in Norway. The control group was composed of first-degree relatives of *LRRK2* mutation carriers who were not carrying *LRRK2* mutations. PD clinical diagnoses were made by experienced senior clinicians based on guidelines described by Gelb and colleagues \[[@CR20]\] and disease stage was assessed according to the Hoehn and Yahr (H&Y) scale. All patients with sPD were screened and confirmed to be negative for known *LRRK2* mutations. Patients with an age at onset of ≤50 years were confirmed to be negative for known pathogenic mutations in *Parkin* and *PINK1*. All family members of LRRK2 patients were examined for clinical features of PD by movement disorder specialists and found to be asymptomatic, although a few had mild premotor signs and an increased Unified Parkinson's Disease Rating Scale (UPDRS) score (three cases scored \> 10 on UPDRS-III) \[[@CR21]\]. The *LRRK2*-mutant PD patients were taking levodopa, and some were taking dopamine agonists and/or monoamine oxidase-B (MAO-B) inhibitors. To collect CSF, lumbar punctures were performed in overnight fasted patients between 8 and 10 am. CSF samples were aliquoted in 1.2--1.5 ml low-binding tubes, and one vial was sent for routine laboratory analysis (i.e., white and red blood cell count, total protein and glucose levels, according to the Parkinson's Progression Markers Initiative \[PPMI\] protocol), whereas the majority of the vials were frozen fewer than 15 min after collection following centrifugation at 2000 g at 4 °C then sub-aliquoted and stored at − 80 °C until further analysis. All patients provided signed informed consent, and the study was approved by the Regional Committee for Medical and Health Research Ethics. Measurement of alpha-synuclein species {#Sec4} -------------------------------------- All immunoassays used to measure the different species of α-syn were developed in-house and described in previous reports \[[@CR9], [@CR22], [@CR23]\] . Briefly, to capture t- or pS129-α-syn, a 384-well ELISA microplate was coated with 0.1 or 0.5 μg/ml Syn-140 (a sheep anti-α-syn polyclonal antibody) in 200 mM NaHCO~3~, pH 9.6 (50 μl/well) by overnight incubation at 4 °C, while 0.2 μg/ml of our mouse conformation-specific antibody, Syn-O2, was used to capture o-α-syn. After incubation with 100 μl/well of blocking buffer (PBS-T containing 2.25% gelatin) for 2 h at 37 °C, 50 μl/well of the CSF samples (diluted 1:2 in artificial CSF) along with serial dilutions of recombinant human t-, pS129-or o-α-syn (50 μl) were dispensed in each well, and the plate was incubated at 37 °C for 2.5 h. After washing with PBS-T, 50 μl/well of 11D12 (a mouse anti-α-syn monoclonal antibody) \[[@CR9]\], PS129 (a mouse anti-pS129-α-syn monoclonal antibody) \[[@CR9]\], or FL-140 (rabbit polyclonal antibody, Santa Cruz Biotechnology, Santa Cruz, CA, USA), for measuring t-, pS129-α-syn, or o-, respectively, were added to the corresponding wells and incubated at 37 °C for 2 h. Next, the plates were washed and incubated for 2 h at 37 °C with 50 μl/well of species-appropriate secondary antibody (goat anti-rabbit IgG HRP or donkey anti-mouse IgG HRP, Jackson ImmunoResearch Laboratories Inc., USA, 1:20,000 dilution). After washing, the plates were incubated with 50 μl/well of enhanced chemiluminescent substrate (Super Signal ELISA Femto, Pierce Biotechnology, USA). The chemiluminescence, expressed in relative light units, was immediately measured using a PerkinElmer Envision multi-label plate reader (PerkinElmer, Finland). CSF samples were measured in a blinded fashion and randomized for analysis, with all LRRK2 symptomatic/asymptomatic, PD and HC samples being tested together on the same ELISA microplates. A series of internal controls was also run to check for run-to-run variations. The concentrations of α-syn species in the samples were calculated using the corresponding standard curves. Measurement of AD biomarkers {#Sec5} ---------------------------- CSF Aβ42, Aβ40, total tau (tTau), and phosphorylated threonine 181 tau (pTau) were measured using MILLIPLEX® MAP Human Amyloid Beta Tau Magnetic Bead Panel (Luminex xMAP) run on the Bio-Plex® 3D instrument (Bio-Rad Laboratories, Hercules, CA) according to manufacturer's instructions. The kit allows simultaneous quantification of Aβ40, Aβ42, tTau, and pTau. Measurement of inflammatory markers {#Sec6} ----------------------------------- A magnetic human chemokine bioplex assay (Bio-Rad Laboratories, Hercules, CA) was used to measure 40 chemokines from human CSF samples (6Ckine/CCL21, BCA-1/CXCL13, CTACK/CCL27, ENA-78/CXCL5, Eotaxin/CCL11, Eotaxin-2/CCL24, Eotaxin-3/CCL26, Fractalkine /CX3CL1, GCP-2/CXCL6,GM-CSF, Gro-α/CXCL1, Gro-β/CXCL2, I-309/CCL1, IFN-ϒ, IL-1β, IL-2, IL-4, IL-6, IL-8/CXCL8, IL-10, IL-16, IP-10/CXCL10, I-TAC/CXCL11, MCP-1/CCL2, MCP-2/CCL8, MCP-3/CCL7, MCP-4/CCL13, MDC/CCL22, MIF, MIG/CXCL9, MIP-1α/CCL3,MIP-1δ/CCL15, MIP-3α/CCL20,MIP-3β/CCL19, MPIF-1/CCL23, SCYB16/CXCL16, SDF-1α + β/CXCL12, TARC/CCL17, TECK/CCL25, TNF-α). Bioplex assays was run according to the manufacturer's instructions using the recommended 1 in 2 dilution of CSF. Plates were read using the Bio-Plex® 3D suspension array system; the next generation multiplexing platform based on xMAP technology. Enclosed standards were used to generate an 8-point standard curve to which a 5-parameter logistic curve was fitted and were used to quantify unknown concentrations using BioPlex manger software. The coefficient of variance between duplicates was mostly \< 10%, as was the variance between standard curves run on separate plates. Only chemokines with robust readings above background were considered for further analysis. Statistical analysis {#Sec7} -------------------- IBM SPSS software (version 24.0, Chicago, IL, USA) was used for the statistical analyses of the data, whereas R 3.6.1 was used for plotting the box and whisker plots. Demographic and clinical characteristics were compared between study groups using chi-square tests, analysis of variance with post hoc Bonferroni tests or Kruskal- Wallis tests followed by Mann-Whitney U tests, where appropriate. All datasets were tested for normality and the presence of outliers. As data were considered inappropriate for parametric analyses, Spearman rank-order correlation coefficients were used to examine correlations within the study group. For all CSF biomarkers that showed robust readings above background (i.e. α-syn species, AD markers and 18 inflammatory markers), differences between the diagnostic groups were assessed using general linear models (GLMs) corrected for age and gender with post-hoc Bonferroni corrections for multiple comparisons. Only biomarkers that varied significantly among the study groups were included in the next Discriminant Function analyses. Since our main objective was to identify predictive markers for the early detection of PD, both symptomatic *LRRK2* mutation carriers and sPD groups were combined for discriminant function analysis. Discriminant function analysis evaluates canonical discriminant functions based on combinations of the selected markers which contribute maximally to group separation and assesses how well these canonical discriminant functions discriminate the diagnostic groups. CSF biomarker data were Z-transformed prior to discriminant function analysis. We also performed multivariate logistic regression analyses with backward stepwise selection (separate analyses for each comparison). The control group was entered as a reference category and t-, o- and pSer129-α-syn, TNF-α and IL-16 were enrolled as predictors. For the resulting models, we report AUC, sensitivity, specificity and OR (95% CI) of the individual model. Results {#Sec8} ======= Patient population and demographics {#Sec9} ----------------------------------- Demographics, clinical characteristics and CSF biomarkers levels of the study groups are summarized in Table [1](#Tab1){ref-type="table"}. Twenty-three of the 74 subjects (30%) with *LRRK2* mutations analyzed had a manifest PD and were carrying either the most common *LRRK2* mutation, G2019S, or a different *LRRK2* mutation, N1437H. The symptomatic *LRRK2* mutation carriers had a mean age of 60 ± 11 years. 51 subjects of the 74 subjects (70%) had no symptoms of PD at the time of CSF sample collection and these cases had a mean age of 57 ± 14 years. There was no significant difference in disease duration between the symptomatic *LRRK2* mutation carriers and the sPD patients. Moreover, no difference was found between groups with regard to the routine CSF levels of white-red cell count as well as the total protein, albumin, and glucose levels, including the plasma glucose level. Controlling for age and gender did not significantly alter the results. Table 1Demographics and CSF biomarkers by diagnostic groupCtrl (*n* = 43)sPD (*n* = 60)Asymptomatic *LRRK2* mutation carriers (*n* = 51)Symptomatic *LRRK2* mutation carriers (*n* = 23)Age (y), mean ± SD^a^49 ± 1857 ± 1057 ± 1460 ± 11Gender (male), n(%)^a^19 (38.0%)36 (52.9%)26 (50%)6 (25%)Disease duration (y)^b^NA4 (1--6)NA7 (5--19)MoCA^c^27 (27--29)27 (25--28)27 (26--28)25 (23.7--27)H&YNA2 (2--2)0 (0--0)2 (2--3)UPDRS-III^d^NA24 (19--29)3 (0.25--5)24 (19.5--27)t-α-syn (pg/mL)^e^816 (596--1112)573 (466--710)^\*\*\*^617 (431--803)^\*\*^608 (432--740)^\*\*^o-α-syn (pg/mL)^e^161 (148--186)187.5 (170.5--219.8)^\*\*^183 (160--230)^\*\*^182 (146--196)pS129-α-syn (pg/mL)^e^116 (103--145)139 (114.25--163)121 (94--150)122 (106--145)tTau (pg/mL)^f^141.2 (89.6--201.3)134.7 (88.3--224.9)123.6 (86.7--236.4)113.5 (94.9--167.8)pTau (pg/mL)^f^12 (9--16)14.3 (9.6--21.7)14.7 (10.7--20.8)15.9 (10.5--22.4)Aβ-40 (pg/mL)^f^2506 (1967--3120)2595.7 (2193.7--3098.2)2776.2 (2287.6--4020.6)3048.9 (1959.3--3810.8)Aβ-42 (pg/mL)^f^517.9 (404.77--738)535.4 (405.11--662.6)580.3 (457.8--905.3)539.9 (410.9--880.9)IL-16^g^6.5 (4.9--8)5.1 (4--6.6)^\*^5.1 (3.9--6.8)4.3 (2.8--5.3)TNF-α^g^3.3 (2.6--5.5)4.2 (2.4--6.5)5.5 (4.2--7.4)^\*^5.5 (2.5--7.3)Data are expressed as mean ± SD, median (IQR), or n (%). Demographical differences between groups were analyzed using analysis of variance with post hoc Bonferroni tests (age), X2 tests (gender), and Kruskal-Wallis with post hoc Mann-Whitney U tests (MoCA, H&Y, UPDRS and disease duration). Differences in CSF biomarker levels between groups were assessed with a GLM adjusted for age and gender. T-, o-, and pS129-α-syn, IL-16, TNF-α, tTau, ptau, Aβ40 and Aβ-42 were log-transformed, yet presented as raw data (\* *p* \< 0.05, ^\*\*^*p* \< 0.01, ^\*\*\*^*p* \< 0.001 compared with Ctrl)Aβ1--42, amyloid β1--42; Ctrl, healthy controls; H&Y, Hoehn and Yahr scale; IL-16, interleukin-16; MoCA, Montreal Cognitive Assessment; NA, not applicable; o-α-syn, oligomeric α-synuclein; pSer129-α-synuclein, phosphorylated α-synuclein protein at serine 129; pTau, tau phosphorylated at threonine 181; sPD, sporadic PD; TNF-α, tumor necrosis factor-alpha; tTau, total tau protein; t-α-syn, total α-synuclein; UPDRS-III, Unified Parkinson's Disease Rating Scale^a^Ctrl, *n* = 43; sPD, *n* = 58; Asymptomatic *LRRK2* mutation carriers, *n* = 51, Symptomatic *LRRK2* mutation carriers, *n* = 23^b^ Ctrl, n = NA; sPD, *n* = 59; Symptomatic *LRRK2* mutation carriers, *n* = 21^c^ Ctrl, *n* = 15; sPD, *n* = 59; Asymptomatic *LRRK2* mutation carriers, *n* = 48, Symptomatic *LRRK2* mutation carriers, *n* = 22^d^ Ctrl, *n* = 15; sPD, *n* = 59; Asymptomatic *LRRK2* mutation carriers, *n* = 48, Symptomatic *LRRK2*mutation carriers, *n* = 21^e^ Ctrl, *n* = 43; sPD, *n* = 60; Asymptomatic *LRRK2* mutation carriers, *n* = 51, Symptomatic *LRRK2* mutation carriers, *n* = 23^f^ Ctrl, *n* = 28; sPD, *n* = 40; Asymptomatic *LRRK2* mutation carriers, *n* = 26, Symptomatic *LRRK2* mutation carriers, *n* = 11^g^ Ctrl, *n* = 28; sPD, *n* = 42; Asymptomatic *LRRK2* mutation carriers, *n* = 24, Symptomatic *LRRK2* mutation carriers, *n* = 12 CSF biomarkers levels in diagnostic groups {#Sec10} ------------------------------------------ The levels of CSF t-, o- and pS129-α-syn in the different diagnostic groups are presented in Table [1](#Tab1){ref-type="table"} and Fig. [1](#Fig1){ref-type="fig"}. The GLM showed differences between diagnostic groups in CSF t-, o- and pS129-α-syn, TNF-α and IL-16 levels (*P* \< 0.01, adjusted for age and gender; Table [1](#Tab1){ref-type="table"}). Post-hoc tests revealed that in comparison to Ctrl (median and IQR = 816 (596--1112) pg/ml, *n* = 43), levels of CSF t-α-syn were significantly lower in sPD group (median and IQR = 573 (466.5--710.5), *n* = 60) (*P* \< 0.001, Fig. [1](#Fig1){ref-type="fig"}a), asymptomatic *LRRK2* mutation carriers (median and IQR = 617 (431--803) pg/ml, *n* = 51) (*P* \< 0.01, Fig. [1](#Fig1){ref-type="fig"}a), and symptomatic *LRRK2* mutation carriers (median and IQR = 608 (432--740) pg/ml, *n* = 23) (*P* \< 0.01, Fig. [1](#Fig1){ref-type="fig"}a). There were no significant differences in the levels of CSF t-α-syn between the groups of sPD, symptomatic and asymptomatic *LRRK2* mutation carriers. On the other hand, and as shown in Fig. [1](#Fig1){ref-type="fig"}b, both sPD (median and IQR = 187.5 (170.5--219.8) pg/ml, *n* = 60) (*P* \< 0.001, Fig. [1](#Fig1){ref-type="fig"}b) and asymptomatic LRRK2 (median and IQR = 183 (160--230) pg/ml, *n* = 51) (*P* \< 0.01, Fig. [1](#Fig1){ref-type="fig"}b) groups had CSF profile with higher levels of o-α-syn compared with Ctrl (median and IQR = 161 (148--186) pg/ml, *n* = 43). Examination of CSF levels of pS129-α-syn revealed a trend of an increase in sPD group (median and IQR = 139 (114.25--163) pg/ml, n = 60) compared to Ctrl (median and IQR = 116 (103--145) pg/ml, *n* = 43), however the difference did not reach statistical significance. (Fig. [1](#Fig1){ref-type="fig"}c). The ratios of o-α-syn/t-α-syn % and pSer129-α-syn/t-α-syn were both higher in sPD, symptomatic and asymptomatic *LRRK2* mutation carriers compared with Ctrl (*P* \< 0.01; Fig. [1](#Fig1){ref-type="fig"}d, e). Fig. 1Box-and-whiskers plots of CSF levels of α-syn forms in sPD, symptomatic and asymptomatic *LRRK2* mutation carriers, and Ctrl. Box-and-whiskers plots of CSF levels of α-syn species in sPD, asymptomatic LRRK2 carriers, symptomatic LRRK2 carriers and Ctrls. **a** CSF levels of t-α-syn, **b** CSF levels of o-α-syn, **c** CSF levels of pSer129-α-syn, **d** ratio of o-α-syn/ t-α-syn %, **e** ratio of pSer129-α-syn/ t-α-syn %. The line through the middle of the boxes corresponds to the median and the lower and the upper lines to the 25th and 75th percentiles, respectively. The whiskers extend from the 5th percentile on the bottom to the 95th percentile on top. Differences between groups were assessed with the GLM compared to Ctrl group and adjusted for age and gender. \**P* \< 0.05, \*\**P* \< 0.01, \*\*\**P* \< 0.001 The age- and gender-adjusted GLM revealed no significant differences in levels of AD biomarkers between the study groups. To determine whether inflammatory biomarkers were altered among the different study groups, a panel of 40 proinflammatory cytokines, chemokines, and growth factors were assessed in CSF samples from all study participants. Of the 40 markers measured, 18 showed robust readings above background and were further analyzed (6Ckine, IL-6, SDF-1 α + β, IL-16, MDC, MIF, TNF-α, MPIF-1, Eotaxin, IP-10, MCP-1, IL-8, MCP-2, SCYP16, MIP-1Delta, CTACK, MIP-3β, Fractalkine). GLM Bonferroni corrected analysis revealed that of the remaining 18, only IL-16 and TNF-α were significantly different. A univariate, post-hoc analysis corrected for age and gender as covariates demonstrated that IL-16 levels were significantly lower in sPD group (median and IQR = 5.1 (4--6.6) pg/ml, *n* = 42) (*P* \< 0.05) compared with Ctrl (median and IQR 6.5 = (4.94--8) pg/ml, *n* = 28). More interestingly, analysis showed a significant increase of TNF-α in asymptomatic *LRRK2* mutation carriers (median and IQR = 5.5 (4.2--7.4) pg/ml (*P* \< 0.05), *n* = 24) compared to Ctrl (median and IQR = 3.3 (2.6--5.5) pg/ml (*P* \< 0.05), *n* = 28). The inflammatory profiles of symptomatic *LRRK2* mutation carriers and sPD cases were not significantly different (data not shown). By use of Spearman correlations, we evaluated associations between different CSF α-syn forms and AD core biomarkers (Supplementary Table [1](#MOESM1){ref-type="media"}). For the Ctrl group, but not for any of the other groups, we found a positive association between o-α-syn and pSer129-α-syn (*r* = 0.39, *P* \< 0.01). We also noted an inverse correlation between t-α-syn and o-α-syn (*r* = 0.31, *P* \< 0.05) that was only present in asymptomatic *LRRK2* mutation carriers. When we explored correlations between α-syn species and the AD biomarkers, we found that t-α-syn positively correlated with Aβ-40 and Aβ-42 in the asymptomatic LRRK2 group (*r* = 0.568, *P* \< 0.01, *r* = 0.485, *P* \< 0.05). No other correlations were noted. Correlations between CSF alpha-synuclein levels and clinical parameters {#Sec11} ----------------------------------------------------------------------- Correlational analyses of CSF levels of α-syn species with clinical parameters (age, disease duration, UPDRS-III, H&Y and MoCA) are shown in Supplementary Table [2](#MOESM1){ref-type="media"}. In summary, higher levels of CSF t-α-syn correlated with: worse cognitive function as assessed by the Montreal Cognitive Assessment (MoCA) score (*r* = − 0.44, *P* \< 0.01) in the sPD group. Similarly, a weak correlation of t-α-syn and age was noted in PD group (*r* = 0.29, *P* \< 0.01). While pS129-α-syn positively correlated with age in PD group (*r* = 0.384, *P* \< 0.01), aging was also associated with increased levels of CSF TNF-α in both PD and asymptomatic carriers groups (*r* = 0.362, *P* \< 0.01, *r* = 0.492, *P* \< 0.01, respectively). Discriminant function analysis {#Sec12} ------------------------------ In an attempt to identify the optimal panel that can serve as predictive markers for PD at the prodromal stage, we performed a discriminant function analysis of biomarkers that were significantly different between the groups. Canonical discriminant function classification results are presented in Table [2](#Tab2){ref-type="table"}. In the analysis, both groups of sPD and symptomatic LRRK2 carriers were combined as one PD group. A panel of t-, o- and pS129-α-syn, TNF-α and IL-16 together correctly classified 60% of all cases in the asymptomatic *LRRK2* mutation carriers, PD, and Ctrl groups (lambda = 0.644, *P* \< 0.001). The discrimination plot of the two canonical discriminant functions for discrimination of the three groups is presented in Fig. [2](#Fig2){ref-type="fig"} and the loadings of individual predictors on each discriminant function are shown in Supplementary Table [3](#MOESM1){ref-type="media"}. Canonical discriminant function 1 strongly correlated with t-α-syn (*r* = − 0.694^**\***^), o-α-syn (*r* = − 0.499\*) and pS129-α-syn (*r* = 0.390\*) and discriminated both asymptomatic *LRRK2* mutation carriers and PD groups from Ctrl group; we will subsequently refer to this function as the *Disease* function. Canonical discriminant function 2 strongly correlated with TNF-α (*r* = 0.678\*) and IL-16 (*r* = − 0.554\*) and further discriminated asymptomatic *LRRK2* mutation carriers group from Ctrl group; we will subsequently refer to this function as the *Prodromal* function. Asymptomatic *LRRK2* mutation carriers' centroid is located at the intersection of both the *Disease* axis and the *Prodromal* disorders axis. Table 2Canonical Discriminant functions classification results ^a^GroupPredicted Group MembershipTotalctrlPDAsymptomatic carriersOriginalCountCtrl203427PD7331454Asymptomatic carriers591024%Ctrl**74.1**11.114.8100.0PD13.0**61.1**25.9100.0Asymptomatic carriers20.837.5**41.7**100.0^a^ 60.0% of original grouped cases correctly classifiedFig. 2Discriminant function plot of canonical discriminant functions. CSF biomarkers were z-transformed before analyses. Discriminant function plot of canonical discriminant functions for discrimination of asymptomatic *LRRK2* mutation carriers, PD and controls. Yellow dots indicate individual data of control subjects, green dots indicate individual data of Parkinson's disease patients (sporadic and symptomatic *LRRK2* mutation carriers) and red dots indicate individual data of LRRK2 asymptomatic carriers. The golden stars represent the group centroids We employed backward-elimination multiple logistic regression analyses to identify optimal biomarker panels for bilateral comparisons between: \[[@CR1]\] asymptomatic *LRRK2* mutation carriers and Ctrl, and \[[@CR2]\] PD patients and Ctrl, for which t-, o- and pS129-α-syn, TNF-α, and IL-16 were entered as predictors and the Ctrl group was used as the reference group in each comparison. A summary of the final models are shown in Table [3](#Tab3){ref-type="table"}. The combination of t-α-syn, o-α-syn, and TNF-α discriminated the asymptomatic *LRRK2* mutation carriers group from the Ctrl group: low levels of t-α-syn (OR, 0.997; 95% CI, 0.994--0.999), high levels of o-α-syn (OR, 1.029, 95% CI, 0.999--1.060), and high levels of TNF-α (OR, 1.418; 95% CI, 0.998--2.014) indicate that individuals are at higher risk developing PD. Examining PD and Ctrl groups only, we found that low levels of t-α-syn (OR, 0.996; 95% CI, 0.994--0.999), high levels of o-α-syn (OR, 1.031; 95% CI,1.005--1.056), high levels of pS129-α-syn (OR, 1.035; 95% CI, 1.010--1.059), low levels of IL-16 (OR, 0.785; 95% CI, 0.603--1.022) differentiates the PD group from the Ctrl group. Receiver operating characteristic curves for both models are illustrated in Fig. [3](#Fig3){ref-type="fig"}. Each of the models generated an area under the curve (AUC) of \> 0.80. Table 3Logistic regression analysis of multiple CSF biomarkersCtrl GroupPredictorsOR (95% CI)Accuracy of model*P*Asymptomatic carrierst-α-syno-α-synTNF-α0.997 (0.994--0.999)1.029 (0.999--1.060)1.418 (0.998--2.014)AUC: 0.843 (0.724--0.961)Sens: 87.5%,Spec: 66.7%0.000PD (i.e. sPD & symptomatic *LRRK2* mutation carriers)t-α-syno-α-synpS129-α-synIL-160.996 (0.994--0.999)1.031 (1.005--1.056)1.035 (1.010--1.059)0.785 (0.60--1.022)AUC: 0.896 (0.823--0.969)Sens: 87.0%Spec: 78.6%0.000*AUC* area under the curve, *HC* Healthy controls, *NPV* negative predictive value, *OR* odds ratio, *o-α-syn* oligomeric α-synuclein, *PD* sporadic and symptomatic *LRRK2* mutation patients, *PPV* positive predictive value; pSer129-α-synuclein, phosphorylated α-synuclein protein at serine 129, *Sens* sensitivity, *Spec* specificity, *TNF-α* tumor necrosis factor-alpha, *t-α-syn* total α-synucleinFig. 3Receiver operating characteristic curves (ROC) showing the diagnostic accuracy of the final logistic regression models. Asymptomatic *LRRK2* mutation carriers vs Ctrl subjects, (B) PD patients vs Ctrl subjects Discussion {#Sec13} ========== *LRRK2* mutations are well-described as a cause of genetic familial parkinsonism, and drugs inhibiting LRRK2 kinase activity are already in clinical trials \[[@CR24]\]. However, as *LRRK2* mutation carriers are at high risk of developing PD, asymptomatic individuals with *LRRK2* mutations are an excellent group for discovery of biomarkers of prodromal PD based on the premise that they are highly likely to develop PD in future. In the current study, we measured 47 different candidate biomarkers and found evidence that low levels of CSF t-α-syn, and high levels of CSF o-α-syn and TNF-α potentially differentiate asymptomatic *LRRK2* mutation carriers, i.e. PD subjects at the prodromal phase, from healthy controls. As these biomarkers in combination had greater discriminant power than those in isolation, these findings emphasize the value of combining multiple markers for early detection of PD. In line with previous cross-sectional studies assessing the diagnostic power of CSF α-syn forms \[[@CR8], [@CR9]\], we report that sPD patients had significantly decreased levels of CSF t-α-syn and increased levels of CSF o- and pS129-α-syn compared to Ctrl group. Importantly, our results also show that asymptomatic *LRRK2* mutation carriers displayed significantly higher CSF o-α-syn than Ctrls. These findings support the hypothesis that α-syn oligomerization is an early event in the pathophysiology of PD and add further weight to evaluation of CSF o-α-syn as a candidate biomarker for detection of prodromal PD. A previous study showed a similar trend, although the results were not significant \[[@CR16]\]. One possible explanation for disparate results between both studies is that the use of conformation-specific antibodies and oligomeric-specific ELISA in the present study, minimized the overlap between groups and improved the discriminant power between the study groups. Our results partially contradict with Vilas et al.*,* 2016 study, where CSF t-α-syn levels were lowered only in idiopathic (i.e. sporadic) PD group in contrast to other diagnostic groups; however, the differences in CSF tTau, pTau, Aβ-40 or Aβ-42 levels didn't reach statistical significance. Immunoassays employed to measure AD biomarkers in both studies used Luminex xMAP but employed different antibodies, which may underlie different results. If this is the case, this would have important implications for the design of future biomarkers studies. Several biomarkers and brain tissue studies have shown that neuroinflammatory process precedes neurodegeneration in PD \[[@CR25]--[@CR27]\], whilst very few studies have evaluated inflammatory biomarkers in relation to α-syn forms \[[@CR15]\]. In the current study, we found CSF TNF-α to be significantly higher in asymptomatic *LRRK2* mutation carriers compared to Ctrls. In a recent 11-year study in over 4 million Norwegians, an inhaled asthma medication with anti-TNF activity, a brain-penetrant drug, was associated with lower levels of t-α-syn and decreased the risk of PD \[[@CR28], [@CR29]\]. While the authors of the study didn't claim a causative association, their findings correspond with our present findings, as asymptomatic *LRRK2* mutation carriers showed significantly lower CSF t-α-syn and higher TNF-α and o-α-syn. However, the control subjects in our study were significantly younger than sPD, symptomatic and asymptomatic *LRRK2* mutation carriers (*p* \< 0.05), a notable finding given that previous studies have demonstrated an age-dependent increase in CSF TNF-α levels \[[@CR30]\], which we also observed (Supplementary Table [2](#MOESM1){ref-type="media"}). However, CSF TNF-α levels were still significantly increased in asymptomatic *LRRK2* mutation carriers compared with Ctrl subjects when CSF TNF-α values were analyzed after age adjustment (Table [1](#Tab1){ref-type="table"}). We observed no effect of gender on biomarker levels in all study groups, and analysis was performed with and without gender adjustment with no difference in overall findings. Therefore, we concluded that gender bias did not affect the results in this study. It is important to note that *LRRK2* mutations (G2019S, N1437H) have incomplete penetrance, meaning that not all of the asymptomatic *LRRK2* mutation carriers in our cohort would develop PD. This may explain the overlap between asymptomatic *LRRK2* mutation carriers and both Ctrl and sPD groups (Fig. [1](#Fig1){ref-type="fig"}), falling almost at the intersection in the discriminant function analysis (Fig. [3](#Fig3){ref-type="fig"}). In light of the incomplete penetrance of *LRRK2*, discriminant function analysis correctly classified 60.0% of original grouped cases, where 74.1% Ctrl, 61.1% PD and 41.7% cases of asymptomatic *LRRK2* mutation carriers were correctly classified (Table [2](#Tab2){ref-type="table"}). It is tempting to speculate that the lower discriminant function analysis levels for asymptomatic *LRRK2* cases reflects the incomplete penetrance of this mutation, and could suggest that the cases correctly discriminated are those most likely to develop PD. Future studies could further explore this question in *LRRK2* patients in large prospective studies, such as the ongoing Parkinson's progression markers initiative, to determine their diagnostic utility in predicting PD in *LRRK2* mutation carriers. In a recent study by Halliday and colleagues, using CSF and serum samples from the Michael J. Fox Foundation LRRK2 cohort consortium \[[@CR15]\], 28 cytokines were measured in CSF and compared between Ctrl (*n* = 22) and asymptomatic *LRRK2* mutation carriers (*n* = 25). However, in this study, none of the markers, including TNF-α, distinguished between the two groups but CSF TNF-α levels combined with 5 other cytokines significantly differentiated these cases from sPD (*n* = 29) and symptomatic *LRRK2* mutation (G2019S) carriers. The discrepancy seen in our results and the above mentioned study could be due to the differences in the inclusion and exclusion criteria of the subjects, and/or time difference in processing or storing the samples. Admittedly, the size of our cohort is relatively small, particularly considering the number of subjects in which all the biomarkers' measurements were available. Furthermore, the only control group in this study was composed of first-degree relatives of *LRRK2* mutations carriers and, although they were confirmed negative for *LRRK2* mutations, including another standard control group shall be considered in further studies. Previous studies have described heterogeneous pathologies in *LRRK2* mutation carriers, where different brain areas and methods were studied and used \[[@CR31]--[@CR33]\]. Such differences in neuropathological changes elicited by *LRRK2* mutations necessitates the need for cohorts where neuropathological examination of *LRRK2* mutation carriers is conducted both to provide neuropathological confirmation of clinical diagnoses, and to better define and validate differential profiles of biomarkers. Conclusions {#Sec14} =========== Our study, by demonstrating lower CSF t-α-syn levels, higher o-α-syn and TNF-α in asymptomatic *LRRK2* mutation carriers highlights the power of those biomarkers at providing an early detection of PD. Future studies are necessary to confirm the potential sensitivity and specificity of combing α-syn species with inflammatory biomarkers as predictive and perhaps progression biomarkers of PD. Comparison of CSF and blood levels of the current biomarkers in larger cohorts with longitudinal follow-up of *LRRK2* mutation carriers, and other "at risk" groups, are of great importance. Such studies are crucial to select biomarkers that could identify individuals at high risk to convert to PD, who would then be the target group for the development of preventive treatments. Supplementary information ========================= {#Sec15} **Additional file 1: Supplementary Table 1.** Associations between CSF biomarkers. No correlation between CSF biomarkers were present in symptomatic *LRRK2* mutation carriers group alone. Both sPD and symptomatic *LRRK2* mutation carriers groups were combined as one PD group. Associations between CSF biomarkers were assessed with Spearman correlation coefficients. Data shown as r. Significance: \*\*\* *p* \< 0.001; \*\* *p* \< 0.01; \* *p* \< 0.05 Aβ1--42, amyloid β1--42; Ctrl, Healthy controls; o-α-syn, oligomeric α-synuclein; pSer129-α-synuclein, phosphorylated α-synuclein protein at serine 129; pTau, tau phosphorylated at threonine 181; PD, Parkinson's disease patients; tTau, total tau protein; and t-α-syn, total α-synuclein. **Supplementary Table 2.** Associations between CSF α-syn species and clinical parameters. Associations between CSF biomarkers were assessed with Spearman correlation coefficients. Data shown as r. Significance: \*\*\* *p* \< 0.001; \*\* *p* \< 0.01; \* *p* \< 0.05. Both sPD and symptomatic *LRRK2* mutation carriers groups were combined as one PD group. Aβ1--42, amyloid β1--42; Ctrl, healthy controls; H&Y, Hoehn and Yahr scale; MoCA, Montreal Cognitive Assessment; NA, not applicable; o-α-syn, oligomeric α-synuclein; pSer129-α-synuclein, phosphorylated α-synuclein protein at serine 129; pTau, tau phosphorylated at threonine 181; sPD, sporadic PD; TNF-α, tumor necrosis factor-alpha; tTau, total tau protein; t-α-syn, total α-synuclein; UPDRS-III, Unified Parkinson's Disease Rating Scale. **Supplementary Table 3.** Discriminant loadings for each individual predictor. The correlation coefficient represents the relative contribution for each predictor to group separation. IL-16, interlukin-16; o-α-syn, α-synuclein oligomers; pS129-α-syn, phosphorylated Ser 129 α-synuclein; t-α-syn, total α-synuclein; TNF- α, tumor necrosis factor- α. PD : Parkinson's disease CSF : Cerebrospinal fluid TNF-α : Tumor necrosis factor-alpha AUC : Area under the curve PET : Positron emission tomography AD : Alzheimer's disease UPDRS : Unified Parkinson's disease rating scale MAO : Monoamine oxidase inhibitor PPMI : Parkinson's Progression Markers Initiative ELISA : Enzyme Linked Immunosorbent Assay PBS : Phosphate-buffered saline PBS-T : Phosphate-buffered saline with tween-20 HRP : Horseradish peroxidase HC : Healthy control OR : Odds ratio CI : Confidence interval GLM : General linear model IL : Interleukin IQR : Interquartile range NPRP : National Priorities Research Program Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s40035-020-00192-4. The authors thank all patients and healthy subjects, lab manager Dr. Houari Abdesselem, funding agencies, Michael J Fox Foundation and Norwegian Parkinson Foundation This work would have not been possible without their valuable contribution. Disclosures {#FPar1} =========== The authors have no financial disclosures or conflicts of interest to report. Study design and supervision: N.M, O.E. Conduction of the experiments: N.M, M.T. and N.V. Clinical data collection: J.A. and E.H. Data analysis: N.M. and N.E. Writing the first draft of the manuscript and incorporating revisions from other authors: N.M. Data interpretation and critical revision of the manuscript for important intellectual content: N.M, J.A., W.B., T.T., B.M., H.B., and O.E. Final review of the draft and approval to submit for publication: O.E. All authors read and approved the final manuscript. This study was supported by Strat-up Funding to OE from Qatar Biomedical Research Institute (SF 2007--007) and Qatar National Research Fund (NPRP No.: 8--517--3-112). The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request. All patients provided signed informed consent, and the study was approved by the Regional Committee for Medical and Health Research Ethics. Not applicable. The authors declare that they have no competing interests.

1.. Introduction {#s1} ================ Metabolic syndrome (MetS) refers to a cluster of metabolic disorders such as hyperglycemia, insuline resistance, type II diabetes mellitus, and hyperlipidemia which cause obesity and cardiovascular diseases [@b1]. Diet quality is a strong predictor of this syndrome [@b1],[@b2]. A diet, which is high-carbohydrate and low fibre, increases directly postprandial levels of blood glucose and insuline, which leads to insuline resistance [@b3]. Insuline resistance is the cause of type II diabetes mellitus and cardiovascular diseases [@b4]. Therefore, eating well-quality diets would be well-advised for controlling metabolic syndrome [@b5]. White rice is the most commonly starchy food staple, especially in developing countries of Asia and Africa [@b6]. However, its high glycemic index leads to white rice could not be frequent consumption food for diabetes patients [@b7]. Pre-germinated brown rice (PGBR) may be more benefits than white rice. PGBR, a newly developed type of rice, was made by soaking brown rice kernels in water to slightly germinated them [@b8]. There are few studies proved the effects of PGBR on the prevention and treatment of metabolic syndrome [@b9]--[@b11]. The aim of this study was to investigate the effects of PGBR used to treat metabolic syndrome patients. 2.. Materials and methods {#s2} ========================= 2.1.. Settings and subjects study {#s2a} --------------------------------- The protocol of this study was approved by the ethics committee of the National Institute of Nutrition (NIN), Hanoi, Vietnam (number 1656/QĐ-VDD) and also approval from all participants, before recruiting subjects. 80 subjects with the metabolic syndrome (MetS) were recruited in the intervention. All of them live in Bac Ninh City. Bac Ninh City is in the Red River Delta of northern Vietnam, approximately 40 km from the capital, Hanoi. All subjects were fully informed concerning the process of the study, and the study was designed in with the Helsinki Declaration on Human Studies. Eligibility criteria were: age from 55 to 70 years old, live in Bac Ninh City, and were diagnosed having MetS. A subject was defined as having MetS if he/she was any three of the following six factors: waist circumference \> 90 cm for men and \> 80 cm for women; blood triglyceride level ≥ 1.69 mmol/L; serum HDL-cholesterol concentration \< 1.04 mmol/L for men and \< 1.3 mmol/L for women; fasting plasma glucose \> 6.1 mmol/L; systolic blood pressure ≥ 130 mmHg and/or diastolic blood pressure ≥ 85 mmHg. Exclusion criteria included the following: subjects who were taking medicines that controlled blood glucose or lipid metabolic; subjects who were suffering from serious heart diseases, brain diseases, kidney diseases, liver diseases, or gastrointestinal diseases; subjects who were consuming PGBR daily; subjects who had body mass index (BMI) under 18.5; subjects who were not feeling well after starting the intervention. After screening, 86 subjects met the requirements. During 3 months of the intervention, 3 subjects in each group were excluded because of the individual health status. The subjects were divided randomly into two groups: a PGBR group (n = 43) and a placebo group (n = 43). The subjects in PGBR group received 1 PGBR capsule (2.0 kg) containing 200 g PGBR/day for 10 days/time, and subjects in the placebo group took 1 capsule (2.0 kg) containing 200 g white rice/day for 10 days/time. The amount of energy, lipid, and protein is almost equivalent to the placebo and PGBR capsules. All subjects were advised to avoid change in their lifestyle, and no subjects were taking any medicines that affected cholesterol levels or body fat reduction during the time of the study. We measured blood samples at the laboratory of NIN, Hanoi. Blood samples were taken an over-night fast. The PGBR and placebo groups were measured on the same day. 2.2.. Anthropometric measurements {#s2b} --------------------------------- Weight, height, waist and hip circumferences were measured two times. Bodyweight and height were measured in light clothing and without shoes. BMI was calculated as body weight per height squared (kg/m^2^). Waist circumference was measured mid-way between the lower rib margin and the iliac crest, while hip circumference was measured at the broadest circumference around the buttocks. 2.3.. Nutrition survey {#s2c} ---------------------- We used the Semi-Quantitative Food Frequency Questionnaire for a nutrition survey at baseline and completion. Energy and nutrient intake were measured based on the Vietnamese Food Composition Table 2007 [@b12]. 2.4.. Physical activity survey {#s2d} ------------------------------ We used the structured questionnaire for a physical activity survey at baseline and completion. The levels of physical activity were built based on the Global Recommendations on Physical Activity for Health [@b13]. 2.5.. Blood sample collection {#s2e} ----------------------------- Blood samples were taken in the morning and after an overnight fast and were measured two times. Fasting glucose was measured by Accu-Chek glucometer. Venous blood samples were kept frozen at 8 °C for analysis: serum total cholesterol, HDL cholesterol, LDL cholesterol, concentrations of insuline, fasting triglyceride. 2.6.. Blood pressured {#s2f} --------------------- Blood pressure was measured by using the mercury sphygmomanometer. All subjects were measured in a quiet, air-conditioned room (the room temperature at 25 °C). The participants had to rest before and during measurements for about 20 min. 2.7.. Homeostatic Model Assessment for Insuline Resistance (HOMA-IR) {#s2g} -------------------------------------------------------------------- HOMA-IR was calculated using the following formula: fasting insuline (mU/l) × fasting glucose (mmol/l)/22.5. Fasting glucose and insuline were measured using an enzymatic method and chemiluminescence immunoassay, respectively. 2.8.. Statistical analysis {#s2h} -------------------------- Data were analyzed by using SPSS for windows 16.0. We calculated the change in percentage for each variable at completion and baseline. All variables of the PGBR and placebo groups were compared using an unpaired *t*-test, and data at baseline and completion were compared using a paired *t*-test. Varibales with confirmed homoscedasticity were compared by Student\'s *t*-test. *p* values of less than 0.05 were considered statistically significant for all analyses. Mann-Whitney test and Wilconxon test were used in non-normal distribution. Mc Nemar test -- non parametrict test -- was used on paired nominal data. 3.. Results {#s3} =========== 3.1.. Subjects {#s3a} -------------- Eighty subjects completed the intervention; 3 subjects from the placebo group and 3 subjects from the PGBR group withdrew from the study, due to personal reasons. All subjects completed over 90% of the study process. No subjects in either group had any symptoms related to the intake of test substances. 3.2.. Nutrition facts for PGBR and placebo capsules {#s3b} --------------------------------------------------- [Table 1](#publichealth-07-01-005-t01){ref-type="table"} shows the amount of energy, protein, lipid, carbohydrate, ash, sodium, and γ-aminobutyric acid (GABA) per 100 g GPBR and per 100 g white rice. ###### Nutrition facts per 100 g of PGBR and white rice. Nutrition facts per 100 g PGBR White rice --------------------------- -------- ------------ Energy (kcal) 357 345 Protein (g) ≥ 7 ≥ 5 Lipid (g) ≥ 2 ≥ 0.1 Carbohydrate (g) ≥ 60 ≥ 75 Fibre (g) 3--4 0.5--0.9 Inositol (mg) ≥ 10 \- GABA (mg) 12--20 1--3 3.3.. Baseline characteristic {#s3c} ----------------------------- [Table 2](#publichealth-07-01-005-t02){ref-type="table"} shows the baseline parameters of both groups. There were no significant differences in parameters for age, sex, education levels, height, weight, BMI, abdominal circumference, systolic blood pressure, diastolic blood pressure, fasting glucose, insuline, HOMA-IR, total cholesterol, triglyceride, LDL cholesterol, HDL cholesterol, percentage of constipation, physical activity levels, diet incides between the PGBR and the placebo groups. ###### Baseline characteristics of PGBR and placebo groups. Total sample PGBR group Placebo group --------------------------------- ------------------------------------- -------------- --------------- ------------ Age(y) 65.1 ± 3.81 65.2 ± 3.78 65.0 ± 3.85 Sex Male 16 (20%) 8 (20%) 8 (20%) Female 72 (80%) 36 (80%) 36 (80%) Education Primary and secondary 38 (47.5%) 18 (45.0%) 20 (50.0%) High school 31 (38.7%) 16 (40.0%) 15 (38.0%) Colleges/University/Graduate School 11 (13.8%) 6 (15.0%) 5 (12.0%%) Occupation Workers/Employees 14 (17.5%) 8 (20%) 6 (15.0%) Retirement 46 (57.5%) 20 (50.0%) 26 (65.0%) Business 12 (15.0%) 7 (17.5%) 5 (12.5%) Housewife 8 (10.0%) 5 (12.5%) 3 (7.5%) Height (cm) 156.7 ± 5.3 156.5 ± 5.6 156.9 ± 6.3 BMI (kg/m^2^) 25.7 ± 2.2 25.9 ± 2.4 25.5 ± 2.3 Abdominal circumference (cm) Male 92.9 ± 2.4 93.2 ± 3.1 92.6 ± 2.3 Female 87.6 ± 3.1 88.1 ± 4.6 87.1 ± 3.2 Systolic blood pressure (mmHg) 124.8 ± 15.4 125.0 ± 15.7 124.7 ± 16.5 Diastolic blood pressure (mmHg) 81.9 ± 12.4 82.4 ± 12.2 81.7 ± 13.6 Fasting glucose (mmol/l) 6.4 ± 0.8 6.4 ± 1.2 6.3 ± 0.9 Insuline (mU/l) 9.5 ± 5.4 9.6 ± 6.0 9.4 ± 5.6 HOMA-IR 2.7 ± 1.7 2.7 ± 1.9 2.6 ± 1.8 Total cholesterol (mmol/l) 5.5 ± 0.8 5.5 ± 1.0 5.6 ± 0.9 Triglyceride (mmol/l) 2.4 ± 1.4 2.4 ± 1.6 2.3 ± 1.3 LDL cholesterol (mmol/l) 3.0 ± 0.8 3.0 ± 1.2 3.0 ± 0.6 HDL cholesterol (mmol/l) 1.12 ± 0.2 1.11 ± 0.2 1.12 ± 0.3 Constipation status (%) 22 (27.5%) 12 (30.0%) 10 (25.0%) Physical activity levels Moderate 56 (70.0%) 28 (70.0%) 27 (67.5%) Vigorous 9 (11.3%) 4 (10.0%) 5 (12.5%) Intensity 38 (47.5%) 18 (45.0%) 20 (50.0%) 3.4.. Physical characteristics {#s3d} ------------------------------ In [Table 3](#publichealth-07-01-005-t03){ref-type="table"}, we indicate the anthropometrics and blood pressures at all intervention times. No significant difference was observed between the placebo and PGBR groups in weight, BMI, systolic blood pressure or diastolic blood pressure. There were significant reductions in abdominal circumferences of both sexes of the PGBR group between baseline and completion (p \< 0.05). ###### Physical characteristics throughout the intervention. Baseline End ---------------------------------------- --------- -------------- ------------------ Weight (kg) PGBR 63.5 ± 8.2 61.8 ± 8.6 Placebo 62.1 ± 7.9 62.6 ± 8.4 BMI (kg/m^2^) PGBR 25.9 ± 2.4 25.0 ± 2.3 Placebo 25.5 ± 2.3 25.8 ± 2.6 Abdominal circumference in male (cm) PGBR 93.2 ± 3.1 92.1 ± 3.7 *^a^* Placebo 92.6 ± 2.3 92.7 ± 2.6 Abdominal circumference in female (cm) PGBR 88.1 ± 4.6 86.2 ± 5.8 *^a^* Placebo 87.1 ± 3.2 86.2 ± 4.0 *^a^* Systolic blood pressure (mmHg) PGBR 125.0 ± 15.7 120.1 ± 16.4 Placebo 124.7 ± 16.5 121.9 ± 13.2 Diastolic blood pressure (mmHg) PGBR 82.4 ± 12.2 74.6 ± 11.1 Placebo 81.7 ± 13.6 77.3 ± 15.3 ^*a*^ *p \< 0.05, compare to baseline, paired t-test*. 3.5.. Blood biochemical parameters {#s3e} ---------------------------------- [Table 4](#publichealth-07-01-005-t04){ref-type="table"} shows the results of blood tests at month 3 of the intervention. No significant difference was observed between the placebo and PGBR groups in fasting glucose, insuline, HOMA-IR, total cholesterol, LDL cholesterol, triglyceride. HDL cholesterol, and LDL-C/HDL-C ratio were significantly different between the placebo and PGBR groups. Both groups were significantly different in insuline when comparing baseline and completion. Fasting glucose, HOMA-IR, of the PGBR group showed a significant reduction from baseline to completion. 3.6.. Energy and nutrient intakes and effects on patients with MetS {#s3f} ------------------------------------------------------------------- [Table 5](#publichealth-07-01-005-t05){ref-type="table"} shows the amount of energy and nutrient intake of both groups at baseline and completion; there were no significant differences in energy, protein, carbohydrate, lipid, and fiber between the PGBR and placebo groups at baseline or completion. Fibre was significantly different between the placebo and PGBR groups. Fibre of the PGBR group also showed a significant increase from baseline to completion. No significant difference was observed between placebo and PGBR groups in the physical activity level. At the end of the intervention, there were significant differences in the percentage of patients with MetS between placebo and PGBR group. ###### Blood test results throughout the intervention. Baseline End ---------------------------- --------- ------------ --------------------- Fasting glucose (mmol/l) PGBR 6.4 ± 1.2 5.7 ± 0.6 *^a^* Placebo 6.3 ± 0.9 6.2 ± 0.9 Insulin (mU/l) PGBR 9.6 ± 6.0 8.3 ± 4.6 *^a^* Placebo 9.4 ± 5.6 8.9 ± 4.1 *^a^* HOMA-IR PGBR 2.7 ± 1.9 1.5 ± 0.9 *^a^* Placebo 2.6 ± 1.8 2.4 ± 1.5 Total cholesterol (mmol/l) PGBR 5.5 ± 1.0 5.2 ± 0.9 Placebo 5.6 ± 0.9 5.5 ± 0.9 HDL cholesterol (mmol/l) PGBR 1.11 ± 0.2 1.44 ± 0.4^¥,\ *t*^ Placebo 1.12 ± 0.3 1.11 ± 0.3 LDL cholesterol (mmol/l) PGBR 3.0 ± 1.2 2.9 ± 0.8 Placebo 3.0 ± 0.6 3.0 ± 0.9 LDL-C/HDL-C ratio PGBR 2.9 ± 1.1 2.3 ± 0.6^¥,\ *t*^ Placebo 2.8 ± 0.9 2.9 ± 0.9 Triglyceride (mmol/l) PGBR 2.4 ± 1.6 2.0 ± 1.4 Placebo 2.3 ± 1.3 2.2 ± 0.9 ^*a*^ *p \< 0.05, compare to baseline, paired t-test;* *^¥^ p \< 0.05, compared with the control group, Mann-Whitney test; ^t^ p \< 0.05, compare to baseline, Wilconxon test*. ###### Energy and nutrient intakes. Baseline End ----------------------------------- ---------- ---------------- --------------------- Energy (kcal) PGBR 1260.4 ± 387.7 1263.1 ± 401.0 Placebo 1264.0 ± 337.9 1271.8 ± 331.0 Carbohydrate (g) PGBR 220.4 ± 79.9 221.3 ± 72.1 Placebo 225.6 ± 78.2 227.6 ± 78.2 Protein (g) PGBR 51.0 ± 18.6 51.3 ± 14.9 Placebo 50.1 ± 12.8 50.0 ± 13.4 Lipid (g) PGBR 25.0 ± 10.7 24.9 ± 9.0 Placebo 24.5 ± 12.7 25.4 ± 13.8 Dietary Fibre (g) PGBR 5.3 ± 2.0 9.6 ± 1.8\*^,\ *a*^ Placebo 5.2 ± 1.8 5.4 ± 1.9 Percentages of MetS reduction (%) PGBR 0 30 (12)^Φ,\ γ^ Placebo 0 10 (4) *\* p \< 0.05, compared with the control group, t-test; ^a^ p \< 0.05, compare to baseline, paired t-test; ^Φ^ p \< 0.05, compared with the control group, χ2 test; ^γ^ p \< 0.05, compare to baseline, χ2 Mc Nemar test*. 4.. Discussion {#s4} ============== The study examined the effects of PGBR on patients with MetS in 3-month-intervention in Vietnam. The effects of PGBR on MetS patients were shown in the reduction significantly of MetS percentages in the PGBR group compared to the placebo group. Another intervention in Japanese males also showed that the increases significantly in glucose were lower after the ingestion of brown rice than after the ingestion of white rice [@b14]. In the present study, abdominal circumferences in the PGBR group was significantly reduced after the 3-month intervention, while those in the placebo group were unchanged with no differences in physical activity levels. Similar to our results, waist circumference (90.3 ± 10.3 cm) and systolic blood pressure (134 ± 13 mmHg) of Japanese males were also lower at the end of the 8-week brown rice diet period [@b14]. It might prove to be the case that PGBR reduces waist circumference over longer periods of time. Improvement in lipid metabolism and reduction of the abdominal circumference were observed in some studies that have been conducted with Taiwanese [@b8] and Vietnamese [@b9] subjects by replacing white rice with PGBR in two of the subjects\' meals per day. However, the effect of PGBR on body weight and BMI are also not clear in present intervention. In the present intervention, MetS definition of the National Cholesterol Education Program Treatment Panel III -- Adults: 3^rd^ Report (NECP ATP III) was used to define patients. MetS was diagnosed when patients have ≥ 3 of 5 following criteria which are about waist-line, triglycerides, HDL-cholesterol, fasting blood glucose, and blood pressure. Thus, although only HDL-cholesterol concentrations were significantly increased from 1.11 (mmol/l) to 1.44 (mmol/l) compared to the placebo group, the number of patient with MetS was reduced from 40 patients to 28 patients. Then the percentages of this reduction were higher significantly compared to the reduction of the placebo group (from 40 patients to 36 patients). Moreover, for triglyceride and blood pressure, significant differences between groups were not observed at endpoint. However, in PGBR group, abdominal circumference in both male and female, and fasting glucose decreased significantly compared to its baseline. Future research with longer time intervention is needed. Our main findings were that PGBR is correlated with HDL cholesterol during comparison between the PGBR and placebo group. Several interventions showed LDL cholesterol in the PGBR group was significantly reduced after the 6-month intervention[@b11] and after 12-week intervention [@b15] while non-significant finding was showed in our study. Meanwhile, HDL cholesterol in the PGBR group was significantly increased after the 3-month intervention compared to placebo group and baseline. Fish -- Rich diet should be a reason. Omega 3 from fish meat including α-Linolenic acid, eicosapentaenoic acid (EPA), and docosahexaenoic acid (DHA) has been suggested to be similar to that of fibric acids. EPA and DHA, as well as fibric acids, stimulate the β-oxidation of themselves or of other fatty acid substrates, by this mechanism, also elicit peroxisomal proliferation. Thus, polyunsaturated fatty acids may lead to an increase in the HDL cholesterol concentrations. Further researches need to be discovered this different result. Although the difference of BMI was not significant, reductions significantly of fasting glucose and insuline compared to baseline showed that white rice may be a major cause of diabetes mellitus [@b16],[@b17], especially for East Asians, such as Vietnamese who take about 70% of their energy from rice [@b18]. In previous study, effects of PGBR on fasting glucose was proved by bioactives (γ-aminobutyric acid (GABA), acylated steryl glycoside, oryzanol, and phenolics) involve in PGBR\'s downregulation gluconeogenic genes (*Fbp1* and *Pck1*) [@b19]. In this study, the fibre concentration of PGBR group was significantly different from placebo group and baseline while energy, carbohydrate, protein, and lipid were no differences. Fiber slows carbohydrate digestion and glucose absorption. In another intervention (2014), PGBR contains a higher amount of fiber help control blood glucose and lipid concentration in Vietnamese women with Impaired Glucose Tolerance [@b9]. Effects of fiber to prevent and treatment MetS and diabetes mellitus were also proved in several interventions [@b20],[@b21]. However, during the 3-month intervention, percentages of constipation among participants were decreased (the data not shown). This situation may cause a reduction in vegetable consumption\'s subjects. Thus, the decrease of vegetable intake in PGBR group should be a confounder. 5.. Conclusion {#s5} ============== The results of the present study suggest that PGBR benefits of MetS patients, and may provide long-term benefits of diabetes mellitus. Further studies with longer durations of intervention are warranted to examine the effects of substituting PGBR with white rice on MetS for future cardiovascular and blood glucose diseases. Thanks for the support from Loc Troi Company who provided the PGBR samples. This study was financially supported by the National Institute of Nutrition, Vietnam. **Conflict of interests:** The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Introduction {#Sec1} ============ There is a clinical need for efficient and cost-effective grafts in urethral reconstructive surgery. Current surgical techniques require harvesting from autologous sites eg. buccal mucosa and penile skin tissue^[@CR1]^. This increases the risk of complications, the possible lack of tissue availability and patient discomfort. Certain oral pathologies are contraindications for buccal mucosal graft harvesting^[@CR2]^. Both cell-free and cell-seeded urethral grafts have been investigated. To date, cell-based grafts have shown better functional results in preclinical large animal models and inconclusive results in subsequent clinical trials^[@CR3]^. In 2015 and 2017, the clinical outcome of a commercial cell-seeded tissue engineered buccal mucosa product, was published^[@CR4],[@CR5]^. This product requires a buccal biopsy to be taken, followed by *ex-vivo* cell culturing in a GMP facility and a 3-week manufacturing time. The cost-effectiveness and efficacy of these grafts have been debated^[@CR6],[@CR7]^. Therefore, it is unlikely that this approach will become standard clinical practice for the patient with limited financial resources or insurance coverage. To overcome these problems we have targeted the development of a cell-free urethral graft that is cost-effective and adequate also to be used in third world countries. *In vitro* the mechanical niche is an important parameter to be considered for controlling cell fate. For example, mesenchymal stem cells are known to differentiate into different cell lineages, depending on the substrate stiffness^[@CR8]^. Previous studies have shown that fluid content or collagen fiber density relates to the stiffness and influenced material properties of the collagen gels^[@CR9],[@CR10]^. The more extracted fluid from the collagen gels, the stiffer the collagen material gets. In this manuscript we address the question if collagen fiber density of a collagen graft influences the functional regeneration of the rabbit urethra? Two different cell-free bovine tubular collagen grafts were manufactured, one with polarized and low collagen fiber density distribution (LD-graft) and one control with uniform and high collagen fiber density distribution (HD-graft). Polarization was achieved by specific compression and the collagen density increased towards the luminal side of the tube. These grafts were implanted in rabbits and were examined post-implantation by histology and immunohistochemistry, and the functional outcome was assessed by voiding cystourethrography and mating. Materials and Methods {#Sec2} ===================== Preparation of collagen grafts {#Sec3} ------------------------------ Tubular collagen grafts were fabricated under sterile conditions, utilizing liquid type I bovine collagen (5 mg/mL, Symatese, F). 8.5 mL of collagen solution were added to 0.8 mL of 10 × MEM and neutralized with 1 M NaOH (approx. 1.85 mL). The neutralized solution was set in a mold previously used for manufacturing tubular constructs of 3 mm inner diameter^[@CR10]^. To fabricate the LD-graft, the gelated collagen was manually compressed by rolling it on filter paper, followed by an air-drying step (Fig. [1A](#Fig1){ref-type="fig"}). To fabricate the HD-graft, only air-drying was applied to achieve a desired liquid content of 1% w/w (Fig. [1A](#Fig1){ref-type="fig"}). Monitoring of the water content of collagen gels/grafts were done with a balance (Mettler Toledo, CH). The fabricated grafts were kept in PBS supplemented with 1% Penicillin/Streptomycin (Gibco, CH) and 2.5 mg/mL Fungizone (Gibco) until used.Figure 1Manufacturing technique of the low-density graft (LD-graft) and high-density graft (HD-graft) (**A**) Two methods were utilized to manufacture the density controlled collagen grafts by monitoring the water content removed either by rolling compression or air-drying. The polarized collagen fiber distribution characteristics of the LD-graft were achieved by utilizing both a rolling compression followed by air-drying. The uniform collagen fiber distribution characteristics of the HD-graft were achieved by only air-drying the collagen. (**B**) By weighing and monitoring the collagen gel during production, we could achieve our polarized (compressed and air-dried: LD-graft) and uniform (only air-dried: HD-graft) collagen fiber distribution characteristics. Note: Images in Fig. 1A are Sirius red stained paraffin sections with scale bars of 200 µm. Scanning electron microscopy {#Sec4} ---------------------------- The samples were fixed with 1% tannic acid and 1.25% glutaraldehyde, then washed with 0.1 M cacodylate, and dehydrated in increasing ethanol concentrations prior to critical point drying. Thereafter coated with gold/palladium and imaged at a voltage of 10 kV using a scanning electron microscope (SEM, XLF30, Philips). Mechanical evaluation {#Sec5} --------------------- Burst pressure, ultimate tensile strength (UTS) and Young's modulus of grafts (N = 4 in each group) were measured with an electronic manometer (Extech instruments HD750) or an Instron tensile machine (Norwood, MA, USA) as described previously^[@CR10]^. Surgical implantation of the graft and evaluation in rabbits {#Sec6} ------------------------------------------------------------ Following approval by the Animal Ethics Committees of the Canton of Vaud (Authorization number: VD-2740), Switzerland and of the Faculty of Medicine of the University of Malaya (Authorization number: 2013-07-19/SUR/R/TCR), Kuala Lumpur the experiments were performed on New Zealand white male rabbits (2.5--3.5 kg; Charles River Laboratories France, and Harlan and Bred, Singapore) in Lausanne and Kuala Lumpur. All experiments were done in compliance with national directives for the care and use of laboratory animals. The surgical steps and post-operative monitoring were done as previously described^[@CR10]^. Rabbits from the long-term study group were involved in the in-house breeding program of the Animal Experimental Unit of the Faculty of Medicine, University Malaya. Cysto-urethrography {#Sec7} ------------------- Animals were submitted under general anesthesia for macroscopic evaluation and voiding cysto-urethrography (Visipaque 270 mg/mL). All images were collected with a Philips BV Pulsera. The diameter of the urethra was measured utilizing a scale. Knowing that the graft was sutured at 0.5 cm proximal to the base of the glans and it measured 2 cm in length, the position of the graft could be determined on the radioscopic images. It was then possible to estimate the potential presence of stenosis at the anastomotic sites. Stenosis was defined as a persistent 50% reduction of the diameter of the urethra at the same location. Histology and immunohistochemistry {#Sec8} ---------------------------------- Entire penises were harvested and fixed in 4% PFA. Specimens were embedded in paraffin, and 8 µm thick sections were prepared. Antibodies used: Mouse anti-α-smooth muscle actin (1:150, Abcam, CH), Goat anti-uroplakin-2 (1:150, Labforce, CH), Mouse anti-caldesmon (1:400, Sigma, D), Rabbit anti-CD31 (1:100, Abcam, CH), Donkey anti-mouse Alexa 647 (1:800, Abcam, CH), and Donkey anti-goat Alexa 546 (1:800, Abcam, CH). Images were taken with a Leica DM5500 microscope (Leica, D) and with a LSM 700 microscope (Zeiss, D). Alpha smooth muscle actin (α-SMA) expression was quantified with Fiji imaging program (ImageJ) as previously described^[@CR10]^. Statistical analysis {#Sec9} -------------------- Statistical analyses were performed using Prism v5.0a (GraphPad), using the test mentioned in the figure legends. A p-value of less than 0.05 was considered significant. All error bars in diagrams represent the standard deviation (SD). Data Availability {#Sec10} ----------------- The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request. Results {#Sec11} ======= Manufacture of HD- and LD-grafts {#Sec12} -------------------------------- By applying a fast compression step, resulting in a less than 50% liquid loss followed by a slow air-drying step, we could control the liquid content of our produced grafts and therefore also the collagen grafts final density (Fig. [1B](#Fig1){ref-type="fig"}). This procedure yields a controlled polarized collagen fiber distribution, with a more dense structure at the compressed outer surfaces while the internal luminal part of the collagen tubular wall was left with a less dense collagen structure as shown by SEM (Fig. [2A](#Fig2){ref-type="fig"}). This graft hereafter is referred to as the LD-graft. In comparison, in the only air-dried collagen grafts, the latter become highly dense and with a uniform non-polarized collagen structure (Fig. [2B](#Fig2){ref-type="fig"}). This graft is hereafter referred to as the HD-graft. The denser HD-grafts showed significantly better mechanical properties, in terms of measured burst pressure, UTS and Young's Modules compared to the LD-graft (Supplementary Figure [1](#MOESM1){ref-type="media"}).Figure 2Description and characterization of manufactured collagen graft with controlled spatial orientation. (**A** and **B**) SEM images of LD- and HD-grafts with three different magnifications. Short-term rabbit evaluation of implanted HD- and LD-grafts {#Sec13} ----------------------------------------------------------- The LD-grafts were associated with a fistula formation rate of 15.4% (2 out of 13 rabbits) compared to one of 33% (3 out of 9 rabbits) in the HD-graft (Fig. [3A](#Fig3){ref-type="fig"}). No urethral stricture formation was observed. No mortality was seen as a result of graft implantation. At 1 month, it was noticed that the HD-graft still had not been completely remodelled by the surrounding tissue as shown by the persistance of implanted collagen visualized by Hemotyxlin & Eosin (H&E) and Masson's Trichrome (MT) staining (Fig. [3B](#Fig3){ref-type="fig"}). In LD-grafts, the remaining collagen could be identified to a lesser extent in the histological samples at 1 month (Fig. [3C](#Fig3){ref-type="fig"}). The luminal side of both grafts had shown overgrowth of urothelial cells within the first month of implantation (Fig. [3B,C](#Fig3){ref-type="fig"}). Pronounced ingrowth of smooth muscle cells was still lacking at 1 month, irresepective of the graft type (Fig. [3B,C](#Fig3){ref-type="fig"}). The urothelial cells had covered the grafted region in the HD-graft at 3 months with an organized multilayered urothelial cell distribution (Fig. [3B](#Fig3){ref-type="fig"}). This was still lacking in the LD-graft at 3 months (Fig. [3C](#Fig3){ref-type="fig"}). At 6 months, the urothelium, irrespective of the graft type, showed normal stratification (Fig. [3B,C](#Fig3){ref-type="fig"}). H&E and MT stained sections showed signs of vascular structures already at 1 and 3 months in both grafts, this was confirmed by immunohistochemistry for a vascular marker, CD31 (Supplementary Figure [2](#MOESM1){ref-type="media"}). More smooth muscle cell ingrowth from the native tissue into the LD-graft was seen at 3 and 6 months when compared to the HD-graft (Fig. [3B,C](#Fig3){ref-type="fig"}). Image analyis of regenerated smooth muscle, using anti-α-smooth muscle actin (α-SMA) antibodies, confirmed the difference between the two grafts to be significant (Fig. [3D,F](#Fig3){ref-type="fig"}). A second smooth muscle cell marker, caldesmon, confirmed the α-SMA result of smooth muscle cell ingrowth (Supplementary Figure [3](#MOESM1){ref-type="media"}). Uroplakin-2, a specific antibody for terminal urothelial differentiation, also confirmed an earlier differentiation of urothelium in the HD-graft compared to the LD-graft area at 1 and 3 months (Fig. [3G,H](#Fig3){ref-type="fig"}). At 6 months, the LD-graft also showed normal Uroplakin-2 expression.Figure 3*In vivo* evaluation of high-density (HD) and low-density (LD) collagen grafts in a rabbit urethral defect model. (**A**) Functional surgical outcome analyzed by micturating cysto-urethrography of rabbit implanted with LD- and HD-grafts for 1, 3 and 6 months. (**B**) H&E and MT stained sections of the grafted area of HD-graft 1, 3 and 6 months after implantation. (**C**) H&E and MT stained sections of the grafted area of low-density graft 1, 3 and 6 months after implantation. (**D**) Quantification of SMA expression in grafted areas for rabbits implanted with LD- and HD-graft after 1, 3, and 6 months. (**E**) Immunohistochemistry for compared to the HD-grafts-SMA of LD-graft implanted for 1, 3 and 6 months. (**F**) Immunohistochemistry for SMA of HD-graft implanted for 1, 3 and 6 months. (**G**) Immunohistochemistry for Uroplakin-2 (Up2) of LD-graft implanted for 1, 3 and 6 months. (**H**) Immunohistochemistry for Up2 of HD-graft implanted for 1, 3 and 6 months. Note: areas circled in dashed black line indicating the remaining collagen pieces of the grafts after 1 month (Scale bar H&E and MT 250 µm, Scale bar SMA and Up-2 50 µm). (Error bars represent the standard deviation of four independent samples. \*\*p \< 0.01, Student t-test). Long-term rabbit evaluation of implanted LD-grafts {#Sec14} -------------------------------------------------- LD-grafts were implanted in 5 rabbits for 9 months and 2 rabbits for 11 months. No complications were noted in these rabbits (Fig. [4A](#Fig4){ref-type="fig"}). At 9 and 11 months, the urothelial layer appeared comparable to the native rabbit urethra (Figs [4B,E and H](#Fig4){ref-type="fig"}) and there was pronounced ingrowth of smooth muscle cells, however, the appearance was not comparable to native rabbit urethra (Figs [4C,D,F,G,I and K](#Fig4){ref-type="fig"}). All animals in the long-term group were able to mate normally and had produced offspring (Fig. [4L](#Fig4){ref-type="fig"}).Figure 4Long-term *in vivo* evaluation of low-density collagen grafts in a rabbit urethral defect model. (**A**) Functional surgical outcome analyzed by micturating cysto-urethrography of rabbits with an artificially created, circumferential urethral defect that was bridged with a LD-graft (N = 5 rabbits for 9 months, N = 2 rabbits for 11 months). (**B**,**C**,**E**,**F**, and **H**,**I**) H&E and MT stained sections of LD-graft implanted for 9 and 11 months, and a control native rabbit urethra. (**D**,**G** and **J**) Immunohistochemistry for α-SMA of LD-graft implanted for 9 and 11 months, and a control native rabbit urethra. (**K**) Quantification of SMA expression in LD-graft compared to a control native rabbit urethra. (**L**) A photo of rabbit offspring from fathers implanted with a LD-graft. Note: green arrow pointing at offspring (Scale bar H&E and MT 250 µm, Scale bar SMA 50 µm). Error bars represent the standard deviation of four independent samples. Discussion {#Sec15} ========== In order to answer the question if collagen fiber density of a graft influences the urethral regeneration *in vivo*, we had to develop two fabrication methods to produce both a polarized low- density and a uniform high-density collagen graft. Our rabbit results show that the low-density graft had an improved smooth muscle regeneration compared to the implanted high-density graft. Previous studies have shown that fluid content or collagen fiber density relates to the stiffness and influences material properties of the collagen gels^[@CR9],[@CR10]^. The more extracted fluid from the collagen gels, the stiffer and more mechanical strong the collagen material gets. We experienced the same for our two grafts, where the surgeons reported that the HD-graft was more mechanically strong compared to the LD-graft. The HD-graft was designed to serve as a control, to represent a dense and stiff matrix (approximately 0.1 MPa), although it was not in the similar high range as the well studied SIS grafts (Small-intestine submucosa, approximately 7.2 MPa)^[@CR11]^, it was still two times stiffer than the LD-graft. Importantly, the SIS is reported to have been tried with patients for urethral reconstruction but has failed and is therefor not recommended to use in urethral reconstruction^[@CR12]^. According to literature the native rabbit urethra seems to have a stiffness of approximately 0.2 MPa, which is more in the range of our engineered scaffolds as compared to the SIS urethral scaffolds^[@CR13]^. We would like to challenge the current opinion in urethral reconstructive surgery that a cell-free graft is ineffective to bridge a urethral defect, and to rephrase this statement to: A cell-free graft, which is too dense or stiff, is ineffective for urethral defect grafting. Published studies have already demonstrated the failure of using stiff cell-free grafts in this clinical indication^[@CR12]^. However, when these graft are seeded with cells they have shown promising result in pre-clinical models but ultimately only limited success in clinical applications^[@CR3]--[@CR5],[@CR14]--[@CR17]^. In our graft implantation study in rabbits, the LD-graft exhibited more ingrowth of smooth muscle cells with formation of muscle bundles, due to its less dense collagen structure or, whereas the HD-graft demonstrated faster urothelial regeneration due to its denser collagen structure. This confirms similar reports where denser/stiffer grafts have been implanted in animals and showed good urothelial regeneration but has lacked an adequate smooth muscle regeneration^[@CR18]^. The accelerated ingrowth of smooth muscle cells into the low-density grafts is likely due to the lower surface area of collagen that needs to be degraded and remodeled by the infiltrating cells. Examining closer the HD-graft, it showed limited smooth muscle cell ingrowth in the center of grafted areas, but a normal muscle bundle structure was observed at the interface of the graft and native tissue (Supplementary Figure [3F and G](#MOESM1){ref-type="media"}). Both implanted grafts showed no signs of urine erosion. This is most likely due to the fact that all grafts showed early urothelial regeneration protecting the underlying graft. Interestingly, the onset of fully differentiated urothelium, is delayed in the less dense grafts compared to the highly dense collagen grafts. It can be speculated that this difference in density influencing the "mechanical niche" favoring either urothelial or smooth muscle cell growth. Improved smooth muscle regeneration is an advantage, as the primary pathology of urethral stricture is due to muscular fibrosis^[@CR19]^. By 6 months, the LD-graft had significantly more expression of α-SMA compared to the HD-grafts. At 11 months, cystourethrography showed urethral patency and normal urine flow and bladder voiding. Successful mating resulting in progeny could also be seen proving clinical relevant functional regeneration of the grafted areas. Our study has limitations. The rabbit model used had an artificially created urethral defect, with healthy urethral tissue on the edges. The graft was implanted to replace a totally excised urethral segment, what is not always mandatory as in clinical practice only partial circumferential replacement is often performed^[@CR20],[@CR21]^. Although the grafts were prepared under sterile conditions, the production needs to be transferred into a GMP setting to allow translation of this technology to human application. This translation is allowed if the bioburden is lower than the sterility assurance levels set by governing authorities. To further reduce the bioburden, final sterilization can be attempted, however, it is known that sterilization can change the material properties of the grafts which might compromise the final outcome of the graft^[@CR22]^. An off-the-shelf urethral graft without the incorporation of cells and growth factors, showing adequate urothelial and smooth muscle cell ingrowth into the implanted graft, represents an important pre-clinical breakthrough. This was demonstrated by the comparable encouraging results of experimental studies conducted in two research centers in Switzerland and Malaysia opening the way for clinical translation of this cost effective, cell-free grafting for application in urethral reconstruction. Electronic supplementary material ================================= {#Sec16} Supplementary material H. M. Larsson and G. Vythilingam contributed equally to this work. **Electronic supplementary material** **Supplementary information** accompanies this paper at 10.1038/s41598-018-27621-9. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. The authors would like to thank: Dr. Haryanti Azura (UM), Prof. Andrea Superti-Furga, Marco Bürki, and Jacques Barraud (CHUV) for their advice and support. This work was funded by the University Malaya with two research grants (UMRG) (RP005G-13HTM and RU028-2015 UM.0000216/HRU.OP), an international short-term research fellowship (SNSF: IZK0Z3-171101) and a grant from the Commission of Technology and Innovation of the Swiss Government (CTI project no-16627.1 PFLS-LS). H.M.L. Performed experiments, analyzed data, prepared figures, co-wrote manuscript. G.V. Performed experiments, analyzed data, co-wrote manuscript. K.P. Performed experiments. E.V. Performed experiments. E.M.E. Analyzed data and revised manuscript. S.S. Performed experiments. R.C.T. Performed experiments, analyzed data and revised manuscript. T.K. Analyzed data and revised manuscript. J.A.H. Analyzed data and revised manuscript. P.F. Analyzed data, co-wrote manuscript and financed study. Competing Interests {#FPar1} =================== The authors declare no competing interests.

Background ========== The methylenetetrahydrofolate reductase enzyme (MTHFR) catalyzes a reaction that produces 5-methyltetrahydrofolate (5-methylTHF), the methyl donor for homocysteine in the synthesis of methionine. The 677C\>T mutation of the *MTHFR*gene has been associated with a thermolabile enzyme with decreased activity that may cause an increase in plasma homocysteine concentrations \[[@B1]\] when folate status is poor. This polymorphism is one of the most widely studied clinically relevant polymorphisms in humans, as it is related to cardiovascular disease \[[@B2]\] and neural tube defects (NTD; 601634) \[[@B3]\]. A large number of studies have provided a broad overview of the prevalence of the 677C\>T polymorphism in different human populations, showing that the distribution of frequencies is diverse \[[@B4]\]. These differences have been also observed between groups of different ages in the same Spanish population (older and younger than 24 years) \[[@B5]\] and in a Swiss population (older and younger than 60 years) \[[@B6]\], as well as in a Japanese population \[[@B7]\]. In some populations, such the Toscanians in Italy \[[@B8]\] and Mexicans \[[@B9]\], the homozygous mutated genotype (TT) has reached frequencies greater than 30%. On the other hand, in Africans the frequency of the TT genotype is very low (less than 1%) \[[@B10],[@B11]\], but, in African-Americans, it has already reached 2% \[[@B12]\]. Studies based on the distribution of genotypic and allelic frequencies of the 677C\>T polymorphism and the 1298A\>C polymorphism in the *MTHFR*gene in Israeli, Japanese and Ghanaian Africans populations \[[@B13]\] concluded that the 677T mutation in the *MTHFR*gene emerged as a founder haplotype with some selective advantage. Recently, preliminary evidence of genetic selection of this polymorphism related to folate intake has been reported \[[@B14]\]. The aim of the present study is to analyze the changes in frequencies of the 677C\>T polymorphism during the 20^th^century and particularly the evolution of the frequencies during the decade of 1980--1989, by comparing the genotype frequencies between living subjects born in this period versus samples of spontaneous abortions (SA) that occurred during in the same time period. Methods ======= Subjects -------- This study was approved by the Ethics Committee at the University Hospital \"Virgen de la Victoria\" (Málaga). One of the study groups consisted of 344 fetal tissue samples from SA, obtained from the Department of Pathology of the University Hospital Carlos Haya (Málaga). These samples were selected after checking the clinical history and by the inclusion criteria of containing histologically confirmed fetal tissue collected in the 1980s from SA at less than 3 months (11 ± 1.70 week) and of unknown etiology. These fetal samples were compared with a control population of 276 subjects born in the 1980s with an average age of 22 ± 4.58. Another population of subjects born in the south of Spain in the 20th century were genotyped (1305 subjects, 697 women and 608 men) and divided into four groups according to birth date: 1900 to 1925 (n = 206); 1926 to 1950 (n = 320), 1951 to 1975 (n = 408), 1976 to 2000 (n = 371). Individuals were selected randomly from different areas of the province of Malaga, in southern Spain, and from different social statuses to avoid a selection bias. All the selected individuals were Caucasian and residents of the study area. The parents of all subjects included in the study were also Caucasian and born in Spain. The possibility of a founder effect or genetic drift was investigated and rejected. All the selected individuals were also genotyped for an insertion/deletion polymorphism in the angiotensin converting enzyme (ACE) gene and/or the 2756A\>G polymorphism in the methionine synthase gene (MTR), in order to determine whether or not our adult and young populations were genetically homogeneous. No significant differences were observed in allelic or genotypic frequencies for these genes between the different groups. The population studied was randomly selected according to age. Subjects 0--12 years old were selected from dried blood spots from neonatal screening papers; subjects 10--24 years old were recruited from students in primary and secondary schools and in university; subjects 25--50 years old and \>51 years old were recruited using their Andalusia Health Service identity cards. After approval by the University Hospital Ethical Committee, all the subjects were contacted, and, from those whose written consent was obtained, 10 ml of blood was taken. The investigation in this study conforms to the principles outlined in the Declaration of Helsinki. Genetic analysis ---------------- The fetal samples were extracted from the archived formalin-fixed, paraffin-embedded tissue sections. Genomic DNA was extracted from fetal tissue using the method described by Coombs et al. (1999) \[[@B15]\]. Genomic DNA of the second and third groups was extracted from peripheral leukocytes using the AquaPure Genomic DNA Blood Kit (Bio-Rad). Genotyping was performed using Real Time PCR with allele specific Taqman^®^probes and primers described by Ulvik et al. (2001) \[[@B16]\] and the following optimized protocol for 45 cycles: 10 s -- 94°C, 40 s -- 54°C, 15 s -- 72°C. The PCR mix (25 μl total volume) consisted of 5 μl of genomic DNA, 0.5 μl of sense primer, 0.62 μl of anti-sense primer, 0.85 μl Taqman^®^probe FAM, 0.43 μl Taqman^®^probe TET, 20 μl PCR-buffer iQ-SupermixTM (Bio-Rad) (containing 100 mM KCl, 40 mM Tris-HCl, (pH 8.4) 1.6 mM dNTP (dATP, dCTP, dGTP and dTTP), iTaq^®^polymerase (50 units/mL) and 6 mM MgCl2) and 17.75 μl H2O. Statistical and mathematical analysis ------------------------------------- All samples were genotyped, and the allelic and genotypic frequencies were compared. Differences were analyzed statistically using the chi-square test or Fisher\'s exact test. Correlations are expressed using Pearson\'s coefficient (r). Compliance of genotype distributions with Hardy-Weinberg (HW) equilibrium was evaluated by chi-square analysis. For all tests, a p-value \< 0.05 was considered to be statistically significant. Values are expressed as the mean ± SD. The genetic selection model was calculated for the evolution of the 677C\>T genotypes. The genetic selection could be classified as codominant or incompletely dominant and directional with the heterozygous genotype having an intermediate fitness. For this kind of selection, the most appropriate mathematical model is dq = \[spq(2hp+q-h)\]/\[p^2^+ 2pq(1-hs) + q^2^× (1-s)\], where dq is the change of frequency of the allele with lower fitness, s is the fraction of that genotype lost to selection, h is the degree of dominance (between 0, for no dominance and 1, for complete dominance), and p is the frequency of the allele with higher fitness. Results ======= We analyzed the genotype frequencies of the 677C\>T polymorphism in a population born during the 20th century. A total of 1305 subjects were divided into four groups of 25 years according to birth date. The genotype frequencies were compared between the four quarters of the century and showed very significant changes (p \< 0.001) in the group born in the last quarter of the 20th century (1976--2000), when compared to any of the other groups. The changes show a decrease of the CC genotype and an increase of the TT genotype in the last 25 years of the 20th century. (Table [1](#T1){ref-type="table"}) ###### Distribution of C677T genotype frequencies in the 20th century. **Frequencies by date of birth** -------------- ---------------------------------- ----- ------- ----- ------- ----- ---------- ----- **CC** 0.393 81 0.394 126 0.37 151 0.299\* 111 **CT** 0.495 102 0.488 156 0.49 200 0.461 171 **TT** 0.112 23 0.119 38 0.14 57 0.24\*\* 89 **C allele** 0.641 264 0.638 408 0.615 502 0.53\*\* 393 **T allele** 0.359 148 0.363 232 0.385 314 0.47\*\* 349 Note: \*p \< 0.05; \*\*p \< 0.01 (significance values in the last quarter of the century are relative to any other quarter of the century) Considering that each 25 year period corresponds to a generation, allelic frequencies did not change during the first 75 years of the century (HW equilibrium). However, we found that allelic and genotypic frequencies for the 677C\>T polymorphism in the last quarter of the century are significantly different compared to the previous generation (1951--1975). The genotype frequencies in the last quarter of the century are not the expected by a HW calculation using the allelic frequencies of the previous generation. This could be described as a consequence of genetic selection found in this population, in the absence of other causes. Applying the mathematical model described above to our population, the calculated fitness (s) is 0.5, and it can be predicted that both alleles will be approximately at a frequency of 50% in the next generation and allele T will be at 90% after seven generations (Figure [1A](#F1){ref-type="fig"}). Another possibility is that a scenario could be predicted in which both alleles will have frequencies of about 50% in the next generation and that they will maintain this stability while conditions remain unchanged. (Figure [1B](#F1){ref-type="fig"}) {#F1} The comparison of the genotype frequencies between a group of fetal samples from SA that occurred during the 1980--1989 decade and living subjects born in the same decade showed significant differences in genotype frequencies (p \< 0.001). CC genotypes were almost absent in abortion samples compared to living subjects (0.03 vs 0.33), while CT and TT genotypes were overrepresented in the same group. When 3-year periods are studied in the decade, we detected a significant increase of the mutated subjects during the decade (CT p \< 0.05; TT p \< 0.01). Allele frequencies showed the same pattern (p \< 0.05). Controls showed the same tendency but without statistical significance. (Table [2](#T2){ref-type="table"}) ###### Allelic and genotypic frequency comparison between fetal and control groups. -------------------------------------------------------------------------------------------------------------- **C677T-*MTHFR***\ **Frequencies by date of birth** **Genotype-Allele** --------------------- ---------------------------------- ------------- ------------- ----- ------------- ----- ***Fetal Group*** *(n = 110)* *(n = 129)* *(n = 105)* \% n \% n \% n **CC** 0.073\*\* 8 0.008\*\* 1 0.019\*\* 2 **CT** 0.727\*\* 80 0.705\*\* 91 0.6\*\# 63 **TT** 0.2 22 0.287 37 0.381\#\# 40 **C allele** 0.436\*\* 96 0.36\*\* 93 0.319\*\*\# 67 **T allele** 0.564\*\* 124 0.64\*\* 165 0.681\*\*\# 143 ***Control Group*** *(n = 88)* *(n = 93)* *(n = 95)* \% n \% n \% n **CC** 0.33 29 0.301 28 0.305 29 **CT** 0.477 42 0.452 42 0.421 40 **TT** 0.193 17 0.247 23 0.274 26 **C allele** 0.568 100 0.527 98 0.516 98 **T allele** 0.432 76 0.473 88 0.484 92 -------------------------------------------------------------------------------------------------------------- Note: \*p \< 0.05; \*\*p \< 0.01 (significant values relative to controls). \#p \< 0.05; \#\#p \< 0.01 (significant values relative to the 1982--1984 period) The evolution of genotype frequencies during the 1980--1989 decade of the TT genotypes correlates well in both living populations as well as fetal samples r = 0.98 (p = 0.11). Discussion ========== Different reports show that the prevalence of the 677C\>T polymorphism of the *MTHFR*gene differs dramatically among human populations. Evidence of this dynamism can be observed in many reports: frequency variations between populations that are geographically very close, even in the same country \[[@B8]\]; changes found in the same race or ethnic group such as Africans \[[@B10],[@B11]\] and African-Americans \[[@B12]\]; the high prevalence of the 677C\>T poymorphism in populations with special nutritional features such as Mexicans \[[@B9]\] and Japanese \[[@B13]\]; and changes in frequencies between generations of the same population, as has been observed in Spain \[[@B5]\], Switzerland \[[@B6]\] and Japan \[[@B7]\]. There are numerous interpretations of this great diversity, and most tend to be related to adaptation to external conditions such as climate or nutritional status. Dependence of folate degradation on skin pigmentation \[[@B17]\], nutritional habits or human intervention periconceptional periods could explain this genetic variation. Definitely, external factors in combination with different levels of MTHFR enzyme activity, conditioned by polymorphisms, could influence the fetal viability of certain genotypes. In 1998, we suggested the possibility of genetic selection in Spain in favor of the mutants of the 677C\>T polymorphism in the *MTHFR*gene based on the fact that treatment with vitamins and folates during pregnancy increased the viability of fetuses with the TT homozygous genotype. This hypothesis was based on the increase in the number of mutated individuals found in our population since the mid-1970s \[[@B5]\] and the coincident increased intake of vitamins and folate by pregnant women in Spain \[[@B18],[@B19]\]. In 2002, a new study found changes in genotype frequencies for the 677C\>T and 1298A\>C polymorphisms in different age groups. Total homocysteine (tHcy) levels in plasma were also analyzed according to the different genotype interactions \[[@B20]\]. That study hypothesized about fetal viability and about a genetic selection model on the basis of non-linkage disequilibrium between both polymorphisms. Recently, a study with fetal and control populations showed the strong influence of these polymorphisms, though mainly of the 677C\>T polymorphism, on spontaneous early abortion \[[@B21]\]. In the present study, significant changes in allelic and genotypic frequencies are detected, as is Hardy-Weinberg disequilibrium, at the 677C\>T polymorphism. We hypothesize that there is a dynamic process of genetic selection that favors the T allele. This process of selection started during the last quarter of the 20th century, during which the frequency for mutant homozygous (TT) rose significantly from 14% to 24%. We propose that this increase in mutants is due to the inclusion of an external factor that enhances mutant fetal viability. If we apply the mathematical model for dynamic selection developed for diploid organisms with sexual reproduction, the T allele could reach to 90% in seven generations in our population (Figure [1A](#F1){ref-type="fig"}). However, this model assumes selection in a constant environment that applies to all individuals in the population studied. In our case, we suggest that the external factor is related to an increase in folate and vitamin intake in women in periconceptional period and does not affect to all individuals \[[@B18],[@B19]\]. We assume that prediction of a classic selection model in this case is only theoretical. On the basis of a competition between alleles in which an environmental factor favors one allele versus the other, the final result would be that predicted by the previous mathematical model. However in this case, the environment is not selecting against the wild type allele but rather allowing the survival of more mutated alleles. Therefore, the expected result would be not a systematic increase of the mutated allele but the creation of an allelic balance dependent on vitamin and folate abundance conditions. In this case, the mutation would have a lower influence on fetal viability (Figure [1B](#F1){ref-type="fig"}). The results showed an increase in mutated genotypes (CT and TT) and a strong protection against abortion by the wild type genotype (CC), which is practically non-existent in the SA group. The frequency of the CC genotype shows no change over the decade studied (1980--1989), which indicates that folate does not exert a visible effect on this genotype. However, the frequency of the mutated allele increases during this decade, especially in fetuses from abortions, and this increase correlated with the increase of the T allele in the control population. This finding suggests that the effect of folate is crucial to viability during the early stages of embryonic development, but, even with folate, not all embryos will survive until birth. In this population, the mutant allele with lower enzymatic activity has higher fitness than the wild type. In the folate cycle, it can be observed that 5,10-methyleneTHF availability may be important. 5,10-methyleneTHF is the substrate for several reactions in the cycle, but two of them (5-methylTHF and thymidilate synthesis) might be essential for embryo development in folate deficiency conditions. In both cases, complete or limited MTHFR activity will produce higher or lower 5,10-methyleneTHF availability, which might be an essential factor for embryo development, such that a greater folate levels can compensate the lower enzymatic activity of the mutant. The implications of this polymorphism in nucleotide synthesis have not yet been determined, but certain data, such as high levels of uric acid found in mutated subjects \[[@B22],[@B23]\], suggest that there are different turnover rates associated with different polymorphisms. Conclusion ========== We suggest that there is genetic selection in our population for the T allele of the *MTHFR*-- 677C\>T polymorphism, whose origin could be an increase in fetal viability during the early stages of embryonic development because of an increase in folate and vitamin intake by women in the periconceptional period that began to be established in Spain in the last quarter of the 20th century \[[@B18],[@B19]\]. Higher frequencies for the T allele and TT genotype in our population are observed in the living and SA populations. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= AMO performed the statistical analysis, helped to draft the manuscript and revised it for publication. GC is the corresponding author, participated in the acquisition of samples and carried out the genotyping. ARP carried out the bibliographic search and helped to draft the manuscript. AJJ participated in the selection and the processing of samples. MJG coordinated the laboratory work and selected the genotyping method. AR selected the control subjects and designed the consent form. MR helped in the interpretation of data and tables performance. ARE conceived the study and is the guarantor of this work and the general coordinator. All authors read and approved the final manuscript. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2350/9/104/prepub> Acknowledgements ================ This study was supported by the Ministerio de Educación y Ciencia (Spain), Grants (SAF2008-03314).

Background ========== Papaya mosaic virus (PapMV) is a member of the large family of *Flexiviridae* in the genus *Potexvirus*. The virus has a flexuous rod shape of 500 nm in length and 13 nm in diameter \[[@B1]\]. The CP, made mostly of alpha helices \[[@B2]\], is composed of 215 amino acids and has an estimated molecular weight of 23 kDa \[[@B3]\]. We showed previously that non-infectious nanoparticles made of recombinant PapMV CP are similar in shape and appearance to wild-type virus purified from plants \[[@B3]\]. PapMV nanoparticles were used previously as a vaccine platform technology to improve the immunogenicity of a peptide antigen fused to the nanoparticle structure \[[@B4]-[@B8]\]. The PapMV vaccine platform can induce a long-lasting memory response to an antigen fused on its surface \[[@B4]\]. Previous studies showed the capacity of PapMV nanoparticles to trigger a CTL response, in both *in vitro* and *in vivo* models, when the CTL epitope was fused to the C-terminus of the CP \[[@B6],[@B7],[@B9]\]. Although PapMV tolerates insertion of several peptides to its C-terminus \[[@B4]-[@B7],[@B10]\], a recent study revealed that N-terminal fusion of some peptides is also tolerated \[[@B8]\]. Depending on the nature of the amino acid sequence, some peptides can interfere with the CP assembly or with nanoparticle stability, which can affect their ability to stimulate an humoral response. A modification of the fusion site on the CP can help to resolve this issue. In this study, we compared the efficacy of nanoparticles harbouring fusion of a CTL epitope at either the N- or the C-terminus to trigger a cellular immune response. The crystalline and highly ordered structure of the nanoparticles is critical to triggering an efficient humoral response, as also reported by many other groups \[[@B11]-[@B13]\]. However, it is still unknown if assembly into the highly ordered nanoparticle structure, made of several hundreds of PapMV CP, is more efficient than assembly of a smaller disc-like structure (aggregate of 20 subunits) in triggering the CTL response. Since the mechanisms of induction of humoral and CTL immune responses rely on different immune cells and mechanisms, we also evaluated the importance of highly ordered assembly of recombinant PapMV CP into nanoparticles in triggering a CTL response by comparing the immunogenicity of nanoparticles and discs. Results and discussion ====================== Engineering of PapMV nanoparticles fused to the influenza CTL epitope --------------------------------------------------------------------- In this study, we used the CTL epitope NP~147-155~, a 9-mer H-2Kd epitope specific for Balb/C mice derived from influenza virus to search for the optimal position for fusion of this epitope to the vaccine platform. The CTL epitope was flanked by 5 native residues of influenza NP protein at both the N- and C-termini to favour natural processing of the peptide, and was fused genetically to either the N- (before F13) or C-terminus of PapMV CP \[Figure [1](#F1){ref-type="fig"}A\]. Following expression in *E. coli*, recombinant PapMV particles harbouring the peptide on their surface (named PapMV NP-12 and PapMV NP-C) were affinity purified on a Ni^2+^ column via the 6xHis tag located at the C-terminus \[Figure [1](#F1){ref-type="fig"}B\]. Purified proteins were subjected to ultracentrifugation to pellet the nanoparticles. The lower molecular weight forms, composed of discs (20 subunits of CP) remained in the supernatant. As shown by dynamic light scattering (DLS), two structures were obtained with this purification protocol: (1) typical long rod-shaped structures of approximately 90 nm in length for PapMV NP-12 and PapMV NP-C \[Figure [1](#F1){ref-type="fig"}C--D\], and (2) smaller discs of approximately 20 nm in length for PapMV NP-12 and PapMV NP-C \[Figure [1](#F1){ref-type="fig"}D\]. The DLS method provides a global view of the size of the particles in solution that fits our qualitative observations by electron microscopy, and revealed the difference in size between discs (black line) and nanoparticles (dotted line) for both constructs (Figure [1](#F1){ref-type="fig"}D). It also shows that structures formed by NP-12 and NP-C constructs are similar in size, confirming observations by transmission electron microscopy showing that PapMV NP-12 and NP-C nanoparticles are similar in length, structure and appearance \[Figure [1](#F1){ref-type="fig"}C\]. {#F1} We previously reported that residue F13 of PapMV CP is critical for the interaction between the PapMV CP subunits when assembling into nanoparticles \[[@B14]\]. We showed that this hydrophobic residue fits snugly inside the hydrophobic pocket of the neighbouring CP \[[@B2]\]. Interestingly, insertion of the NP~147-155~ epitope just before F13 in the N-terminal fusions clearly does not interfere with the interaction between PapMV CP monomers that is crucial for self-assembly of nanoparticles. PapMV NP-12 nanoparticles are better inducers of the CTL response ----------------------------------------------------------------- To evaluate the potential of PapMV nanoparticles to induce a CD8+ mediated cellular response, we immunised 6- to 8-week-old Balb/C mice three times at 2-week intervals by the intraperitoneal route with 100 μg of PapMV (without fusion), NP-12 or NP-C nanoparticles. Two weeks after the second boost, spleens were harvested and ELISPOT assays using the NP~147-155~ peptide were performed to quantify the level of IFN-γ secreted by CD8+ cells \[Figure [2](#F2){ref-type="fig"}\]. Secretion of IFN-γ is proportional to the level of precursors of CD8+ cytotoxic lymphocytes specific to the fused CTL epitope in vaccinated mice. The result showed that, compared to all the other treatments, mice immunized with PapMV NP-12 nanoparticles secrete significantly more IFN-γ \[Figure [2](#F2){ref-type="fig"}\]. {#F2} Fusion of a peptide to the PapMV vaccine platform could affect its stability, and potentially the ability to mount an immune response to the fused epitope \[[@B8]\]. As temperature can affect protein stability, we thus measured the influence of temperature on the aggregation of recombinant nanoparticles using DLS. Upon heating, proteins initiate partial denaturation through exposure of their hydrophobic residues to the solvent. This conformational change triggers formation of aggregates that can be measured easily by DLS. We found that PapMV NP-C nanoparticles initiated aggregation at 25°C while PapMV (without fusion) and NP-12 nanoparticles were more stable and initiated aggregation at 37°C or higher \[Figure [3](#F3){ref-type="fig"}A\]. Therefore, the higher stability of NP-12 nanoparticles at 37°C or higher appears to correlate with an optimal CTL response in mice. {#F3} Based on this observation, it was anticipated that stabilization of NP-C nanoparticles by chemical cross-linking should improve their immunogenicity. Therefore, we compared the capacity of NP-C and NP-12 nanoparticles either cross-linked with glutaraldehyde or not to induce a CTL response. The cross-linked nanoparticles were very stable even at temperatures exceeding 37°C \[Figure [3](#F3){ref-type="fig"}B\]. Mice (5 per group) were immunized three times at 2-week intervals with 100 μg of PapMV NP-C and NP-12 nanoparticles that were either cross-linked with glutaraldehyde or not. Surprisingly, cross-linking NP-C did not lead to an improved CTL response in mice \[Figure [4](#F4){ref-type="fig"}\]. The quantity of IFN-γ secreted by specific splenocytes remained similar to the response obtained with untreated nanoparticles. However, cross-linking of NP-12 nanoparticles did affect the efficacy CTL response induction as compared to native NP-12 \[Figure [4](#F4){ref-type="fig"}\]. We hypothesize that the cross-link made the nanoparticles too rigid, which consequently decreased their susceptibility to the proteases responsible for releasing the CTL epitope. To test this hypothesis, we resolved trypsin digests of cross-linked NP-12 and NP-C nanoparticles by SDS-PAGE (Figure [5](#F5){ref-type="fig"}). NP-12 and NP-C nanoparticles were only partially susceptible to the digest and lost 3 kDa on the gel as compared to non-digested nanoparticles. Cross-linked NP-C and NP-12 appeared resistant to digestion. The PapMV CP subunits of the cross-linked nanoparticles remained tightly attached to each other, leading to a very large molecular weight multimer that can barely enter the acrylamide gel. This experiment highlighted the likely difficulty of digesting cross-linked nanoparticles by host proteases involved in the excision of CTL epitopes for loading onto the MHC class 1 complex---an essential step in triggering a CTL response. In brief, nanoparticles must be sufficiently stable at animal body temperature (37°C) but flexible enough to allow their digestion by cellular proteases for loading onto MHC class 1 molecules. Therefore, an optimal balance between these two properties is crucial to triggering the CTL response efficiently. {#F4} {#F5} PapMV NP-12 nanoparticles but not discs are able to trigger a CTL response -------------------------------------------------------------------------- In this experiment, our objective was to compare the efficacy of PapMV nanoparticles and discs in triggering a CTL response. PapMV NP-12 discs and nanoparticles were used to immunize (3 immunizations) mice (5 per group) with 100 μg of protein. We measured the production of IFN-μ after stimulation of splenocytes harvested from immunized mice 2 weeks after the second boost using the NP~147-155~ peptide. The level of IFN-γ secreted by splenocytes specific to PapMV NP-12 nanoparticles was significantly higher than that specific to PapMV NP-12 discs \[Figure [2](#F2){ref-type="fig"}\], suggesting that assembly into a highly ordered structure, i.e. nanoparticles, is critical to triggering the CTL response efficiently. We also noted that NP-12 discs appeared less stable than nanoparticles, and initiated aggregation at 34°C as compared to 37°C for nanoparticles \[Figure [3](#F3){ref-type="fig"}\]. Discs have the same diameter but are shorter than nanoparticles (30 nm vs 90 nm) and also less stable. These differences in size and stability could account for the observed differences in immunogenicity. Another difference between discs and nanoparticles is the RNA that they contain. Discs are associated with only very small amounts of RNA but nanoparticles contains ssRNA of bacterial origin \[[@B3]\]. It is possible that the ssRNA found in nanoparticles plays a role in the efficacy of the measured immune response. It is known that ssRNA of bacterial origin, as found in PapMV nanoparticles, can be recognized as pathogen associated molecular patterns (PAMPs) by several nucleic acid sensors like RIG-I, MDA-5, TLR7 or TLR8 that are at the interface between the innate and the adaptive immune response \[[@B15]-[@B17]\]. We are currently investigating if these sensors play a role in the CTL response. Our results are consistent with findings obtained with peptide fusions made at the N-terminus of the CP of potato virus X (PVX)---another member of the potexvirus family. It was shown that production of recombinant PVX virus particles *in planta* can elicit either an humoral \[[@B18]\] or a CTL immune response \[[@B19]\]. The use of the N-terminus for fusion of peptides on this type of vaccine platform can, however, be problematic if the recombinant virus particles are produced in planta because the fusion may interfere with long-distance transport of the virus throughout the plants and thus affect yield \[[@B20]\]. This is one of the main reasons why we chose to produce our nanoparticles in a bacterial expression system that does not depend on the replication or cell-to-cell transport of the virus. It is well accepted that the N-terminus of PapMV and potexvirus CP is exposed at the surface of the virus particle \[[@B2],[@B8],[@B21],[@B22]\]. The recently published 3D structure of PapMV CP revealed that the N-terminus is involved in the interaction between two CP subunits in the virus particle \[[@B2]\], and that the 12 N-terminal residues upstream of F13 are directly exposed on the surface \[[@B14]\]. The availability of the CTL epitope located at this position, as compared to the C-terminus, probably facilitates its cleavage by host proteases and favours loading of the MHC class I pocket in the immune cells \[[@B6],[@B9]\]. Conclusions =========== The results of this study improve our understanding of the PapMV vaccine platform and highlight the importance of nanoparticle stability in triggering a CTL response. We can now beneficiate of two different points of fusion for a CTL epitope on the PapMV CP. The fusion at the N-terminus was clearly superior for the NP~147-155~ peptide but it does not mean that this will be the case for another CTL epitope. The amino acid sequence of the CTL epitope and its influence on the structure on the PapMV CP can have a major impact on its stability and their immunogenicity. Those results are increasing the versatility of the vaccine platform and provide more options for production of stable constructs. Because it is well established that the trigger of a CTL response to conserved epitopes is a valuable approach in the design of prophylactic or therapeutic vaccines to chronic diseases \[[@B23]-[@B27]\], we believe that the PapMV vaccine platform will be a very useful tool. Methods ======= Ethics statement ---------------- All the work with animals adhered to the Institution-approved ethics protocol of the "Comité de Protection des Animaux" -- CHUQ (CPA-CHUQ). The approval of this project is listed under the authorization number 2010148--1. Cloning and production of PapMV NP constructs --------------------------------------------- The PapMV CP construct (CPΔN5) used for this study has been described previously (Tremblay et al., 2006). To generate the PapMV NP constructs, PapMV CPΔN5 was used to introduce the following oligonucleotides by PCR at position 12 at the C-terminus of the coat protein \[Table [1](#T1){ref-type="table"}\]. The linear vector harbouring the NP~147-155~ coding region fused to the PapMV was digested with BsiWI and ligated using T4 DNA ligase (New England BioLabs). The resulting PapMV NP clones harbour a fusion of the nucleoprotein epitope from Influenza virus at position 12 or at their C-terminus followed by a 6xHis tag for the purification process. We retained five amino acids on either side of the H-2Kd CTL epitope to ensure efficient processing. The integrity of the PapMV NP clones was confirmed by DNA sequencing. Expression and purification of PapMV particles fused to the NP~147-155~ peptide was performed as described previously \[[@B5]\]. Levels of expression for each recombinant nanoparticle were determined by SDS-PAGE. LPS contamination was always less than 50 endotoxin units (EU)/mg of protein. The size and structure of the nanoparticles were confirmed by observation on a TEM (JEOL-1010, Tokyo, Japan). ###### Forward and reverse oligonucleotides used to produce PapMV CP recombinant proteins ***Name*** ***Oligonucleotide sequence*** ------------ ----------------------------------------------------------------------------- **NP-12** Forward 5^′^-AGCT***CGTACG***CGTGCGCTGGTTCGTACCGGTATGGACTTCCCCGCCATCACCCAGGAAC-3^′^ Reverse 5^′^-TCGA***CGTACG***CTGGTAGGTCGCGTCGTTCAGGTTGGCTATGTTGGGTGTGGATGCC-3^′^ **NP-C** Forward 5^′^-ACGT***CGTACG***CGTGCGCTGGTTCGTACCGGTATGGACACGCGTCACCATCACCATCAC-3^′^ Reverse 5^′^-TCGA***CGTACG***CTGGTAGGTCGCGTCGTTCAGGTTACTAGTTTCGGGGGG-3^′^ Dynamic light scattering ------------------------ The size of nanoparticles and discs was determined using a ZetaSizer Nano ZS (Malvern, Worcestershire, United Kingdom) at a temperature of 4°C and at a concentration of 0.1 mg/ml in PBS 1x for nanoparticles and at a concentration of 0.25 mg/ml in Tris--HCl 10 mM for discs. The thermal stability of PapMV nanoparticles was measured under the same experimental conditions at temperatures from 24°C to 40°C. Chemical cross-linking with glutaraldehyde ------------------------------------------ Cross-linking was performed using 0.1% glutaraldehyde in 10 mM Tris, 50 mM NaCl pH 7.5 in a final volume of 50 μl. The optimal concentration of protein for cross-linking was 150 ng/ml. After addition of glutaraldehyde, the mixture was incubated at room temperature for 30 minutes in the dark. The reaction was stopped with 15 μl of loading dye and heated for 10 minutes at 95°C before separating the proteins by 8% SDS-PAGE. The cross-linked proteins used to immunize mice were stored at 4°C until immunization without adding loading dye. SDS-PAGE and trypsin digest --------------------------- Prior to SDS-PAGE, samples were mixed with one-third of the final volume of loading buffer containing 5% SDS, 30% glycerol and 0.01% bromophenol blue and heated for 10 minutes at 95°C. For the trypsin digest, we incubated 10 μg of proteins at 37°C in a volume of 50 μl for 120 minutes in 100 mM Tris--HCl pH 8.5 with 0.2 μg trypsin (Roche, 1418475). The reaction was stopped by adding 10 μl of loading dye. Samples were heated for 10 minutes at 95°C prior to loading on SDS-PAGE \[[@B3]\]. Immunization ------------ Five 6- to 8-week-old Balb/C mice (Charles River, Wilmington, MA) were immunized by the intraperitoneal route with: (i) 100 μg of PapMV NP-12 nanoparticles; (ii) 100 μg of PapMV NP-C nanoparticles; (iii) 100 μg of PapMV CP nanoparticles; (iv) 100 μg of PapMV NP-12 discs, and (v) 100 μg of PapMV NP-C discs. Primary immunization was followed by two booster doses given at 2-week intervals. Blood samples were obtained before each injection and 2 weeks after the last one and stored at -20°C until analysis. ELISPOT ------- Two weeks after the last boost, mice were sacrificed and spleens were recovered for ELISPOT assay performed as described previously \[[@B25]\]. The precursor frequency of specific T cells was determined by subtracting the background spots in media alone from the number of spots seen in wells reactivated with NP~147-155~ peptide. Data were analyzed with a parametric or a non-parametric ANOVA test when the variances differed significantly and with a Tukey or a Dunn post-test to compare difference among groups of mice. Values of \*p \< 0.05, \*\*p \< 0.01 and \*\*\*p \< 0.001 were considered statistically significant. Statistical analyses were done with GraphPad PRISM 5.01. Competing interest ================== DL is founder and shareholder in the company FOLIA BIOTECH INC., a start-up company with the mandate to exploit commercially the PapMV nanoparticle technology. The patent is issued (U.S. Patent No. 7,641,896). This does not alter the authors' adherence to all the BMC policies. Authors' contributions ====================== CB carried out all the experiments presented in this manuscript. NM and DL participated in the design of the study and supervision of CB. CB help to make the first draft of the manuscript. DL coordinated the study and completed the writing of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This research project was funded by the Canadian Institute of Health Research Canada (CIHR) (grant number: 185160). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. We also thank Dr Helen Rothnie for English editing of the manuscript.

**PURPOSE:** The face and craniofacial skeleton (CFS) make up a complex 3D structure that is critical to human function and cosmesis. Traumatic injury to the CFS requires fracture treatment to both allow the recovery of mechanical function and forms a foundation for the restoration of soft tissue anatomy. CFS reconstruction aims to restore pre-injury appearance, however in severe injuries shape information for the skull and facial bones may be missing. This presents a particular challenge in bi-frontal injuries and pan-facial fractures where the mirror imaging of the intact side of the head cannot be used to guide reconstruction. The reconstruction of 3D facial surface geometry from pre-injury 2D photographs has recently been established through large scale morphable face modeling.^1^ As well, in forensic sciences, models with variable soft-tissue depths^2^ are used to determine face shape from skull geometry. This study aims to 'reverse-engineer' a forensics' tissue depth model to determine pre-injury CFS shape from reconstructed 3D facial geometry. It is hypothesized that 3D forensics data can be used to fill in missing gaps in CFS geometry with sufficient accuracy to guide pre-operative planning for CFS reconstruction. **METHODS:** The forensics' tissue depth model was applied to 3D facial geometries acquired through segmentation of head CT data. Age, sex and BMI were used as input parameters to guide the application of the forensics' tissue depth model data to each face. The tissue depths between the face and CFS were determined by finding the Euclidian distance transform (nearest neighbor) employed by the original forensics study and via calculation using normal vectors generated from the face surface. Calculated tissue depth was evaluated against measured thickness on the head CT between the segmented CFS (bone) and the skin. **RESULTS:** Tissue depth determined by nearest neighbor and normal vector measurements yielded accurate reconstructions of the frontal and zygoma bones (within 1mm, or +/- 2 voxels). However, only the normal vector technique succeeded in estimating tissue depth in bone regions where the face and skull have differing concavity (i.e. eye sockets, maxilla). Agreement was more limited in the lower facial skeleton where greater variation of soft tissue structures occur. **CONCLUSION:** The reversed forensics tissue depth model was found to appropriately infer bony anatomy for the upper CFS from 3D face geometry. The 3D skull shaping provided by this work yields sufficient accuracy to warrant its inclusion into a translational pipeline of tools for pre-operative planning for CFS reconstruction. **References:** 1\. Booth J. et al. Int J Comput Vis. 2017, 1--22. 2\. Shrimpton S. et al. *Forensic Sci. Int.* 2014, *234*, 103--110.

1. Introduction {#sec1} =============== In wild sika deer, *Cervus nippon yesoensis*, in Hokkaido, Japan, sarcocysts of protozoan parasites of the genus *Sarcocystis* are highly prevalent, and four *Sarcocystis* species have been identified in the region to date: *S. ovalis*, *S. pilosa*, *S. tarandi*-like, and *S. truncata*-like ([@bib28]; [@bib19]; [@bib12]). Among these species, *S. ovalis* has been reported to have a corvid as its definitive host in the region ([@bib13]). Although the definitive hosts of the remaining species remain unknown, phylogenetic and epidemiological evidence seems to indicate that members of the Felidae (or unknown animals) are the likely definitive hosts of *S. tarandi*-like and *S. truncata*-like ([@bib4]; [@bib10]). The remaining species, *S. pilosa*, was originally isolated from *C. nippon* in Lithuania ([@bib21]), where seven *Sarcocystis* species were characterized in farmed sika deer by means of morphological and molecular methods ([@bib22]). Seven partially different *Sarcocystis* species, including *S. pilosa*, were also characterized in wild sika deer (*C. nippon centralis*) in mainland Japan ([@bib2]). The definitive host of *S. pilosa* is suspected to be a member of the Canidae, because *S. pilosa* falls phylogenetically within a clade that includ *Sarcocystis* spp. using Canidae as their definitive host. Further, sarcocysts that are morphologically similar to *S. pilosa* have been described in *C. nippon centralis* and *C. nippon yesoensis* in Japan, and this type of sarcocyst is experimentally able to infest and reproduce in dogs ([@bib25], [@bib26]; [@bib3]). The red fox, *Vulpes vulpes schrencki*, is a very common canid in Hokkaido, and has been observed feeding on deer carrion ([@bib29]). It therefore seems likely that red foxes serve to maintain *S. pilosa* in sika deer in Hokkaido. To clarify the *Sarcocystis* life cycle and the cause of the high prevalence of these parasites in deer, a survey of fecal sporocysts in red fox fecal samples was conducted. 2. Materials and methods {#sec2} ======================== 2.1. Red fox fecal sample collection {#sec2.1} ------------------------------------ As described by [@bib18], red fox fecal samples were collected along road verges, agricultural fields, and paths that were likely to be utilized by red foxes in eastern Hokkaido. Fecal collection was conducted in May 2018 (n = 44) and in December 2018 (n = 21). To aim to collect feces from different foxes, the sites where feces were picked up were separated by at least 2 km. To confirm that fecal samples belonged to red foxes, fecal DNA was extracted as described by [@bib20], and the 12S ribosomal RNA (rRNA) gene was analyzed as described by [@bib31]. To inactivate *Echinococcus multilocularis* eggs, which are highly prevalent in red foxes in the study area, fecal samples were incubated at 70 °C for 12 h before being stored at −30 °C until use. 2.2. Fecal sporocyst examination {#sec2.2} -------------------------------- Fecal sporocysts were examined in 1 g of feces using a modified sucrose flotation method (specific gravity, 1.27) ([@bib14]). For species identification, sporocysts were collected from the supernatant of the centrifuge tube by simple sedimentation in saline. 2.3. DNA extraction and PCR sequencing of collected sarcocysts {#sec2.3} -------------------------------------------------------------- Genomic DNA from collected sporocysts was extracted using a PowerSoil DNA Isolation Kit (Mobio Laboratories, Solana Beach, CA) according to the manufacturer\'s instructions. The 18S rRNA and cytochrome *c* oxidase subunit I (COI) genes were amplified and sequencing was performed using previously described primers ([@bib12]). Direct sequencing was performed using a BigDye Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems, Carlsbad, CA), and the obtained sequences were compared with those deposited in GenBank/EMBL/DDBJ. 3. Results and discussion {#sec3} ========================= Of the 65 fecal samples analyzed, one (collected in December 2018) contained *Sarcocystis* sporocysts at approximately 500 sporocysts per gram of feces. The fecal sample was confirmed to belong to *V. vulpes* based on the 12S rRNA gene sequence. To confirm the presence of sporocysts in the feces, fecal examination was repeated again, and revealed reproducibility of the examination. The fecal sample was also positive for eggs of *E. multilocularis* and Capillariidae, and negative for *Cystoisospora* oocysts. The Capillariidae species might be *Calodium hepaticum* (syn. *Capillaria hepatica*), which reproduces in the liver of vole, and the eggs pass thorough the intestines of foxes that prey on vole. In other feces, eggs of hookworm, *Toxocara canis*, *Dibothriocephalus nihonkaiensis*, and oocysts of unidentified *Cystoisospora* were detected by the flotation. The detected *Sarcocystis* sporocysts contained four sporozoites and were 15.0 μm (SD = 0.2) long and 9.3 μm (SD = 0.4) wide ([Fig. 1](#fig1){ref-type="fig"}) (mean values for 100 sporocysts). 18S rRNA (1669bp) and COI (1029bp) gene sequences of the sporocysts were deposited in GenBank/EMBL/DDBJ with accession numbers LC496069 and LC496070, respectively. Both sequences were 99.95%--100% identical to those of *S. pilosa* obtained from sarcocysts from sika deer in Hokkaido and Lithuania, and 99.23%--99.87% identical to such samples from mainland Japan ([Supplemental Table S1](#appsec1){ref-type="sec"}).Fig. 1Morphology of detected sporocysts under light microscopy. Scale bar: 10 μm.Fig. 1 Although *Sarcocystis* is considered highly endemic in sika deer from Hokkaido ([@bib26]; [@bib12]), very little is known about the life cycles of these *Sarcocystis* spp. Of the four prevalent species in the region, we focused on *S. pilosa*, the definitive hosts of which are suspected to belong to the Canidae. We therefore collected and analyzed red fox fecal samples and we detected *S. pilosa* sporocysts in one sample. Although it was previously experimentally demonstrated that red foxes can act as a definitive host for *S. hjorti*, which is closely related to *S. pilosa* ([@bib5]), this is the first record of *S. pilosa* sporocysts in feces excreted by red foxes in the wild. The finding indicates that the red fox serves as a definitive host of *S. pilosa*, and also that red foxes could be an infection source for deer in the region. In this study, only one fecal sample from red foxes contained sporocysts, representing 1.5% of the analyzed fecal samples and 4.8% of fecal samples collected in the winter. Prevalence of *Sarcocystis* sporocysts in feces of red foxes has been evaluated in many studies: 1.9% in Bulgaria ([@bib16]), 3.8% in Ireland ([@bib30]), 10.1% and 17.9% in the USA ([@bib7]; [@bib6]), and 84.4% in Newfoundland ([@bib15]). Species were unfortunately not identified in those studies, and thus intermediate host animals were also not determined. Although simple comparison might not be appropriate because of biological and geographical differences, the rate of sporocyst-positive feces in the present study might relatively be low. However, the number of fecal samples examined in our study was statistically insufficient, and further investigation is necessary to evaluate the true prevalence. In addition, it is indispensable in future work to confirm that *S. pilosa* can reproduce in fox intestine by histological and/or intestinal scraping examinations, to eliminate the possibility that we have observed pseudoparasitism caused by ingestion by red foxes of carcasses of other animals that do serve as a definitive host of *S. pilosa*. Red fox feeding habits in Hokkaido were inferred by analyzing residues in collected feces, which revealed that the annual percentage occurrence of material derived from deer was 16.7%, with the highest utilization of deer occurring in May ([@bib29]). Considering the feeding habits of red foxes, the prevalence of *Sarcocystis* infection in red foxes may be higher than that observed in our study. Moreover, given that *S. pilosa* endemicity among deer in the region is \>90% ([@bib12]), difference in the prevalence of these parasites between the definitive and intermediate hosts is considered unlikely. Excretion of sporocysts peaks during the early patent period, before dramatically decreasing in just a few weeks in the late patent period ([@bib24]). In addition, in fecal samples of arctic foxes (*V. lagopus*), lower detectability of fecal sporocysts using a flotation technique was reported than that by qPCR analysis of fecal suspension (3% and 16%, respectively) ([@bib9]). Consequently, fecal sporocyst detectability might be markedly lower than the actual prevalence in the wild. Among the *Sarcocystis* species that use deer as an intermediate host in Germany (e.g. *S. tenella*/*S. capracanis*, and *S. gracilis*), the prevalence among red foxes was relatively high (6%--10% of surveyed animals) ([@bib17]). Comparison of *Sarcocystis* detectability based on fecal examination and mucosal scraping examination revealed that the latter exhibited higher sensitivity ([@bib27]). Thus, to more accurately evaluate *Sarcocystis* prevalence, such mucosal examination of small intestines from the definitive host animal is necessary. Further, sporocysts are known to be resilient in field conditions (especially in cold temperatures), and can remain infective for extended periods of time ([@bib23]). In the same way that the red fox serves as the definitive host for some *Sarcocystis* spp. ([@bib8]; [@bib5]; [@bib17]), so too does the raccoon dog ([@bib11]; [@bib17]). Indeed, raccoon dogs have been experimentally infested with *Sarcocystis* by feeding with sarcocysts from sika deer in Japan, and they excreted sporocysts in their feces ([@bib26]). The *Sarcocystis* species used in that experiment was not clear due to the lack of molecular information at the time. Sarcocysts of *S. pilosa* were also detected in *C. nippon centralis*, which exists on the main island of Japan ([@bib2]), and slight molecular diversity was observed compared with *S. pilosa* isolates from Hokkaido and Lithuania ([Supplemental Table S1](#appsec1){ref-type="sec"}). This might reflect a difference in the definitive host animals; the red fox population is suspected to be smaller on the main island of Japan, and, therefore, red foxes may not play an equally important role in the transmission of *S. pilosa* in other regions of Japan as they presumably do in Hokkaido. Moreover, the raccoon dog could also be a definitive host of *Sarcocystis* species, which could be determined by a survey of raccoon dogs in Hokkaido and the main island of Japan. In conclusion, this is the first study to find *S. pilosa* sporocysts in red fox feces in Hokkaido, which indicates that red foxes could be a definitive host of this *Sarcocystis* species on this island. Because of the low detectability of sporocysts by fecal examination, examination of intestinal mucosal scrapings from red foxes is considered necessary to more accurately reveal the prevalence of this parasite in the region. In addition, a survey of other canids (e.g. raccoon dogs) should be undertaken to identify the primary definitive host(s) for *S. pilosa* in this region. Financial supports {#sec4} ================== Part of this work was supported by Health and Labour Sciences Research Grants (H30-Shokuhin-Wakate-003 and H30-Shokuhin-Ippan-011) that organaized by Ministry of Health, Labour and Welfare (Japan). Declaration of competing interest ================================= The authors received no support and have no relationships that could be perceived as constituting a conflict of interest. Appendix A. Supplementary data {#appsec1} ============================== The following is the Supplementary data to this article:Multimedia component 1Multimedia component 1 We thank Mallory Eckstut, PhD, from Edanz Group ([www.edanzediting.com/ac](http://www.edanzediting.com/ac){#intref0010}) for editing a draft of this manuscript. Supplementary data to this article can be found online at <https://doi.org/10.1016/j.ijppaw.2019.12.001>.

1. Introduction =============== The occurrence and undesirable complications from healthcare-associated infections have been well recognized in the literature for the last several decades \[[@B1-ijerph-09-03330]\]. The most common sources of infectious agents causing healthcare associated infections, described in a scientific review of 1,022 outbreak investigations \[[@B2-ijerph-09-03330]\] are (listed in decreasing frequency): the individual patient, medical equipment or devices, the hospital environment, the healthcare personnel, contaminated drugs, contaminated food, and contaminated patient care equipment. Although the person-to-person transmission route is the most likely, the role of the environment should not be ignored and hospital linen may contribute to the spread of nosocomial infections \[[@B3-ijerph-09-03330],[@B4-ijerph-09-03330]\]. Healthcare textiles include bed sheets, blankets, towels, personal clothing, patient apparel, uniforms, gowns, drapes for surgical procedures \[[@B5-ijerph-09-03330]\]. Contaminated textiles and fabrics often contain high numbers of microorganisms from body substances, including blood, skin, stool, urine, vomitus, and other body tissues and fluids. Although contaminated textiles in healthcare facilities can be a source of substantial numbers of pathogenic microorganisms, reports of healthcare associated diseases linked to contaminated fabrics are few, therefore the overall risk of disease transmission is very low \[[@B5-ijerph-09-03330]\]. Cleaning in general has two main functions: first: non-microbiological, to improve or restore appearance, and prevent deterioration. Second, microbiological, to reduce the numbers of microbes present, together with any substances that support their growth or interfere with disinfection \[[@B4-ijerph-09-03330]\]. The purpose of laundering hospital textiles is therefore to ensure clean and safe textiles for patients and staff and thus enable uninterrupted implementation of healthcare \[[@B5-ijerph-09-03330],[@B6-ijerph-09-03330]\]. The most common found microorganisms on hospital textiles are: Gram negative bacteria, coagulase negative staphylococci, *Bacillus* sp. and typical skin flora \[[@B7-ijerph-09-03330]\]. Most people working in hospitals assume that laundry returned to them is in fact clean and therefore safe. Laundry may certainly have had the dirt removed, but it is far from sterile and experience encourages infection control teams to take laundering very seriously in outbreaks that seem to have no obvious cause \[[@B8-ijerph-09-03330]\]. 2. Reports on the Survival of Microorganisms on Hospital Textiles after Laundering ================================================================================== Literature in the field of survival of microorganisms on hospital textiles after laundering is very diverse and perhaps even confusing and contradictory. Each publication states a different laundering temperature as appropriate. It is therefore important to note that a successful laundering procedure is dependent on several factors and each much be optimized. These factors with a possible synergistic effect include: duration of laundering procedure, mechanical action of laundering procedure, dosage and type of added detergents and disinfection agents, bath ratio, type of linen, filling ratio, *etc*. According to Sinner the four basic interconnected factors of the laundering procedure are: duration, mechanical action, chemicals and temperature \[[@B9-ijerph-09-03330]\]. If one of these factors is decreased such as for example temperature, then other factors such as chemicals, mechanical action or time must be increased to achieve the same laundering and disinfecting effect. This also explains the differences in the published efficient laundering conditions. The exact correct optimized combination of all the mentioned factors is therefore important in order to achieve a hygienic laundering procedure for hospital textiles. Wilcox and Jones \[[@B10-ijerph-09-03330]\] stated that many isolates of *Enterococcus faecium* survived exposure to laundering temperatures specified in the U.K. Department of Health guidelines for disinfecting foul and used and infected linen (60 °C for 10 min). Another published report by Orr and co-workers \[[@B11-ijerph-09-03330]\] even confirms survival of certain strains of enterococci at laundering temperatures as high as 71 °C. They therefore concluded that hospital linen is a possible source of enterococcal cross-infection. The survival of enterococci on textiles at laundering temperatures as high as 60 °C was also confirmed by a study \[[@B12-ijerph-09-03330]\] where biomonitors (*Enterococcus faecium* inoculated onto textile swatches with pre-inoculated defibrinated sheep blood) were washed in a simulated common hospital laundry procedure. It was found that *Enterococcus faecium*, as well as *Staphylococcus aureus*, *Enterobacter aerogenes* and *Pseudomonas aeruginosa* all survived the chosen laundering conditions at 60 °C, but none of the challenge organisms survived laundering at 75 °C. On the other hand other studies \[[@B13-ijerph-09-03330],[@B14-ijerph-09-03330]\] confirm that optimizing the laundering procedure including using high-tech environmental detergents and innovative disinfection agents renders an appropriate disinfection effect, even at laundering temperatures as low as 30 °C as noted in the study \[[@B13-ijerph-09-03330]\] with challenge organisms *Enterococcus faecium* and *Enterobacter aerogenes.* In another study \[[@B14-ijerph-09-03330]\] the optimum laundering temperature was found to be at 40 °C. Walter and co-workers \[[@B15-ijerph-09-03330]\] reported that *Staphylococcus aureus* survived a 10 min laundering at 54 °C followed by drying. *Klebsiella pneumoniae* also survived the same laundering procedure, but did not survive the drying procedure. The same research also indicated that the challenge bacteria *Staphylococcus aureus* was not found after a 60 °C laundering procedure; thus recommending a 60 °C laundering procedure for linen in healthcare facilities. In the report by Smith and co-workers \[[@B16-ijerph-09-03330]\] it was found that soiled hospital terry towels initially contaminated with Gram-positive rods (predominantly *Klebsiella*, *Enterobacter* and *Serratia* spp.) and Gram-positive bacteria (predominantly *Staphylococci*) in the range between 10^7^ to 10^9^/100 cm^2^ were washed in different laundering procedures. It was found that the washing cycle with a temperature of 60 °C followed by the drying cycle at 93.3 °C was sufficient to maintain linen hygiene. Christian and co-workers \[[@B17-ijerph-09-03330]\] also conducted experimental research on low temperature laundering of hospital textiles using economically reasonable chemicals and wash conditions. They examined the disinfection effect of laundering procedures against aerobic chemoorganotrophs, staphylococci and total coliforms. They found that low temperature washing procedures (47.8 °C) using increased concentrations of bleach eliminated all bacterial groups as effectively as the high temperature procedures (77 °C). In the research by Blaser and co-workers \[[@B7-ijerph-09-03330]\] a comparison of laundering procedures at 71.1 °C and 22 °C was conducted. The argument of such low-temperature washing was the vast amount of energy used for laundering at 71.1 °C. The 22 °C laundering procedure included the use of low-temperature chemicals. The initial counts on the use soiled terry towels and sheet were between 10^6^ to 10^8^/100 cm^2^ with predominantly Gram-negative rods (especially *Enterobacteriaceae* and *Pseudomonadaceae*) and *Staphylococcus* species as the most common Gram-positive organisms. It was found that the bacterial counts from low-temperature and high-temperature washed fabrics were comparable. The authors therefore concluded that low-temperature washing for eliminating pathogenic bacteria from hospital laundry is as effective as high-temperature laundering. It has been reported that *Clostridium difficile* \[[@B18-ijerph-09-03330]\] spores can survive temperatures and chemical treatment of typical hospital laundering cycles and that cross-contamination of *Clostridium difficile* spores can occur on bed linen during a wash cycle. Therefore the persistent nature of this organism must be considered by infection control personnel when implementing programs for laundering soiled and contaminated hospital linen. Articles reporting on the survival of microorganisms on hospital textiles after laundering, together with their main conclusions are summarized in [Table 1](#ijerph-09-03330-t001){ref-type="table"}. ijerph-09-03330-t001_Table 1 ###### Reports on the survival of microorganisms on hospital textiles after laundering. Described laundering conditions Added disinfection agent or bleach Surviving microorganism Reference ---------------------------------------- -------------------------------------------------------------------------------- ---------------------------------------------------------------------------------------------------- ----------------------------------------------------- 10 min at 60 °C No *Enterococcus faecium* Wilcox & Jones, 1995 \[[@B10-ijerph-09-03330]\] 10 min at 60 °C or 3 min at 71 °C No Certain strains of *Enterococcus faecalis* and *Enterococcus faecium* Orr *et al*. 2002 \[[@B11-ijerph-09-03330]\] less than 10 min at 60 °C 3 mL Peroxyacetic acid/ kg textiles *Enterococcus faecium, Staphylococcus aureus, Pseudomonas aeruginosa* and *Enterobacter aerogenes* Fijan *et al*. 2007 \[[@B12-ijerph-09-03330]\] 20 min at 30 °C 10 mL Sodium hypochlorate/kg textiles or 12.5 mL peroxyacetic acid/kg textiles *Enterococcus faecium* and *Enterobacter aerogenes* Fijan *et al*. 2010 \[[@B13-ijerph-09-03330]\] 43 min at 30 °C 10 mL Sodium hypochlorate/kg textiles *Enterococcus faecium* 13 min at 49 °C Added chlorine bleach (without specifications) *Staphylococcus aureus* and *Klebsiella pneumoniae* Walter *et al*. 1975 \[[@B15-ijerph-09-03330]\] 66 °C Added chlorine bleach cycle (without specifications) *Staphylococci*, *Klebsiella,* and *Enterobacter* species Smith *et al*. 1987 \[[@B16-ijerph-09-03330]\] 8 min at 47.8 °C 0.58 Chlorine bleach/kg Predominantly aerobic bacteria, staphylococci and total coliforms Christian *et al*. 1983 \[[@B17-ijerph-09-03330]\] 77.2 °C 0.11 Chlorine bleach/kg 22.2 °C Low temperature bleach (without specifications) Predominantly *Enterobacteriaceae*, *Pseudomonadaceae* and *Staphylococcus* species Blaser *et al*. 1984 \[[@B7-ijerph-09-03330]\] 71.1 °C High temperature bleach (without specifications) Typical program for hospital bed linen 50 ppm Chlorine, 54 ppm peracid, 100 ppm peroxid *Clostridium difficile* spores Hellickson & Owens, 2007 \[[@B18-ijerph-09-03330]\] 3. Reports on the Presence of Microorganisms on Hospital Textiles ================================================================= [Table 2](#ijerph-09-03330-t002){ref-type="table"} summarizes reports of articles on the presence of microorganisms on hospital textiles. ijerph-09-03330-t002_Table 2 ###### Reports on the presence of microorganisms on hospital textiles. Surviving microorganism Hospital textile Time Reference --------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------ ---------------------------------------- ------------------------------------------------------- Moulds Sheets, pyjamas After use by patients Bureau-Chalot *et al*. 2004 \[[@B3-ijerph-09-03330]\] Coagulase-negative staphylococci, *Bacillus* spp., *Corynebacterium* spp., saprophytic Gram negative bacilli Sheets, pyjamas, uniforms After laundering in hospital laundry Fijan *et al*. 2005 \[[@B6-ijerph-09-03330]\] *Staphylococcus aureus, Clostridium difficile* and vancomycin resistant enterococci Nurses' uniforms After 24 h shift Perry *et al*. 2001 \[[@B19-ijerph-09-03330]\] *Acinetobacter baumannii* Bed linen and curtains After use Hota *et al*. 2004 \[[@B20-ijerph-09-03330]\] MRSA Bed linen and uniforms Coagulase negative Staphylococci, *Corynebacterium* spp., *Micrococcus* spp., *Bacillus* spp., *Enterococcus* spp., saprophytic Gram negative bacilli, moulds Sheets, pyjamas and uniforms After laundering in hospital laundries Fijan *et al*. 2005 \[[@B21-ijerph-09-03330]\] Rotaviral RNA Sheets, pyjamas and uniforms After laundering in hospital laundries Fijan *et al*. 2008 \[[@B22-ijerph-09-03330]\] Parainfluenza virus Hospital gown 4 h after inoculation Brady *et al*. 1990 \[[@B23-ijerph-09-03330]\] Vancomycin resistant enterococci Bed linen 11 weeks after inoculation Hochmuth *et al*. 2005 \[[@B24-ijerph-09-03330]\] The report by Brady \[[@B23-ijerph-09-03330]\] indicates that the parainfluenza virus can survive 4 h on clothing; thus suggesting the need to consider fomites as a possible source of transmission of the virus. In the report by Perry and co-workers \[[@B19-ijerph-09-03330]\] microbiological sampling of nurses' uniforms yielded the detection of *Staphylococcus aureus*, *Clostridium difficile* and vancomycin-resistant enterococci (VRE) before and after the span of duty. The authors recommended provision and frequent changing of nurses' uniforms. The report by Hochmuth and co-workers \[[@B24-ijerph-09-03330]\] noted that VRE strains can survive for 11 weeks on linen and plastic with a 4 log cfu reduction after 7 weeks. They concluded that VRE can survive for prolonged periods on inanimate surfaces that are frequently encountered in a healthcare setting and that the proper disinfection of these surfaces is important in the prevention of nosocomial transmission of VRE. In the report by Bureau-Chalot and co-workers \[[@B3-ijerph-09-03330]\] over 200 samples of hospital linen (sheets, pyjamas) as well as linen rooms and trolleys for transporting linen were collected. The most common found microorganisms were of human origin (coagulase-negative staphylococci) and of environmental origin (*Bacillus* spp., moulds). It was found that clean linen can become a vector for transmission of pathogens or that pathogens present on linen may become airborne during bed-making and may then contaminate surfaces. The report by Hota \[[@B20-ijerph-09-03330]\] also reviews the presence of microorganisms on hospital textiles. In her survey of the literature *Acinetobacter baumannii* was found on bed linen and curtains, as well as other parts of the surrounding inanimate environment \[[@B25-ijerph-09-03330]\], MRSA was found on uniforms worn by health workers and on bed linen \[[@B26-ijerph-09-03330],[@B27-ijerph-09-03330]\]. Other environmental sites that included VRE were gowns worn by patients and health workers \[[@B28-ijerph-09-03330]\]. There are several published articles which show results of investigating the microbial counts of laundered hospital linen using contact plates with RODAC agar and swabbing over period of 5 years between 2004 and 2008 \[[@B6-ijerph-09-03330],[@B21-ijerph-09-03330],[@B22-ijerph-09-03330]\]. The following microorganisms at various occasions were found in hospital laundries on cleaned, folded laundry prepared for reuse: coagulase negative *Staphylococci*, *Corynebacterium* spp., *Micrococcus* spp., *Bacillus* spp., non-fermentative Gram negative bacilli, *Enterococcus* spp., saprophytic Gram negative bacilli, moulds and rotaviral RNA. Although the results seem alarming, after the initial microbiological-sanitary surveillance, all laundries underwent systematic sanitary measures and the results of microbial investigations yielded very low counts on the clean and folded hospital textiles. 4. Reports of Microorganisms from Hospital Textiles as a Possible Source of Infection of Patients ================================================================================================= Reports on hospital textiles as possible source of infection of patients are summarized in [Table 3](#ijerph-09-03330-t003){ref-type="table"}. ijerph-09-03330-t003_Table 3 ###### Reports on hospital textiles as possible source of infection of patients. Microorganism Hospital textile Reference --------------------------------------- --------------------------------------------------- ------------------------------------------------------ *Streptococcus pyogenes* Babies' vests (contamination of dryers) Brunton, 1995 \[[@B8-ijerph-09-03330]\] *Bacillus cereus* Cleaned hospital linen Barrie *et al*. 1994 \[[@B29-ijerph-09-03330]\] Cleaned hospital linen Barrie *et al*. 1992 \[[@B30-ijerph-09-03330]\] Cleaned infants' nappies Birch *et al*. 1981 \[[@B31-ijerph-09-03330]\] Reused towels Dohmae *et al*. 2008 \[[@B32-ijerph-09-03330]\] Towels and bedsheets Sasahara *et al*. 2011 \[[@B33-ijerph-09-03330]\] MRSA Bed linen Creamer & Humphreys, 2008 \[[@B34-ijerph-09-03330]\] Linen Shiomori *et al*. 2002 \[[@B35-ijerph-09-03330]\] *Pseudomonas aeruginosa* Patients' clothes, bed linen Panagea *et al*. 2005 \[[@B36-ijerph-09-03330]\] VRE Drawsheet Bonten *et al*. 1996 \[[@B37-ijerph-09-03330]\] *Staphylococcus aureus* Mattress Ndawula & Brown, 1991 \[[@B38-ijerph-09-03330]\] Antibiotic resistant coliform bacilli Blankets, mattresses Kirby *et al*. 1956 \[[@B39-ijerph-09-03330]\] *Trichophyton interdigitale* Contaminated socks English *et al*. 1967 \[[@B40-ijerph-09-03330]\] An extensive investigation \[[@B8-ijerph-09-03330]\] of what seemed to be a recurring outbreak of streptococcal infection associated with a maternity unit was conducted. On each occasion, extensive environmental and epidemiological investigations were carried out, which indicated that babies were being infected very shortly after birth. The infection team decided to look at the laundering of the vests usually given to new-born children. Investigation of the laundry and in particular the hot air dryers, revealed extensive contamination with the MT type of *Streptococcus pyogenes* involved in the outbreak. After all babies' vests had been autoclaved the outbreaks ceased. An investigation into two cases of post-operative *Bacillus cereus* meningitis \[[@B29-ijerph-09-03330],[@B30-ijerph-09-03330]\] revealed that hospital linen laundered by a batch continuous washing machine was heavily contaminated by *Bacillus cereus* spores. It was found that the linen introduced into the washing machine had a high *Bacillus cereus* spore content and that this was still present after the wash process. In a maternity unit 44% of umbilical swabs from neonates contained an unusual serotype of *Bacillus cereus* \[[@B31-ijerph-09-03330]\]. On further investigation the same serotype could be isolated from air samples, the hands of members of staff and 'clean' nappies from the hospital laundry. It was found that the nappies appeared to be the primary vehicle of *Bacillus cereus* dissemination among the infants. Creamer and Humphreys \[[@B34-ijerph-09-03330]\] emphasized that bed linen can rapidly become heavily contaminated with colonised skin scales and may contribute to the spread of infections. They also stated that precautions such as changing of linen after discharge, using national or standard laundering procedures, storage of clean linen in clean linen storage presses, clean trolleys, *etc*. are sufficient. In their literature survey they found MRSA \[[@B35-ijerph-09-03330]\], *Pseudomonas aeruginosa* \[[@B36-ijerph-09-03330]\], VRE \[[@B37-ijerph-09-03330]\] to have been associated with the spread of pathogens by bed linen as one of the possible environmental routes. Ndawula and Brown \[[@B38-ijerph-09-03330]\] found mattresses were reservoirs of epidemic methicillin-resistant *Staphylococcus aureus*. In a study \[[@B39-ijerph-09-03330]\] of the cause of urinary tract infections the authors were unable to determine the source of the resistant bacteria and the exact mode of infection; the catheters themselves, and the solutions used to irrigate them, could not be incriminated. Blankets, mattresses, and possibly the nasopharyngeal flora of hospital personnel appeared more likely possibilities. Several *Bacillus cereus* nosocomial infections in Japan were investigated \[[@B32-ijerph-09-03330]\] and novel multilocus sequence types were found in patients. After eliminating food-poisoning as a causative agent it was found that the similar strains were found on dried and streamed reused towels and that towels represent an important source of contamination. In the investigation of a *Bacillus cereus* bacteremia outbreak \[[@B33-ijerph-09-03330]\] it was found that hospital linens and the washing machine were highly contaminated with *B. cereus*, which was also isolated from the intravenous fluid of symptomatic patients. All of the contaminated linens were autoclaved, the washing machine was cleaned with a detergent, and improved hand hygiene was promoted among the hospital staff. The number of patients per month that developed new *B. cereus* bacteremia rapidly decreased after implementing these measures. The source of this outbreak was identified as *B. cereus* contamination of hospital linens, and *B. cereus* was being transmitted from the linens to patients via catheter infection. The authors concluded that their findings demonstrated that bacterial contamination of hospital linens can cause nosocomial bacteremia. In a study of a significantly higher incidence of *Trichophyton rubrum* along with a common incidence of *Trichophyton interdigitale* in a long-stay hospital for mentally retarded men it was found that a significant number of crippled patients who had never walked acquired tinea pedis \[[@B40-ijerph-09-03330]\]. The appropriate fungus was isolated before laundering from the worn socks of three patients with *T. interdigitale* infection and of one patient with *T. rubrum* infection. After laundering the fungus was recovered from the socks of one of the patients with *T. interdigitale* infection. In view of the failure of laundering to eliminate the fungus from worn socks, it was suggested that infected socks were the most important route of cross-infection among the crippled patients. All these research publications emphasise that correct laundering procedures of hospital textiles are an important measure for preventing health-acquired infections especially when other more common sources of infections have been ruled out \[[@B8-ijerph-09-03330]\]. 5. Reports on Microorganisms on Textiles as a Cause for Nosocomial Infections of Hospital Workers ================================================================================================= Reports on hospital textiles as possible source of infection of hospital workers are summarized in [Table 4](#ijerph-09-03330-t004){ref-type="table"}. ijerph-09-03330-t004_Table 4 ###### Reports on hospital textiles as possible source of infection of hospital workers. Microorganism Source Employee Reference ----------------------------------------- ---------------------------------- --------------------------------------------- ---------------------------------------------------- *Sarcoptes scabiei* Handling unclean hospital linen Hospital laundry personnel Thomas *et al*. 1987 \[[@B41-ijerph-09-03330]\] *Microsporum canis* Handling contaminated laundry Hospital staff Shah *et al*. 1988 \[[@B42-ijerph-09-03330]\] *Salmonella typhimurium* Handling unclean hospital sheets Hospital laundry personnel Datta & Pridie, 1960 \[[@B43-ijerph-09-03330]\] *Salmonella hadar* Handling unclean hospital linen Hospital laundry personnel Standaert *et al*. 1994 \[[@B44-ijerph-09-03330]\] Hepatitis A virus Handling unclean hospital linen Hospital laundry personnel and nurses' aids Borg & Portelli, 1999 \[[@B45-ijerph-09-03330]\] Keeffe, 2004 \[[@B46-ijerph-09-03330]\] Wilson and co-workers \[[@B47-ijerph-09-03330]\] did not find any evidence to support the hypothesis that uniforms could be a vehicle for the transmission of infections as no studies demonstrated the transfer of microorganisms from uniforms to patients in a clinical situation. They state however that there is an epidemiological link between contaminated clothing and healthcare associated infection when clothing is highly contaminated in an industrial laundry. In an outbreak of scabies among the employees of a hospital laundry \[[@B41-ijerph-09-03330]\] it was found that the most probable cause of this outbreak was transmission via unclean hospital bed linen. During the time compatible with the outbreak a patient with Norwegian scabies was hospitalized. It was concluded that improper handling of the dirty laundry by laundry workers (alleged lack of use of protective gloves) led to the outbreak among the laundry workers. Shah and co-workers \[[@B42-ijerph-09-03330]\] reported of an unusual nosocomial outbreak among staff and patient infected with *Microsporum canis*. It was established that likely modes of subsequent disease transmission from a single infected patient included person-to-person contact and handling of contaminated laundry. In an outbreak of infection with *Salmonella typhimurium* \[[@B43-ijerph-09-03330]\] in a general hospital extensive research was conducted to find the source of transmission. There was no evidence of food poisoning and it was found that several laundry workers who handled sheets from infected persons were excreting *S. typhimurium*. These laundry workers did not have any direct contact with infected patients. Therefore handling dirty laundry was the most likely cause of infection among these laundry workers. The same conclusion was reached by Standaert and co-workers \[[@B44-ijerph-09-03330]\], who investigated an extensive outbreak of salmonella gastroenteritis in a nursing home among residents as well as employees (nurses and laundry personnel). Three laundry personnel who had no contact with residents were infected. Due to the delayed onset of symptoms of these laundry personnel; a secondary transmission was suggested. It was concluded that linen soiled with faeces was the source of nosocomial *Salmonella hadar* infection among the laundry workers. It was found that most of these laundry personnel did not use protective clothing and gloves when handling dirty and infected laundry. The authors stressed the importance of using appropriate precautions when handling dirty linen. Borg and Portelli \[[@B45-ijerph-09-03330]\] investigated laundry personnel and nursing aids in pediatric and infectious disease wards for seropositivity to hepatitis A. It was found that the ratio for seropositivity to hepatitis A between laundry personnel consistently handling dirty linen as compared with colleagues handling only clean items was 16.5. The authors concluded that the increased exposure of hospital laundry workers to potentially infected linen can constitute a risk of occupational hepatitis A for this group of employees. Keeffe \[[@B46-ijerph-09-03330]\] also listed laundry workers as one of the at-risk occupations for a hepatitis A infection. Oliphant and co-workers \[[@B48-ijerph-09-03330]\] also investigated an outbreak of Q fever among laundry workers handling material from a laboratory and found that it was presumably transmitted from contaminated clothing. From all these publications it is obvious that it is necessary to implement infection control practices including proper handling of dirty linen by laundry workers in order to prevent possible health-acquired infections \[[@B49-ijerph-09-03330],[@B50-ijerph-09-03330]\]. Although soiled linen may contain large numbers of pathogenic microorganisms, the risk of actual disease transmission is very low and hygienic and common-sense storage and processing of clean and soiled linen are recommended \[[@B51-ijerph-09-03330]\]. 6. Discussion and Conclusions ============================= Healthcare institutes are obligated to ensure all necessary measures to prevent or limit the spread of healthcare associated infections. One of the possible vehicles of transmission is inanimate fomites such as textiles. When textiles are heavily contaminated with potentially infective body substances, they can contain bacterial loads of 10^6^ to 10^8^ cfu/100 cm^2^ of fabric. However, the incidence of healthcare associated infections transmitted from hospital linen is very low especially when evaluated in the context of the volume of items laundered in healthcare settings (estimated to be five billion pounds annually in the United States) \[[@B5-ijerph-09-03330]\]. It is obvious that the various existing control measures for hospital laundry are effective in reducing the risk of disease transmission to patients and staff. Therefore, use of current control measures should be continued to minimize the contribution of contaminated laundry to the incidence of healthcare associated infections. These control measures are based on principles of hygiene, common sense, and consensus guidance. According to the Slovenian Public Gazette \[[@B52-ijerph-09-03330]\] the correct hygienic management of hospital textiles is achieved by minimal technical requirements regarding sorting, transport and laundering textiles. These requirements according to the Slovenian Expert background and guidelines for management and prevention of health-associated infections include \[[@B53-ijerph-09-03330]\] the following measures: 1. Correct collecting and sorting of contaminated hospital textiles; 2. Correct transporting of contaminated hospital textiles; 3. Correct division of clean and unclean area in laundry; 4. Correct sorting, laundering, drying and ironing of hospital textiles; 5. Correct transport and storage of clean hospital textiles. Very similar conditions are defined in the U.S. Recommendations of C.D.C. and Healthcare Infection control practices Advisory Committee \[[@B5-ijerph-09-03330]\], the USA APIC text of infection control and epidemiology \[[@B54-ijerph-09-03330]\] as well as the standard EN 14065 RABC: Risk Analysis and Biocontamination Control System for textiles in Europe \[[@B55-ijerph-09-03330]\]. These conditions are also implemented in the German Quality and Hygiene assurance for hospital textiles RAL-GZ 992 \[[@B56-ijerph-09-03330]\]. The RABC standard and the RAL-GZ 992/2 quality and hygiene assurance are valid as a system of quality hygiene assurance for hospital textiles in Slovenia. Experience has been shown that proper implementing of any of these chosen Guidelines results in clean and properly disinfected linen in healthcare facilities. The key elements in the laundering process according to the APIC text \[[@B54-ijerph-09-03330]\], supported by the U.S. C.D.C. \[[@B5-ijerph-09-03330]\], include water temperature, type of detergents, disinfectant, rinsing and finishing as well as supplementing the process with common sense and hygienic approaches to collection and transport. According to the C.D.C. guidelines \[[@B5-ijerph-09-03330]\] it is also important to acknowledge that hospital textiles (especially high tough surfaces such as bed linen and pajamas) should not only be appropriately cleaned, but also disinfected and in certain cases (such as surgical drapes and reusable gowns, and in some cases linens in neonatal intensive care units as well as linens in burn therapy units) even sterilization of textiles is necessary \[[@B5-ijerph-09-03330]\]. In the future more research should be conducted in the area of the adherence of microorganisms onto textiles and the likelihood of shedding from the textiles during use thus making them airborne. Another important theme for future research is to study the infectivity of microorganisms after being adhered onto textiles for certain periods of time. This information would give more insight in the transfer of microorganisms from textiles to patients in a clinical situation. Another important future focus is using particles with antimicrobial activity for textile modification in order to enhance antimicrobial properties of medical textiles without adding antimicrobial agents into textiles which can have possible harmful or toxic effects \[[@B57-ijerph-09-03330]\].

Introduction {#sec1-1} ============ Knowledge of internal dental morphology is an extremely important step in planning and administering endodontic therapy.\[[@ref1]\] The numerous anatomical variations existing in the root canal system may contribute to the failure of root canal therapy. Failure to explore and instrument even one of the canals results in improper cleaning of root canal system and can lead to endodontic treatment failure.\[[@ref2]\] The mandibular first molar, the earliest permanent posterior tooth to erupt, seems to be the tooth that most often requires root canal treatment. The usual canal distribution is two canals in the mesial root and one or two canals in the distal root. Baugh D and Wallace J. also described the presence of a middle mesial canal.\[[@ref3]\] Additionally, Stoner *et al*. and Beatty and Iterian\[[@ref4][@ref5]\] have reported on more obscure cases in which a third canal was located in the distal root. Martinez and Bandaneli\[[@ref6]\] showed two cases with six canals. Astonishingly, Reeh\[[@ref7]\] has even reported a case with seven canals, consisting of four canals in the mesial root and three in the distal root. The role of advanced diagnostic tools cannot be overlooked in the diagnosis and management of such a complex canal system. It has been postulated that secondary dentin apposition during tooth maturation would form dentinal vertical partitions inside the root canal cavity, thus creating root canals. A third root canal may also be created inside the root canal cavity of mandibular molars by this process. Such third canals are usually situated centrally between the two main root canals, the buccal and lingual of the mesial and distal roots. The diameter of these third middle canals is usually smaller than that of the other two.\[[@ref6]\] This case report presents the management of a mandibular first molar with six root canals, three in mesial and three in distal root. Additionally, this case report highlights the use of Cone Beam Computed Tomography (CBCT) as a diagnostic tool in Endodontics. Case Report {#sec1-2} =========== A 38 year-old female patient presented with a chief complaint of pain in the mandibular right posterior region for the past two weeks. Clinical examination revealed a carious left mandibular first molar (36). The clinical and radiographic findings led to a diagnosis of chronic irreversible pulpitis, necessitating endodontic therapy. Preoperative radiographic evaluation of the involved tooth pointed towards the presence of more than one canals in the distal root \[[Figure 1](#F1){ref-type="fig"}\]. Anaesthesia of mandibular left first molar was achieved with inferior alveolar nerve block using 2% Lignocaine. The tooth was isolated using a rubber dam and an endodontic access cavity preparation was done. Clinical examination revealed four distinct orifices, two located mesially (mesiobuccal, and mesiolingual) and two distally (distobuccal and distolingual). However, on careful examination of the access cavity with endo-explorer, an additional orifice was detected between the two main canals, both mesially and distally. \[Figures [2a](#F2){ref-type="fig"} and [b](#F3){ref-type="fig"}\] {#F1} {#F2} {#F3} Glide path and patency was achieved using no. 6, 8, and 10 k- files. Working-length radiographs were taken at different angulations with a file placed in each of the three mesial and three distal orifices \[[Figure 3](#F4){ref-type="fig"}\], and confirmed with electronic apex locator (Raypex 5). Cleaning and shaping was performed using a crown down preparation with rotary Protaper instruments (Maillefer, Dentsply, Ballaigues, Switzerland) under profuse irrigation with 3% sodium hypochlorite solution. After drying the root canals with sterile paper points, obturation was carried out with Protaper gutta percha cones (Maillefer, Dentsply, Ballaigues, Switzerland) using zinc oxide eugenol sealer. The access cavity was temporarily restored with Cavit \[Figure [4a](#F5){ref-type="fig"} and [b](#F6){ref-type="fig"}\]. {#F4} {#F5} {#F6} CBCT imaging confirmed the presence of six canals in the concerned tooth. However, the middle mesial and middle distal canals were found confluent with their respective mesial/distal buccal canals at the junction of middle and apical one third, indicating towards the presence of three orifice and two apical foramina in each root \[Figure [5a](#F7){ref-type="fig"} and [b](#F7){ref-type="fig"}\]. {#F7} Discussion {#sec1-3} ========== A thorough knowledge of root canal morphology and canal configuration of the teeth plays an important role in the success of endodontic therapy.\[[@ref8]\] Studies have described the presence of aberrant canals in the mandibular first molar with three canals in the mesial as well as distal roots.\[[@ref3][@ref9][@ref10][@ref11]\] The third mesial and distal canal is defined as being independent when a distinct coronal orifice and apical foramen are observed, or confluent when converging into one of the other two main canals and terminating at a common apical foramen.\[[@ref7]\] Many authors have agreed on the presence of three foramina in the mesial root; however, only a few reported the presence of three independent canals, which presents itself as a rare anatomical variant. In a study of 760 mandibular molars, Fabra *et al*.\[[@ref12]\] found that 20 molars (2.6%) had three canals in the mesial root. Endodontic success in teeth with the aforementioned number of canals requires a careful clinical and radiographic inspection. Diagnostic aids such as CBCT, Dentascan, multiple preoperative radiographs, examination of the pulp chamber floor with a sharp explorer, troughing of the grooves with ultrasonic tips, staining the chamber floor with 1% methylene blue dye, performing the sodium hypochlorite "champagne bubble test," and visualizing canal bleeding points are all important aids in locating the root canal orifices.\[[@ref13]\] The search for an extra orifice is also aided by the use of microscopes, magnifying loupes and fiber-optic trans-illumination to locate the developmental line between the mesiobuccal and mesiolingual orifices.\[[@ref14]\] A significant constraint in conventional radiography is that it produces a 2D image of a 3D object, resulting in the superimposition of the overlying structures. Therefore, such radiographs are of limited value in cases with complex root canal anatomy. Interpretation and appraisal based on a 2D radiograph may alert the clinician to the presence of aberrant anatomy; however, may not be able to present the variable and complex morphological structure of the root canals and their interrelations. Hence, CBCT has been specifically designed to produce undistorted three dimensional non invasive information of the root canal anatomy. Gopikrishna *et al*.\[[@ref15]\] used spiral computerized tomography for the confirmatory diagnosis of a morphological aberration in the maxillary first molar. Matherne *et al*. (2008)\[[@ref16]\] conducted an ex vivo investigation to compare a Charge Coupled Device Photostimuable Phosphor Plates (CCDPSP) digital radiography system with CBCT to detect the number of root canals in 72 extracted teeth. They found that with digital radiography, endodontists fail to identify at least one root canal in 40% of teeth. Treating extra canals may be challenging; however, the inability to find and properly treat the root canals may cause failures. With advance diagnostic aids like CBCT and Dentascan, these challenges can be overcome. Although the incidence of root and canal variations is rare, every effort should be made to find and treat all the root canals for successful clinical results. **Source of Support:** Nil. **Conflict of Interest:** None declared.

Sir, A forty-nine year-old male underwent live related renal transplantation thirteen years ago for end stage renal disease secondary to bilateral vesicoureteric reflux. He was on prednisolone 10 mg per day and azathioprine 125 mg per day for maintenance immunosuppression. He presented with pain in his abdomen and vomiting that had been prevalent since the past day. Evaluation was suggestive of hollow viscous perforation which was confirmed by computed tomography of the abdomen. As laparotomy revealed a perforation with growth in the jejunum, a jejunal segment resection was done followed by jejunojejunal anastamosis. Histopathological investigation revealed that the growth was a diffuse, large B cell lymphoma \[[Figure 1](#F0001){ref-type="fig"}\]. Immunohistochemistry confirmed it to be a B cell variant of diffuse large cell NonHodgkin\'s lymphoma of the small bowel. Viral serology showed significant Epstein barr virus (viral capsid antigen) IgM and Epstein barr virus (nuclear antigen) IgG levels being normal. The patient was maintained on CHOP (Cyclophosphamide, Adriamycin, Vincristine, and Prednisolone) regimen by a medical oncologist. {#F0001} Renal transplant recipients carry a three to five fold higher risk of malignancy as compared to healthy people. Posttransplant lymphoproliferative disease (PTLD) is second only to skin cancers in contributing to 12% of posttransplant malignancies and its incidence increases with time.\[[@CIT1]\] The risk of developing PTLD is highest within the first year after the transplant, and the probability decreases thereafter.\[[@CIT2]\] The incidence increases with the age of the patient and the duration and cumulative dose of immunosuppressive therapy. It is associated with high morbidity and mortality. Cumulative doses and longer duration of systemic glucocorticoids and azathioprine increase the incidence of nonHodgkin\'s lymphoma (NHL).\[[@CIT3]\] The single most important risk factor for developing PTLD is EBV infection. This association is well established and 80--90% of PTLD cases are associated with primary EBV infection or reactivation of previously acquired EBV.\[[@CIT4]\] Most cases of PTLD occur as nodal diseases, but they can present with localized symptoms involving organs such as the gastrointestinal tract, lungs, skin, liver, central nervous system, and infiltrative lesions in the allograft. Extranodal involvement occurs in more than 50% of the cases. The gastrointestinal tract is predominantly involved with an increased propensity for ulceration and perforation.\[[@CIT5]\] In our patient, the points of interest were: a) presentation as perforation, b) the neoplastic growth resected turning out as a diffuse B cell variant of nonHodgkin\'s lymphoma with positive Epstein barr viral serology.

Lead is one of the most useful elements in industry, but serves no useful function in the human body. Environmental and industrial lead exposures continue to pose major public health problems in the exposed population.\[[@CIT1]\] Over the years, it has become increasingly evident that low-level lead exposure resulting in blood lead levels between 10 and 15 μg/dL can lead to deleterious effects like cognitive impairment and behavioral deficits, high blood pressure (BP) and impaired renal function.\[[@CIT2][@CIT3]\] Lancereaux\[[@CIT4]\] provided the first description of kidney disease and interstitial nephritis by postmortem examination of a lead-poisoned artist. It was not until the late 1920s when an epidemic of chronic nephritis in Queensland, Australia, was linked to childhood lead poisoning that the full spectrum of lead-induced nephropathy became apparent.\[[@CIT5][@CIT6]\] This was followed by cases of renal diseases from the US in individuals consuming lead-contaminated illegally distilled moonshine whisky.\[[@CIT7]\] ENVIRONMENTAL LEAD EXPOSURE {#sec1-1} =========================== Environmental lead exposure continues to be a public health problem. In the past, lead-based paint was a major source of lead poisoning among children. The painted surfaces of old houses contained significant amounts of lead. Direct ingestion of lead paint, lead-contaminated house dust and water by children has been identified as a major contributor to lead poisoning among the children. Many studies have confirmed that lead-contaminated dust is a major determinant of lead concentrations in blood.\[[@CIT8]\] Similarly, a highly significant correlation between lead concentration in drinking water and blood lead concentrations has been reported.\[[@CIT9]\] Children are more susceptible to the effects of environmental lead than adults because of the increased gastrointestinal absorption of lead in children. Children are more vulnerable because they absorb lead 5--10 times more effectively than adults and have a greater exposure because of their exploratory behavior and frequent hand to mouth activity.\[[@CIT10]\] Adults at the highest risk are those exposed to lead fumes or dust in the industry.\[[@CIT11][@CIT12]\] OCCUPATIONAL LEAD NEPHROPATHY {#sec1-2} ============================= An association between lead poisoning and renal diseases in humans has been recognized and documented by several studies.\[[@CIT8][@CIT12][@CIT13]\] Elemental lead and inorganic lead compounds are absorbed by ingestion or inhalation, but organic lead compounds, e.g. tetraethyl lead, may also be absorbed by skin contact. Organic lead compounds are the most toxic. Absorption of lead from the lungs is very efficient, especially if the particles are less than 1 μm in diameter. The gastrointestinal absorption of lead varies with the age of the individual; children absorb around 50% of what they ingest but adults only absorb 10--20% of what they ingest. Lead is very similar to calcium chemically. Thus, once in the body, it is handled as if it were calcium. Lead serves no useful purpose in the human body and its presence in the body can lead to toxic effects, regardless of the exposure pathway. The kidney is the critical organ after long-term occupational or environmental exposure to lead. Excessive exposure to lead may cause acute or chronic nephrotoxic effects. Two types of nephropathy, acute and chronic nephropathy, have been observed in humans. Acute Pb nephropathy is characterized functionally by a generalized deficit of tubular transport mechanisms (Fanconi syndrome) and morphologically by the appearance of degenerative changes in the tubular epithelium and the nuclear inclusion bodies containing Pb protein complexes.\[[@CIT15][@CIT16]\] These effects, which are usually reversible with chelation therapy, have been reported mainly in children manifested by glycosuria and aminoaciduria. Chronic occupational exposure to lead has also been linked to a high incidence of renal dysfunction, which is characterized by glomerular and tubulointerstitial changes, resulting in chronic renal failure, hypertension and hyperuricemia. Chronic lead nephropathy is an irreversible renal disease that develops over months or years of excessive exposure.\[[@CIT17][@CIT18]\] This has been reported in adults who had ingested leaded paint during childhood and those who consumed illicitly distilled alcohol (moonshine whisky).\[[@CIT14][@CIT15]\] In the chronically exposed adults, Pb nephropathy occurs as a progressive tubulointerstitial nephritis that is difficult to diagnose at the early stage. Incipient Pb nephropathy is not associated with urine abnormalities easily detected by dipsticks. The tests evaluating the glomerular filtration rate (GFR) (creatinine clearance, blood urea nitrogen, serum creatinine) are the only ones that can be used to detect the renal effect caused by the occupational exposure to Pb.\[[@CIT14][@CIT19][@CIT20]\] But, when these tests are abnormal, the nephropathy has already reached the irreversible phase that may lead to renal sufficiency.\[[@CIT19]\] Chronic low-level exposure to lead is also associated with an increased urinary excretion of low molecular weight proteins and lysosomal enzymes.\[[@CIT21]\] Epidemiologic studies have shown an association between blood lead levels and BP, and hypertension is a cardinal feature of lead nephropathy.\[[@CIT13][@CIT20][@CIT22][@CIT24]\] Most lead-associated renal effects or toxicity are a result of the ongoing chronic or current high acute exposure. They can also be attributable to a previous chronic lead exposure. The lowest level at which Pb has an adverse effect on the kidney remains unknown. Both glomerular and tubular effects have been reported.\[[@CIT20]\] The glomerular effects range from an increased prevalence of high molecular weight proteinuria to a nephrotic syndrome.\[[@CIT18][@CIT23]\] The reported tubular changes consist of an enhanced urinary excretion of enzymes. BIOMARKERS OF NEPHROTOXICITY {#sec1-3} ============================ The prevention of renal diseases induced by exposure to industrial or environmental Pb largely relies on the capability to detect nephrotoxic effects at a stage when they are still reversible or at least not yet compromising the renal function. During the past decade, various tests have been proposed for the early detection of the toxic effects at different sites on the nephron. Some of these tests have been validated and some need epidemiological validation. Currently, there are some early and sensitive indicators available that are considered predictive or indicative of renal toxicity from lead exposure. Recent studies have shown more than 20 potential markers of renal effects that can be arbitrarily classified into three broad categories.\[[@CIT25]--[@CIT28]\] [Table 1](#T0001){ref-type="table"} shows the different biomarkers used in Pb-induced nephrotoxicity. ###### Showing the different biomarkers used in Pb-induced nephrotoxicity. No. Reference Exposure Exposure duration (years) B Pb conc. (μg/dL) Biomarkers used ----- -------------------------------- --------------- --------------------------- -------------------- ------------------------ 1 Lilis *et al*.\[[@CIT25]\] Occupational \> 10 79 GFR, SCr 2 Cramer *et al*.\[[@CIT26]\] Occupational 9 103 GFR, HP 3 Wedeen *et al*.\[[@CIT27]\] Occupational 5 48 GFR, HP, TMPAH 4 Wedeen *et al*.\[[@CIT28]\] Occupational 14 51 GFR, HP 5 Hong *et al*.\[[@CIT29]\] Occupational 7 68 GFR, TMG 6 Lilis *et al*.\[[@CIT30]\] Occupational 12 80 SCr, BUN 7 Dekort *et al*.\[[@CIT31]\] Occupational 12 47 SCr, BUN 8 Verschoor *et al*.\[[@CIT33]\] Occupational \< 2--10 47 UNAG, URBP 9 Staeseen *et al*.\[[@CIT34]\] Environmental NA 10 SCr 10 Omae *et al*.\[[@CIT35]\] Occupational 0.1--26 37 CCr, CUA, Uβ2μG, Cβ2μG 11 Hu\[[@CIT36]\] Environmental NA 6 Ccr 12 Kim *et al*.\[[@CIT37]\] NA NA 10 Scr 13 Fels *et al*.\[[@CIT38]\] Environmental NA 13 SCr, UE, UP, ULMWP 14 Hsiao *et al*.\[[@CIT39]\] Occupational 13 40 SCr 15 Sonmez *et al*.\[[@CIT40]\] Occupational 0.14 25 UNAG 16 Muntner *et al*.\[[@CIT41]\] NA NA 5 SCr B Pb conc., blood lead concentration; NA, not available; GFR, glomerular filtration rate; SCr, serum creatinine; HP, histopathology; TMPAH, transport max. for PAH; TMG, transport max for glucose; BUN, blood urea nitrogen; UNAG, urine N-acetyl-β-D-glucosaminidase; URPB, urine retinal binding protein; CCr, creatinine clearance; CUA, uric acid clearance; Uβ2μG, urine β2μG; UP, urine proteins; ULMWP, urinary low molecular weight proteins. Functional markers {#sec2-1} ------------------ Creatinine in serum (crt-S).Creatinine in urine (crt-U).Urinary proteins of low or high molecular weight.3.1.High molecular weight proteins -- albumin, transferrin, immunoglobulin G.3.2.Low molecular weight proteins -- retinol-binding protein (URBP), β~2~ -microglobulin (β~2~ -m), urinary α-1-microglobin (Uα1m).Urinary enzymes -- N-acetylglucosaminidase (NAG).Alkaline phosphatase (ALP).γ-glutamyl transferase (γ-GT). These are the biological markers of tubular damage, which are characterized by enhanced urinary excretion of α-1-microglobulin, β~2-~ m, RBP, NAG, APL and γ-GT. Cytotoxicity markers {#sec2-2} -------------------- These include: The brush border tubular antigens (BBA, BB~50~ and HF~5~).Enzymes -- B-galactosidase. Exposed Pb workers show an increased leakage of tubular antigens and several enzymes as a sign of renal toxicity. This is in all likelihood a reflection of the damage to the proximal tubular cells. Biochemical markers {#sec2-3} ------------------- These include: Eicosanoids -- 6-keto-prostaglandin F~1~α (6-keto-PGF~1~α), prostaglandin F~2~α (PGF~2~α), prostaglandin E~2~(PGE~2~).Thromboxane (TxB~2~).Fibronactin.Urinary sialic acid activity, sialic acid in plasma or in RBCs.Urinary kallikrein activity.Urinary glycosaminoglycans (GAG).Intestinal alkaline phosphate (IAP). The most outstanding effect found in the workers exposed to Pb is an interference with the renal synthesis of eicosanoids, resulting in a lower urinary excretion of 6-keto-PGFα and an enhanced excretion of TXB~2~. It is generally accepted that the urinary 6-keto-PGF~1~α and TXB~2~ primarily reflect the glomerular synthesis of prostacyclin and TXA~2~, whereas urinary PGE~2~ and PGF~2~α are largely contributed by the renal medulla. The decrease in PGE~2~, PGF~2~α and enhanced excretion of TXB~2~ resulting from biochemical or cytotoxic effects in the medulla and glomeruli represent the earliest renal changes associated with exposure to Pb.\[[@CIT29][@CIT30]\] The early effects on urinary excretion of 6-keto-PGFα and TXB~2~ suggest that the initial insult in Pb nephropathy might also involve the vasculature and glomeruli and is not exclusively localized in the tubulointerstitial compartment. The changes in the renal synthesis of eicosanoids raises the question of their relevance to health and are indicative of the degenerative process that may lead to a loss of renal function.\[[@CIT32]\] Together with the changes in the urinary excretion of eicosanoids, the increased excretion of Tamm--Horsfall glycoprotein (THG) appears as an early renal effect induced by exposure to Pb. This increase could reflect an injury to the epithelial cells of the ascending limb of the Loop of Henle and the most proximal part of the distal convoluted tubules where this glycoprotein is localized. The physiological function of THG is still obverse. It might have several important functions, such as rendering the ascending limb of Henley\'s loop impermeable to water, transport to sodium, defense against infection or the immuneregulation of several cytokines.\[[@CIT22][@CIT32]\] Urinary kallikrein is a serine proteases synthesized by the distal tubular cells, which might serve as an index of distal nephrotoxicity. As most of the kallikreine is associated with the membranes that face the urinary compartment, an increased urinary excretion of kallikrein could result from toxic damage in the distal tubular cells. The increased urinary excretion of sialic acids appears as a rather early effect of exposure to lead. The GAGs are polysacchrides composed of repetitive disaccharide units. They are found in the glomeruli and the tubules and their leakage into the urine has been suggested to be a marker of injury to the nephron. An increased excretion of GAG has also been suggested to be an indicator of damage to the renal papilla, which is rich in GAG.\[[@CIT24]\] GLOMERULAR FILTRATION RATE {#sec1-4} ========================== Creatinine clearance, blood urea nitrogen (BUN) and serum creatinine are some of the parameters that can be used to detect the renal effects caused by occupational exposure to Pb. But, when these tests are found abnormal, the nephropathy has already reached the irreversible phase that may lead to renal insufficiency.\[[@CIT17]\] The renal effects of Pb, consisting mainly in a decline of the GFR without proteinuria, have been reported in workers with a longstanding exposure to Pb, with a Pb-B of 600 μg/L or more.\[[@CIT20]\] So far, studies conducted on populations of workers with a lower level of exposure to Pb have disclosed no renal effect or only infraclinical changes of marginal significance.\[[@CIT12][@CIT13]\] In humans, a reduced GFR (i.e. indicated by decreases in the creatinine clearance or increases in the serum creatinine concentration) has been observed in association with exposures resulting in average PbBs \< 20ug/dL. However, some studies have shown an increased GFR with Pb exposure. This may represent hyperfiltration, which may contribute to adverse renal effects. Decrements in GFR may contribute to an elevation in the BP, and an elevated BP may predispose people to glomerular disease. These effects may be mechanistically related and, furthermore, can be confounders and covariables in epidemiological studies.\[[@CIT28]--[@CIT30]\]

1. Introduction {#sec1-jcm-08-01780} =============== While numerous guidelines for the management of mild traumatic brain injury (TBI) exist, there is still controversy regarding the treatment of head trauma patients with antithrombotic therapy (ATT). The management of mild TBI on ATT is complicated by the heterogeneity of patients with different medications and varying patient characteristics. Most authors define mild TBI based on a Glasgow Coma Scale (GCS) of 13 to 15, others include any impact to the brain, not necessarily causing symptoms \[[@B1-jcm-08-01780]\]. For patients using ATT, several studies show an increased risk for abnormal computed tomography (CT) findings, even with a normal neurological exam and a history lacking neurological symptoms \[[@B2-jcm-08-01780],[@B3-jcm-08-01780],[@B4-jcm-08-01780]\]. Therefore, in most centers patients on ATT receive a routine CT at presentation, even if common definitions of mild TBI are not fulfilled and head trauma is merely reported, or visible signs of head trauma are present. Vitamin K antagonists (VKA) were shown to increase the risk for clinically significant TBI and mortality \[[@B5-jcm-08-01780],[@B6-jcm-08-01780],[@B7-jcm-08-01780]\] and numerous studies indicate an increased risk and mortality for patients on all other kinds of ATT \[[@B8-jcm-08-01780],[@B9-jcm-08-01780],[@B10-jcm-08-01780],[@B11-jcm-08-01780],[@B12-jcm-08-01780]\]. Delayed traumatic intracranial hemorrhage (DIH) can occur up to several weeks after trauma to the head \[[@B13-jcm-08-01780]\] and was reported to occur more frequently in patients with ATT, ranging from 0.2% to 6% \[[@B14-jcm-08-01780],[@B15-jcm-08-01780],[@B16-jcm-08-01780],[@B17-jcm-08-01780]\]. Due to studies showing a high number of DIH, international guidelines and recommendations in literature suggest admitting patients with ATT for observation after negative CT \[[@B18-jcm-08-01780],[@B19-jcm-08-01780],[@B20-jcm-08-01780],[@B21-jcm-08-01780]\]. Many trauma centers adopted management protocols highly cautious of DIH and performed repeat CT after head trauma for asymptomatic patients. These extensive management protocols with in-hospital observation and repeat CT were evaluated in numerous studies, and most authors concluded that a routine repeat CT is not necessary \[[@B22-jcm-08-01780],[@B23-jcm-08-01780]\]. Some authors even question the necessity for clinical observation after negative CT \[[@B18-jcm-08-01780],[@B24-jcm-08-01780],[@B25-jcm-08-01780]\]. 2. Materials and Methods {#sec2-jcm-08-01780} ======================== Our level I trauma center follows a high level of precaution for head injury patients. CT are performed based on the Canadian CT Head Rule, which, however, excludes patients with ATT \[[@B26-jcm-08-01780]\]. At the time of data collection, all cases of head trauma with ongoing ATT regardless of clinical signs for TBI received a CT and were admitted for a minimum of 24 h of in-hospital observation. Patients receiving VKA, Clopidogrel or direct oral anticoagulants (DOAC) received a routine repeat CT before discharge to detect delayed hemorrhages. Patients using acetylsalicylic acid (ASA) or low molecular weight heparin (LMWH) underwent clinical observation but did not receive a routine repeat CT. After two years of said management, we analyzed our clinical protocol to determine the frequency of delayed intracranial hemorrhage in patients with head trauma and antithrombotic therapy, adjusting our practice and thereby contributing to the ongoing international debate on the management of head trauma patients on ATT. The study was performed in a level I trauma center with authorization by the local Institutional Review Board (1632/2014). Between January 2012 and April 2014 patients aged 18 years or older were retrospectively included if they were admitted for observation after blunt head trauma with ongoing ATT and no pathologies in their initial CT. Management of these patients followed the described standard clinical protocol. This included an initial CT, clinical and GCS assessment including history of unconsciousness and laboratory tests including S100 and coagulation studies at time of admission. We did not routinely perform laboratory tests for evaluation of therapeutic levels during the observation period (viscoelastic tests, platelet function or anti-Xa assays) for ATT other than VKA. In-hospital observation for a minimum of 24 h followed, and patients received their applicable protocols:Patients using ATT with an expected higher risk for DIH based on the literature, such as vitamin K antagonists (VKA), direct oral anticoagulants (DOAC) and Clopidogrel received a routine repeat CT prior to discharge from hospital.Patients using ATT with an expected low risk for DIH including acetylsalicylic acid (ASA) and prophylactic doses of low molecular weight heparin (LMWH) did not receive a routine repeat CT and were discharged after observation only. Due to the greater number of patients receiving ASA compared to other ATT we included only patients from January 2013 until December 2013 in this study. The primary endpoint of this study was the occurrence of delayed intracranial hemorrhages. Data was collected and analyzed using SPSS version 24 and descriptive statistics were performed. Due to the low number of delayed intracranial hemorrhages the variables age, GCS and prothrombin time were tested using the Mann--Whitney--U test while the remaining nominal variables were tested using Fisher's exact test. The significance level alpha for all implemented tests was set to α \< 0.05. 3. Results {#sec3-jcm-08-01780} ========== During the study period 793 patients fulfilled the inclusion criteria, with a majority of 453 (57.1%) women and 340 (42.9%) men. A routine repeat CT was performed in 395 cases and in-hospital observation without routine repeat CT in 398 patients. The average patient age at presentation was 81 years (range 32--102). The most prevalent ATT was acetylsalicylic acid in 368 patients (46.4%), followed by vitamin K antagonists in 255 (32.2%) and Clopidogrel in 86 patients (10.8%), see [Table 1](#jcm-08-01780-t001){ref-type="table"}. Since patients using ASA were included from only one year whereas all other patients were collected from a two-year observation period, the distribution of different types of ATT is not representative of the total population at our institution. Only blunt trauma was included in the study, with low energy trauma due to falls accounting for 95.2% of all cases. Most patients presented with a normal neurological status, only 16.5% of patients showed any neurological symptoms at presentation. The average GCS at presentation at the hospital was 15 and in 75.5% there was no history of unconsciousness or amnesia reported by either the patients or others. Lesions to the head such as abrasions and lacerations were present in 57.4%. The mean prothrombin time of the 255 patients using VKA was 32%, and 91.8% of the VKA patients were in their therapeutic range at time of admission. The timing of the routine repeat CT as well as discharge from hospital followed the clinical protocol and occurred after a minimum observation time of 24 h, which resulted in an average of two nights of in-hospital observation, depending on the time of admission, with discharge after morning rounds. In total, there were 11 cases (1.2%) of delayed intracranial hemorrhages. The routine repeat CT group showed nine DIH, resulting in 2.3% of cases detected with routine repeat CT. In the observation-only group, 16 patients showed a worsening of GCS or other symptoms indicating TBI. Two of these repeat CTs based on clinical judgement found minor DIH (0.5% of the observation group). One of the 11 patients with DIH needed an urgent decompressive craniectomy due to subdural hematoma with midline shift on the second day of observation. This was an 84-year-old female with vitamin K antagonist therapy, who showed no clinical signs of traumatic brain injury, no exterior injury to the head and an international normalized ratio of 2.9 at admission. The repeat CT was performed due to neurological deterioration with reduced Glasgow Coma Scale 27 h after admission. The patient underwent immediate decompressive craniectomy and was consecutively discharged to a neurological rehabilitation facility with a mild left-sided hemiparesis. The other patients with small delayed intracranial hemorrhage did not undergo surgical intervention and were discharged from hospital after an average observation period of 12 days (range 5--23). A review of the cases of delayed intracranial hemorrhages in the repeat CT group by a radiologist revealed, that small epi- and subdural hematomas, minimal intracerebral and subarachnoid hematomas were visible but not described in the initial CT report in four of the eleven cases. Excluding the four cases of initially undiagnosed pathologies in the CT report, an adjusted number of seven DIH (1.8%) were found in the repeat CT group, and in 0.9% overall. There were no significant differences between patients with or without delayed intracranial hemorrhage regarding age, sex, mechanism of injury, extent of external head injury, S100 level and coagulation studies or neurological status at admission. Characteristics of patients with DIH in our study population can be seen in [Table 2](#jcm-08-01780-t002){ref-type="table"}. Four patients died during the in-hospital observation, all due to non-TBI-related causes such as pneumonia or heart failure. 4. Discussion {#sec4-jcm-08-01780} ============= No feasible diagnostic or observation protocol will be able to exclude all cases of fatal delayed bleeding as DIH can occur up to several weeks after head trauma. Most clinically significant DIH will be detected within an observation period of 24 h, but there will always be cases that occur unusually late or in surprisingly neurologically intact patients \[[@B14-jcm-08-01780],[@B15-jcm-08-01780],[@B16-jcm-08-01780],[@B17-jcm-08-01780],[@B19-jcm-08-01780],[@B20-jcm-08-01780],[@B21-jcm-08-01780]\]. The results of our study, with only one clinically significant case of delayed intracranial hemorrhage (0.3% in the repeat CT group, 0.1% in total), support other investigations which concluded that a routine repeat CT is not necessary for patients with antithrombotic therapy, due to the low clinical significance of most detected DIH \[[@B27-jcm-08-01780],[@B28-jcm-08-01780],[@B29-jcm-08-01780],[@B30-jcm-08-01780],[@B31-jcm-08-01780],[@B32-jcm-08-01780]\]. The single case of a clinically significant DIH in our study showed an altered neurological status and would therefore have received an additional CT during the observation period regardless of clinical protocol for routine repeat CT. In our study population the routine repeat CT was effective in detecting DIH, but was superior to clinical observation only for the detection of cases of little clinical significance and devoid of therapeutic consequence. Based on our results and on the growing international consensus eliminating repeat CT, we have changed our clinical protocol to a 24-h observation period after the initial negative CT, and discharge from hospital without repeat CT. Due to the low number of DIH in the observation group using ASA and/or LMWH with 0.5%, we no longer admit these patients for in-hospital observation after a negative CT. Since we only included patients using ASA from one year compared to a two-year observation period for all other ATT, the total impact of not routinely admitting these patients after a negative CT is even greater, uses less resources and can improve patient satisfaction. Most studies addressing the management of head injury patients on ATT do not explicitly discuss the degree of head injury. Most patients in our study would have not received a CT based on the commonly applied Canadian CT Head Rule had they not been treated with ATT, as the majority of patients had a normal neurological exam at presentation and only 57.4% of patients in our study had visible signs of a head injury. Because of the potential higher risk for intracranial pathologies in patients with all types of ATT we support the recommendation of performing an initial CT as suggested by the Scandinavian Guidelines \[[@B18-jcm-08-01780]\] and the recent Austrian consensus statement \[[@B33-jcm-08-01780]\]. However, the increasing number of patients with ATT in a time of limited resources raises questions about a clear definition of head trauma to determine risk factors for DIH. Whether a reported minor impact to the head without visible injuries justifies the cost and patient inconvenience of enduring a CT and in-hospital observation cannot be answered by our study and remains a decision based on clinical judgement. Despite all guidelines, it is still necessary to make decisions based on risk stratification while considering the consequences for the individual patient. Chenoweth et al. \[[@B28-jcm-08-01780]\] found a 0.3% rate of DIH in their prospective study including patients with and without ATT, and concluded that "this (the low risk of DIH and fact that they can occur later than 24 h) highlights the importance of clinical judgment regarding the severity of trauma, additional injuries, and ability to monitor the patient for deterioration when making decisions about admission for older patients after blunt head trauma." However, our patient requiring surgical intervention, while only representing 0.1% of our study population, initially showed no clinical signs of a massive trauma to the head and her only risk factors for significant TBI were her age and her vitamin K therapy. This patient could have died at home if she were discharged without observation by relatives or nurses after the initial CT. Clearly, the admission of patients must be indicated based on economic factors as well as on aspects of the patient's will and ethical considerations for elderly patients, who might experience massive stress due to the changed environment in the hospital \[[@B34-jcm-08-01780]\]. A relevant but unanswered question underlying this study is, how does the fear of legal consequences affect clinical management, although a potentially fatal DIH may not lead to an escalation of care due to advanced patient age? If a patient admitted for observation were to incur a DIH, would the same patient be operated on if he or she was 95 years old, not living independently and suffering from dementia? The average age of patients in our study was 81 years. Unfortunately, data on dementia, degree of dependency or existence of a living will was not available for our retrospective study design. In our study some 91.8% of patients using VKA were in a therapeutic range and we did not perform further coagulation assays specific to other ATT during the observation period. Assessing additional therapeutic effects of ATT at time of presentation could possibly reduce admissions and save resources, while informing patients of their insufficient hypocoagulative protection. Limitations of the study are explained by the retrospective study design and lack of data about DIH after discharge. Furthermore, the number of patients using DOAC was relatively low at the time of data collection but is now a controversial topic in literature. Since there may have been cases of undetected DIH in the group without routine repeat CT and the low number of patients using DOAC this study cannot address the relative risk of various ATT for DIH. The large number of patients and relative homogenous population constitute a significant strength of our study, but multicenter prospective trials are needed to further investigate this matter. Consistent with current studies, our results indicate that routine repeat CT seem to be no more effective than in-hospital observation, due to the fact that clinically significant DIH reveal themselves by neurological deterioration. Furthermore, we conclude that in-hospital observation for head trauma patients using acetylsalicylic acid is not necessary due to the rarity of clinically relevant DIH. But neither our study nor current literature can answer the ethical questions behind the data. They must be answered by individual centers and countries based on existing resources and their cultural environment. We thank Daniel Toth for reviewing the CT images of delayed intracranial hemorrhage cases, Claudia Gahleitner for support with statistics and Leonard Höchtl-Lee for his contribution as native English proof reader. Conceptualization, S.H., E.S., A.A.; methodology, S.H., E.S., A.A.; software, A.A.; validation, A.A.; formal analysis, A.A.; investigation, A.A., H.B., M.S., P.D.; resources, A.A.; data curation, A.A., H.B., M.S., P.D.; writing---original draft preparation, A.A.; writing---review and editing, S.H., E.S., H.B., M.S., P.D.; visualization, A.A.; supervision, S.H.; project administration, A.A. This research received no external funding. The authors declare no conflict of interest. jcm-08-01780-t001_Table 1 ###### Type and frequency of antithrombotic therapy in total and in the subgroups of the study. ------------------------------------------------------------------------------------------------------------------ Antithrombotic Therapy Total\ Repeat CT Group\ Observation Group Only\ n (%) n (% of Group) n (% of Group) ------------------------------------------------------- ------------- ------------------ ------------------------- Acetylsalicylic acid 368 (46.4%) 0 368 (92.5%) Vitamin K antagonists 255 (32.2%) 255 (64.6%) 0 Clopidogrel 86 (10.8%) 86 (21.8%) 0 Clopidogrel and acetylsalicylic acid 22 (2.8%) 22 (5.6%) 0 DOAC (dabigatran and rivaroxaban) 32 (4.0%) 32 (8.1%) 0 Low molecular weight heparin 26 (3.3%) 0 26 (6.5%) Low molecular weight heparin and acetylsalicylic acid 4 (0.5%) 0 4 (1.0%) ------------------------------------------------------------------------------------------------------------------ jcm-08-01780-t002_Table 2 ###### Characteristics of patients with delayed intracranial hemorrhage in our study population including sex and age, antithrombotic therapy, reported unconsciousness and amnesia, prothrombin time at admission, neurological symptoms during in-hospital observation, type of delayed intracranial hemorrhage (EDH = epidural hematoma, SDH = subdural hematoma, SAH = subarachnoid hematoma, ICH = intracerebral hematoma) and necessity of surgery. ASA = acetylsalicylic acid. Patient Antithrombotic Therapy Unconsciousness Amnesia Glasgow Coma Scale Head Wound Prothrombin Time Neurological Symptoms Delayed Intracranial Hemorrhage Surgery --------- ------------------------ ----------------- --------- -------------------- ------------ ------------------ -------------------------------------------------------------------------- --------------------------------- --------- m, 72y Vitamin K antagonist yes yes 15 yes 56 no EDH, SDH no f, 93y Vitamin K antagonist no no 15 yes 47 no SAH no f, 83y Vitamin K antagonist no no 15 yes 20 no SAH no f, 82y Vitamin K antagonist no no 15 yes 48 no SAH no m, 92y Vitamin K antagonist no no 15 yes 33 no SAH no f, 84y Vitamin K antagonist no no 15 no 27 yes SDH yes m, 90y Clopidogrel + ASA no no 15 yes 89 no Hygroma no m, 54y Clopidogrel + ASA no yes 15 yes 105 no ICH no m, 89y Dabigatran no no 15 yes 89 no SDH no f, 82y ASA no no 15 yes 104 no (repeat CT was recommended by radiologist after review of initial CT) EDH no f, 79 ASA yes yes 14 yes 88 yes EDH no

1. Introduction {#sec1-biomolecules-10-00575} =============== Aflatoxin is a type of secondary metabolite produced mainly by microscopic fungal species *Aspergillus flavus* and *Aspergillus parasiticus* in the environment of high temperature and humidity (temperature 25--30 °C, moisture \> 15%) \[[@B1-biomolecules-10-00575]\]. According to the International Agency for Research on Cancer (IARC) \[[@B2-biomolecules-10-00575]\], aflatoxins have been classified as a grade I carcinogenic substance. Among them, aflatoxin B~1~ (AFB~1~) is the most toxic, with strongest carcinogenicity; it contaminates more than 100 kinds of foods such as grain, oils, milk, condiments, nuts, tea and dairy products \[[@B3-biomolecules-10-00575],[@B4-biomolecules-10-00575]\]. Since AFB~1~-caused food contamination comprises about 75% out of total mycotoxin contaminations \[[@B5-biomolecules-10-00575]\], maximum residue limits (MRLs) for AFB~1~ in grains have been set (from 2 to 20 μg kg^−1^) in many countries, including the European Union (EU), the United States of America and China \[[@B6-biomolecules-10-00575],[@B7-biomolecules-10-00575],[@B8-biomolecules-10-00575]\]. To better monitor the threat of AFB~1~ contamination, various methods have been developed in the past few decades \[[@B9-biomolecules-10-00575],[@B10-biomolecules-10-00575],[@B11-biomolecules-10-00575],[@B12-biomolecules-10-00575]\]. Although the results are reliable and accurate, instrumental techniques \[[@B13-biomolecules-10-00575]\] need expensive equipment and complicated sample pretreatment. Biosensors based on the antibody immunoprobes such as enzyme-linked immunosorbent assay (ELISA) \[[@B14-biomolecules-10-00575]\] and fluorescence-linked immunosorbent assay (FLISA) \[[@B15-biomolecules-10-00575],[@B16-biomolecules-10-00575]\] can achieve quantitative detection with good performance of specificity, sensitivity and simplicity, but the heterogeneous immunoassays require multiwashing procedures and long analysis times. To address the above issues, lateral flow immunochromatography assays have been considered as a promising method for onsite screening of mycotoxins \[[@B17-biomolecules-10-00575],[@B18-biomolecules-10-00575],[@B19-biomolecules-10-00575]\]. Moreover, immunochromatography assays based on fluorescent markers (time-resolved fluorescent nanobeads (TRFN), quantum dot nanobeads (QB) and quantum dots (QD), etc.) have gradually become a popular research field in recent years for their advantages of sensitivity, accuracy, automated detection, shorter detection time, and so on \[[@B20-biomolecules-10-00575],[@B21-biomolecules-10-00575],[@B22-biomolecules-10-00575]\]. Several fluorescence immunochromatography assays for highly sensitive detection of AFB~1~ have been reported \[[@B20-biomolecules-10-00575],[@B21-biomolecules-10-00575],[@B23-biomolecules-10-00575],[@B24-biomolecules-10-00575],[@B25-biomolecules-10-00575]\]. Although many methods based on immune interactions have been developed for the detection of toxic and harmful substances, it is impossible to compare the performance of those methods for identifying the most appropriate approach due to the utilization of distinct antibodies/antigens, markers and the detection conditions. In recent years, only a few reports have used comparative methods under the same conditions \[[@B26-biomolecules-10-00575],[@B27-biomolecules-10-00575],[@B28-biomolecules-10-00575],[@B29-biomolecules-10-00575],[@B30-biomolecules-10-00575],[@B31-biomolecules-10-00575]\]. For instance, Xie et al. \[[@B27-biomolecules-10-00575]\] established flow immunochromatography to detect *Escherichia coli O157:H7* in milk, in which fluorescent microspheres and colloidal gold were compared in terms of antibody labeling efficiency, sensitivity, antibody consumption and coefficient of variation. Wu et al. \[[@B28-biomolecules-10-00575]\] systematically and comprehensively compared the performance of fluorescent microsphere and quantum dot immunochromatographic strips for quantitative detection of aflatoxin M~1~ (AFM~1~) in milk. However, to the best of our knowledge, among the widely used fluorescent labeling materials of TRFN, QB and QD, there are no clear statements on which labeling material is better for AFB~1~ detection in foods by immunochromatography. In this paper, in order to find a more suitable fluorescent detection method for quantitative detection of AFB~1~ in grains, TRFN, QB and QD were used as labels to establish fluorescent immunochromatography (TRFN-FICA, QB-FICA and QD-FICA) for the first time by comparing antibody labeling efficiency, detection sensitivity, antibody and antigen consumption, and accuracy under the same conditions ([Figure 1](#biomolecules-10-00575-f001){ref-type="fig"}). 2. Materials and Methods {#sec2-biomolecules-10-00575} ======================== 2.1. Materials and Apparatus {#sec2dot1-biomolecules-10-00575} ---------------------------- ### 2.1.1. Materials {#sec2dot1dot1-biomolecules-10-00575} Time-resolved fluorescent nanobeads (TRFN, 1%, solid content, *w*/*v*; carboxylate-modified Eu (III)-chelate-doped polystyrene nanobeads; excitation = 365 nm, emission = 610 nm) were purchased from Bangs Laboratories, Inc. (Fishers, Hamilton, IN, USA). Carboxylated quantum dot nanobeads (QB, 1 uM, *w*/*v*, excitation = 365 nm, emission = 610 nm) and quantum dots (QD, 1.0 mg/mL, *w*/*v*; carboxylate-modified CdSe/ZnS core/shell nanocrystals with amphiphilic polymer coating; excitation = 365 nm, emission = 610 nm) were purchased from NanoGen (Beijing, China). Anti-AFB~1~ monoclonal antibody (mAb) and coating antigen (AFB~1~-CMO-BSA) were donated by Beijing WDWK Biotech Co., Ltd., (Beijing, China). N-hydroxysuccinimide (NHS) and 1-ethyl-3-\[3-(dimethylamino) propyl\] carbodiimide (EDC) were obtained from Aladdin (Shanghai, China). AFB~1~, aflatoxin B~2~ (AFB~2~), AFM~1~, aflatoxin M~2~ (AFM~2~), aflatoxin G~1~ (AFG~1~), aflatoxin G~2~ (AFG~2~), zearalenone (ZEN), ochratoxin A (OTA), deoxynivalenol (DON) and bovine serum albumin (BSA) were purchased from Sigma (St. Louis, MO, USA). Chicken IgY and rabbit antichicken IgY-IgG were obtained from Biodragon Immunotechnologies Co., Ltd. (Beijing, China). Other chemical substances were purchased from Beijing Chemical Reagent Company (Beijing, China). All solvents and other chemicals were of analytical reagent grade and did not require further purification. A working standard of AFB~1~ was prepared from the 2 mg mL^−1^ stock solution by serial dilution with a sample buffer solution (0.3 M Tris-HCl containing 0.5% polyvinyl pyrrolidone and 0.4% Tetronic 1307, pH 8.0). The nitrocellulose (NC) membrane (Unistart CN95) was acquired from Sartorius Stedim Biotech GmbH (Goettingen, Germany). The sample pad (glass fiber) and the absorbent pad were supplied by Shanghai Liangxin Co., Ltd. (Shanghai, China). The microtiter plates were supplied by Guangzhou JET BIOFIL Co., Ltd. (Guangzhou, China). ### 2.1.2. Apparatus {#sec2dot1dot2-biomolecules-10-00575} An XYZ3060 dispensing platform was purchased from Bio Dot Inc. (Irvine, CA, USA). The CM4000 guillotine-cutting module was purchased from Kinbio Tech Co., Ltd. (Shanghai, China). The fluorescence immunochromatography quantitative analyzer was purchased from WDWK Bio Co., Ltd. (Beijing, China). Ultrapure water was purified with the Milli-Q system from Millipore Corp. (Bedford, MA, USA). The size distributions and surface morphologies of the three labels were determined by transmission electron microscope (JEM 1200EX, Tokyo, Japan). The mAb labels were characterized with a particle size analyzer (Malvern Instruments Ltd., Worcestershire, UK). 2.2. Preparation of Three Labeled Antibody Probes {#sec2dot2-biomolecules-10-00575} ------------------------------------------------- The TRFN-mAb was prepared based on the procedures described in the previous literature with slight modification \[[@B26-biomolecules-10-00575],[@B30-biomolecules-10-00575]\]. Briefly, 5 μL of TRFN was dissolved in 45 μL of activation buffer (50 mM MES (2-Morpholinoethanesulfonic Acid), pH 6.0) and then centrifuged at 20,000× *g* for 15 min at 4 °C. Subsequently, 40 μL of activation buffer, 5 μL of NHS solution (1 mM) and 5 μL of EDC solution (1 mM) were added to the tube and stirred for 15 min; the solution was centrifuged at 20,000× *g* for 15 min and the precipitate was resuspended in 25 μL boric acid buffer (40 mM, pH 8.0). Next, 25 μL of anti-AFB~1~-mAb was added to the suspension and incubated at room temperature for 2 h, then centrifuged, and the precipitate was dissolved in 100 μL of blocking buffer (20 mM PBS, 50 mM ethanolamine, 4% BSA, pH 7.4) for 2 h. After the end of the blocking, the mixture was centrifuged at 20,000× *g* for 15 min at 4 °C, and the precipitate was resuspended in 50 μL of complex solution (10 mM Tris, 1% BSA, 2% sucrose, 2% trehalose, pH 8.5) at 4 °C until use. Ultrasonic dispersion was required for 3 min after each resuspension by centrifugation. The preparation of TRFN-IgY, QB-mAb and QD-mAb probes was identical to the preparation of TRFN-mAb, and the only differences were that chicken IgY was used instead of anti-AFB~1~-mAb, and QB and QD were used instead of TRFN, respectively. All the labeled antibody probes were stored at 4 °C until use. 2.3. Preparation of the Fluorescence Immunochromatography Assay Strips {#sec2dot3-biomolecules-10-00575} ---------------------------------------------------------------------- The fluorescence quantitative immunochromatographic strips consisted of four parts: absorbent pad, NC membrane, sample pad and adhesive plastic-backing sheet ([Figure 1](#biomolecules-10-00575-f001){ref-type="fig"}B). The procedures for making test strips were the same as our previously reported work with some modifications \[[@B32-biomolecules-10-00575],[@B33-biomolecules-10-00575]\]. Briefly, a proper amount of AFB~1~-CMO-BSA and rabbit antichicken IgY-IgG were separately sprayed onto the NC membrane as capture reagents to form *T* line and *C* line. The distance between *T* and *C* line was 1.2 cm and the dispense rate was 0.7 μL cm^−1^. Afterward, the dried NC membrane, sample pad and absorbent pad were laminated and cut into 4.7 mm wide test strips. Finally, the PVC sheet and strip were installed onto a plastic plate and stored in dry conditions at 4 °C until use. 2.4. Sample Preparation and Detection {#sec2dot4-biomolecules-10-00575} ------------------------------------- The sample preparation procedure was applied for corn, soybeans, sorghum, wheat, rice and oats. First, all samples were ground into powder and sieved through 20 mesh; then 1.00 ± 0.05 g of the pulverized samples were extracted with 4 mL of methanol/water solution (70/30, *v*/*v*); the mixture was vortexed for 5 min and centrifuged at 4000× *g* for 5 min at room temperature. Afterwards, 1 mL of the supernatant was diluted with 9 mL of sample buffer solution (0.3 M Tris-HCl containing 0.5% polyvinyl pyrrolidone and 0.4% Tetronic 1307, pH 8.0) to obtain a sample treatment solution. Finally, an appropriate amount of fluorescent probes was added and incubated with 120 μL of sample treatment solution for 5 min at room temperature (25 °C) in the microwell; 85 μL of incubated working solution was added into the test area. The fluorescence intensity ratio of *T* line and *C* line were defined as *F~T~* and *F~C~*. The fluorescence values of *F~T~*, *F~C~* and *F~T~*/*F~C~* were collected for quantification. 2.5. Establishment of Quantitative Calibration Curves {#sec2dot5-biomolecules-10-00575} ----------------------------------------------------- The quantitative calibration curves were established by plotting *B/B~0~* (the concentration of the analyte was 0 μg L^−1^, the value of *F~T~*/*F~C~* was marked as *B*~0~; while the concentration of the analyte was at other concentrations, the value of *F~T~*/*F~C~* was marked as *B*) against the logarithm of AFB~1~ concentration. Different concentrations of AFB~1~ (0, 5 × 10^−4^, 1 × 10^−3^, 5 × 10^−3^, 0.01, 0.05, 0.1, 0.5 and 1 μg L^−1^) were prepared by diluting in sample buffer solution; each piece of data was repeated for 6 times and fit to a four-parameter logistic equation using Origin (version 8.5, OriginLab, USA) software packages, $${y =}\left( {A - D} \right)/\left\lbrack {{1 +}\left( {x/C} \right)^{B}} \right\rbrack{+ D}$$ where A is the response value at high asymptote, B is the slope at the inflection point, C is the x value at the inflection point (corresponding to concentration resulting in 50% inhibition), D is the response value at low asymptote. 2.6. Validation of FICAs {#sec2dot6-biomolecules-10-00575} ------------------------ For validation of TRFN-FICA, QB-FICA and QD-FICA, 60 different field grain samples (10 samples for each of corn, soybeans, sorghum, wheat, rice and oats) were analyzed by the three FICAs and liquid chromatography--tandem mass spectrometry (LC-MS/MS); the LC-MS/MS procedures were performed according to the standard method of "GB5009.22-2016" \[[@B34-biomolecules-10-00575]\]. The detection performances of the three FICAs were compared to that of the LC-MS/MS to assess reliability. The LOD was calculated as the mean value of 20 blank samples plus three times the standard deviation (mean + 3SD). The accuracy of the method was investigated by spiking blank samples with single or multiple analytes at three concentrations (LOD, 2LOD, 4LOD). The recovery was calculated by the following equation: Recovery (%) = (measured concentration / fortified concentration) × 100%. The intra-assay and interassay precisions were represented by the coefficient of variation (CV); each sample was tested 6 times in duplicate and on three consecutive days. 3. Results and Discussion {#sec3-biomolecules-10-00575} ========================= 3.1. Principle of Three Fluorescence Labels for Detection of Aflatoxin B~1~ {#sec3dot1-biomolecules-10-00575} --------------------------------------------------------------------------- Three fluorescence labels were selected for the determination of AFB~1~ by direct competition reaction in general ([Figure 1](#biomolecules-10-00575-f001){ref-type="fig"}). Specifically, the rabbit antichicken IgY-IgG was immobilized on *C* line, and exhibited a constant *C* line fluorescence signal since the independent TRFN-chicken IgY was specially prepared for it. AFB~1~-CMO-BSA (coating antigen) was immobilized on *T* line, and when the fluorescence probes (QD-mAb, QB-mAb and TRFN-mAb) were not bound to free AFB~1~ molecules, they could be specifically captured by coating antigen as a reference signal in FICAs; otherwise, it would flow past both *T* and *C* lines with no signal. According to this principle, the adopted dual system (independent *T* and *C* lines) can maintain a comparatively stable *C* line fluorescence intensity with no interference; the fluorescence intensity of *T* line decreased with increased concentration of AFB~1~. Compared with previous studies of coating secondary antibodies to form *T* line \[[@B35-biomolecules-10-00575],[@B36-biomolecules-10-00575]\], this dual system achieved better performance and could be applied in later reported immunochromatographic assays \[[@B37-biomolecules-10-00575]\]. Overall, quantitative relationships can be established between the concentrations of AFB~1~ and *F~T~/F~C~* ratios, and can be further quantitatively calculated by the portable reader. 3.2. Characterization of Fluorescence Labels {#sec3dot2-biomolecules-10-00575} -------------------------------------------- The surface morphology and size of the three labels (TRFN, QB, QD) were characterized by transmission electron microscope (TEM), showing that TRFN, QB, QD had relatively uniform size distribution ([Figure 2](#biomolecules-10-00575-f002){ref-type="fig"}). TRFN are composed of rare earth lanthanide chelates (such as Eu(III), Tb(III) and Dy(III)) and exhibit longer (microsecond) lifetimes, allowing fluorescence decay to be monitored over time. This technique provides a means to separate the "true" fluorescence signal from short-lived background fluorescence, and an opportunity to improve assay sensitivity \[[@B38-biomolecules-10-00575]\]. QD are new fluorescent labels with great prospects, and have been widely used to improve the detection sensitivity of FICA because of their narrow emission spectra, broad excitation range and highly fluorescent quantum yields \[[@B20-biomolecules-10-00575]\]. Furthermore, QB are tens of thousands of quantum dots wrapped in inorganic materials such as silicon dioxide by self-assembly, which is easy to mass produce; they have stronger fluorescence stability and intensity than the corresponding single QD \[[@B28-biomolecules-10-00575]\]. These labels were distributed uniformly in the low magnification image and scattered well in the magnified view; the magnified TEM image in [Figure 2](#biomolecules-10-00575-f002){ref-type="fig"}(B2) revealed that the single quantum dots were embedded uniformly when compared to [Figure 2](#biomolecules-10-00575-f002){ref-type="fig"}(C2). After chemically binding to the surfaces of the antibody, these fluorescence labels provided a high degree of long-term stability in sample detection \[[@B28-biomolecules-10-00575]\], and the particle size analyzer indicated that the average hydrodynamic diameters of TRFN-mAb, QB-mAb and QD-mAb were significantly increased from 90 (TRFN) to 113 nm, 110 (QB) to 136 nm, and 15 (QD) to 42 nm ([Figure S1](#app1-biomolecules-10-00575){ref-type="app"}), respectively. This proved that the three fluorescent probes were successfully synthesized, and all the probes were used for fluorescence immunochromatography detection. 3.3. Optimization and Establishment of Standard Calibration Curve {#sec3dot3-biomolecules-10-00575} ----------------------------------------------------------------- To achieve the best performance of FICAs, parameters such as coupling pH, lateral flow immune response time, working concentration of labeled mAb immunoprobes (anti-AFB~1~-mAb) and working concentration of coating antigens (AFB~1~-CMO-BSA) were taken into consideration as important factors that affected the sensitivity of the FICAs. Therefore, all the FICAs needed to be introduced at optimum parameters. In this assay, the fluorescence intensity of *C* line was almost constant under the same reaction conditions (1.6 μg mL^−1^ of rabbit anti-chicken IgY-IgG as coating antigen and 3.4 μg mL^−1^ of TRFN-IgY as immunoprobe). The competitive inhibition ratio was observed by investigating appropriate fluorescence intensity of *T* line and *C* line, which was chosen as a factor to reflect the sensitivity of FICAs. As seen in [Figure S2A](#app1-biomolecules-10-00575){ref-type="app"}, the fluorescence intensity of TRFN-FICA was enhanced with an increase of pH, and the highest competitive inhibition ratio was observed at pH 7.0; therefore, pH 7.0 was regarded as the optimal pH for coupling with TRFN. Using the same reasoning, pH 6.0 and 7.0 were the optimal labeling pH for QB-mAb and QD-mAb, respectively ([Figure S2](#app1-biomolecules-10-00575){ref-type="app"}). In this study, the concentration parameters of labeled mAb immunoprobes (anti-AFB~1~-mAb) were 3.0, 4.5 and 4.5 μg mL^−1^ for TRFN-FICA, QB-FICA and QD-FICA, respectively ([Figure S3](#app1-biomolecules-10-00575){ref-type="app"}). Coating antigens (AFB~1~-CMO-BSA) were 0.3, 0.65 and 0.65 μg mL^−1^ for TRFN-FICA, QB-FICA and QD-FICA, respectively ([Figure S4](#app1-biomolecules-10-00575){ref-type="app"}). The optimum immunochromatography durations were 25, 30 and 35 min for TRFN-FICA, QD-FICA and QB-FICA, respectively ([Figure S5](#app1-biomolecules-10-00575){ref-type="app"}). 3.4. Validation of FICAs {#sec3dot4-biomolecules-10-00575} ------------------------ ### 3.4.1. Sensitivity {#sec3dot4dot1-biomolecules-10-00575} Under optimum conditions, with the increasing concentration of AFB~1~ diluted in sample buffer solution, the fluorescence intensity of the corresponding test line gradually decreased. The calibration curves of three fluorescent label-based FICAs were constructed by plotting *B*/*B*~0~ against the logarithm of AFB~1~ concentrations ([Figure 3](#biomolecules-10-00575-f003){ref-type="fig"}); we then fit the data using linear equations. The sensitivity of TRFN-FICA, QB-FICA and QD-FICA were evaluated using the values of IC~50~ obtained from the calibration curves, which were 0.0133, 0.0442 and 0.0848 μg L^−1^, respectively. The dynamic linear ranges, determined as the concentrations causing 20%--80% inhibition of *B*/*B*~0~, were 0.00368--0.04804, 0.01621--0.09775 and 0.03756--0.16776 μg L^−1^, respectively. ### 3.4.2. Specificity {#sec3dot4dot2-biomolecules-10-00575} To examine the specificity of proposed FICAs, three structurally-related mycotoxins, such as AFB~2~, AFM~1~ and AFG~1~, and non-structurally-related mycotoxins, including DON, OTA and ZEN, were tested individually by the FICAs to evaluate specificity ([Table S1](#app1-biomolecules-10-00575){ref-type="app"}). Data were obtained from six replicates. All results clearly demonstrated that the three FICAs (TRFN-FICA, QB-FICA and QD-FICA) have negligible cross reactivity (CR \< 20%) with the other mycotoxins, and the proposed three FICAs can be applied to detect AFB~1~ with high specificity. 3.5. Application to Grain Samples {#sec3dot5-biomolecules-10-00575} --------------------------------- Detection performance of the three FICAs was investigated in real samples. The LOD was calculated as the mean value of 20 blank grain samples plus three times the standard deviation (mean + 3SD). Each of the 20 blank grain samples (corn, soybean, sorghum, wheat, rice and oats) were extracted and analyzed according to the sample preparation and detection procedure. The LODs for TRFN-FICA, QB-FICA and QD-FICA were 0.04, 0.30 and 0.80 μg kg^−1^, respectively. TRFN-FICA possessed the advantages of sensitivity, rapidity, antibody and antigen consumption, and accuracy when compared with QB-FICA and QD-FICA ([Table 1](#biomolecules-10-00575-t001){ref-type="table"}). Moreover, in comparison with most available immunoassay methods for comprehensive performance ([Table 2](#biomolecules-10-00575-t002){ref-type="table"}), the detected performances of QD-FICA and QB-FICA were in accordance with the reported fluorescence immunochromatography in real samples or buffer solution \[[@B23-biomolecules-10-00575],[@B39-biomolecules-10-00575]\]; TRFN-FICA had the best LOD and reached 125%--150% better sensitivity than the reported multiple time-resolved fluorescence immunochromatography assay \[[@B21-biomolecules-10-00575],[@B24-biomolecules-10-00575]\]. Therefore, fluorescence immunochromatography assay, especially TRFN-FICA, possessed the obvious advantages of sensitivity, rapidity and cost-effectiveness for onsite screening of AFB~1~ in grains \[[@B36-biomolecules-10-00575],[@B40-biomolecules-10-00575],[@B41-biomolecules-10-00575],[@B42-biomolecules-10-00575]\]. Furthermore, in order to verify and compare the reliability of FICAs, 60 grain samples were analyzed by liquid chromatography--tandem mass spectrometry (LC-MS/MS) \[[@B34-biomolecules-10-00575]\], TRFN-FICA, QB-FICA and QD-FICA. A total of 12 samples were confirmed as positive samples, while others (48 samples) were negative by LC-MS/MS and three FICAs, and the representative mass chromatograms (highest and lowest concentrations for positive samples) are listed in [Figure S6](#app1-biomolecules-10-00575){ref-type="app"}. There were no false negative or false positive results reported by the three FICAs, and analysis of field grain samples by FICAs were in accordance with that of LC-MS/MS ([Figure 4](#biomolecules-10-00575-f004){ref-type="fig"}). These results indicate that all three FICAs are reliable methods for the determination of AFB~1~ residues in grains, and that TRFN-FICA obtained the best fit. 3.6. Accuracy and Precision of Three Label-Based FICAs {#sec3dot6-biomolecules-10-00575} ------------------------------------------------------ We performed recovery experiments to assess the accuracy and precision of the three FICAs using six kinds of blank grain samples (corn, soybeans, sorghum, wheat, rice and oat) with a series of known concentrations of AFB~1~. The choices of low, medium and high concentrations with AFB~1~ were the same as in the previous reported literature \[[@B32-biomolecules-10-00575],[@B33-biomolecules-10-00575]\], which were represented by LOD, 2LOD and 4LOD, respectively. Data were obtained from six replicates and on three consecutive days. The intraday and interday recovery of TRFN-FICA ranged from 86.48% to 114.10% and 83.64% to 125.61%, respectively; the coefficient of variation were all less than 10%. Meanwhile, TRFN-FICA had better recovery than QB-FICA and QD-FICA ([Figure 5](#biomolecules-10-00575-f005){ref-type="fig"}), confirming that the accuracy and precision of TRFN-FICA were better than QB-FICA and QD-FICA. 4. Conclusions {#sec4-biomolecules-10-00575} ============== In this study, three FICAs (TRFN-FICA, QB-FICA and QD-FICA) were systematically compared for the quantitative detection of AFB~1~ in grains successfully. Under optimum conditions, six types of grain samples were analyzed, showing that TRFN-FICA was the most consistent with LC-MS/MS. Moreover, TRFN-FICA had the lowest LOD, shortest immune duration (25 min), and less coating antigen consumption (0.30 μg) and antibody consumption (0.015 μg). Overall, compared with QB-FICA and QD-FICA, TRFN-FICA had a unique advantage in quantitative detection of AFB~1~ in grain, providing a reference for the selection of markers in detection methods. We sincerely thank Beijing WDWK Biotech Co., Ltd., (Beijing, China) for providing experimental antibodies and coating antigens. The following are available online at <https://www.mdpi.com/2218-273X/10/4/575/s1>. Figure S1: Particle size of three FICA labels. Figure S2: Optimization of coupling pH in preparing labeled mAb probes: (A) TRFN-mAb; (B) QB-mAb; (C) QD-mAb. Figure S3: Optimization of the concentration of anti-AFB~1~-mAb in preparing labeled mAb probes: (A) TRFN-mAb; (B) QB-mAb; (C) QD-mAb. Figure S4: Optimization of the concentration of AFB~1~-CMO-BSA for fluorescent immunochromatography: (A) TRFN-FICA; (B) QB-FICA; (C) QD-FICA. Figure S5: Immunoreaction dynamics of the three FICAs: (A) TRFN-FICA; (B) QB-FICA; (C) QD-FICA. Table S1: Cross reactivity (CR) of analytes with antibody detected by FICAs. ###### Click here for additional data file. Conceptualization, X.T.; methodology, Z.L.; software, X.W. (Xuan Wu); validation, X.W. (Xin Wang); data curation, X.W. (Xuan Wu); writing: original draft preparation, X.W. (Xin Wang); writing: review and editing, Z.L. and X.T.; visualization, X.W. (Xin Wang); supervision, X.T.; project administration, X.T.; funding acquisition, X.T. All authors have read and agreed to the published version of the manuscript. This research was funded by the National Natural Science Foundation of China, grant number 31672605; Natural Science Foundation of Chongqing, China, grant number cstc2018jcyjAX0242 and cstc2017jcyjAX0313; China Postdoctoral Science Foundation, grant number 2016M590855; Chongqing Postdoctoral Science Foundation Special Funded Project, grant number Xm2017074. The authors declare no conflicts of interest. {#biomolecules-10-00575-f001} {#biomolecules-10-00575-f002} {#biomolecules-10-00575-f003} {#biomolecules-10-00575-f004} {#biomolecules-10-00575-f005} biomolecules-10-00575-t001_Table 1 ###### Performance of TRFN-FICA, QB-FICA and QD-FICA in 6 grains. Parameter TRFN-FICA QB-FICA QD-FICA ------------------------------------------ ----------------- ----------------- ----------------- LOD (μg kg^−1^) 0.04 0.30 0.80 Antibody usage per test card (μg) 0.015 0.09 0.03 The best coating for AFB~1~-CMO-BSA (μg) 0.30 0.65 0.65 Immunoassay duration (min) 25 30 35 Recovery (%) 83.64%--125.61% 80.29%--129.45% 64.53%--133.86% Coefficient of variation (%) 3.10%--6.75% 2.88%--7.16% 2.34%--8.96% biomolecules-10-00575-t002_Table 2 ###### Comparison of immunoassays for determination of AFB~1~. -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Detection Method Marker Target Substance Sample Detection Limit of AFB~1~ (μg kg^−1^) ----------------------------------------------------------------------------- ---------------- ------------------ ---------------------------------------------- --------------------------------------- Immunoadsorption \[[@B15-biomolecules-10-00575]\] Enzyme AFB~1~ Feed samples 2.0 Multiplex immunochromatography \[[@B36-biomolecules-10-00575]\] Colloidal gold AFB~1~, ZEN, OTA Corn 0.10 Rice 0.12 Fluorescent immunochromatography \[[@B24-biomolecules-10-00575]\] TRFN AFB~1~ Corn 0.06 Fluorescent immunochromatography \[[@B20-biomolecules-10-00575]\] QB AFB~1~ Buffer solution 0.005\ (When the inhibition is 10) Multiplex fluorescent immunochromatography \[[@B23-biomolecules-10-00575]\] QB AFB~1~, ZEN Buffer solution 0.00165\ (When the inhibition is 10%) Multiplex fluorescent immunochromatography \[[@B21-biomolecules-10-00575]\] TRFN AFB~1~, ZEN Buffer solution 0.05 Fluorescence resonance energy transfer \[[@B39-biomolecules-10-00575]\] QD AFB~1~ Rice 0.04 Fluorescent immunochromatography (this study) TRFN AFB~1~ Corn, soybean, sorghum, wheat, rice and oats 0.04 QB 0.30 QD 0.80 -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- [^1]: These authors contributed equally to this work.

In the most recent version of the projections of the Office of the Actuary in the Centers for Medicare & Medicaid Services, national health spending growth is expected to average 5.5% per year for 2017-2026 in the United States (US): approximately 1.0% higher than projected gross domestic product (GDP) growth.^[@bibr1-2192568219878133]^ This results in an increase in the health-related costs as a percentage of GDP from 17.9% in 2016, to nearly 20% by 2026, reaching a total of \$5.7 trillion by 2026.^[@bibr1-2192568219878133]^ When looking granularly at specific medical fields, a significant portion of the rising costs of health care in the US relates to the diagnosis and treatment of spinal pathology. It is estimated that 12% to 30% of US adults have an active back problem with approximately 6% having made a visit to a physician for these conditions at one point in their lives, costing upward of \$100 billion to the system each year.^[@bibr2-2192568219878133],[@bibr3-2192568219878133]^ Specifically with regard to spine surgery, fusions and laminectomies were the third and fifth most commonly performed surgical procedures in the United States in 2015, respectively.^[@bibr4-2192568219878133]^ Given the rising costs associated with spine surgery and an aging population, it becomes increasingly clear that the current trajectory is not sustainable, and further scrutiny will be placed on the field in assessing the effectiveness, efficiency, and safety of care delivered. As more healthcare systems invest in healthcare analytics and "big data" (large, complex datasets such as those found in electronic medical records), the opportunity arises to employ predictive analytics via machine learning (ML)/artificial intelligence (AI) approaches to improve quality, reduce waste and error, and minimize cost.^[@bibr5-2192568219878133],[@bibr6-2192568219878133]^ Recent developments in the technologies related to healthcare data collection and analytics have led to a rapid rise in the application of AI within health-related fields. One such application is ML, a branch of AI that involves the construction and application of statistical algorithms that continuously learn and make observations from existing data, and then create a predictive model based on that data.^[@bibr7-2192568219878133]^ With advances in computer processing capability, data storage, and networking, these computer-based algorithms can perform the intricate and extremely complex mathematical operations of classification or regression (specifically nonlinear regression) on immense amounts of data to detect intricate and potentially previously unknown patterns in that data.^[@bibr8-2192568219878133]^ ML algorithms have been able to analyze complex and large volumes of electronic medical record data to produce predictions for a wide range of clinical problems.^[@bibr9-2192568219878133]^ For example, Rajkomar et al^[@bibr9-2192568219878133]^ demonstrated that ML models outperformed traditional, clinically used models in predicting mortality, unexpected readmission, and increased length of stay (LOS) in a study cohort of all admissions in 2 major hospitals from 2009 to 2016. Various investigators have been developing image analysis methods using ML algorithms that have shown promising results in fields such as dermatology, radiology, and ophthalmology. For example, Esteva et al^[@bibr10-2192568219878133]^ have trained ML algorithms to classify skin cancer with a level of competence comparable to dermatologists. These early examples provide insight into early contemporary use of AI in medicine and provide a view of technology that may transform the medical field over the decades to come. While not the first medical field to adopt an "AI approach" to problem solving, the spine surgery field has recently seen an outpouring of publications related to research in this area. An initial topic of focus by researchers was related to the cost of spine care, as there has been heightened emphasis on moving to a value-based (quality/cost) health care market. For example, as Medicare payments are standardized by procedures performed regardless of hospital LOS, ML systems have been designed with the ability to accurately predict spine surgery-related LOS, discharge to nonhome facility, and early unplanned readmissions using only presurgical or predischarge variables.^[@bibr11-2192568219878133][@bibr12-2192568219878133][@bibr13-2192568219878133]-[@bibr14-2192568219878133]^ These models can help identify/target certain high-risk patients and the variables that contribute to that risk status, allowing hospitals to allocate specific clinical and social resources to reduce costly LOS and readmissions. This can help to maximize efficiency of care delivered, while also keeping constant or even increasing the quality of care delivered. As spinal surgery has evolved with an explosion of new techniques and technologies in recent decades, there still remains a lack of quality, high-level evidence to support much of the spine care rendered in the US, especially with the cost associated with many of the treatments and devices. As there are numerous surgical treatments in spine surgery that do not easily lend themselves to traditional randomized controlled trials (due to either cost or ethical considerations, among other reasons), an opportunity arises that is ripe for solutions derived from ML approaches. Multiple clinical registries are being collected that contain large quantities of high-quality, spine health care data, such as the 1000-patient Spinal Laminectomy versus Instrumented Pedicle Screw (SLIP) II study.^[@bibr15-2192568219878133][@bibr16-2192568219878133]-[@bibr17-2192568219878133]^ These registries contain demographics, surgery-related variables, patient-reported and complication outcome measures, and notably, they even contain digital imaging with metadata. Leveraging of these vast data repositories can help develop predictive algorithms that are able to incorporate the full range of variables (including complex imaging) in order to guide treatment recommendations. Because of this lack of high-level evidence, there remains much heterogeneity in the current surgical treatment of spinal disorders, with significant clinical and economic implications.^[@bibr18-2192568219878133][@bibr19-2192568219878133]-[@bibr20-2192568219878133]^ For instance, national surveys of US spine surgeons conducted by Mroz et al^[@bibr21-2192568219878133]^ found 69% disagreement for recurrent lumbar disk herniation, while another study demonstrated 75% disagreement among surgeons on the approach to treat patients with lower back pain,^[@bibr22-2192568219878133]^ implying that 2 similar patients with the same pathology could receive entirely different care. Furthermore, a cost analysis based on the results of the national survey mentioned above revealed that there is also a variation in costs based on spine surgeon specialty, practice type, surgical volume and geographical location.^[@bibr23-2192568219878133]^ Recent ML/AI approaches to this problem have been published that attempt to assist surgeons' decisions with predictions of patient outcomes. Utilizing data from repositories created from AOSpine prospective, multicenter studies, Merali et al^[@bibr17-2192568219878133]^ developed a supervised ML model that accurately predicts a positive outcome on an individual patient after surgery for degenerative cervical myelopathy, with an average area under the curve of 0.70, classification accuracy of 77%, and sensitivity of 78% on an independent testing cohort. Shah et al^[@bibr24-2192568219878133]^ were able to build an ML model that predicts probability of failure of nonoperative management in spinal epidural abscess, while Karhade et al^[@bibr25-2192568219878133]^ successfully developed an ML algorithm that predicts in-hospital and 90-post discharge mortality in this patient group. The same group was able to predict short-term postoperative mortality in individual patients with spinal metastatic disease with an ML model, aiding in decision-making and informed discussions with the patient regarding surgical intervention this challenging patient population.^[@bibr26-2192568219878133]^ All of these previously mentioned studies have now published their prognostic tools in an open-access, digital interface to be integrated into practice, supporting clinicians in developing treatment plans that are more standardized across the world. Along with prediction of positive patient outcomes, clinician researchers have also used AI/ML to forecast negative outcomes as well, as recent publications have explored the likelihood of complications from spine surgery. In multiple articles, the same group led by Cho et al utilized an artificial neural network-based ML algorithm to predict surgical complications in patients undergoing elective anterior cervical discectomy and fusion, posterior lumbar fusion, and adult spinal deformity surgeries. Their models were able to specifically predict the risk of cardiac-related, wound-related, venous thromboembolism--related, and mortality in these patients, outperforming the American Society of Anesthesiologists Physical Status Classification scoring in predicting individual risk prognosis.^[@bibr27-2192568219878133],[@bibr28-2192568219878133]^ Another publication by Sheer et al^[@bibr29-2192568219878133]^ describes their method to create a ML model that successfully predicts major intraoperative/perioperative complications following adult spinal deformity surgery with an accuracy of 87%. Utilizing large databases of patient information, Han et al^[@bibr30-2192568219878133]^ were able to analyze over 1 000 000 patients that had previously undergone spine surgery and developed multiple ML predictive models that identify risk factors for postoperative complications. Karhade et al^[@bibr24-2192568219878133]^ were even able to predict prolonged opioid prescription after surgery for lumbar disc herniation in an ML algorithm. These surgery- and patient-specific models can help to aid in surgical planning, as well as patient counseling and shared decision making. If these models identify modifiable risk factors in the preoperative setting of a nonurgent surgery, time and effort could be dedicated to improved medical management of that comorbidity prior to surgery, in effect reducing the risk of complications and increasing the probability of a good outcome. In deciding if a patient is indicated for surgery, one area where a surgeon's subjectivity may still reign supreme is review of the spine imaging. Utilizing classification techniques from radiology literature, new research is revealing the applicability of AI and ML algorithms to the analysis of spine imaging. One technique involves the use of ML models utilizing natural language processing to distinguish specific words and phrases from unstructured radiology reports in order to classify patients by imaging findings, as Tan et al^[@bibr31-2192568219878133]^ were able to do in a cohort of patients with low back pain. However, in more recent publications, other groups were able to utilize the imaging itself to detect and classify a variety of pathologies. Hopkins et al^[@bibr32-2192568219878133]^ were able to predict both the diagnosis of cervical spondylotic myelopathy and its severity with high sensitivity and specificity (90.25% and 85.05%, respectively), utilizing magnetic resonance imaging alone in an artificial neural network model. Further, work has been done to develop ML models in the detection and grading of lumbar spinal stenosis^[@bibr33-2192568219878133]^ and fracture detection and classification with various types of imaging modalities.^[@bibr34-2192568219878133]^ AI/ML imaging analysis can even assist real time in the outpatient clinic, where Sharif Bidabadi et al^[@bibr35-2192568219878133]^ were able to accurately identify foot drop of an L5 origin and classify patients into various recover stages with an 85% accuracy. While there is much work to be done, this initial work which was all published in the past year, shows the feasibility of using AI/ML-based approaches to analyzing spine imaging. Common themes among large institutions and large spine centers are tighter financial margins, less resources, and heightened payer scrutiny on indications, outcomes, and postprocedural treatments. This collectively creates real strain on the departmental workforce (ie, secretaries, advanced practice providers, physicians). An AI platform that successfully predicts patient and surgeon performance from financial, outcome, and electronic medical record databases across an entire book of business stands to provide the leverage to homogenize outcome and cost. This, in turn, positions said organization optimally for contract negotiations and population health initiatives. Furthermore, a fully integrated AI platform can also automate much of what currently strains department assets. Postsurgical checks, ordering medications and imaging, patient reminders, and scheduling follow-up visits, are all some examples of how such a platform can enhance overall spine center efficiencies and performance, patient satisfaction (eg, more automated touch points), and employee engagement. Challenges Ahead {#section1-2192568219878133} ================ Even though current research described above highlights the promise and potential of AI in spine surgery, the field as a whole still face many challenges. First, in order to create an AI-driven decision platform, very large and appropriately labeled data sets are required, which the majority of centers in the United States still lack. This becomes even more difficult with imaging-based analysis. Second, some ML models require manual labeling of the data for classification and learning to occur. This presents a clear challenge in the analysis of spine surgery pathologies, where there is still widespread disagreement about what constitutes normal versus abnormal with regard to certain exam/imaging findings, and subsequently the appropriate treatment(s). This can be circumvented by allowing the model itself to do the analysis and classification, such as is the case with unsupervised algorithms. Given the vast quantity of data analyzed, this can reveal links between variables that experts would not have otherwise expected. However, it is difficult to backtrack and get precise information regarding the specifics of the data sorting in these types of models. And with poor quality or quantity of data to learn from, the model may make erroneous associations and/or can be "overfitted" to the training dataset, producing a lack of external validity. Furthermore, many ML algorithms thus far are typically trained and validated internally within one institution. Further work needs to be carried out to examine if a predictive model is transferable from one site to another, and what implications this holds as a "live" ML model undergoes continuous calibration and evolution based on new sets of data. As exponential expansion of computing capacity converges with unsustainable healthcare spending, a hopeful opportunity has emerged: the use of AI to enhance healthcare quality and safety. AI-based, ML approaches to spinal pathologies are already distinguishing relevant from irrelevant data regarding a particular patient, assisting with appropriate hospital-based care, interventions or even surgeries, predicting cost of care, and predicting future outcomes on a variety of anchored measures. While many shortcomings still exist as the technology is in early development, extrapolating from today's progress and fully implemented into the healthcare system, AI could help solve a number of problems in spine surgery by improving outcomes, minimizing cost, standardizing care for a given pathology, and driving efficiencies within a spine service line in large centers. These types of approaches could deliver on the value equation while serving as a resource for improving physician performance and promoting appropriate, efficient care in this era of financial uncertainty in health care. **Declaration of Conflicting Interests:** The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. **Funding:** The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This supplement was supported by funding from AO Spine North America. **ORCID iD:** Matthew M. Grabowski  <https://orcid.org/0000-0001-8550-0124>

See related research by Switzer et al., <http://breast-cancer-research.com/content/14/5/R125> Unlike the constitutive nitric oxide (NO) synthase isoforms, the inducible isoform of NO synthase - nitric oxide synthase 2 (NOS2) - is capable of producing sustained intracellular levels of NO, and it is increasingly appreciated that protein S-nitrosylation, the covalent modification of cysteine thiol by NO, is important for NOS2-dependent signal transduction. In the previous issue of *Breast Cancer Research*, Switzer and coworkers delineate a novel pathway for S-nitrosylation in regulation of estrogen receptor (ER)-negative breast cancer invasion \[[@B1]\]. This work adds to the growing appreciation that S-nitrosylation can regulate myriad pathways important for tumorigenesis, including gene transcription, apoptosis and DNA repair \[[@B2]-[@B4]\], and it complements another recent study by these coworkers showing that S-nitrosylation and activation of epidermal growth factor receptor is associated with the induction of epithelial-to-mesenchymal transition as well as chemoresistance in ER-negative breast cancer \[[@B5]\]. More generally, these data begin to explain why expression of NOS2 correlates with aggressive tumor phenotypes and poor clinical outcomes in a variety of malignancies, including breast cancer, lung cancer, colon cancer and prostate cancer \[[@B6]-[@B9]\]. The small GTPase Ras was one of the earliest described regulatory targets of S-nitrosylation. Modification of a single cysteine residue that is located in the nucleotide-binding region of wild-type Ras and is conserved among all Ras isoforms (Cys118 in human H-Ras) stimulates guanine nucleotide exchange and downstream pathways, including activation of mitogen-activated protein kinase signaling \[[@B10]\]. S-nitrosylation of wild-type Ras by endothelial NO synthase has been shown to promote pancreatic tumor growth \[[@B11]\]. Switzer and colleagues identify a role for S-nitrosylation of wild-type Ras in ER-negative breast cancer \[[@B1]\]. They find that the only elements common to genes upregulated in high NOS2-expressing breast cancer are binding sites for the Ets-1 transcription factor. Using a model ER-negative breast cancer cell line, they further show that NO induces S-nitrosylation of wild-type Ras, leading to phosphorylation and activation of Ets-1 through the Ras/MEK/ERK pathway. Knockdown of Ets-1 inhibits the NO-dependent expression of basal-like breast cancer markers and attenuates NO-dependent cancer cell invasion. These findings delineate a mechanism by which NOS2 promotes an aggressive tumor phenotype in ER-negative breast cancer. While NOS2 expression is high in ER-negative tumors \[[@B9]\], it is not constitutively expressed in the triple-negative breast cancer cell line MDA-MB-468 (lacking ER, progesterone receptor and epidermal growth factor receptor 2) that was utilized in this and related studies \[[@B1],[@B5]\]. Indeed, the loss of NOS2 expression in cultured tumor cells *ex vivo*is not uncommon and could reflect a deficiency in cytokine signaling and/or an inability to replicate the hypoxic tumor environment. Regardless of mechanism, this deficit in NOS2 expression can make challenging the study of the (patho)physiological functions of the enzyme in cell culture systems. To model NOS2-based signaling as it occurs in the solid tumor, Switzer and colleagues employ a long-lived NO donor as well as adenoviral expression of NOS2 \[[@B1]\]; in earlier work, they also co-cultured these cells with macrophages stimulated to produce high levels of NOS2-derived NO \[[@B5]\]. While these conditions may adequately recapitulate pathophysiological NO levels within the tumor, further work is needed to determine the extent to which these manipulations fully reconstitute NOS2-dependent signal transduction. These questions may in part be addressed by defining the precise mechanisms underlying enhanced expression of NOS2 in breast cancer and other malignancies. In addition to the environmental factors of cytokine secretion and hypoxia, aberrant regulation of NOS2 transcription, mRNA stability, and proteasomal degradation in the tumor cells present potential mechanisms. There is increasing recognition that denitrosylases - in particular, S-nitrosoglutathione reductase (GSNOR) and thioredoxin (Trx) - are critical modulators of S-nitrosothiol (SNO) homeostasis \[[@B12]\]. As the name implies, GSNOR metabolizes S-nitrosoglutathione, a small molecule intermediate in S-nitrosylation and denitrosylation. This enzyme may have general housekeeping functions, one of which is to protect against indiscriminate protein S-nitrosylation (nitrosative stress) that may occur as a result of excessive NOS2 activity. In this regard, GSNOR deficiency has been tied to inhibition of DNA repair and to hepatocellular carcinoma \[[@B3]\], and the enzyme is also deficient in some lung cancers \[[@B13]\]. Trx reduces protein SNOs directly, and inhibition of thioredoxin reductase (TrxR) can increase protein SNOs by blocking Trx turnover. Trx has another major role as a modulator of oxidative stress, and it is in this capacity that chemotherapeutics targeting TrxR are believed to affect cancer cell killing. Given the potential for protein S-nitrosylation to stimulate tumor growth and invasion, TrxR inhibitors may have unexpected deleterious effects. More generally, the balance of SNO formation/degradation is clearly an important factor in malignant cell transformation and thus warrants further investigation. Given the potential role of NOS2 in tumor chemoresistance and aggressiveness, it is important to note that NOS2 inhibitors have had some success in inhibiting tumor growth \[[@B14],[@B15]\]. At the same time, interventions that increase SNO levels, either through application of NO donors and low-mass SNOs or through inhibition of SNO-metabolizing enzymes, are being proposed as therapeutics in numerous diseases \[[@B16]\]. Given recent findings in the cancer field, it may be necessary to weigh the risk-benefit of therapies designed to increase protein S-nitrosylation for those individuals at high risk of developing cancer. Abbreviations ============= ER: estrogen receptor; GSNO: S-nitrosoglutathione reductase; NO: nitric oxide; NOS2: nitric oxide synthase 2; SNO: S-nitrosothiol; Trx: thioredoxin; TrxR: thioredoxin reductase. Competing interests =================== The authors declare that they have no competing interests. Acknowledgements ================ This work was supported in part by National Institutes of Health grants HL106121 (to MWF and HEM) and HL092994 (to HEM).

Background {#Sec1} ========== Over the past decade, there has been an increasing emphasis among university educators on globalization and internationalization, and global health programs and international experience have become key areas of focus for university professionals. According to China's Ministry of Education, in 2017, the number of Chinese students studying abroad exceeded 600,000 for the first time, reaching 608,400, which was an increase of 11.74% compared with 2016 \[[@CR1]\]. Researchers may benefit from experiencing the cultural differences inherent in exchange programs, and they may bring an awareness of these differences into their curricula, thereby broadening international training opportunities through work, education, and research activities \[[@CR2]--[@CR7]\]. After returning home, many health professionals work at medical universities or affiliated hospitals. As an implicit responsibility of these roles, health professionals must mentor medical students through engagement in clinical work and scientific research. The research ability of medical students during their postgraduate studies depends on many aspects, including the background and extent of their advisors' education and scientific research ability. Our previous research \[[@CR8]\] has shown that overseas study experience improved health professionals' scientific research ability, but no previous research has examined the impact of overseas experience on those professionals' students' scientific research ability. Is there a gap in the research ability of students of health professionals who studied or trained overseas ("returning" professionals) and that of students of health professionals who did not train overseas ("resident" professionals)? Do returning professionals, as postgraduate advisors, have a greater positive impact on their students' scientific research ability, compared with resident professionals? Does overseas experience play a positive role in the development of the research ability of medical students in China? In this study, we set out to answer these questions and to encourage the construction of a health professionals' overseas-experience database to help postgraduate students choose highly effective advisors. We also hope that our research results will be helpful in forming better training programs for medical students and in improving medical education policy. Methods {#Sec2} ======= Harbin Medical University (HMU) is the only Western medical institution in Harbin, in China's Heilongjiang Province. It is a government institution that offers a five-year medical bachelor's degree course, a three-year master's degree course, and a three-year PhD degree course. HMU has recently begun to restructure and reform its medical education program to offer comprehensive solutions to national medical education problems. This study was conducted from September 2016 to April 2018. During this period, only 1--2% of master's students (*n* = 1561) published a Science Citation Index (SCI) article prior to graduation. In contrast, around 70% of PhD students (*n* = 1083) published SCI articles before graduating. Thus, we included only PhD students in our analyses. PhD students in China are differentiated by their year of enrollment. PhD students in the class of 2015 graduated in June 2018 and their articles may not have been published by the end of the research period, so we selected PhD students enrollment in 2012--2014 for our study. Health professionals with a minimum of 6 months study-abroad experience were categorized as "returning," because we designated 6 months as the minimum experience required to be an independent researcher. Therefore, health professionals with less than 6 months study-abroad experience were excluded from this study. In the study, both "returning" and "resident" professionals had PhD degrees. These professionals were 56 scientific researchers from HMU and 76 clinical doctors from HMU's affiliated hospitals. We analyzed 257 students of returning health professionals (Group A)-106 from HMU (Group A~1~), who were trained to be scientific researchers, and 151 from HMU's affiliated hospitals (Group A~2~), who were trained to be clinical doctors. To collect information effectively, we constructed the questionnaire presented in Additional file [1](#MOESM1){ref-type="media"} Table S1, which was distributed to each of the 257 students. To ensure confidentiality and a high response rate, three investigators meet the students face to face and distributed the questionnaires to them. The students were required to complete the questionnaires while the investigator was present, and the investigators were not allowed to disclose any information or data collected. The SCI is internationally recognized as an authoritative scientific literature search tool. We used the SCI impact factor (IF) as an indicator of scientific and research capability. Because authorship contribution is determined differently across institutions and countries and there is no international system to weigh these differences against, we ranked the authors to reflect contribution differences according to HMU's 2013 official promotion system. This system uses the following formulae: first author or corresponding author = IF \* 100%; second author = IF \* 50%; and third or later author = IF \* 25% \[[@CR9]\]. We compared the publishing histories of the students of returning professionals with those of the students of resident professionals to estimate the impact of advisors' overseas experience on their students' research capacity. Other relevant advisor information was included, such as total IF, number of articles published while abroad, duration of overseas study, and age at travel abroad. We tried to find out the correlations between these overt factors and students' scientific research ability and identify which factor should be considered most for students when choosing advisors. The following data were collected from 257 students of returning professionals: (1) student's and advisor's names; (2) student's and advisor's ages; (3) student's and advisor's sexes; (4) student's enrollment year (Grade); (5) student's school and department; (6) student's total IF for articles published during their PhD study; (7) student's number of articles published during their PhD study; (8) advisor's duration of study or training abroad (months); (9) advisor's age when they went abroad; (10) advisor's total IF for articles published while abroad; and (11) advisor's number of articles published while abroad. We used SPSS, Version 19 (IBM Corp., Armonk, NY) to estimate multiple linear regression models to identify the factors associated with student's IF for articles and with the number of articles students published during their PhD study. To explore the relationship between health professionals' international experience and the academic output of their students, we selected 257 age-, enrollment year-, and specialty-matched students of resident professionals (Group B)---106 scientific research students from HMU (Group B~1~) and 151 clinical medicine students from affiliated hospitals (Group B~2~). We compared the total IF and the number of articles published between the returning and resident groups. We selected the control group (Group B) based on exact matches for age, year, and specialty with Group A participants. And there were 78 students of Group A who did not have a match. In total, there were 694 students enrolled in 2012--2014, 335 of whom were students of returning professionals. To avoid arbitrarily selecting matching controls, when there was more than one match, we randomly selected one control, without considering student's research backgrounds and advisor's name. We replaced study cases for whom we could not find an exact match with another student who had at least one match. We distributed the same questionnaires to the 257 control students of resident professionals, and questions regarding the advisor's information of studying abroad were not filled. Results {#Sec3} ======= In this study, the questionnaire response rate is 100%. Of the 257 students of returning professionals (Table [1](#Tab1){ref-type="table"}, Group A), there were 82 students in the 2012 graduating class, 87 in the 2013 class, and 88 in the 2014 class. Students' ages ranged from 26 to 34 years (mean = 31.29 years). In Group A, there were 106 scientific research students from HMU (Group A~1~: 37 students in the 2012 class, 32 students in the 2013 class, and 37 students in the 2014 class), with an age range of 26--34 years (mean = 31.32 years), and 151 clinical medicine students from affiliated hospitals (Group A~2~: 45 students in the 2012 class, 55 students in the 2013 class, and 51 students in the 2014 class), with an age range of 26--34 years (mean, 31.26 years). Similarly, there were 257 age-, enrollment year-, and specialty-matched students of resident professionals with the same grade and age proportion as those in Group A (Table [2](#Tab2){ref-type="table"}, Group B), consisting of 106 scientific research students (Group B~1~: 37 students in the 2012 class, 32 students in the 2013 class, and 37 students in the 2014 class), with an age range of 26--34 years (mean = 31.32 years), and 151 clinical medicine students (Group B~2~: 45 students in the 2012 class, 55 students in the 2013 class, and 51 students in the 2014 class), with an age range of 26--34 years (mean = 31.26 years). Table 1Data of 257 students of returning professionals (Group A), including 106 scientific research students (Group A~1~) and 151 clinical medicine students (Group A~2~)MaximumMinimumMeanTotal IF of papers published during PhDGroup A 58.23Group A 0Group A 4.01Group A~1~ 58.23Group A~1~ 0Group A~1~ 4.17Group A~2~ 26.82Group A~2~ 0Group A~2~ 3.91Number of papers published during PhDGroup A 10Group A 0Group A 1.16Group A~1~ 10Group A~1~ 0Group A~1~ 1.15Group A~2~ 4Group A~2~ 0Group A~2~ 1.16Tutor's duration of studying abroad (months)Group A 192Group A 6Group A 38.54Group A~1~ 192Group A~1~ 6Group A~1~ 42.16Group A~2~ 120Group A~2~ 12Group A~2~ 36Tutor's age of going abroadGroup A 46Group A 30Group A 36.39Group A~1~ 46Group A~1~ 32Group A~1~ 36.51Group A~2~ 45Group A~2~ 30Group A~2~ 36.30Tutor's total IF of papers published while abroadGroup A 94.69Group A 0Group A 7.59Group A~1~ 94.69Group A~1~ 0Group A~1~ 11.35Group A~2~ 60.22Group A~2~ 0Group A~2~ 4.94Tutor's number of papers published while abroadGroup A 30Group A 0Group A 2.58Group A~1~ 30Group A~1~ 0Group A~1~ 4.17Group A~2~ 13Group A~2~ 0Group A~2~ 1.46 Table 2Data of 257 students of resident professionals (Group B), including 106 scientific research students (Group B~1~) and 151 clinical medicine students (Group B~2~)MaximumMinimumMeanTotal IF of papers published during PhDGroup B 21.66Group B 0Group B 2.83Group B~1~ 21.66Group B~1~ 0Group B~1~ 2.71Group B~2~ 12.13Group B~2~ 0Group B~2~ 2.92Number of papers published during PhDGroup B 5Group B 0Group B 0.89Group B~1~ 5Group B~1~ 0Group B~1~ 0.76Group B~2~ 4Group B~2~ 0Group B~2~ 0.97 The IF for articles published by individual students of returning professionals during their PhD study (Table [1](#Tab1){ref-type="table"}, Group A) ranged from 0.00 to 58.23 (mean = 4.01), and the number of articles ranged from 0 to 10 (mean = 1.16). The total IF for articles published during PhD study in Group A was 1031.68, and the total number of articles was 297. The total IF for articles published during PhD study was 441.61 in Group A~1~ and 590.07 in Group A~2~. The number of articles was 122 for Group A~1~ and 175 for Group A~2~. In contrast, the individual IF for resident professionals (Table [2](#Tab2){ref-type="table"}, Group B) ranged from 0.00 to 21.66 (mean = 2.83), and the number of articles ranged from 0 to 5 (mean = 0.89). The total IF for articles published in Group B was 727.65, and the total number of articles was 228. A wide gap was observed between Group A and Group B for both total IF and number of articles. The total IF for articles published was 286.79 in Group B~1~ and 440.86 in Group B~2~. The number of articles was 81 in Group B~1~ and 147 in Group B~2~. We compared the mean scores using a *t*-test for three paired samples (Group A and Group B, Group A~1~ and Group B~1~, and Group A~2~ and Group B~2~; Fig. [1](#Fig1){ref-type="fig"}). We found statistically significant differences between Group A and Group B for total IF (*P* = 0.001) and total number of articles (*P* = 0.040). There were significant differences in the mean scores for total IF (*P* = 0.040) and the total number of articles (*P* \< 0.001) between Group A~1~ and Group B~1~. Group A~2~ and Group B~2~ had significantly different mean scores for total IF (*P* = 0.009), but the difference between these groups for the total number of articles (*P* = 0.061) was not statistically significant. Fig. 1Total IF and number of papers of returning professionals' students and resident professionals' students during their PhD. **a** Total IF of returning professionals' students and resident professionals' students. **b** Number of papers of returning professionals' students and resident professionals' students. Group A/A~1~/A~2~: returning professionals' students; Group B/B~1~/B~2~: resident professionals' students. For 3 paired samples (Group A and Group B, Group A~1~ and Group B~1~ and Group A~2~ and Group B~2~) statistical differences were found for total IF of papers (\*\**P* \< 0.01, \**P* \< 0.05) using a t -test. But for number of papers statistical differences (\*\**P* \< 0.01, \**P* \< 0.05) were only found for 2 paired samples (Group A and Group B and Group A~1~ and Group B~1~) using a t -test Given the results of our multiple linear regression analysis of all 257 students of returning professionals (Table [1](#Tab1){ref-type="table"}), we concluded that the IF earned during a student's PhD study was not related to their advisor's study-abroad duration (*P* = 0.719), their advisor's age when they studied abroad (*P* = 0.088), or their advisor's number of articles published while abroad (*P* = 0.843). However, the total IF for articles published by the 151 clinical medicine students of returning professionals during their PhD study was positively correlated with their advisor's total IF for articles published while abroad (*P* = 0.019). A strong linear association was observed between the total IF for articles published by the clinical medicine PhD students of returning professionals and the total IF for their advisor's articles published while abroad. However, for the full sample, the number of articles published by returning professionals' students during their PhD study was not associated with their advisor's study-abroad duration (*P* = 0.883), their advisor's age when they studied abroad (*P* = 0.310), their advisor's total IF for articles published while abroad (*P* = 0.408), or their advisor's number of articles published while abroad (*P* = 0.880). Discussion {#Sec4} ========== Engagement in international training experiences provides significant benefits to health professionals, including an appreciation of cultural diversity, the capacity to adapt to societal change, knowledge of alternative approaches to health and disease, and an understanding of public health and its implications for underserved populations. It is often assumed that a health professional's study-abroad experience has a positive influence on the scientific research capacity of their students; however, such an influence had not previously been evaluated in China. Our study revealed that, for both scientific research students and clinical medicine students, there is a wide gap between the research abilities of students of returning professionals and students of resident professionals. These data strongly suggest that not only scientific research students but also clinical medicine students can benefit from their advisors' international experience, presumably because their research capacity is indirectly elevated by their advisors' broadened horizons. These findings suggest that advisors with overseas experience will cultivate more capable medical students. How can China promote this positive influence? There is no precedent for establishing legislation or other official measures that incentivize medical trainees and professionals to gain research experience overseas \[[@CR10]\]. It is important for clinical faculty, administrators, and government officials engaged in medical education to integrate resources into training programs that support advisors' overseas training for the purpose of improving medical students' research capacity. We found a significant relationship between the total IF for articles published by the clinical medicine PhD students of returning professionals and the total IF for advisor's articles published while abroad. Furthermore, our previous work has shown that this relationship was tied to the advisor's duration of study abroad but not to their age when they studied abroad \[[@CR8]\]. In contrast, the total IF for articles published by the 106 scientific research students during their PhD study was not associated with their advisor's total IF for articles published while abroad (*P* \> 0.05). On the basis of our findings, we conclude that medical education administrators and the government could encourage clinical professionals to study overseas and to prolong the duration of their study abroad to a minimum of 24 months to improve the IF of their work while overseas. For example, the China Scholarship Council, which currently requires traveling scholars to cover their own living and tuition expenses for prolonged overseas stays, could consider offering additional support for up to 2 years to the most outstanding scholars. Another conclusion we drew from our findings is that medical students should consider their potential advisors' overseas experience when choosing a mentor. Advisors with overseas experience tend to cultivate medical students with higher research ability. Health professionals cannot produce articles worthy of a high IF without state-of-the-art laboratory equipment and facilities. Health professionals studying at overseas institutions have access to fully equipped and up-to-date laboratories that allow them to engage with advanced technology and move their research ideas forward. After spending time overseas, these professionals import these experiences and skills upon returning to China, and they incorporate them into their research and teaching. In contrast, the innovative thinking and research capacity of resident professionals is hampered by the limited infrastructure in China, possibly leading to resident professionals not cultivating an equivalent research capacity in their students, compared with returning professionals. During our research with returning professionals and their students, many reported finding more innovative conditions and attitudes abroad, compared with Chinese institutions. This seemed to be the main factor that motivated them to study abroad. This difference is not only because of poor research conditions in China, but also because of insufficient investment in medical research. We argue that officials at higher levels of the political hierarchy should take bold steps to improve the national research capability and to cultivate more innovation among medical experts. Besides, in the near future, we are going to do further research on how international experiences of advisors in English language countries versus non-English language countries, and low-resource environments versus high-resource environments, would impact the academic output of their students in order to find out the optimization training model for health professionals in China. Study limitations {#Sec5} ----------------- Our work is an initial effort to better understand the impact of health professionals' study-abroad experience on their PhD students' research productivity, using medical institutions in Harbin as a case study. We were only able to recruit a small sample from a single medical school and its four affiliated hospitals; therefore, we cannot generalize our findings to the entire medical education situation in China. Additionally, because authors make non-uniform decisions about assigning authorship for research contributions and there is no international standard for ranking authors in a universal manner, we ranked authors using HMU's promotion system (2013); however, this system cannot be applied to SCI articles worldwide. Furthermore, professionals who undertook 6 months or more of international training are likely to have high-level administrative titles, so more candidates apply for PhD positions with these professionals. Only one or two students can obtain PhD positions with each professional per year, so the students of these professionals are likely to be particularly competitive among their peers. Consequently, these students may have higher capability and motivation at baseline, when they enter the PhD program, so comparisons of students of returning and resident professionals may be biased unless the students' capability to conduct research and write articles at baseline can be quantified and matched. Furthermore, professionals who are selected for overseas training may be more accomplished because of prior experience or articles, which may explain their selection for this type of training. This selection may therefore be a "marker" for their capabilities, which may have remained the same if they had not gone overseas, meaning that they would still have had a positive benefit on their students without overseas training. In addition, the data used in this study were collected using self-administered questionnaires, which may have omissions or inaccuracies, compared with independent observation of publications in the literature. Conclusions {#Sec6} =========== This study suggests that the research ability of medical students in China is higher among those whose advisors studied abroad because the total IF for articles published by the clinical medicine PhD students of returning professionals was associated with their advisors' total IF for articles published while abroad. Furthermore, returning clinical professionals can improve their total IF while overseas by prolonging their study-abroad duration, which will, in turn, benefit their future students. We suggest that medical students consider potential advisors' overseas backgrounds when choosing a mentor. Finally, we identified a strong need to upgrade research facilities and to increase investment in medical research in China. Supplementary information ========================= {#Sec7} **Additional file 1: Table S1.** Questionnaire which was administered to 257 students of 'returning' professionals. HMU : Harbin Medical University IF : Impact factor SCI : Science Citation Index **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s12909-019-1853-y. Not applicable. TJL analyzed and interpreted the data. TTZ and NC performed the collection of the data. LMZ edited the English text of a draft of this manuscript. TJL was a major contributor in writing the manuscript. All authors read and approved the final manuscript. This research was funded by the Medical Education Branch of the Chinese Medical Association (2016B-KY009) in the design of the study and collection, analysis, and interpretation of data and in writing the manuscript. The datasets used and analyzed for the current study are available from the corresponding author on reasonable request. We obtained ethical approval from the Harbin Medical University Research Centre's ethics committee. This work was carried out in accordance with the Declaration of Helsinki, including but not limited to the anonymity of participants being guaranteed and the informed consent of participants being obtained. All participants received written and verbal information about the research project before providing written consent to participate. Not applicable. The authors declare that they have no competing interests.

All relevant data are within the manuscript and its Supporting Information files. 1 Introduction {#sec001} ============== The formation of stable aggregates is very common in nature. For example, long-range attraction through chemotaxis can lead to aggregation of Dictyostelium cells \[[@pone.0222371.ref001]\] or eukaryotic cells during development to form organs and blood vessels \[[@pone.0222371.ref002]\]). But the formation of aggregates can also arise from Brownian motion and contact adhesion. Numerous examples can be cited, from inert particles such as colloids \[[@pone.0222371.ref003]\] to living cells, but also in ecology where animals like mussels produce stable patterns by clustering \[[@pone.0222371.ref004]\]. It has been shown that the living entities can, through this process, optimize at the same time protection against predation and access to food. In cancer, tumor cells circulating in the blood stream form aggregates that will become a metastatic tumor when settling in an organ \[[@pone.0222371.ref005], [@pone.0222371.ref006]\]. The merging of metastatic lumps, forming a larger aggregate, can also occur \[[@pone.0222371.ref007], [@pone.0222371.ref008]\]. It is now recognized that cells cultured in 2D at the bottom of a plastic Petri dish do not behave as they would do in their natural environment. For example, in vitro, an organization in 3D-clusters makes the aggregates more resistant to treatments compared to the same cells plated in 2D, in a Petri dish \[[@pone.0222371.ref009]\]. Several factors can explain these different behaviors \[[@pone.0222371.ref010]\]: first, the fact that the dimensionality is not the same (2D versus 3D) is important; second, cell-plastic interactions are often very strong and prevail over cell-cell interactions; finally, the plastic dish has a very high stiffness, often non realistic (for brain cells for example). Therefore, new approaches that allow cells to grow in 3D aggregates *in vitro* are being pursued. Aggregates and spheroids *in vitro*, formed on non-adhesive substrates, are considered as pseudo-tumors that can be used to study tumor development in more realistic conditions. Recently, in the context of brain tumors (gliomas), we developed a new PEG-based hydrogel that allows the formation of a tumor-like structure, which can be used to study the effect of drugs in conditions more realistic than those of a 2D Petri dish. In some of those gels, we grafted poly(L-lysine) (PLL), because of its ability to promote unspecific cell adhesion via electrostatic interactions between the polyanionic cell surfaces and the polycationic layer of adsorbed polylysine. Addition of PLL in the gels allows us to modulated cell-substrate adhesion. The stiffness of the gel can also be tuned, in order to mimic the stiffness of the natural matrix (in our case, the brain) corresponding to a specific cell line. In \[[@pone.0222371.ref009]\], we chose to study the behavior of two glioblastoma (which is the most aggressive type of gliomas) cell lines. Glioblastomas are currently non-curable and the development of an *in vitro* system that could mimic the development of these tumors could be used in order to test new drugs or radiotherapeutic strategies. Compared to other tissues, the stiffness of the brain is low (lower than 1 kPa \[[@pone.0222371.ref011]\]), so we chose to design soft gels. We observed a significant difference in cell growth between PLL-containing (adhesive substrate) and PLL-free soft PEG hydrogels (non-adhesive substrate), showing the role of non-specific adhesion factors such as PLL in the migration, proliferation and aggregation in two glioblastoma cell line cultures. More precisely, we showed that on a non-adhesive substrate, the aggregates are larger and less numerous than on an adhesive substrate. The formation of aggregates has been studied theoretically with a perikinetic equation, first in the context of colloid aggregation \[[@pone.0222371.ref012]--[@pone.0222371.ref015]\]. In \[[@pone.0222371.ref016]\], the same concepts are used to study and fit the evolution of the number of cell aggregates on a non adherent substrate. The evolution of the mean projected aggregate area is more difficult to model, especially when aggregates are composed of cells that can deform, modulate their cell-cell adhesion and reorganize in 3D. For example, it has been observed that after their formation, cell aggregates often go through a compaction phase \[[@pone.0222371.ref010], [@pone.0222371.ref017], [@pone.0222371.ref018]\] that reduces their projected area. Other theoretical approaches consider the formation of aggregates under the point of view of phase separation: like two immiscible liquids, when mixed in a liquid medium cells move and seek a lower energy state through adhesion with other cells. The evolution of a system from a state where the concentration of particles is uniform to a final state where patterns appear is a spontaneous phase transition driven by motion of particules, the latter being either passive by diffusion (for example, in colloids), or active (as for mussels or cells) and adhesion. This phase separation corresponding to the formation of aggregates has been described by mean-field models, based on the Cahn-Hilliard \[[@pone.0222371.ref019], [@pone.0222371.ref020]\]. With chemotaxis, Keller-Segel equations can also be used to describe pattern formation \[[@pone.0222371.ref021]\]. In development biology, discrete approaches, in particular with cellular Potts models \[[@pone.0222371.ref022]\] have been used to model the formation of patterns or the segregation of two cell types in aggregates. However, to our knowledge, a model that describes the formation of aggregates from the early stages of a population of individual migrating cells to their aggregation and the late aggregate compaction, does not exist. For instance, in \[[@pone.0222371.ref016]\], the decrease of the mean aggregate area is modeled with an exponential function, but there is no direct connection with processes at the cellular level. In order to be able to describe the individual cells, an agent-based model should be chosen. One advantage of agent-based models is that one can easily implement the local rules of cell-cell interaction (\[[@pone.0222371.ref023]--[@pone.0222371.ref026]\], for a good review, see \[[@pone.0222371.ref027]\]). In \[[@pone.0222371.ref009]\], we presented the snapshots of already formed aggregates. Here, we add new experimental results by following the whole process, from the early stage where the cell population is composed only of individual cells, to their aggregation, and later to the compaction of the aggregates. We confirm the differential migration and aggregation of cells on the substrates with different adhesivity and for two different cell lines. We combine these experimental results with a theoretical study based on two models: first, we show that a spaceless model of perikinetic aggregation can reproduce the experimental evolution of the number of aggregates. Second, we developed a minimal off-lattice agent-based model, whose rules are defined in order to reproduce the important phenomena that drive the behavior of cell assemblies: cell and aggregate motion, cell-cell adhesion, cell proliferation and aggregate compaction. We show that this model reproduces very well the experimental temporal evolution of both the number of aggregates and their area, on adhesive and non-adhesive soft gels, for the two cell lines and that it gives access to quantitative values of three parameters. 2 Materials and methods {#sec002} ======================= 2.1 Preparation of the hydrogels and glioma cell lines {#sec003} ------------------------------------------------------ In \[[@pone.0222371.ref009]\], we showed that the PEG concentration of our artificial substrate optimal for the survival and growth of the two glioma cell lines is around 3% PEG. This concentration corresponds to an elastic modulus around 300 Pa, close to the value measured for brain tissue \[[@pone.0222371.ref028]\]. We use this concentration in all the following experiments. All the experimental methods can be found in \[[@pone.0222371.ref009]\]. Briefly: poly(L-lysine) hydrobromide (PLL-HBr 30,000 Da, Sigma-Aldrich, Saint-Quentin Fallavier, France) was first functionalized with an acrylate residue. Hydrogels were prepared from 3% (w/v) PEG-DA 6 kDa precursor (Sigma-Aldrich), dissolved in DPBS with 0.01% (w/v) of DMPA solubilized in VP. Precursor solutions were photopolymerized under UV (UV-LED LC-L1; Hamamatsu, 2 W/cm^2^, λ = 365 nm) for 40 s in homemade cylindrical dishes. Photopolymerized hydrogels were then incubated during 1 day in a high volume of DMEM for the hydrogel structure to be hydrated and thermodynamically stable before cell seeding. After this day of hydration and two rinsings with fresh medium, cells were seeded upon hydrogels at 2 10^5^ cells/well (or 10^6^ for the high-density experiments). To avoid cell medium acidification, the cell culture medium was replaced by fresh medium every day. The two glioma cell lines, F98 from rat model and U87-MG from human glioma, were provided by ATCC (CRL 2397 and HTB-14, respectively). Cells were maintained in Dulbecco's Modified Eagle's Medium (DMEM; Life Technologies, Courtabœuf, France) added with phenol red as pH indicator, supplemented with 4.5 g/L of D-glucose and pyruvate, 10% (v/v) fetal bovine serum (Life Technologies), and 1% (v/v) of penicillin and streptomycin antibiotics (Pen-Strep, Life Technologies). Cells were disposed into sterile culture flasks with anti-fungal filters to limit contamination and maintained in culture at 37°C, under a humidified atmosphere of 95% relative humidity with 5%CO~2~. Cells were replicated when they attained 80%--90% cell confluence. In what follows, we call "non adhesive gels" gels composed of PEG only and "adhesive gels" gels containing PEG and PLL. 2.2 Microscopy and image processing {#sec004} ----------------------------------- Microscopy image acquisition was performed using a 10x Nikon water-immersion objective placed on an Eclipse 80i Nikon microscope (Scop Pro, Marolles-en-Hurepoix, France) which was equipped with Differential Interference Contrast (DIC) device. The sample, the stage and the objective were completely enclosed in a chamber that allows the fine control of temperature, humidity and CO~2~ pressure on the living sample (Box and Cube system by Life Imaging Services, Basel, Switzerland). Cells were deposited on the gel surface and rapidly placed under the microscope. We performed time-lapse imaging (1 picture every minute) to follow the formation of the aggregates. Micrographs series were obtained using a Zyla 5.5 MPX Andor SCMOS cooled camera (Scop Pro, France) and MetaMorph acquisition software (Molecular Devices, Sunnyvale, CA). Image processing (from experiments and simulations) was performed with the Fiji software. Customized Fiji macros were developed to detect cells and aggregates in experiments and simulations. We used Python for further analysis of the data. More details are available in the [S1 Appendix](#pone.0222371.s001){ref-type="supplementary-material"}. 2.3 The spaceless model {#sec005} ----------------------- The number of aggregates during a perikinetic aggregation process can be predicted by the model of Smoluchowski \[[@pone.0222371.ref012]\]: $$\begin{array}{r} {\frac{dN_{k}}{dt}\left( t \right) = \frac{1}{2}\sum\limits_{i = 1}^{k - 1}K_{i,k - i}N_{i}\left( t \right)N_{k - i}\left( t \right) - \sum\limits_{i = 1}^{\infty}K_{ik}N_{i}\left( t \right)N_{k}\left( t \right)} \\ \end{array}$$ where *N*~*k*~ is the number of aggregates of size *k*, (*K*~*ij*~) is the kernel aggregation rate, where the element *K*~*ij*~ is the aggregation rate between clusters of size *i* and clusters of size *j*. We tried two different scenarii for the aggregation process. In the first scenario, all the aggregates move and interact with other aggregates with the same constant rate *K*. In this case, the model is solvable analytically and the evolution of the total number of aggregates at time *t*, *N*(*t*), is described by the equation: $\frac{dN}{dt}\left( t \right) = - KN\left( t \right)^{2}$, and thus *N*(*t*) = *N*~0~/(1 + *K N*~0~ *t*). This model was used in \[[@pone.0222371.ref016]\] to fit the evolution of aggregate number on a non-adhesive substrate. In the second scenario, only individual cells move and can interact with other individual cells or with aggregates (*K*~1*j*~ = *K*~*j*1~ = *K*~1*i*~ = *K*~*i*1~ = *K*~1~ ∀*i*, *j* and *K*~*ij*~ = 0 if *i* \> 1 and *j* \> 1). The total number of aggregates, *N*(*t*) is given by the following equations, that we solved numerically: $$\begin{array}{r} {\frac{dN}{dt} = K_{1}N_{1}\left( \frac{1}{2}N_{1} - N \right)} \\ \end{array}$$ $$\begin{array}{r} {\frac{dN_{1}}{dt} = - K_{1}c_{1}N} \\ \end{array}$$ $$\begin{array}{r} {\frac{dN_{2}}{dt} = K_{1}\left( - \frac{1}{2}{N_{1}}^{2} - N_{1}N_{2} \right)} \\ \end{array}$$ $$\begin{array}{r} {\frac{dN_{i}}{dt} = K_{1}N_{1}\left( N_{i - 1} - N_{i} \right)\text{for}\mspace{720mu} i \geq 3.} \\ \end{array}$$ where *N* is the total number of aggregates. 2.4 The spatial model {#sec006} --------------------- We define an agent-based model that involves a collection of agents evolving in a continuous 2D surface. Each agent is a cell, modeled by a disk. The disk radius is the same for all cells in a given simulation and its value stays constant during the simulation. The simulation space is lattice-free, i.e. the position of each cell is an ordered pair of real numbers corresponding to the coordinates of the center of the sphere. In experiments, the field of view is a rectangle of size 1280 per 1080 pixels (each pixel is a square of side length 0.658 *μ*m). In the simulations, we choose a square as 2D surface by simplicity. Since the experimental field of view represents a small part of the whole surface of the substrate, it is devoided of boundary effects; therefore, we choose periodic boundary conditions in the simulations (walls or other closed boundaries would induce strong artefacts). The unit of length in simulations is set so that the side of the square has the same length as the length of the region observed in experiments (842.24 *μ*m). At each iteration, all the cells are updated, one by one and in a random order in order to avoid undesirable correlations. At each iteration, each cell participates in the following processes: motion, both individual and collective, influenced by cell-cell adhesion and aggregate compaction, and proliferation. One iteration corresponds to one minute. For the sake of clarity, in the following, we call "individual cells" cells that are not part of an aggregate (individual cells have no neighbors). An aggregate is thus an assembly of at least two cells. 2.5 Rules of the spatial model {#sec007} ------------------------------ ### 2.5.1 Cell motion {#sec008} At each iteration, individual cells choose a direction uniformly at random and move by a step *a*~0~ in that direction. For a given simulation, this step length is constant during time and is the same for all the cells. Once cells are part of an aggregate, they continue to move, and their motion is the composition of two motions, the motion of each cell inside the aggregate and the motion of the whole aggregate. The diffusion coefficient of a spherical particle of size *r* in a viscous medium is proportional to 1/*r*; thus, for a 2D aggregate comprising *N* cells, it is proportional to $1/\sqrt{N}$. We keep the same dependency here: the motion of the whole aggregate is assumed to be random and the step of this motion is chosen so as to decrease as $1/\sqrt{N}$. Regarding the motion of each cells inside the aggregate, the length of the step is chosen as a decreasing function of the number of neighbors *n*: *a* = *a*~0~/(1 + *n*^2^). The choice of this function is purely phenomenological. Another choice could have been that of \[[@pone.0222371.ref029]\] where the probability of migration decreases with the number of neighbors, being proportional to (1 − *q*)^*n*^, where 0 \< *q* \< 1 is the adhesion parameter and *n* is the number of neighbors. It turns out that the precise choice is irrelevent, since the results do not depend on the exact dependence of the step length on the number of neighbors (for reasonable choices of the corresponding expression). In the non-adhesive case, individual cells are sometimes subject to a hydrodynamic flux (due to gel local heterogeneity) that can bias the cell motion: when the flux is "on" in the simulation, the direction of the motion in the simulation is limited to a half plane (defined by the direction of the flux), both for individual cells and aggregates. ### 2.5.2 Superimposition {#sec009} In order to model the fact that cells are deformable and can be organized in three dimensions in aggregates, cells are allowed to partially superimpose in the model. The maximum superimposition is quantified by a parameter *α*~max~ that is defined as the ratio of the overlapping length (i.e. the difference between the diameter of a cell and the minimum possible distance between two cell centers) to the diameter of a cell. If *α*~max~ is close to 1, two cells can superimpose almost completely, whereas if *α*~max~ is close to 0, the cells stay well-separated. The value of this parameter is constant during a given simulation but can vary from one simulation to the other, in order to model different cell lines. This parameter is similar to the stacking index used in \[[@pone.0222371.ref013]\] to indicate the formation of aggregates with some vertical stacking, forming multilayered clumps. ### 2.5.3 Cell-cell adhesion {#sec010} If during its motion, the position of a moving cell would break the superimposition rule, the motion is prematurely stopped and the two cells adhere to each other. There is no break-up mechanism and so cells cannot detach from their neighbors (except if its step leads it to a position where it has at least the same number of neighbors). Therefore, if the cell chooses a direction of motion that would lead to a smaller number of neighbors after performing the step, this step is canceled and the cell does not move. ### 2.5.4 Compaction {#sec011} We do not have precise details on what happens inside aggregates during compaction, so we chose to model the effect of cell compaction and 3D organization with a simple yet efficient empirical rule: we bias the individual cell motion in an aggregate towards the center of mass of the aggregate. Since this reorganization is much more visible for larger aggregates, we decided to modify accordingly the cell motion. When the aggregate is very small, its cells may move towards any direction. However, when the aggregate becomes more massive, the cell motion is biased towards the center of mass of the ensemble. At the limit of a very massive aggregate, the motion is possible only in a ± 90° sector around the line joining the cell with the center of mass. In practice, at each iteration and for each cell, a random angle for the cell motion is drawn in the angular sector ±90(1 + *e*^−*n*/40^) (in °) with *n* being the number of cells in the aggregate. The overall effect does not depend crucially on the precise form of the function. This bias in the direction is similar to the one in \[[@pone.0222371.ref030]\] where the direction choice is biased towards the direction with the higher number of cells within a distance corresponding to several cell diameters. ### 2.5.5 Proliferation {#sec012} We model the cell division by the following process: after the cell has moved it has a certain probability to try to proliferate. That probability per iteration is later on referred as "proliferation rate". If the cell division process is engaged, a random position around the dividing cell is chosen. If that position is compatible with the superimposition rule (meaning that the distance between the daughter cell and all the other cells should be larger than the minimum distance allowed by the superimposition coefficient), the daughter cell is created. If not, the division process is aborted. 2.6 Typical workflow {#sec013} -------------------- Simulations yield the positions of the cells, that are used to produce images representing the same observed area as in experiments. The size of individual cells in simulations is chosen equal to the mean size of cells in the experiments, so that the simulation results can be analyzed the same way as the experimental results, and we can then directly compare the dynamics of the mean area and of the numbers of aggregates in the two approaches. More details are available in the [S1 Appendix](#pone.0222371.s001){ref-type="supplementary-material"}. 3 Results {#sec014} ========= 3.1 Experimental observations {#sec015} ----------------------------- On [Fig 1](#pone.0222371.g001){ref-type="fig"} top and on the different movies of aggregation, we observed several phases in the aggregation process ([S1 Video](#pone.0222371.s005){ref-type="supplementary-material"} for F98 and [S2 Video](#pone.0222371.s006){ref-type="supplementary-material"} for U87). {#pone.0222371.g001} The first phase is the random motion of individual cells. Cells remain round, do not polarize, and do not adhere strongly to the substrate. However, they move, extend small filopodia, and perform a random motion (see [S1](#pone.0222371.s005){ref-type="supplementary-material"} and [S2](#pone.0222371.s006){ref-type="supplementary-material"} Videos). After analyzing the motion of 25 cells (F98) during one hour, we plotted the mean square distance covered versus time and we deduced that the cells have a diffusion coefficient of 1.6 ± 0.58*μ*m^2^ min^−1^ on the adhesive substrate (see [S3 Fig](#pone.0222371.s004){ref-type="supplementary-material"}), close to the value found in \[[@pone.0222371.ref031]\]. The second phase is the formation of aggregates: during this random motion, individual cells encounter other cells or already formed aggregates and new aggregates begin to form. When this happens, the individual cells stick to the other cells or to the aggregates; becoming unstuck is so rare that we neglected this phenomenon. The third phase corresponds to the dynamics of formed aggregates: they move as a whole, with small aggregates exhibiting a global motion larger than bigger ones. Moreover, the inside of the aggregates is also dynamic: the cells inside move and reorganize constantly (see [S1](#pone.0222371.s005){ref-type="supplementary-material"} and [S2](#pone.0222371.s006){ref-type="supplementary-material"} Videos). During the aggregation phase, which lasts around 2 hours, a few events of proliferation are visible among individual cells. This proliferation continues within aggregates, increasing their size continuously \[[@pone.0222371.ref009]\]. The last phase corresponds to the compaction of the already formed aggregates: a few hours after the formation of aggregates, they compact and reorganize into a three-dimensional shape. The projection of this shape in 2D is close to a disk. Experimentally, compaction occurs only for the F98 cell line. For the U87 cell line, this aggregate compaction does not occur and aggregates stay in a 2D configuration, see [Fig 1](#pone.0222371.g001){ref-type="fig"}. We measured, in the experimental field of view, the mean area of the aggregates and their number as a function of time, for adhesive and non-adhesive substrates, for the F98 and the U87 cell lines. We define the normalized number of aggregates as the raw number of aggregates divided by the number of individual cells at initial time. In [Fig 2](#pone.0222371.g002){ref-type="fig"}, the normalized number of aggregates (top) and the mean area of aggregates (bottom) are represented as a function of time, for non-adhesive (blue curves) and adhesive (brown curves) substrates, for the F98 cell line. The number of aggregates (respectively mean aggregate area) decreases (resp. increases) faster in the case of the non-adhesive substrate. {#pone.0222371.g002} 3.2 Comparison between experimental data and the spaceless model {#sec016} ---------------------------------------------------------------- We compared our experimental results with the theoretical model for aggregation developed by Smoluchowski \[[@pone.0222371.ref012]\]. In the case of a non-adhesive substrate, we found that the best fit was obtained with a constant aggregation kernel *K*~*ij*~ = *K*. From the fitting procedure, we found *K* = 2.6 10^−13^ m^2^s^−1^, see [Fig 1](#pone.0222371.g001){ref-type="fig"}, top (the dashed blue curve is obviously a better fit of the experimental data than the dotted blue curve). In the case of an adhesive substrate, we found that the best fit is obtained with the solution of the equations corresponding to the scenario where only individual cells can move and interact with the other aggregates. In this case, we found *K* = 6.4 10^−14^ m^2^s^−1^ = 3,8 *μ*m^2^min^−1^, see [Fig 2](#pone.0222371.g002){ref-type="fig"}, top (the dotted brown curve is obviously a better fit of the experimental data than the dashed brown curve). This value is of the same order of magnitude as the cell's diffusion coefficient on adhesive substrate calculated above. 3.3 Comparison between experimental data and the spatial model {#sec017} -------------------------------------------------------------- We model two cell lines: F98 cells (13.2 *μ*m mean diameter) and U87 cells (21.1 *μ*m mean diameter). ### 3.3.1 Qualitative comparison {#sec018} The rules of our model (sketched in [Fig 3(a)](#pone.0222371.g003){ref-type="fig"} have been defined in order to mimic what happens in the experiments: therefore, in the model, cells can move, adhere to other cells, form aggregates that can compact subsequently, and proliferate. {#pone.0222371.g003} In [S1 Video](#pone.0222371.s005){ref-type="supplementary-material"}, it is clear that cells move inside an aggregate, and since aggregates also have a motion on their own, the motion of each cell should be a composition of the two motions, see [Fig 3(a)](#pone.0222371.g003){ref-type="fig"}, left. It is well known that aggregation limited by diffusion only, without any surface tension, leads to clusters with fractal shape \[[@pone.0222371.ref014], [@pone.0222371.ref015], [@pone.0222371.ref032]\]. In our case the number of cells is not large enough to lead to fractals, but without any rule of compaction, especially in simulations with a high initial cell density, aggregates have the shape of long and branched thick filaments and do not organize themselves into more compact shapes. But in experiments at high cell densities, after about 12 hours after the beginning of the aggregation process, the aggregates compact and become more circular (in 2D) (see [Fig 4(a)](#pone.0222371.g004){ref-type="fig"}, middle). The effect of the compaction rule in the case of a high initial cell density is visible on [Fig 4](#pone.0222371.g004){ref-type="fig"} and leads to aggregates that have a compact shape, close to the experiments. {#pone.0222371.g004} The qualitative comparison of the experimental and the simulation results can be performed from [Fig 5](#pone.0222371.g005){ref-type="fig"}, which shows side by side an experiment and a typical simulation result, obtained with our model. {#pone.0222371.g005} 3.4 Quantitative comparison between experimental data and the results of the spatial model {#sec019} ------------------------------------------------------------------------------------------ The experimental results consist of 9 24-hour experiments: 6 experiments for the F98 cell line (3 in the adhesive and 3 in the non-adhesive condition), and 3 experiments for the U87 cell line (one in the adhesive and two in the non-adhesive condition). We decided to fit each experimental curve (not only the mean), so the parameters were independently set for each experiment. ### 3.4.1 Choice of parameters {#sec020} The number of initial cells *N*~0~ was deduced from the first image taken in each 24-hour experiment and was used as the initial condition of the corresponding simulation, a uniformly random distribution of *N*~0~ individual cells. The superimposition parameter was chosen constant for all the experiments with the same cell line. There is a net qualitative difference between the behavior of the two cell lines: the aggregates of F98 cells compact and clearly organize in three dimensions, whereas the U87 cell aggregates keep an almost two-dimensional organization. So the superimposition parameter *α*~max~ should be larger for the F98 than for the U87 cells, see Figs [1](#pone.0222371.g001){ref-type="fig"} and [3](#pone.0222371.g003){ref-type="fig"}. We can clearly detect when the parameter is too small, because then, the mean aggregate area in the simulations is too large, even for very small aggregates at the beginning of the experiment and even if the evolution of the number of aggregates is correct, see the red stars with *α*~max~ = 0.2 in [S1(c) Fig](#pone.0222371.s002){ref-type="supplementary-material"}. It is more difficult to detect when the parameter is too large: the difference between the cyan stars with *α*~max~ = 0.95 and the green stars with *α*~max~ = 0.7 in [S1(c) Fig](#pone.0222371.s002){ref-type="supplementary-material"} is not obvious. Suppose that *α*~max~ = 1, all the cells in an aggregate could in theory superimpose and the area of any aggregate could be reduced to the area of a single cell. But since the motion step length of cells diminishes with the number of neighbors in aggregates, this process takes a lot of time, and it is not possible to see the complete superimposition of all the cells in an aggregate, during the time of experiments. We thus decided to infer the value of *α*~max~ from images of superimposition of two cells, see [Fig 3(b)](#pone.0222371.g003){ref-type="fig"}. From these images, it is clear that F98 cells allow a minimal distance between the cell centers smaller than the U87 cells, and we chose the value of *α*~max~ = 0.7 for the F98 cell line and *α*~max~ = 0.2 for the U87 cell line. Two parameters still need to be set: the step length of individual cells and the proliferation rate. We determined the step length of individual cells so that the decrease of the number of aggregates corresponds to experimental data: if cells move too slowly, this number decreases also too slowly compared to experimental data (see [Fig 6](#pone.0222371.g006){ref-type="fig"}, cyan stars). If the step length is too large, the number of aggregates decreases too fast compared to experimental data (see [Fig 6](#pone.0222371.g006){ref-type="fig"}, red stars). The green stars correspond to the best value of the step. {#pone.0222371.g006} The proliferation rate was chosen so that the mean area in simulations would fit the corresponding experiment at large times (the increase in the aggregate area after formation is only due to proliferation), see [S1(b) Fig](#pone.0222371.s002){ref-type="supplementary-material"}. The value *κ* = 10^−4^ min^−1^ (cyan stars) is too small, the value *κ* = 1.4 10^−3^ min^−1^ is too large and the value *κ* = 7 10^−4^ min^−1^ (green stars) is correct. We added the bias of the flux if visible in the experiments. Without any flux, aggregates still move but their motion is very small and the distance between them is too large to allow any collision. When the flux on individual as well as on aggregates is on, that corresponds to the green stars in [S1(a) Fig](#pone.0222371.s002){ref-type="supplementary-material"}, collisions are possible between large aggregates and the number of aggregates decreases even at large times. We managed to reproduce the dynamics of both the mean area and the number of aggregates of each experiment and for both cell lines used. Results are shown in [Fig 7](#pone.0222371.g007){ref-type="fig"} for the F98 cell line and in [S2 Fig](#pone.0222371.s003){ref-type="supplementary-material"} for the U87 cell line and for both adhesion conditions. {#pone.0222371.g007} 4 Discussion {#sec021} ============ We present here a combination of experimental and simulation results on the behavior of a cell population on soft hydrogels. On these soft gels, cells stay round, move and stick to each other to create aggregates, within one day (this is much shorter than the formation of glioma cell aggregates on an adherent rigid substrate such as plastic, where the time scale is one month \[[@pone.0222371.ref033]\]). The shape and the size of these aggregates depend on the nature of the gels (adhesive or not adhesive), but also on the cell line: U87 aggregates are less cohesive than F98 aggregates, and for both cell lines, aggregates are smaller and more numerous on adhesive substrate (with PLL). First, we compared the experimental data with the solutions of perikinetic equations. We found that the experimental non-adhesive and adhesive cases correspond to two different scenarii: in the non-adhesive dynamics of aggregation, a constant kernel leads to a better agreement with the experimental data, whereas the adhesive case is well fitted by a kernel that is non-zero only for particles of size 1. We found the kernel value *K* = 2.6 10^−13^ m^2^s^−1^ in the non-adhesive case, and a four-time smaller value of *K*~1~ in the adhesive case *K*~1~ = 6.4 10^−14^ m^2^s^−1^. These values are consistent with other studies \[[@pone.0222371.ref031]\], where only the case of a constant kernel is compared to experimental data. The agreement with the theoretical spaceless model is fair, but the model describes only the evolution of the number of aggregates as a function of time, whereas in our experiments, the area was also recorded. We thus developed an agent-based model with simple rules that could reproduce as well the first stages of the experiments, when cells still move as individual cells, as the late stages where cells are in aggregates, and that could reproduce the experimental evolution of both the number and the mean area of the aggregates. To model all these stages, without describing precisely the shape of the cells, we estimated that a cellular Potts model was less adapted to our problem, compared to a classical agent-based cellular automaton. We introduced four rules: the motion rule (for individual cells, for cells inside an aggregate and for aggregates, in the presence or not of a flux), the superimposition rule, the proliferation rule and the compaction rule. The superimposition and the compaction rules may need further justifications: Since the precise 3D organization in aggregates concerns only the F98 cell line, and to follow the approach of \[[@pone.0222371.ref013]\] where a stacking index is defined, we decided to keep our model in 2D and introduce an effective parameter of superimposition, that describes the strength of cell-cell adhesion and their ability to organize in 3D. This approach has the advantage of simplicity, since only one parameter can resume the differences between the behavior of the two cell lines. The compaction that arises after the formation of cellular aggregates is a collective effect of a cell population. In experiments, compaction of aggregates is due to individual cells that deform and flatten their membranes against each other, increasing cell--cell contact and minimizing intercellular spaces. This process is similar to the embryo's compaction, see \[[@pone.0222371.ref034]\]. Compaction may also be due to the possible formation of supracellular stress cables, at the scale of the whole aggregate \[[@pone.0222371.ref035]\]. This made us define a compaction rule that is non-local: the cells' motion is biased towards the center of their aggregate and this bias increases with the size of the aggregate. Actually, the limitation of the motion is not severe: the maximum bias (in very big aggregates) restricts only the cell motion to a half plane towards the center of the aggregate. With this agent-based model, in 2D and with simple rules, we were able to reproduce the behavior of two cell lines, namely the evolution of the number of aggregates and of their projected area, on two different substrates, one adhesive (with PLL) and one not (without PLL). More importantly, by fitting the number of aggregates and the mean area of aggregates as a function of time, we were able to infer quantitatively several properties of the two cells lines, on the two substrates: their speed of motion, their proliferation rate, their superimposition coefficient and the capacity of aggregates to compact. First, our model allowed us to conclude that the effect of the presence of PLL in the gel (more adhesive substrate), for both cell lines, could be modeled as a simple slowing effect on cells. On non-adhesive hydrogels, there is often a flux (probably due to inhomogeneities of the gels) which gives to the cells and aggregates a motion bias (direction in only a half plane), that was taken into account in the model. We found that in the cellular automaton, in order to model the adhesive substrate, we had to decrease the step and remove the flux (i.e. remove the restriction of the motion to a half plane). For F98 cells, the mean speed motion of F98 is 4.7 ± 0.7 pixel length min^−1^ = 3.1 ± 0.4 *μ*m min^−1^ on non-adhesive substrates, whereas it is equal to 1.5 ± 0.3 pixel length min^−1^ = 1.0 ± 0.2 *μ*m min^−1^ on adhesive ones. For U87 cells the mean speed of motion are respectively 6 pixel lengths min^−1^ = 3.9 *μ*m min^−1^ and 2 pixel length min^−1^ = 1.3 *μ*m min^−1^. The surface density of PLL molecules estimated from the volume concentration of 0.001% (w/v) PLL in the hydrogel precursor solution for a hydrogel thickness of 2 mm, is about 5 10^11^ molecules mm^−2^. About 5 10^5^ PLL molecules are found every *μ*m^2^. The F98 and U87 MG cells have a radius between 8 and 20 *μ*m so they move on a quasi-homogeneous surface of PLL molecules. The cells make smaller steps on PLL hydrogels because they are constrained in their motion by the electrostatic interactions they form with the PLL. We also had to change the proliferation rate between the two cell lines. We found that the proliferation rate (in aggregates) is 7.10^−4^ min^−1^ for the F98 and 3.10^−4^ min^−1^ for the U87 cell line. Our results also reveal that the adhesive properties of the substrate do not impact the proliferation rate in aggregates: it is the same in the two conditions (adhesive and non-adhesive substrate), for each cell line. The U87 cells are characterized by a weak adhesion between cells, leading to loose aggregates, whereas the F98 cells are much more cohesive. Moreover, there is no late compaction of the aggregates and the aggregates stay in 2D instead of organizing in 3D as in the F98 case. The results for both lines could be reproduced: in order to describe the U87 cell line we had to use a superimposition parameter smaller (*α*~max~ = 0.2) than the one used for the F98 line (*α*~max~ = 0.7) and to remove the compaction rule. For two cell lines, we show here that by using our spatial model to fit the temporal evolution of the number of aggregates and their mean area, it is possible to infer the quantitative values of speed motion, proliferation and the qualitative abilities of cells to adhere to each other and to deform. It should be possible to study any cell line, providing that the gel stiffness is optimized for this cell line (indeed, carcinomas develop on a much stiffer substrate than gliomas, this was confirmed by a preliminary study of ours on the breast cancer cell line MCF7 which could not form aggregates on soft gels, dying rapidly). It has been shown for example that the cohesivity of aggregates (due to cell-cell adhesion) could be a clinically important parameter, since it seems to be inversely proportional to the *in vitro* invasive potential \[[@pone.0222371.ref036]\]. One promising direction of research is that of the study of cell lines from cancers which are known to develop metastases, such as breast cancer. In that case, our experimental technique should be adapted. Another interesting direction of future studies would be to modulate the adhesion. This could not be done in the present studies using PLL since a higher concentration of this molecule becomes toxic for the cells. Using an other adhesion molecule, such as RGD instead of PLL, we expect to be able to vary adhesion and study the aggregation phenomena for a wide range of values of the latter. We expect to return to address these problems, both experimentally and through modeling, in some future work of ours. Supporting information {#sec022} ====================== ###### Additionnal information about the FIJI macro and the workflow. (PDF) ###### Click here for additional data file. ###### Parameter choice for F98 cell line. Experimental data points (one of the experiment on a non-adhesive substrate) are represented by green triangles (F98 cell line). (a) Flux choice: no flux (cyan stars), flux on individual cell (red stars) and flux on individual cells and small aggregates (green stars). All other parameters are fixed: *a*~0~ = 6 pixel, *α*~max~ = 0.7, *κ* = 7 10^−4^ min^−1^, *N*~0~ = 400. Error bars are chosen equal to the standard deviation over 20 simulations. (b) Proliferation choice: cyan stars, *κ* = 10^−4^ min^−1^; green stars, *κ* = 7 10^−4^ min^−1^); red stars, *κ* = 1.4 10^−3^ min^−1^. All other parameters are fixed: *a*~0~ = 6 pixel length, *α*~max~ = 0.7, *N*~0~ = 400, flux is on. (c) Superimposition parameter choice: cyan stars, *α*~max~ = 0.95; green stars, *α*~max~ = 0.7; red stars, *α*~max~ = 0.2. All other parameters are fixed: *a*~0~ = 6 pixel length, *κ* = 7 10^−4^ min^−1^, *N*~0~ = 400, flux is on. In all the simulations, error bars represent the standard deviation over 20 simulations. (EPS) ###### Click here for additional data file. ###### Experiments and simulations for the U87 cell line. Brown curves are the adhesive case, blue are for the non-adhesive case, triangles are for adhesive experiment and circles are for non adhesive experiment, stars are for simulations. The upper part gives the evolution during time of the number of aggregates while the bottom part gives the mean area of aggregates over time. Simulation parameters: *a*~0~ = 6 pixel length, *N*~0~ = 900, flux is off (cyan); *a*~0~ = 6 pixel length, *N*~0~ = 625, flux is off (blue); *a*~0~ = 2 pixel length, *N*~0~ = 484, flux is off (brown). The parameters *α*~max~ and *κ* are common to all the simulations: *α*~max~ = 0.2, *κ* = 3 10^−4^ min^−1^. Error bars represent the standard deviation over 20 simulations. (EPS) ###### Click here for additional data file. ###### Diffusion coefficient on an adhesive substrate for the F98 cell line. The mean square displacement (msd) is plotted against time. Each point represent the mean over 25 cells. The error bars the standard deviation. (EPS) ###### Click here for additional data file. ###### Aggregation of F98 cells. The aggregation process for the F98 cell line, during the first 400 minutes. (AVI) ###### Click here for additional data file. ###### Aggregation of U87 cells. The aggregation process for the U87 cell line, during the first 600 minutes. (AVI) ###### Click here for additional data file. This work was supported through a grant from Région Ile-de-France ('DIM Problématiques transversales aux systèmes complexes ISC-2014-PME-003") and a grant from the GEFLUC association (association des Entreprises Françaises dans la Lutte contre le Cancer). [^1]: **Competing Interests:**The authors have declared that no competing interests exist.

INTRODUCTION {#s1} ============ Hepatocellular carcinoma (HCC) is the third leading cause of cancer death in the world with more than 788,000 death per year based on data presented in the World Health Organization (WHO). HCC is the most common liver malignancy with risk factors to the liver such as chronic viral hepatitis of B or C, exposure of alcohol abuse and aflatoxin toxins, and aberrant metabolic stress such as nonalcoholic steatohepatitis (NASH) and obesity \[[@R1]\]. Less than 50% of HCC patients were diagnosed at early stage and majority of them were not qualified for curable surgical resection owing to late stage of tumor \[[@R2]\]. The prognosis for untreated HCC patients has a poor average survival rate 6∼20 months and more than 50% of treated patients developed recurrence and metastasis within 5 years of therapy \[[@R3]\]. A multiple kinase inhibitor Sorafenib was firstly approved for systemic therapy of advanced HCC in 2007 with an average response rate of 2∼3% and extending patient survival for a few months \[[@R4]\]. Another multiple kinase inhibitor regorafenib showed 10∼20% response rate as second-line therapy for advanced HCC was approved by FDA in 2017 \[[@R5]\]. Both multiple kinase inhibitors showed anti-angiogenesis activity by targeting angiogenic and oncogenic receptor tyrosine kinases with modest improvement of HCC patient survival \[[@R6]\]. Moreover, with limited success on HCC cancer genome sequencing for common mutations as therapeutic targets for drug development, it is critical to develop other innovative strategies and novel targets and drugs to prolong survival of HCC patients \[[@R7]\]. Several lines of evidence already supported that aberrant translational machinery such as modulation of ribosome biogenesis, Akt1/mTOR signaling and translational initiation play critical roles in tumor progression \[[@R8], [@R9]\]. Translational initiation is the rate-limiting step of protein synthesis that could alter the overall gene expression or selectively enhance translation of oncogeneic mRNAs to impact cancer development and progression \[[@R10], [@R11]\]. EIF4E, the best-studied eukaryotic initiation factor (eIF), is frequently overexpressed in cancers and activates Akt1/mTOR signaling pathway to phosphorylate 4EBP1 (eukaryotic translation initiation factor 4E binding protein 1) for releasing EIF4E to bind to 5′ cap of mRNA to enable translation of oncogenic genes \[[@R12], [@R13]\]. EIF3 translation initiation complex is the largest eIF protein complex composed of 13 subunits from eIF3a to eIF3m and orchestrates formation and stability of 43S preinitiation complexes (PICs) for translational initiation \[[@R14]\]. Aberrant expression of eIF3 subunits were reported in many malignant tumors and play important roles during tumor progression \[[@R15], [@R16]\]. EIF3C also known as EIF3S8 or eIF3-p110 were found to be upregulated in neurofibromatosis type 2 (NF2), colon cancer, glioma, HCC and breast cancer. Silencing of EIF3C via knockdown or interacting with schwannomin could induce cell apoptosis and suppress cell proliferation and tumor growth \[[@R17]--[@R23]\]. Nevertheless, the underlining molecular mechanisms for upregulated EIF3C to mediate tumor progression and serve as therapeutic target to improve patient survival remain unclear. Extracellular vesicles were released as heterogeneous plasma membrane vesicles from majority of cell types into body fluids under normal or disease conditions for intercellular communication \[[@R24]\]. Based on their size and biogenesis, extracellular vesicles in diameter could be mainly divided into exosomes (30∼150 nm), microvesicles (or microparticles) (100 nm∼1 µm) and apoptotic bodies (\>1 µm) as cargos for transmitting proteins, lipids, coding RNAs, noncoding RNAs such as microRNAs or even DNA to target cells for inducing various signaling cascades \[[@R25]\]. Numerous studies have shown that cancer cells utilized extracellular vesicles to communicate with stroma cells in the tumor microenvironment similar to cytokines and growth factors VEGF, FGF, TGFβ and Wnt to sustain favorite microenvironment for tumor progression, angiogenesis and metastasis \[[@R26], [@R27]\]. As part of routing clinical liquid biopsy, extracellular vesicles especially nano-sized exosomes could carry various cargos such as proteins, lipids, mRNAs, and miRNAs, and be easily uptook by recipient cells are emerging for developing tools for diagnostic and therapeutic interventions \[[@R28], [@R29]\]. In this study, we explored the roles of EIF3C upregulation in the tumor progression of HCC. Rather than direct stimulation of tumorigenic features such as cell proliferation and cell migration, we found an interesting mechanism that expression of EIF3C in HCC cells increase secretion of exosomes to promote angiogenesis and tumorigenesis of HCC with cross-validations of markers in human HCC tumor samples. RESULTS {#s2} ======= EIF3C is overexpressed and associated with poor patient survival in multiple HCC cohorts {#s2_1} ---------------------------------------------------------------------------------------- We found that EIF3C is upregulated in HCC tumor samples in comparison with normal tissues in TCGA HCC dataset (Figure [1A](#F1){ref-type="fig"}). Higher expression of EIF3C in tumor samples is associated with poor patient survival in compared to that of lower expression HCC patients (Figure [1B](#F1){ref-type="fig"}). We validated EIF3C RNA expression in another HCC transcriptome dataset and found that EIF3C gradually increased expression to advanced stages and associated with poor patient survival of HCC patients \[[@R30]\] (Figure [1C](#F1){ref-type="fig"} and [1D](#F1){ref-type="fig"}). Consistently, we also detected higher expression of EIF3C protein level at advanced stages of HCC tissues and in association with poor patient survival by using immunohistochemistry (IHC) assays (Figure [1E--1G](#F1){ref-type="fig"}). {#F1} The increasing expression of EIF3C in HCC tissues prompted us to examine its oncogenic properties in HCC cells. We were disappointed that there is no alterations in cell proliferation and expression of tumor progression-related genes including HIF1A, TGFβ1 and VEGF at RNA levels but decrease of trans-well cell migration assays in HCC cells PLC5, SNU449 and Huh7 ([Supplementary Figure 1A--1C](#SD1){ref-type="supplementary-material"}). Overexpressed-EIF3C in HCC cells increased release of exosomes and enhanced angiogenesis *in vitro* and *in vivo* {#s2_2} ----------------------------------------------------------------------------------------------------------------- To investigate the roles of overexpressed EIF3C in HCC progression, we performed proteomic study of EIF3C protein complex with immunoprecipitation by using mass spectrometry. Interestingly, we found that EIF3C pulled down 1,738 proteins that participated in functions such as intracellular trafficking and secretion in the clusters of orthologous groups (COGs) and extracellular exosome in cellular component of gene ontology by DAVID Bioinformatics analysis ([Supplementary Figure 2A](#SD1){ref-type="supplementary-material"} and [Supplementary Table 1](#SD2){ref-type="supplementary-material"}). We hypothesized that EIF3C might enhance secretion of exosomes to promote tumor angiogenesis. We found that incubation of conditioned mediums collected from EIF3C-overexpressed HCC cells PLC5, SNU449 and Huh7 (Figure [2A](#F2){ref-type="fig"}) with HUVEC cells could enhance the tube formation of HUVEC in compared with that of mock control for *in vitro* angiogenesis assays (Figure [2B](#F2){ref-type="fig"}). We speculated that the conditioned mediums especially collected from EIF3C-expressed HCC cells might increase the release of vesicles to promote tube formation of HUVEC cells. Indeed, more vesicles ranged in exosome size were observed in EIF3C overexpressed PLC5 than that of mock control detected by electron microscopy (EM) (Figure [2C](#F2){ref-type="fig"}) \[[@R31]\] and nanoparticle tracking analysis (NTA) (Figure [2D](#F2){ref-type="fig"}) \[[@R32], [@R33]\]. Consistently, when incubated HUVEC cells with PKH26-labelled vesicles, more PKH26-labelled vesicles derived from EIF3C expressed PLC5 were uptook by HUVEC cells than that of mock control (Figure [2E](#F2){ref-type="fig"}). Moreover, PKH26 labelled vesicles of EIF3C-expressed PLC5 were uptook more than that of mock control in terms of fluorescent labelling intensity in various HCC epithelial cells (Huh7, SNU449 and PLC5) and fibroblasts (WI-38 and NIH3T3) (Figure [2E](#F2){ref-type="fig"} and [Supplementary Figure 2B--2C](#SD1){ref-type="supplementary-material"}) \[[@R34]\]. Together, our results demonstrated that EIF3C-overexpressed HCC cells could increase the release of extracellular exosomes and uptake by divergent cell types. {#F2} To confirm EIF3C expression increased release of exosomes in HCC cells, we detected protein expression of exosome biomarkers in exosomes including ALIX, TSG101, CD9, CD63 and CD81 as positive as well as GM130 as negative controls by Western blotting analysis \[[@R35]--[@R37]\]. To evaluate angiogenic functions of EIF3C-enhanced exosomes, we performed tubes formation of HUVEC cells *in vitro* and plugs assays *in vivo* for angiogenesis assays with PLC5 vesicles under treatments of exosome generation inhibitor GW4869 \[[@R38]\]. Our results demonstrated that EIF3C slightly increased expression of divergent exosome biomarkers in vesicles of HCC cells (Figure [2F](#F2){ref-type="fig"}), increased RNA concentration ratios in exosomes ([Supplementary Figure 2E](#SD1){ref-type="supplementary-material"}), enhanced tubes formation of HUVEC cells (Figure [2G](#F2){ref-type="fig"}) and promoted the formation of new blood vessels in the transplanted matrix gel plugs in nude mice (Figure [2H](#F2){ref-type="fig"} and [2I](#F2){ref-type="fig"}). In contrast, treatment of exosome generation inhibitor GW4869 in EIF3C expressed HCC vesicles diminished aforementioned exosome uptook, biomarker expression, and angiogenesis *in vitro* and *in vivo* (Figure [2E, 2G--2I](#F2){ref-type="fig"} and [Supplementary Figure 2B--2D](#SD1){ref-type="supplementary-material"}). We further performed exosome proteomic analysis for links of EIF3C-mediated angiogenic functions by using mass spectrometry and gene ontology analysis by DAVID Bioinformatics. Consistently, our results showed that the 204 EIF3C specific exosome containing proteins might participate in extracellular exosome in cellular component of gene ontology analysis and in VEGF, hypoxia and angiogenesis of BIOCARTA pathway ([Supplementary Figure 2F](#SD1){ref-type="supplementary-material"} and [Supplementary Table 2](#SD3){ref-type="supplementary-material"}). Together, our results demonstrated that EIF3C- overexpressed HCC cells could increase the release of extracellular exosomes to promote *in vitro* angiogenesis by tube formation assays. Inoculation of EIF3C-increased PLC5 exosomes with Huh7 enhanced HCC angiogenesis and tumorigenesis {#s2_3} -------------------------------------------------------------------------------------------------- In addition to enhance HCC angiogenesis, we examined the potential of EIF3C-increased exosomes participated in HCC tumorigenesis. Interestingly, we found that subcutaneous inoculation of mixture of EIF3C-enhanced PLC5 exosomes with Huh7 into nude mice promoted tumor growth (Figure [3A](#F3){ref-type="fig"} and [3B](#F3){ref-type="fig"}). The increased tumor masses was suppressed by treatment of exosome inhibitor GW4869 on the EIF3C enhanced exosomes in the inoculated mixture with Huh7 cells in the subcutaneous tumorigenesis assays. Moreover, we also found that the increased subcutaneous tumor growth of the mixture of EIF3C-enhanced exosomes with Huh7 also potentiated expression of EIF3C and CD31 endothelial marker by IHC assays (Figures [3C-E](#F3){ref-type="fig"}). Treatments with exosome inhibitor GW4869 suppressed expression of CD31 endothelial marker but not EIF3C expression in the subcutaneous tumors. {#F3} To validate the expression of EIF3C in promoting human HCC tumorigenesis and angiogenesis, we performed IHC assays for concordant expression of EIF3C and vessel biomarker CD34 in human HCC tumors. We found that upregulated-EIF3C expression is concordantly increased expression of CD34 endothelial marker in human HCC tumors (*P \<* 0.001 and *r =* 0.5321) (Figure [3F](#F3){ref-type="fig"} and [3G](#F3){ref-type="fig"}). Our results suggested that the increasing expression of EIF3C in HCC cells could promote HCC angiogenesis and tumorigenesis via increasing secretion of exosomes. S100A11 involved in EIF3C-enhanced angiogenesis and tumorigenesis of HCC {#s2_4} ------------------------------------------------------------------------ We found that two members of S100 family proteins S100A11 and S100P were identified in EIF3C exosome proteomic data ([Supplementary Table 2](#SD3){ref-type="supplementary-material"}). Since members of S100 family proteins are known to play critical roles in cancer progression and angiogenesis \[[@R39]--[@R41]\] and since recent quantitative proteomics of exosomes identified S100A11 is a carcinoma-related protein located in the HCC exosomes \[[@R42]\], we speculated that expression of S100A11 might participate in EIF3C-mediated exosomes to promote HCC angiogenesis and tumorigenesis. Indeed, our results demonstrated that EIF3C upregulated expression of S100A11 in HCC cells PLC5 and SNU449 by Western blotting analysis (Figure [4A](#F4){ref-type="fig"}). After demonstration of knockdown efficiency of three S100A11 shRNAs by Western blotting analysis (Figure [4B](#F4){ref-type="fig"}), we found exosomes collected from EIF3C-expressed HCC cells significantly reduced HUVEC tube formation of *in vitro* angiogenesis assays (Figure [4C](#F4){ref-type="fig"}). Importantly, simultaneous high RNA expression of EIF3C and S100A11 in human HCC tumors showed worst survival, and expression of either EIF3C or S100A11 showed poor survival in compared to low expression of both RNAs of HCC patients in TCGA HCC dataset (Figure [4D](#F4){ref-type="fig"}). We also validated concordant protein expression of EIF3C and S100A11 on HCC tumor tissues by IHC assays. Consistently, the concordant high expression of EIF3C and S100A11 in HCC tumors showed the poor survival in compared to longer survival of low expression of both proteins in human HCC patients (Figures [4E](#F4){ref-type="fig"} and [4F](#F4){ref-type="fig"}). {#F4} DISCUSSION {#s3} ========== To extend our understanding how aberrant translational machinery participated in tumorigenesis, we investigated the roles of upregulated-EIF3C in association with poor patient survival and served as theranostic target for improving HCC therapy. We found that EIF3C mediated tumor progression via increasing release of oncogenic exosomes to potentiate angiogenesis and tumorigenesis in tumor microenvironment. Treatment of exosome inhibitor GW4869 or suppression of S100A11 expression to diminish EIF3C-mediated HCC angiogenesis further suggested that up-regulated EIF3C expression in HCC is a theranostic target for HCC therapeutic interventions. Except EIF3E and EIF3F found down-regulated and conferred tumor suppressive activity in cancers \[[@R43], [@R44]\], majority of EIF3 members including EIF3A \[[@R45]\], EIF3B \[[@R46]\], EIF3C \[[@R47]\], EIF3D \[[@R48]\], EIF3H \[[@R49]\], EIF3I \[[@R50], [@R51]\] and EIF3M \[[@R52]\] were up-regulated and played important roles in tumor progression of multiple cancer types. Nevertheless, detail mechanisms of how the aberrant expression of individual EIF3s directly involved in the tumorigenic signaling pathways rather than simply accompanied with aberrant protein synthesis remain critical issues for elucidation. For examples, EIF3E also called *int-6* because of frequent integration by the mouse mammary tumor virus resulted in a truncated EIF3E protein that was shown to be oncogenic to induce mammary tumorigenesis \[[@R53], [@R54]\]. However, ectopic expression of full-length EIF3E did not cause cellular transformation in NIH3T3 cell \[[@R22]\]. EIF3F expression is frequently lower in 70∼100% of tumors in compared to the tumor tissues possibly due to frequent detection of loss of heterozygosity in cancer genome or interacting with and phosphorylated by cyclin-dependent kinase 11 (CDK11) during apoptosis resulted in suppression of protein translation during apoptosis \[[@R55], [@R56]\]. EIF3I can specifically upregulated and interact with phospho-Akt1 (Ser473) to prevent de-phosphorylation by phosphatase PP2A and to sustain oncogenic p-Akt1 activity to facilitate tumor progression \[[@R57]\]. Although we detected no increase of cell proliferation and decrease of cell migration in EIF3C-expressing HCC cells that are inconsistent to previous reports \[[@R17]--[@R19], [@R21], [@R23], [@R58]\], we revealed an EIF3C oncogenic mechanism via increasing release of exosomes in HCC cells to target surrounding HCC and stroma cells to promote HCC angiogenesis and tumorigenesis. With cross-validations of aberrant expression of EIF3C with endothelial markers for angiogenesis and simultaneous high expression of EIF3C and S100A11 in association with poor HCC patient survival in human HCC tissues, some interesting caveats are warrant to explore for better understanding the roles of EIF3C in HCC tumor angiogenesis and progression. First of all, with lines of evidence of supporting EIF3C-stimulated release of exosomes determined by vesicle size and markers, we suspected that the increasing EIF3C exosomes are due to alteration of protein and RNA contents to promote HCC angiogenesis and tumorigenesis. Studies of underline mechanisms of EIF3C-stimulating exosome biosynthesis might lead to future clinical applications. Secondly, although we focused mainly on assays of EIF3C-exosomes mediated angiogenesis, how these EIF3C-exosomes increased EIF3C expression in Huh7 subcutaneous tumors in nude mice to mimic the observation of EIF3C upregulation in human HCC tissues and how EIF3C exosome mediated angiogenesis without affecting expression of HIF1A and VEGF in HCC cells require additional mechanistic studies. Finally, we demonstrated that S100A11 participated in EIF3C-exosomes mediated HCC angiogenesis and simultaneous high expression of EIF3C and S100A11 is associated with poor HCC patient survival. Omics approaches to reveal other tumorigenic components such as proteins, RNAs and non-coding RNAs in EIF3C-exosomes could further dissect the EIF3C upregulation mediated tumor progression. Together, an oncogenic mechanism of enhancing exosome release by ectopic expression of EIF3C in HCC cells to promote HCC angiogenesis and tumorigenesis was revealed in this study. Although the detail molecular mechanisms for individual expression of EIF3C in tumor progression of HCC and other cancer types remain obscure, upregulated-EIF3C served as theranostic marker through treatment of exosome generation inhibitor GW4869 and suppression of S100A11 expression is firmly established. Future studies on using EIF3C as biomarker for anti-angiogenesis therapy in HCC and the combination therapy of anti-EIF3C exosomes with anti-angiogenesis therapies might be important for improvement of HCC intervention. EXPERIMENTAL PROCEDURES {#s4} ======================= Cell culture {#s4_1} ------------ Hepatocellular carcinoma cell lines: PLC/PRF/5(PLC5) and SNU-449 were purchased from ATCC; and Huh7 was obtained from Japanese Collection of Research Bioresources (JCRB) Cell Bank. Cells were all cultured in DMEM (Gibco 12800-058) containing 10% fetal bovine serum, 100 units/mL penicillin, 100 μg/mL streptomycin, and Non-Essential Amino Acids Solution (Gibco 11140-050). Human Umbilical Vein Endothelial Cells (HUVEC) were purchased from ScienCell and cultured in Endothelial Cell Medium (ScienCell). Chemicals and reagents {#s4_2} ---------------------- PKH26 membrane dye and exososme inhibitor GW4869 were purchased from Sigma-Aldrich. Trans-well chambers were purchased from Merck Millipore. PrestoBlue Cell Viability Reagent were purchased from Invitrogen. jetPEI^®^ transfection reagent were purchased from Polyplus-transfection^®^. EIF3C polyclonal antibodies were purchased from Sigma. S100A11,CD63,TSG101,CD31,CD34,CD81,GM130,HA, Actin polyclonal antibodies were purchased from GeneTex; CD9 and ALIX polyclonal antibodies were purchased from Abcam, and horseradish peroxidase-conjugated goat anti-rabbit and anti-mouse secondary antibody were ordered from System Biosciences. pcDNA5/FRT EIF3C plasmid was kindly given by Dr. Hershey JW. Exosome isolation and purification {#s4_3} ---------------------------------- PLC5, SNU-449 and Huh7 cells were cultured to 70% confluence in 10 cm dishes in DMEM medium supplemented with 10% FBS, and transfection for 16 hr, then culture cells in serum free medium for 2 hr to collect their supernatants. Supernatants followed by filtration through a 0.2 μm filter from Millipore. The filtered supernatants were then concentrated by centrifugation using an Amicon® Ultra (AU) filter (Millipore). The final concentrated crude exosomes is ready for downstream analysis except western blot. We cultured cells (2 × 10^7^) in three 10 cm dishes and concentrated supernatants to 700 ul to perform angiogenesis, TEM, uptake and plug assays in each sample in every test. These crude exosomes were used within one week. The filtered supernatants were then mixed with Total Exosome Isolation Reagent (Thermo) and incubated overnight at 2° C to 8° C. The precipitated exosomes were recovered by standard centrifugation at 10,000 × g for 60 min. The pellet was then ready for protein extraction and performed western blot. MASS spectrometry {#s4_4} ----------------- Proteins from Immuno-precipitation or exosomes were fractionated by one-dimensional electrophoresis followed by trypsin in-gel protein digestion and measure the peptide mass and peptide fragment mass by LC-ESI/MS/MS or 2D-LC-ESI/MS/MS and to identify the proteins with matching to database. Nanoparticle tracking analysis (NTA) {#s4_5} ------------------------------------ The number and size of nanoparticles was assessed using NanoSight NS300 nanoparticle characterization system by DKSH in Taiwan. Mice {#s4_6} ---- Male NOD.CB17-*Prkdcscid*/NcrCrlBltw mice were purchased from BioLASCO Taiwan Co., Ltd and used at 4--6 weeks of age. Animal experiments abided by the guidelines for animal care and use issued by Academeia Sinica SPF animal facility Transmission electron microscopy (TEM) {#s4_7} -------------------------------------- Purified exosomes were fixed in 4% paraformaldehyde (w/v) in 200 mM phosphate buffer (pH 7.4). Fixed exosomes were dripped onto Formvar carbon-coated 200 mesh copper grids and absorbed at room temperature (RT) for 10 min. Excess liquid was removed with filter paper. Adsorbed exosomes were negatively stained with 3% phosphotungstic acid at RT for 5 min, dried with an incandescent lamp for 2 min, and observed via TEM (FEI Teccnai G2 F20 TWIN). Images were obtained using a cooled slow CCD camera. Western blotting {#s4_8} ---------------- PLC5-, SNU-449- and Huh7-derived exosomes and cells were lysed in RIPA buffer containing protease inhibitors and proteins were resolved using SDS-PAGE. After transfer to a PVDF membrane, standard immunoblot analysis was performed. Membranes were incubated with primary antibodies HA (1:1000), Actin (1:1000), TSG101 (1:1000), CD63 (1:1000), EIF3C (1:1000), and S100A11(1:1000) at 4°C overnight in a buffer containing 5% skim milk, and then with a horseradish peroxidase (HRP)-conjugated secondary antibody at 37°C for 30 mins. Protein-band densities were analyzed quantitatively using Image Quant TL. Exosome fluorescent labeling and uptake assay {#s4_9} --------------------------------------------- The final concentrated crude exosomes were labeled using the green lipophilic fluorescent dye, PKH26 for 5 min. PKH26-labeled exosomes were pelleted at 100,000 g for 70 min, washed three times with PBS, and re-suspended in Endothelial Cell Growth Medium. HUVEC cells were incubated with labeled exosomes for 16 h. Images were obtained using an inverted fluorescence microscope (PE Ultra VIEW). Cell proliferation {#s4_10} ------------------ Cells were seeded on a new 96-well plate (5,000 cells/well) in DMEM medium with supplements. PrestoBlue^®^ was measured after 72 hr. Eight wells were used for each group at each representative test. Whole experiments were repeated three times. Cell migration {#s4_11} -------------- For the cell migration assays, cell culture was performed using 24-Well Millicell inserts (Millipore) according to the manufacturer's instructions. Cells (1 × 10^5^) in 300 μl serum-free medium were added to the upper chambers and cultured for 24 h. Non-migrating or non-invading cells were removed with cottons swabs. Cells migrated or invaded to the bottom of the membrane were stained with the cell stain buffer and counted under microscope and photographed. Three independent experiments were performed for the same conditions. Tube formation assays in µ-slide angiogenesis {#s4_12} --------------------------------------------- The slides were purchased from ibidi. Ten µl of Geltrex^®^ (Thermo) gel was applied to each inner well. For a final cell number of 10,000 cells per well, we adjusted a cell suspension of 2 × 10^5^ cells/ml. Then mixed exosome and ECM medium (1:1) thoroughly. Applied 50 µl cell suspension to each upper well. Immediately after seeding the cells, position the slide on an inverted microscope equipped with an incubation chamber and then start a time-lapse recording with 4× magnification and a time interval of 20 minutes in between the single images and data acquisition for 16 H (Model- Leica DMI 6000B). Plug assay {#s4_13} ---------- Mixed a total of 50 μl final concentrated crude exosomes with 250 μl ice-cold matrigel (the matrigel maintains as liquid form at 2--8° C and solidifies rapidly at 22--37° C). Subcutaneously injected the 300 μl matrigel mixture into a flank of five male NOD SCID mice (one injection site per mouse) with an ice-cold syringe and a 24 G one inch needle. After inoculation for 14 days, excised the matrigel, fixed with formalin overnight, embedded in paraffin, and sectioned onto slides. *In vivo* xenograft growth {#s4_14} -------------------------- 5 × 10^6^ Huh7 cells in 1:1 ratio of Exosome: Matrigel (Growth Factor Reduced; BD Biosciences) were injected subcutaneously in flanks of 8-week-old male nude mice. Tumors were resected and weighed after 4 weeks. Formalin-fixed and paraffin-embedded tissue samples were examined with IHC stain. IHC and scoring {#s4_15} --------------- IHC was performed on formalin-fixed, paraffin-embedded tissue sections. De-paraffinization, antigen retrieval and antigen-antibody reactions were performed using an automated DAKO Envision with Dual Link system-HRP. Tissue sections were incubated with primary antibodies followed by HRP-conjugated secondary antibody (from DAKO envision kit). Primary antibodies EIF3C (1:100), CD31 (1:00), CD34 (1:100), and S100A11 (1:100) were purchased from Sigma and GeneTex. Staining was developed using counterstained with haematoxylin and evaluated by pathologist. The scores all from the pathologist's visual estimation. We designed 4-grade EIF3C and S100A11 scoring criteria from 0 to 3. Score ranks usually lie in a range from "negative" to "positive", which describing different force of IHC expression in investigated groups, include: "negative" (Score 0), "weak" (Score 1), "moderate" (Score 2), "strong" (Score 3). IHC marker for CD34, were used to establish micro-vessel density (MVD). This parameter is often presented as a number of micro-vessels per square millimeter or mean value with standard deviations. The MVD was measured based on Weidner's method. Each positive endothelial cell cluster of immune-reactivity in contact with the selected field was counted as an individual vessel in addition to the morphologically identifiable vessels with a lumen. The intensity of the staining was scored as 0, 1, 2, 3, indicating absence of staining, weak, moderate, or strong intensity, respectively. Besides, we counted the absolute quantity of CD31 positively stained cells for each investigated in different experimental groups. Results in studies using this method to present as mean values of positively stained cells (and/or structures) among counted experimental groups. Real time PCR {#s4_16} ------------- For real time PCR, the total RNAs of PLC5, SNU-449 and Huh7 cells with overexpression of EIF3C or control were extracted by using TRIzol reagent (Invitrogen) and quantified by spectrophotometry (Nanodrop 2000; Thermo Scientific). The cDNA was synthesized using Revert Aid first strand cDNA synthesis kit (Thermo Scientific). The SYBR Green Supermix (Bio-Rad) was used for real time PCR on a 500 Fast & 7500 Real-Time PCR System (Thermo Scientific). Relative change in gene expression level was determined using the 2−ΔΔCt method. The primers were designed as follows: β-actin-Forward CTGGCACCCAGCACAATG, β-actin-Reversed CCGATCCACACGGAGTACTTG, VE GF-F GGGGGCAGAATCATCACGAA, VEGF-R GC AACGCGAGTCTGTGTTTT, TGF-β1-F CGTGGAG CTGTACCAGAAATA, TGF-β1-R TCCGGTGACATC AAAAGATAA, HIF1A-F CCAGTTACGTTCCTTCGATC AGT and HIF1A-R TTTGAGGACTTGCGCTTTCA. TCGA RNA {#s4_17} -------- The RNA sequencing result of HCC and corresponding clinical data from The Cancer Genome Atlas (TCGA) was acquired from TCGA Data Portal in Nov. 2015. Tumor parts and adjacent normal parts are annotated as TCGA indication. The RSEM expression value are output from level 3 RNAseqV2 result for further analysis. Statistical analysis {#s4_18} -------------------- Quantitative data are presented as means ± standard error of the mean (s.e.m.). Using GraphPad Prism 7 software to test for significant differences in qPCR, proliferation, tumor growth, and migration assays. Data were compared using Student's *t*-test, Log-rank (Mantel-Cox) test, Bartlett's test, or ANOVA test. *P \<* 0.05 was considered significant. SUPPLEMENTARY MATERIALS FIGURES AND TABLES {#s5} ========================================== **Author contributions** H.Y.L. and Y.S.J. designed and performed the research; H.Y.L. and C.K.C. performed IHC, IHC data interpretation and experimental suggestions; C.M.H., C.Y.C. and S.S.L. help on bioinformatics analysis; K.J.C. supports animal experiments; and H.Y.L. and Y.S.J. analyzed the data and wrote the paper. The authors declare no competing financial interests. We thank Common Equipment Core of IBMS and Academia Sinica including microscopy, EM, DNA sequencing, proteomics, pathology and Flow Cytometry for supporting our experiments. We would like to thank RNAi and EM core facilities of Academia Sinica for handling exosome data. **CONFLICTS OF INTEREST** No conflicts of interests to all authors. **FUNDING** Our work has been supported by grants from Academia Sinica and MOST \[106-0210-01-15-02\] and by the MOST, Taiwan with grant numbers MOST1012320B001029MY3, MOST1042320B001009MY3 and MOST1062321B001051.

Introduction {#Sec1} ============ Plate tectonics has reshaped the distribution of the Earth's continental and oceanic lithosphere through time. However, when modern-style plate tectonics started is still debated, with proposals ranging from the Neoproterozoic Era to the Archean Eon^[@CR1]--[@CR7]^. The absence from the geological record of high-pressure and low-temperature metamorphism, including blueschists and ophiolites before the Neoproterozoic Era, presents a challenge to understanding early subduction. Archean mantle is about 300 K hotter than the present mantle and high mantle temperature would have significant effects on the viability and style of the subduction process^[@CR8]^. As a result, early ocean plates may have been more buoyant and richer in MgO but weaker than modern ocean lithosphere. The Mg-rich composition of the Archean oceanic crusts may preclude blueschist formation^[@CR9]^. Popular models^[@CR10],[@CR11]^ for early plate tectonics involve shallow flat subduction and delamination of thick crust, precluding low thermal gradient deep subduction. This interpretation is supported by thermodynamic modeling of metamorphic mineral assemblages that the Archean tonalitic--trondhjemitic--granodioritic (TTG) compositions may be formed by shallow subduction of thick oceanic crust^[@CR12]^. The petrological record also shows that the trace element signatures in TTG magmas can only be produced when melts separate from a garnet amphibolite residue, not an eclogite residue^[@CR13]^. However, low Paleoproterozoic thermal gradients are recorded in blueschist facies metamorphism (\~350 °C GPa^−1^) in West Africa^[@CR3]^ and retrograded amphibolitized eclogite boudins (\~300 °C GPa^−1^) in North America^[@CR6]^. These recent findings challenge the timing of onset of modern-style subduction. A global record of low geothermal gradients in the Paleoproterozoic is a prerequisite to demonstrate a change of plate tectonic processes that dates back to that time. During the Paleoproterozoic Era, there were global-scale continent and mountain building events from West Africa to North America and to North China, leading to Columbia, the first coherent supercontinent in the Earth history^[@CR14],[@CR15]^. The North China craton (NCC, Fig. [1a](#Fig1){ref-type="fig"}) is one of the fundamental Precambrian nuclei of Asia and one of the oldest cratonic blocks in the world (\~3.8 Ga)^[@CR16]^. It preserves a record of a long and complex crustal evolution, but evidence of a thick mantle keel beneath Archean-aged crust is absent^[@CR17]^. The Eastern and Western blocks evolved independently during the Archean and collided along the Trans-North China Orogen (TNCO, Fig. [1b](#Fig1){ref-type="fig"}) during the Paleoproterozoic (\~1.85 Ga) to form the NCC^[@CR16]^. The appearance of Paleoproterozoic carbonatites in the NCC offers an opportunity to evaluate subduction--accretion collision tectonics during the Paleoproterozoic Era. Here, we report pressure--temperature (*P*--*T*) conditions retrieved from an eclogite xenolith in a Paleoproterozoic carbonatite that provides direct petrological evidence of cold oceanic subduction during the Paleoproterozoic Era.Fig. 1Schematic geological map of the North China craton (NCC) with the location of the eclogite xenoliths. The maps are after Zhao et al.^[@CR16]^. **a** The NCC is separated from Yangtze craton by Qinling orogen^[@CR16]^. **b** The Eastern and Western blocks in the NCC were amalgamated by the Paleoproterozoic Trans-North China Orogen (TNCO)^[@CR16]^. The Paleoproterozoic carbonatite and its associated eclogite xenoliths and high-pressure granulites^[@CR17]^ occur in the TNCO. The scale bar is 800 and 400 km in **a** and **b**, respectively Results {#Sec2} ======= Geological background {#Sec3} --------------------- The carbonatites are associated with pyroxenites, pyroxene syenites, and occasionally with granulites (Supplementary Fig. [1a, b](#MOESM1){ref-type="media"}). They occur as dykes and were emplaced in the Fengzhen and Huai'an areas at the crossing between the Khondalite belt and TNCO (Fig. [1b](#Fig1){ref-type="fig"}). These orogens are Paleoproterozoic collisional belts formed by the amalgamation of the Western and Eastern blocks and before that the Ordos and Yinshan blocks, respectively^[@CR16]^. The Khondalite belt is composed of Paleoproterozoic sillimanite--garnet and garnet--biotite gneisses. In the TNCO, late Archean crust is dominated by TTG gneisses. Paleoproterozoic mafic granulites are widely distributed (Fig. [1b](#Fig1){ref-type="fig"}). Their *P*--*T* paths are mostly clockwise, related to collision/exhumation processes, and some of them evolved through eclogite facies^[@CR17]^. The granulites, occurring as lenses and tabular bodies associated with the carbonatites, consist of plagioclase, clinopyroxene, orthopyroxene, amphibole, garnet, and quartz. They are characterized by augen structures consisting of plagioclase and orthopyroxene enveloped by porphyroblastic garnet, indicating rapid exhumation (Supplementary Fig. [1c](#MOESM1){ref-type="media"}). Guo et al.^[@CR18]^ obtained an age of \~1.82 Ga from the Huai'an granulite, with peak *P*--*T* conditions of 750--850 °C and 1.1--1.5 GPa. Mineralogy and geochronology {#Sec4} ---------------------------- The carbonatite is mainly composed of calcite (70--80%), with minor amounts of diopside, apatite, phlogopite, olivine, and spinel (Supplementary Fig. [2](#MOESM1){ref-type="media"} and Supplementary Table [1](#MOESM1){ref-type="media"}). Apatite occurs as well-developed elongate prismatic crystals (typically \~5 mm in length). In-situ LA-ICPMS U--Pb dating of apatite yields weighted U--Pb ages of 1681(±61) Ma (Fengzhen area) and 1765(±35) Ma (Huai'an area) (Supplementary Fig. [2](#MOESM1){ref-type="media"} and Supplementary Table [2](#MOESM1){ref-type="media"}). Within the Paleoproterozoic carbonatites, there are rare but well-preserved cm-sized eclogite xenoliths (Supplementary Fig. [1d](#MOESM1){ref-type="media"}). The eclogite is composed of euhedral garnet porphyroblast up to 1 cm in diameter (\~37 vol%), omphacite (\~36 vol%), kyanite (\~9 vol%), quartz (\~9 vol%), zoisite (\~7 vol%), and phengite (\~2 vol%), with accessory amphibole, biotite, and rutile (Supplementary Fig. [3](#MOESM1){ref-type="media"} and Supplementary Table [3](#MOESM1){ref-type="media"}). The mineral proportions were obtained from a SEM--EDS area-scan of one eclogite mount (Fig. [2a](#Fig2){ref-type="fig"}). Garnet has elevated Si (3.019 ± 0.005 pfu on the base of 12 oxygen atoms) and compositional traverses reveal prograde concentric zoning with a homogeneous core overgrown by a rim of increasing pyrope but decreasing almandine, showing a Mg/(Mg + Fe) ratio increasing from the core to the rim (Supplementary Fig. [4](#MOESM1){ref-type="media"}). The garnet core traps abundant mineral inclusions, including kyanite, quartz, zoisite, omphacite, Ca--Na amphibole, Na amphibole, paragonite, and rutile, whereas the rim is relatively free of inclusions (Fig. [2a](#Fig2){ref-type="fig"}). The jadeite content of omphacite in the matrix and inclusions is 23(±1) and 27(±3) mol%, respectively. Zoisite is poor in Fe^3+^. Subparallel zoisite grains in the matrix define a foliation that is characteristic of orogenic eclogite (Fig. [2a](#Fig2){ref-type="fig"}). Phengite has high Si contents (3.40--3.44 pfu). Large matrix phengite is texturally equilibrated with garnet and omphacite, and contains omphacite and kyanite inclusions, interpreted as peak metamorphic minerals. Minor magnesiohornblende occurs in the matrix, and has higher MgO (\~20 wt%) and lower Na~2~O (\~1.5 wt%) contents than inclusions in the garnet core. The Ca--Na amphibole in garnet is barroisite with elevated Na (Na~B~ = 0.5--0.7 pfu) and is rarely associated with paragonite. Rare sodic amphibole in garnet contains high Mg and Na, and low Ca (\~4.1, 1.2, and 0.01 pfu, respectively). Minor monazite occurs in association with calcite veinlets that crosscut the garnets and primary minerals in the garnet cores. In-situ electron microprobe dating of two groups of monazites (primary and late-stage growth) yields different weighted U--Th--Pb ages of 1839(±26) and 1766(±7) Ma (Supplementary Fig. [5](#MOESM1){ref-type="media"} and Supplementary Table [4](#MOESM1){ref-type="media"}). These ages constrain the eclogite metamorphism to the Paleoproterozoic Era, similar to the Huai'an granulite age in the TNCO^[@CR18]^.Fig. 2Mineral mapping of the fresh eclogite xenolith and *P*--*T* phase diagram. **a** The eclogite contains typical minerals of garnet (Grt) and omphacite (Omp), which do not show any retrogressed feature. The coarse-grained garnets are rimmed by quartz (Q) and phengite (Phen), indicating a relatively high pressure and low temperature metamorphism. Ky kyanite, Zo zoisite, Bt biotite, Rt rutile. The scale bar is 5 mm. **b** *P*--*T* diagram is calculated in the NCKFMASH system^[@CR27]^ according to the effective bulk-rock composition obtained from **a**. The peak pressure and temperature (labeled as MP1) are calculated using the Zr-in-rutile and mineral-pair thermobarometer (Supplementary Table [7](#MOESM1){ref-type="media"}). The peak metamorphic pressure and temperature, labeled as MP2, are independently constrained by the jadeite and Si contents of omphacite and phengite (purple dash line), respectively. The standard deviation (±1.4 kbar) in pressure is adopted on the basis of a Monte Carlo assessment on the phase equilibrium modeling^[@CR53]^. Combining MP1 and MP2 data gives a thermal gradient of 250(±15) °C GPa^−1^. Lws lawsonite, Ep epidote, Grt garnet, Jd jadeite, CPX clinopyroxene, Phen phengite, EC eclogite, PYT pyroxenite, BS blueschist, AMP amphibolite, HBL hornblendite Peak metamorphic *P* and *T* conditions of eclogites {#Sec5} ---------------------------------------------------- We derived the bulk rock composition of the eclogite xenolith by integrating the mineral compositions from electron microprobe analysis and the modes acquired by the SEM--EDS area-scan (Supplementary Table [5](#MOESM1){ref-type="media"}). The eclogite has a similar composition to oceanic gabbro (Supplementary Fig. [6](#MOESM1){ref-type="media"}), but is different from the eclogite xenoliths in kimberlites^[@CR19]^. The latter contains inclusion-free garnet and clinopyroxene, and no quartz. Based on in-situ mineral trace element analyses in the eclogite xenolith (Supplementary Table [6](#MOESM1){ref-type="media"}), the garnets have a strong depletion in the light rare earth elements (LREEs) relative to the heavy (H)REEs, and show a nearly flat chondrite-normalized (cn) HREE pattern with Y/Yb~cn~ of \~1 (Supplementary Fig. [7](#MOESM1){ref-type="media"}). In contrast, zoisite contains high LREE contents with a La/Yb~cn~ of \~266, and is characterized by a positive Eu anomaly. According to these mineral compositions and modal mass, the reconstructed whole-rock shows slightly depleted LREEs (La/Yb~cn~ = 0.7), flat HREE (Y/Yb~cn~ = 1), and positive Eu anomaly. The main hosts for the incompatible trace elements are zoisite \[LREEs, Sr (3574--3988 ppm), Th (2--10 ppm), U (1--2 ppm), Pb (10--11 ppm)\], garnet \[Y (13--31 ppm), HREEs\], and phengite \[Rb (183--201 ppm), Ba (2428--2564 ppm)\]. It is suggested that these trace-element budgets in eclogites are controlled by newly formed minerals such as zoisite, garnet, and phengite, during prograde metamorphism in subduction zones^[@CR20]^. To estimate the peak metamorphic conditions, we first use the Zr-in-rutile thermometer to determine the peak temperature. The Zr concentration in the matrix rutile ranges from 150 to 174 ppm (Supplementary Table [7](#MOESM1){ref-type="media"}). The Zr-in-rutile thermometer^[@CR21]^ yields temperatures of 650 and 670 °C for the minimum and maximum Zr concentrations, respectively (Supplementary Table [8](#MOESM1){ref-type="media"}). For the temperatures of 650 and 670 °C, the garnet--clinopyroxene--phengite (Grt--Cpx--Phen) barometer^[@CR22]^ gives peak pressures of 2.49 and 2.46 GPa, respectively (Supplementary Table [8](#MOESM1){ref-type="media"}). The Grt--Cpx--Phen barometer of ref. ^[@CR22]^ is preferred because of the low jadeite content (\<0.5) of omphacite in the mineral assemblage^[@CR23]^. Using the pressure-dependent Zr-in-rutile thermometer^[@CR24]^, the calculated temperature is 658 and 669 °C at 2.5 GPa for the minimum and maximum Zr concentrations, respectively (Supplementary Table [8](#MOESM1){ref-type="media"}), consistent with the result from the Grt--Cpx--Phen barometer. Further confirmation of the temperature determination comes from the garnet--clinopyroxene (Grt--Cpx) thermometer. Because grossular in the eclogitic garnet is less than 0.35, we use the formula of Powell^[@CR25]^ as suggested by ref. ^[@CR26]^. The compositions of garnet rim and omphacite in the matrix yield a mean temperature of 670 °C at 2.5 GPa. Multiple thermobarometers give consistent estimates of peak pressures (2.5 GPa) and temperatures (650--670 °C). For comparison, Fig. [2b](#Fig2){ref-type="fig"} shows a *P*--*T* phase diagram calculated for the eclogite in the simplified system Na~2~O--CaO--K~2~O--FeO--MgO--Al~2~O~3~--SiO~2~--H~2~O (NCKFMASH) with excess water, using the program THERMOCALC^[@CR27]^. The peak assemblage of garnet + omphacite + kyanite is stable in the *P*--*T* range of 2.6--2.8 GPa and 650--670 °C (Fig. [2b](#Fig2){ref-type="fig"}), based on isopleths of jadeite content in omphacite (22--24) and Si content in phengite (3.40--3.44). This is consistent with the thermobarometric estimates summarized above. These peak conditions correspond to a low apparent thermal gradient of 250(±15) °C GPa^−1^ (Fig. [3](#Fig3){ref-type="fig"}). This is among the coldest subduction recorded in the Paleoproterozoic Era and akin to many Phanerozoic blueschist terranes exhumed from oceanic subduction zones (\<350 °C GPa^−1^)\[[@CR2]\] (Fig. [3](#Fig3){ref-type="fig"}).Fig. 3Peak pressure--temperature condition of this study compared to a global compilation of metamorphic belts. The *P*--*T*--*t* data of these metamorphic belts are indicated by different symbols (see legend) and grouped into three types of metamorphism^[@CR2]^: high- and ultrahigh-pressure metamorphism (\<350 °C GPa^−1^); eclogite and high-pressure granulite metamorphism (\<750 °C GPa^−1^); and granulite and ultrahigh-temperature metamorphism (\>750 °C GPa^−1^). A Paleoproterozoic thermal gradient of \<350 °C GPa^−1^ was recorded in the high-pressure metamorphism from North China (NC), West Africa (WA)^[@CR3]^, and North America (NA)^[@CR6]^ cratons Isotope geochemistry of carbonatites {#Sec6} ------------------------------------ To understand the sources of carbon, we measured the oxygen and carbon isotopic compositions of the carbonatites (Supplementary Table [9](#MOESM1){ref-type="media"}). The δ^18^O~V-SMOW~ and δ^13^C~V-PDB~ values (9.4--15.8‰ and −5.7 to −1.6‰, respectively) are higher, compared to those of normal mantle rocks (5--8‰ and −7 to −5‰, respectively) and typical mantle-derived primary carbonatites (6--10‰ and −8 to −4‰, respectively^[@CR28]^) (Fig. [4a](#Fig4){ref-type="fig"}). While the oxygen isotopic values may be readily modified, the carbon isotopic values are much less susceptible to later processes like fluid infiltration and wall--rock assimilation. The high δ^13^C values recorded in the carbonatites potentially can be accounted for by fractional crystallization, or assimilation of sediments or subducted oceanic crust^[@CR28]^. Fractional crystallization is unlikely to have been the cause for the high δ^13^C values because there is no defined trend between the δ^13^C~V-PDB~ and δ^18^O~V-SMOW~ values. Similarly, only minimal wall--rock assimilation is possible based on a negative Pb anomaly documented in the whole rocks^[@CR29]^. In contrast, heavy-C enrichment in the carbonatites can be easily explained by the involvement of recycled oceanic carbonates, with typical δ^13^C~V-PDB~ values of 0(±1.5)‰^[@CR30]^ (Fig. [4a](#Fig4){ref-type="fig"}), associated with subducted oceanic crust.Fig. 4Isotopic compositions of the carbonatites. **a** The carbonatite samples from Fengzhen and Huai'an show higher oxygen and carbon isotopes compared to normal mantle-derived rocks (box with dashed line) and typical primary carbonatites (gray box with black line) worldwide. The values of the Paleoproterozoic sedimentary carbonates^[@CR30]^ are also shown to illustrate that heavy-C enrichment in the carbonatites is consistent with the involvement of recycled oceanic carbonates. **b** The initial ^143^Nd/^144^Nd and ^87^Rb/^86^Sr ratios of the carbonatites and apatites are compared to the values of the Archean to Proterozoic lithospheric mantle (LM)^[@CR31]^ and the late Archean lower crust in central NCC (LC)^[@CR32]^ with 700-Ma radiogenic addition. The Sr--Nd isotopic variations for the sources of the carbonatites and apatites can be successfully modeled by mixing a 2.5-Ga oceanic crust (OC) with various amounts of 2.5-Ga marine sediments (MS) (Supplementary Table [11](#MOESM1){ref-type="media"}). With a single-stage evolution model^[@CR54]^, the isotopic compositions of the two end-members are calculated as (^87^Rb/^86^Sr)~OC~ = 0.053 and (^147^Sm/^144^Nd)~OC~ = 0.214, and (^87^Rb/^86^Sr)~MS~ = 0.186 and (^147^Sm/^144^Nd)~MS~ = 0.183, respectively. The initial Sr--Nd isotopic compositions of the LC in the central NCC, and those for the bulk Earth (BE) and chondritic uniform reservoir (CHUR) are calculated for 1.8 Ga The isotopic characteristics of Sr and Nd in carbonatites are generally inherited from their source regions^[@CR28]^. In this case, the carbonatites and their apatites show very unusual negative *ε*~Nd~(*t*) values (−5.8 to −7.8) (Supplementary Table [10](#MOESM1){ref-type="media"}), indicating the involvement of a substantial amount of ocean crustal materials in the source. The initial Sr and Nd isotopes of the Archean to Proterozoic lithospheric mantle^[@CR31]^ and TTG crust in the central NCC^[@CR32]^ are very depleted, so that a 700-Ma radiogenic addition cannot produce such an enriched Sr--Nd isotopic signature for the carbonatites. One way to produce the unusually enriched isotopic characteristics of the carbonatites is to add oceanic crust-derived carbonates to a subducted Late Archean (2.5 Ga) basaltic oceanic crust, with a residence time of \~700 Mys. Figure [4b](#Fig4){ref-type="fig"} shows the change of the isotopic compositions of carbonatites by mixing a Late Archean oceanic crust with marine sediment. The isotopic composition with an addition of 10% oceanic crust-derived carbonates matches well with our observed values (Fig. [4b](#Fig4){ref-type="fig"}), providing further evidence for the involvement of the oceanic crust in the genesis of the carbonatites. Discussion {#Sec7} ========== On the basis of U--Th--Pb dating, compositions of the coexisting minerals in the eclogite xenolith, and isotopic compositions of C, O, Sr, and Nd, we have established that cold deep subduction of ocean crust occurred during the Paleoproterozoic Era in the NCC. The cold subduction environment is also consistent with the preservation of the Fe^3+^-rich majorite inclusion in a websterite-derived garnet in the same carbonatites^[@CR27]^. The estimated equilibrium temperatures of the host garnet are in the range of 670--1000 °C for depths of 100--200 km, using the garnet--clinopyroxene Fe--Mg-geothermometer^[@CR33]^. The high Fe^3+^ content (Fe^3+^/∑Fe \> 0.8) of the majorite inclusion is likely a product of a redox reaction involving deep carbonatitic magmas, implying that the Paleoproterozoic subduction in the NCC could have been active for an extended period of time. Reduction of carbonate associated with earlier subduction could have provided an environment in which oxidized Fe^3+^-rich majorite formed. From the regional geology, the low-temperature and high-pressure metamorphism of the carbonated eclogite is the result of the cold Paleoproterozoic oceanic subduction, related to the amalgamation of the Yinshan and Orodos blocks^[@CR16]^. The 1.91--1.88 Ga Halaqin volcanic belt in Yinshan block, located 100-km to the northwestward of the carbonatite outcrop, includes basalts, andesites, and rhyolites with typical geochemical features of an arc sequence^[@CR34]^. A recent study also showed that a Paleoproterozoic garnet websterite in the Yinshan block underwent peak metamorphism at \~1.90 Ga when it was subducted to eclogite facies, and then exhumed back to granulite facies at \~1.82 Ga^[@CR35]^. On a global scale, a semi-continuous record of global kimberlite volcanism starts at \~2 Ga, with a notable increase in kimberlite magmatic activity at 1.2--1.1 Ga^[@CR36]^. This implies that the cooling ambient mantle directly beneath thermally maturing continental keels could have developed into sites of sustained kimberlite melt generation from 2 Ga. Moreover, low thermal gradient Paleoproterozoic subduction has been recently reported in West African and North American cratons^[@CR3],[@CR6]^. Our work directly links Paleoproterozoic carbonatites to cold slab subduction in the NCC. Figure [5](#Fig5){ref-type="fig"} shows the distribution of Paleoproterozoic carbonatites and Paleoproterozoic granulites and eclogites in orogens worldwide. Multiple experimental studies show that melting of a carbonated slab at pressures greater than 2.5 GPa can form carbonate melt^[@CR37],[@CR38]^. There is a clear association between carbonatites^[@CR39]^ and high-pressure metamorphic rocks in Paleoproterozoic orogens^[@CR15]^.Fig. 5Distribution of Paleoproterozoic carbonatites, granulites, and eclogites worldwide. The majority of carbonatite occurrences^[@CR39]^ are associated with high-pressure metamorphism (Supplementary Table [12](#MOESM1){ref-type="media"}) in Paleoproterozoic orogens^[@CR15]^ Although only a few locations of Paleoproterozoic low thermal gradient metamorphism have been reported so far, it is likely that evidence for cold deep subduction in the Paleoproterozoic Era will continue to emerge with expanded exploration to metamorphic terrains, which are intrinsically linked to orogenesis. High-Mg oceanic crust derived from a hotter Archean mantle is less likely to stabilize blueschist-facies mineral assemblages^[@CR9]^. Moreover, the low-temperature records may be easily overprinted by the followed inter-continental collisions with high-temperature metamorphism worldwide. Therefore, we suggest a common occurrence of modern-style plate tectonics back to at least the Paleoproterozoic Era. The early global-scale subduction could be also associated with the formation of Columbia supercontinent^[@CR15]^. Global-scale Paleoproterozoic subduction would provide an efficient pathway for surface sediments to enter the ancient deep Earth. Isotopic and trace element geochemical studies of the ocean island basalts have for many years been used to infer the presence of long-lived (1--2 Ga) compositional heterogeneities in the deep mantle related to recycled oceanic crust together with different types of marine sediments via subduction^[@CR40]^. However, it is still doubted because of the lack of petrological evidence for ancient deep subduction. Our work not only provides a direct record of deep Paleoproterozoic subduction, but also implies that the recycled carbonates may be played a critical role in the change of local redox state of the deep mantle, including the formation of the Fe^3+^-rich majorite^[@CR33]^ related to the subducted sediments. Melting of the recycled carbonates can also produce melt with enriched trace element and isotope compositions, which are important mantle metasomatic agents that interact with peridotite lithologies^[@CR41]^. The Paleoproterozoic carbonatites in the Fengzhen and Huai'an areas of the NCC occur in an orogen setting. The eclogite xenoliths hosted by the carbonatites provide a rare window into the ancient deep-subducted slab, showing the first cold subduction in the NCC during the Paleoproterozoic Era and also recording the deepest Fe^3+^-rich majorite inclusion in a cold subduction environment^[@CR33]^. The characteristics of the Paleoproterozoic slab supports that modern-style plate tectonics has operated since at least the Paleoproterozoic Era. The association of Paleoproterozoic carbonatites with high-pressure metamorphic rocks in Paleoproterozoic orogens worldwide (Fig. [5](#Fig5){ref-type="fig"}) argues for sustained, large-scale deep carbon cycle over a long geological period of time, influencing the evolution of mantle oxidation state and compositional heterogeneities in the deep mantle. Methods {#Sec8} ======= Mineral mapping {#Sec9} --------------- The polished eclogite xenolith sample was analyzed using a Tescan Integrated Mineral Analyzer (TIMA) mineralogy system at the Tescan Orsay Holding Lab, Czech Republic. The TIMA comprises a Tescan Mira Schottky field emission scanning electron microscope with four silicon drift energy dispersive (EDS) detectors arranged at 90° intervals around the chamber. The measurements were performed with the dot-mapping mode, and the BSE image was obtained to determine individual particles and boundaries between distinct preliminary phases. A rectangular mesh of measurements on every distinct phase was obtained with X-ray spectra. The spectroscopic data were matched to mineral definition files, allowing for mineral identification and mapping. The volume and mass ratios of all mineral phases were automatically calculated. The TIMA measurements were performed at 25 kV using a spot size of \~50 nm, a working distance of 15 mm, and a field size set at 1500 μm. Apatite dating {#Sec10} -------------- The ages of apatites from the carbonatites were determined by the U--Th--Pb method using a laser (Geolas) ICPMS (Agilent 7500a) at the Institute of Geology and Geophysics, Chinese Academy of Sciences (CAS). The apatites, separated from the carbonatite samples, were casted into resin mounts together with grains of NIST 610, apatite reference materials NW-1 (1160 Ma), and LAP (474 Ma). The beam diameter of the laser is 30 μm with a repetition rate of 8 Hz. The background of Pb and ^202^Hg is less than 100 cps. The fractionation correction and results were calculated using the program GLITTER 4.0. The ^207^Pb method was applied for common Pb corrections using the upper intercept obtained from a Tera--Wasserburg diagram^[@CR42]^. The ^207^Pb correction method yields more precise and accurate ages than the Tera--Wasserburg Concordia ages because of a small spread in the U/total Pb ratios^[@CR42]^. The U--Pb analyses of the apatite standards also show that the method of laser ablation can effectively date this mineral without significant matrix effects^[@CR43]^. Monazite dating {#Sec11} --------------- The chemical Th--U--total Pb ages of monazite were determined with a wavelength-dispersive X-ray spectrometry (WDS) using a Cameca SX100 electron microprobe at the Masaryk University. Uranium was determined on the U Mβ line (counting time 60 s, detection limit 270 ppm), Thorium on the Th Mα line (counting time 40 s, detection limit 250 ppm), and lead on the Pb Mα line (counting time 240 s, detection limit 130 ppm). Synthetic and natural phases were used as standards, including metallic U for U, PbSe for Pb, CaTh(PO~4~)~2~ for Th, synthetic end-member phosphates (XPO~4~) for REE, Y, sanidine for Si, apatite for Ca, and topaz for F. Data were reduced using the PAP matrix correction routine. Overlapping of peaks and background positions were examined and chosen based on WDS angle scans on natural and synthetic REE-phases. The microprobe was operated at an accelerating voltage of 15 kV and a beam current of 160 nA, with an electron beam defocused to a 2-μm spot to avoid devolatilization, ionic diffusion, and other forms of beam damage on the sample. The monazite age was calculated using the method of refs. ^[@CR44],[@CR45]^. Major element analysis {#Sec12} ---------------------- The chemical compositions of minerals were obtained with a Cameca SX100 electron microprobe at the Masaryk University, and a JEOL field-emission electron microprobe (JEOL JXA 8530F) at the Geophysical Laboratory of the Carnegie Institution of Washington. The microprobes were operated at an accelerating voltage of 15 kV and a beam current of 30 nA. For each mineral or a group of minerals, we chose appropriate matrix-specific standards (both natural and synthetic) and optimal instrument conditions (beam settings, detector type, and counting statistics). In-situ trace element analysis {#Sec13} ------------------------------ In-situ laser-ablation ICPMS analyses of minerals in the eclogite xenoliths were performed at the School of Earth and Space Sciences, Peking University. The diameter of the ablation spot is 44 μm. NIST 610, 612, and 614 glasses were used as calibration standards for all the samples. The elements used for the internal standards include Ca and Si, expressed as CaO and SiO~2~ for garnet, omphacite, zoisite, kyanite, and phengite. Analytical error is ≤5% at the ppm level. In-run signal intensity for indicative trace elements was monitored during analysis to make sure that the laser beam stayed within the selected phase and did not penetrate inclusions. Trace and major element analysis of rutile {#Sec14} ------------------------------------------ We used electron probe microanalyzer (CAMECA SXFive) to analyze eight elements (Si, Fe, Cr, Zr, V, Nb, Ta, and Ti) in rutile at the Institute of Geology and Geophysics, CAS. The microprobe was operated at an accelerating voltage of 20 kV and a beam current of 170 nA. Counting time on the peak of Si, Fe, Cr, Zr, V, Nb, Ta, and Ti is 120, 120, 120, 100, 120, 240, and 10 s, respectively. The rutile R10 was used as standard. The detection limits for the above elements are 23, 35, 32, 35, 12, 54, 69, and 137 ppm, respectively. The eight elements in the standard rutile R10 have been well analyzed by the same electron microprobe^[@CR46]^. Isotope analysis {#Sec15} ---------------- Carbon and oxygen isotopic compositions of calcite from the carbonatites were measured at the Institute of Geochemistry, CAS, using a continuous-flow isotope ratio mass spectrometer (IsoPrime), with reproducibility of ±0.15‰ and ±0.2‰ for carbon and oxygen, respectively. The results are expressed as per mil (‰) variation relative to PDB for carbon and SMOW for oxygen. In-situ Sr--Nd isotopes of apatite from the carbonatites were determined by a Neptune LA-MC-ICPMS at the Institute of Geology and Geophysics, CAS. The laser was focused at a height of 60--120 µm above the sample surface and fired using a repetition rate of 6--8 Hz and an energy density of 10 J/cm^2^. The AP1 and AP2 standard apatites were used as external calibrations. Normalized Sr and Nd isotopic ratios were calculated using the exponential law^[@CR47]^. Sr and Nd isotopic compositions of the carbonatites were analyzed by a MC-ICPMS (VG AXIOM) at Peking University. Mass fractionation corrections for the Sr and Nd isotopic ratios were normalized to ^86^Sr/^88^Sr = 0.1194 and ^146^Nd/^144^Nd = 0.7219, respectively. Repeated measurements for the Nd standard JNdi and Sr standard NBS987 yielded ^143^Nd/^144^Nd = 0.512120 ± 11 (2*σ*) and ^87^Sr/^86^Sr = 0.710250 ± 11 (2*σ*), respectively. To calculate the initial Sr--Nd isotope values, we analyzed the whole-rock Rb, Sr, Sm, and Nd concentrations by solution ICPMS (Thermo Fisher Scientific X-Series II). The data processing procedure includes linear drift correction, internal (matrix) correction, REE and Ba interference corrections, blank subtraction, calibration with international standards, and a dilution correction. Repeat analyses (including well-characterized standards) indicate that the accuracy of trace element measurements is better than 10%. *P*--*T* pseudosections {#Sec16} ----------------------- The *P*--*T* phase diagram was calculated using THERMOCALC v. 3.33^[@CR27]^ with an internally consistent thermodynamic data set 5.5^[@CR48]^ (November 2003 upgrade) in the NCKFMASH system. The effective bulk-rock composition is provided in Supplementary Table [5](#MOESM1){ref-type="media"}. Water was treated in saturation because of the widespread presence of hydrous phases (such as zoisite and phengite). All Fe was assumed as Fe^2+^. Ti was omitted in the calculations, as rutile is the only Ti-rich phase. Following mineral solid solutions are adopted: garnet^[@CR49]^, clinopyroxene^[@CR50]^, amphibole^[@CR51]^, white mica^[@CR52]^, and chlorite^[@CR48]^. Data availability {#Sec17} ----------------- The authors declare that the data supporting the findings of this study are available within the paper and its supplementary information files. Electronic supplementary material ================================= {#Sec19} Supplementary Information Peer Review File **Electronic supplementary material** **Supplementary Information** accompanies this paper at 10.1038/s41467-018-05140-5. **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. We thank Antonin Kopřiva and Y. Chen for helping with chemical analysis. We also thank Michael Brown for discussion and suggestions to improve the manuscript. This research was financially supported by the Chinese National Natural Science Foundation (41773022, 41688103, 41520104004), Czechic CEITEC 2020 (LQ1601), and National Science Foundation Geochemistry grant (EAR-1447311) to Y.F. C.X., J.K., W.S., and Y.L. were responsible for sample collection. C.X., J.K., Y.F., L.Z., Z.L., M.P., and M.V.G were responsible for mineralogy and petrology. W.S., R.T., Y.Y., and Y.L. performed dating and geochemical analysis. C.X. and Y.F wrote the manuscript with inputs from co-authors. Competing interests {#FPar1} =================== The authors declare no competing interests.

Introduction {#S0001} ============ Colon cancer is one of the most common malignant cancers and among the leading causes of cancer-related deaths worldwide.[@CIT0001] The combination therapy of oxaliplatin and fluoropyrimidine has been the standard adjuvant therapy in patients with stage III/IV colon cancer. However, these chemotherapies are toxic and sometimes ineffective. Due to its heterogenicity, multiple genetic mutations were regarded as the underlying causes of this disease, for example, mutations in phosphatidylinositol-3 kinase (PI3K) and the mitogen activated protein kinase (MAPK) pathways. Deregulation of MAPK pathway genes have been found, including the amplification and mutation of KRAS, BRAF, and MEK1.[@CIT0002] Therefore, targeting the downstream MEK in the mutated tumors might be a new strategy for colon cancers, especially the patients with KRAS or BRAF mutations. Several MEK inhibitors have entered clinical trial evaluation. However, clinical activity was poor following the treatment with any single MEK inhibitor, and acquired drug resistance appears inevitable.[@CIT0003]--[@CIT0008] Trametinib is a novel oral MEK inhibitor which has been approved by the FDA (Food and Drug Administration) for BRAF-mutated patients alone or in combination with dabrafenib.[@CIT0009]--[@CIT0011] Combination remedies of MEK inhibitors, rather than single medicine therapy, was considered to be more effective in various tumors.[@CIT0012]--[@CIT0014] Therefore, there is an urgent clinical demand for new synergic agents to cooperate with trametinib to enhance the survival of patients with colon cancer. Interferon-Stimulated Gene 15 (ISG15), a ubiquitin-like protein (UBL),[@CIT0015] is an important oncoprotein and has become a potential diagnostic[@CIT0016] and therapeutic[@CIT0017] target for cancer treatment. The function of ISG15 was largely underestimated mainly due to its low expression in most human malignancies.[@CIT0018] Some studies have found that ISG15 expression was elevated and ISG15-conjugates in many malignant tumors, including melanoma[@CIT0019] and oral squamous cell carcinoma[@CIT0020],[@CIT0021] as well as malignancies of the breast,[@CIT0016] endometrium,[@CIT0018] and bladder.[@CIT0022] However, the roles of ISG15 in tumorigenesis and its responses to anticancer treatments in colon cancer remain largely unknown. In a recent study, Roulois et al[@CIT0023] showed that DNA methylation inhibitor 5-aza-2-deoxycytidine (5-AZACdR) could enhance the expression of ISG15 in LIM1215 colon cancer cells, which implied the nature of ISG15 as a tumor suppressor in colorectal cancer. Controversially, Desai et al[@CIT0018] discovered that ISG15 expression and ISG15-conjugated proteins in two colon cancer cases were up-regulated in comparison to normal colon tissues. Whether ISG15 acts as a tumor suppressor or promotor remains controversial. In this study, we explored the function of ISG15 in colon cancer cell lines. Our results showed that high expression of ISG15 was an intrinsic feature for colon cancer, and ISG15 promoted the cell proliferation and metastasis. Trametinib could increase the expression of ISG15 shown by gene expression analysis. After treatment of a synergistic combination of trametinib with ISG15 siRNA, proliferation and colony formation was inhibited in vitro. Thus, combined targeting of ISG15 and MEK might be a promising therapeutic strategy for colon cancer treatment. Materials and Methods {#S0002} ===================== Tissue Samples and Study Cohort {#S0002-S2001} ------------------------------- Sixty-six pairs of tumor samples and matched adjacent non-tumor tissues were obtained from the Shanghai Outdo Biotech Co., Ltd. (Shanghai, China). All the patients signed informed consent forms. This study was approved by the Ethics Committee of Taizhou Hospital of Zhejiang Province. ISG15 expression was detected in all specimens. Two pathologists were appointed to evaluate the specimens separately without prior knowledge of the clinical statuses of the specimens. Immunohistochemistry {#S0002-S2002} -------------------- Immunohistochemistry (IHC) was performed using the biotin-streptavidin HRP detection system according to the manufacturer's instructions. The tissue chips were incubated with ISG15 antibody (1:100; Abcam, Cambridge, UK) in phosphate-buffered saline (PBS) overnight at 4°C in a humidified container. Biotinylated secondary antibodies (Zhongshan Golden Bridge Biotechnology Co. Ltd., China) were applied. The sections were incubated with HRP-streptavidin conjugates appropriate for detecting ISG15. Proper positive and negative controls were included in each IHC assay. Western Blotting Analysis {#S0002-S2003} ------------------------- Total proteins (30 µg) were subjected to fractionation by SDS polyacrylamide gel electrophoresis and immunoblotting assay. Antibodies used in the study included the following: anti-ISG15, anti-PARP, anti-c-Myc, anti-pERK, anti-ERK (all Abcam, Cambridge, UK), and anti-GAPDH (Zhongshan Golden Bridge Biotechnology Co. Ltd., China). Cell Culture and RNA Interference {#S0002-S2004} --------------------------------- Human colon cancer cell lines were obtained from the State Key Laboratory of Biotherapy, West China Hospital, Sichuan University, including RKO, HCT116 and SW480. And all the cells were tested and authenticated by an AmpFlSTR Identifiler PCR assays in the year of 2019 in TSINGKE Biological Technology Company. The results showed that RKO and SW480 were 100% exact matched, and HCT116 was 98% matched. All the cell lines were confirmed to be mycoplasma negative via a PCR method and directly thawed from the liquid nitrogen jar. Cells were grown in Dulbecco's Modified Eagle Medium supplemented with 10% fetal bovine serum (Gibco, USA), 100mU/mL penicillin, and 100 µg/mL streptomycin in a 5% CO2 atmosphere at 37°C. Cells (1x10 5 cells/well) were seeded into six-well plates and transfected with the indicated constructs using Lipofectamine 2000 (Invitrogen/Life Sciences) according to the manufacturer's instructions. After 72 h, the transfected cells were harvested for further analysis. The sequence of the siRNA sense strand was as follows: si-ISG15\#1: 5-'TCCTGGTGAGGAATAACAA-3ʹ, si-ISG15\#2: 5-'CCAUGUCGGUGUCAGAGCUTT-3ʹ. CCK-8 and Colony Formation Assays {#S0002-S2005} --------------------------------- CCK-8 assay kit was used for colon cancer cell proliferation analysis. Colon cancer cells were seeded in 96-well plate at a density of 5x10^\^3^ cells per well. After incubation, 10 µL CCK8 was added to the wells at different times. The absorbance was measured at 450 nm by a microplate reader (Bio-Rad, Hercules, CA, USA). Colony formation assay was performed to detect the capacity of cell proliferation. RKO, HCT116 and SW480 cells were seeded into 35 mm dishes with 3x10^\^3^ cells per well and cultured for 10 days. Colonies were then fixed with 10% formaldehyde for 30 min and stained for 5 min with 0.5% crystal violet. Wound Healing Assay {#S0002-S2006} ------------------- Colon cancer cells were cultured in 6-well plates and grown until 80--90% confluence. After that, the cell monolayer was scratched with a sterile pipette tip. An Olympus microscope was applied for taking pictures of cell morphology at different time intervals (0,48h). The results were analyzed by ImageJ software. Matrigel Invasion Assays {#S0002-S2007} ------------------------ The capacity invasiveness was determined in a 24-well transwell plate (8 up pore size; Costar). Briefly, 6x10^\^\ 4^ cells were suspended and added to the upper chamber of each insert coated with 200 mg/mL of Matrigel (BD Biosciences, Franklin Lakes, NJ, USA). After 48 h, the cells that traversed the membrane and spread to the lower surface of the filters were stained with haematoxylin and counted.Each experiment was repeated three times independently. RNA Extraction and qRT-PCR {#S0002-S2008} -------------------------- Total RNA was isolated through the guanidinium thiocyanate-phenol-chloroform extraction method using TRIzolsolution (Invitrogen, USA), and cDNA synthesis was done using the Advantage RT-forPCRKit (Takara, Japan). The forward and reverse primers were designed to span two different exons for each gene.The primers for human ISG15 were forward:5ʹ-TGGACAAATGCGACGAACC-3ʹ; reverse: 5ʹ-TTCGTCGTTCACTCGCC-3ʹ.CFX96 Real-Time PCR detectionsystem (Bio-Rad, Hercules, CA, USA) was used to perform qRT-PCR. The threshold cycle number was determinedusing iCycler software version 3.0. Reactions were performed in duplicate and the threshold cycle numbers wereaveraged. The data were analysed using the 2\^^−ΔΔCT^ method. Edu Assay {#S0002-S2009} --------- Quantification of cellular proliferative capacity was determined by 5-Ethynyl-2ʹ-deoxyuridine assay using a Cell-Light TM EdU Apollo^®^488 In Vitro Imaging Kit (Guangzhou RiboBio Co., LTD), according to the manufacturer's instruction. Identification of Differentially Expressed Genes from Public Microarray Data {#S0002-S2010} ---------------------------------------------------------------------------- GSE112282 gene expression profiles were downloaded from the Gene Expression Omnibusdatabase (GEO, [<http://www.ncbi.nlm.nih.gov/geo>]{.ul}). These profiles, deposited by Wyce A et al in 2018, are comprised of two untreated replicates of human colon cells and two replicates of the same cells treated with 25 nM trametinib for 24 h and 96 h. The dataset was analyzed with R BIOCONDUCTOR packages, raw datasets were normalized and DEGs were screened out via the LIMMA package using the cut-off criteria of P-value \< 0.05 and \|log2(fold change)\|\>1.5. Functional Enrichment Analysis {#S0002-S2011} ------------------------------ DAVID ([<https://david.ncifcrf.gov/>]{.ul}) was used to perform functional and pathway enrichment analysis. Gene ontology (GO) analysis, including biological process (BP), cellular component (CC) and molecular function (MF)[@CIT0024] enrichment analysis, was carried out for the upregulated genes. P\<0.001 was regarded as indicating statistical significance. Protein-Protein Interaction Network Construction {#S0002-S2012} ------------------------------------------------ In order to interpret the underlying molecular mechanisms, the online Search Tool for the Retrieval of Interacting Genes (STRING) database was utilized to construct a protein-protein interaction (PPI) network of the DEGs. An interaction score of not \< 0.4 (medium confidence score) was deemed statistically significant, and the PPI was visualized. Subsequently, the hub genes were selected based on connection degree by CYTOSCAPE software. Statistical Analysis {#S0002-S2013} -------------------- Assays for characterizing phenotypes of cells were analyzed by Student's test or One-Way ANOVA, presented as the means ± SDs. In the clinical specimen study, the associations of ISG15expression with categorical variables were analyzed using the χ2 test or Fisher's exact test as appropriate. And other variables were analyzed with Spearman correlation test. P values of \<0.05 were deemed statistically significant. Results {#S0003} ======= Up-Regulation of ISG15 Expression in Colon Cancer Samples {#S0003-S2001} --------------------------------------------------------- cBioPortal was applied for cancer genomics datasets analysis. ISG15 was amplified and the mutation frequency was 5% in colon cancers ([Figure 1A](#F0001){ref-type="fig"}). Among them, the ones with high expression of ISG15 mRNA accounted for 4% ([Figure 1B](#F0001){ref-type="fig"}). Moreover, ISG15 mRNA was over-expressed in colon cancer specimens compared to their non-tumor counterparts. ([Figure 1C](#F0001){ref-type="fig"}, P \< 0.05). Kaplan-Meier analysis demonstrated that patients with high ISG15 expression had significantly shorter overall survival (OS) than patients with low ISG15 expression ([Figure 1D](#F0001){ref-type="fig"}, P \< 0.05). Using Western blot, 83.3% (10/12) of the cases showed relatively higher ISG15 expression than their non-tumor counterpart, consistent with the results of mRNA expression analysis ([Figure 1E](#F0001){ref-type="fig"}). To evaluate whether the expression of ISG15 changes as colon cancer progresses, IHC was performed by using an independent formalin-fixed, paraffin-embedded-based (FFPE-based) tissue microarray (TMA) including 66 pairs of colon cancer patients. The results showed that ISG15 expression in colon cancer tissues was obviously higher than that in adjacent noncancerous colon tissues ([Figure 1F](#F0001){ref-type="fig"}). As indicated in [Figure 1G](#F0001){ref-type="fig"}, the mean immunoreactivity scores of the normal and cancer tissues were 1.02 and 5.01, respectively. Samples with scores above 5 were considered as high ISG15 expression ones, leading to the greatest number of tumors classified as having or not having a clinical outcome. In our study, high ISG15 levels were detected in 30/66 (45.5%) of colon cancer patients. The results showed ISG15 expression did not correlate with patient sex, age or lymph node status (P\>0.05). It was, indeed, associated with poor histological classification (P \< 0.01, [Table 1](#T0001){ref-type="table"}). Overall, our data suggested that ISG15 level was higher in colon cancer samples.Table 1Correlation Between ISG15 Expression and Clinicopathological Features of Colon Cancer Patients P-Value=0.0087\*: the ISG15 Expression Was Associated with Poor Histological ClassificationStatistical Results of ISG15 Expression in 66 Colon Cancer Cells SpecimensVariablesISG15 ExpressionNLow Expression, nHigh Expression, nP-valueAll cases663630Age≥ 504321220.2997^a^\< 5023158GenderMale3718190.3254^a^Female291811Pathology gradeI (well)1293**0.0087^\*\*,b^**I-II (moderate)372314II-III (poor)17413Lymph node statusN-4021190.8015^a^N+361511[^1] Figure 1Upregulation of ISG15 expression in colon cancer samples. (**A, B**) ISG15 status in a colon cancer study obtained from [<http://www.cbioportal.org>]{.ul}. (**C**) Expression of ISG15 in 286 TCGA cancer and 41 normal samples indicated that ISG15 was obviously upregulated in colon cancer. (**D**) Kaplan-Meier analysis indicated that high ISG15 expression predicted poorer prognosis in colon cancer. The error bars in all graphs represent SD, and each experiment was repeated three times (\*\*p\< 0.01  ). (**E**) The difference in protein expression levels of ISG15 in colon cancer tissues was compared with the corresponding adjacent tissues by Western blot in 12 pairs of human colon cancer specimens. (**F, G**) The differential expression of ISG15 in colon cancer samples (n = 66) and adjacent normal colon tissues (n =66) is shown by immunohistochemistry. ISG15 staining scores for tumor tissues and normal tissues are represented by a frequency distribution (0--4: low expression; 5--12: high expression). Knockdown of ISG15 Inhibits Cancerous Proliferation and Migration {#S0003-S2002} ----------------------------------------------------------------- We investigated the effect of ISG15 suppression on the malignant phenotypes of colon cancer cells. Results showed that knockdown of ISG15 with si-SG15\#1 and si-SG15\#2 significantly inhibited cell proliferation in the RKO, HCT116 and SW480 cell lines ([Figure 2A](#F0002){ref-type="fig"}). As indicated in [Figure 2B](#F0002){ref-type="fig"}, knockdown by siRNAs resulted in reduced ISG15 protein in colon cancer cell lines. And we concluded that si-SG15\#1 inhibitory effect was stronger than si-SG15\#2. Colony formation assays and Edu assay also indicated that si-ISG15\#1 could significantly inhibited the proliferation and growth of colon cancer cells ([Figure 2C](#F0002){ref-type="fig"} and [D](#F0002){ref-type="fig"}, [[Figure S1A](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul} and [[B](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}). Next, we examined the migratory abilities of HCT116 and SW480 cells by using transwell migration assay. HCT116 and SW480 cells transfected with si-ISG15\#1 displayed lower efficiency of cell migration compared to the vector control ([Figure 2E](#F0002){ref-type="fig"} and [F](#F0002){ref-type="fig"}, P\<0.01). We also performed a wound healing assay by using RKO, HCT116 and SW480 human colon cancer cells. As shown in [Figure 2G](#F0002){ref-type="fig"} and [H](#F0002){ref-type="fig"} and [[Figure S1C](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}, after 48 hrs of silence, si-ISG15\#1 treatment significantly impeded gap closure compared to the vector control. These results indicated that suppression of ISG15 could inhibit the cell growth and migration in colon cancer cells.Figure 2Knockdown ISG15 inhibits cancerous proliferation and migration. (**A**) Cell proliferation profiling in RKO, HCT116 and SW480 cells infected with si-ISG15\#1 or si-ISG15\#2 continuously for 3 days as analyzed by a CCK8 assay (\*p\< 0.05, \*\*p\< 0.01). (**B**) Protein levels of ISG15 in the RKO, HCT116 and SW480 cell lines infected with si-ISG15\#1 or si-ISG15\#2. GAPDH was used as an internal control. NC, Negative control; si-ISG15, siRNA of ISG15. (**C, D**) Colony formation in RKO, HCT116 and SW480 cells infected with si-NC or si-ISG15. The histogram shows the average colony numbers and represents the mean±SEM of three independent experiments (\*\*\*p\< 0.001). (**E, F**) Transwell invasion assay in si-NC or si-ISG15\#1 transfected HCT116 and SW480 cells. The error bars in all graphs represent the SDs (\*\*\*p\< 0.001). (**G, H**) Wound-healing assay after transfecting with si-NC or si-ISG15\#1 for 48 h in HCT116 and SW480 cells. Data are shown as the mean ± standard error of the mean (\*\*\*p\< 0.001). Trametinib Treatment Activates the Expression of ISG15 by Identification DEGs {#S0003-S2003} ----------------------------------------------------------------------------- The gene expression profiles of GSE112282 were downloaded from the GEO database, to unravel molecular mechanisms underlying trametinib in the tumors of an experimental mouse model. The GSE112282 dataset contained following four groups: DMSO (control) or 30 nM trametinib for 24 or 96 hrs each in the RKO colon cancer cell line. We found 193 and 304 genes differentially expressed between the control and trametinib groups at 24 and 96 hrs, respectively (\>1.5-fold, p\<0.05, [Figure 3A](#F0003){ref-type="fig"} and [B](#F0003){ref-type="fig"}, [[Figure S2A](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul} and [[2B](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}). Gene ontology analysis were applied to discover the functions of the DEGs for 24 hrs ([[Figure S3](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}) and 96 hrs ([[Figure S4](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}). The DEGs of up-regulated genes were significantly enriched in biological processes related to ossification, transforming growth factor beta receptor signaling pathway (i.e., ISG15, TOB1, ID1) at 24 hrs ([Table 2](#T0002){ref-type="table"}) and type I interferon signaling pathway, cellular response to type I interferon (i.e., ISG15, IRF9, IFITM3, IFI6, IFIT/12/3, and IP6K4) at 96 hrs ([[Table S1](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}). The STRING database and Module analysis via Cytoscape software were constructed to identify gene interactions, the significant hub nodes of the network included IP6K2, ISG15, IFI6, IFI27 and IFITM1 at 24 hrs ([Figure 4](#F0004){ref-type="fig"}) and IFITM3, IRF9, ISG15, IP6K2 and HIST1H2BD at 96 hrs ([[Figure S5](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}). Node degree and betweenness were calculated by the DMNC method to obtain hub nodes. The top 15 candidate hub genes were identified which may play a central role in this network ([[Table S2](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}). And no matter which stage for the trametinib groups (24 hrs or 96 hrs), we found ISG15 always appears as a hub gene and was associated with type I interferon signaling pathway and cellular response to type I interferon. Therefore, the PPI network analysis and hub gene selection and analysis showed that treatment of colon cancer cells with trametinib could induced a dependency on ISG15 signaling especially in terms of type I interferon signaling pathway.Table 2Functional Enrichment Analysis of Up-Regulated Genes Were Significantly Enriched in Biological Process Terms (24h)CategoryIDTerm/gene functionGeneIDGene countpvalueGO_TERM_BPGO:0001503OssificationTOB1/FSTL3/ANO6/FGFR3/ID1/ID3/JUND/SMAD3/TP53INP2/BMPR1A/BMPR2/KLF10/TMSB4X/TUFT1/KAZALD1/CAT/IFITM1/DHRS3/EIF2AK3/ISG15201.28E-07GO_TERM_BPGO:0007179Transforming growth factor beta receptor signaling pathwayCDKN2B/ID1/SMAD3/SMAD7/SMAD9/PTPRK/SKIL/BMPR1A/KLF10/CAV1/CAV2/ADAM9123.58E-06GO_TERM_BPGO:0007178Transmembane receptor protein serine/threonine kinase signaling pathwayTOB1/FSTL3/CDKN2B/HES1/ID1/SMAD3/SMAD7/SMAD9/SKIL/BMPR1A/BMPR2/KLF10/CAV1/CAV2/ADAM9166.40E-06GO_TERM_BPGO:0030282Bone MineralizationANO6/FGFR3/SMAD3/BMPR1A/BMPR2/KLF10/TUFT1/EIF2AK3/ISG1596.56E-06GO_TERM_BPGO:0071560Cellular response to transforming growth factor beta stimulusCDKN2B/ANKRD1/ID1/SMAD3/SMAD7/SMAD9/PTPRK/SKIL/BMPR1A/KLF10/CAV1/CAV2/ADAM9138.08E-06GO_TERM_BPGO:0071559Response to transforming growth factor betaCDKN2B/ANKRD1/ID1/SMAD3/SMAD7/SMAD9/PTPRK/SKIL/BMPR1A/KLF10/CAV1/CAV2/ADAM9139.80E-06GO_TERM_BPGO:0050678Regulating of epithelial cell proliferationCDKN1B/CDKN2B/ANG/HOXA5/HES1/ID1/MIR24-2/SMAD3/PTPRK/PEX2/BMPR1A/BMPR2/CAV1/CAV2/MAGED1151.53E-05GO_TERM_BPGO:0060337Type I interferon signaling pathwayIRF9/IFI6/IFI27/IFIT2/IFIT1/IP6K2/IFITM1/ISG1582.15E-05GO_TERM_BPGO:0071357Cellular response to type I interferonIRF9/IFI6/IFI27/IFIT2/IFIT1/IP6K2/IFITM1/ISG1582.15E-05GO_TERM_BPGO:0050673Epithelial cell proliferationCDKN1B/CDKN2B/ANG/HOXA5/HES1/ID1/ID2/MIR24-2/SMAD3/PTPRK/PEX2/BMPR1A/BMPR2/CAV1/CAV2/MAGED1162.63E-05GO_TERM_BPGO:0034340Repose to type I interferonIRF9/IFI6/IFI27/IFIT2/IFIT1/IP6K2/IFITM1/ISG1583.22E-05GO_TERM_BPGO:0030278Regulation of ossificationTBO1/ANO6/ID1/ID3/JUND/SMAD3/BMPR1A/BMPR2/IFITM1/DHRS3/ISG15113.49E-05 Figure 3Trametinib treatment activates the expression of ISG15 as shown by the identification of DEGs. (**A**) Volcanoplot of the 193 identified DEGs at 24 hrs. Red indicates DEGs with a \|log2FC\|\>1.5. DEG, differentially expressed gene; FC, fold change. (**B**) Heat map showing the 193 differentially expressed genes betweenthe trametinib (n=2) and control groups at 24 hrs. (n=2; \>1.5-fold, p\<0.05). The most highly upregulated (red) and downregulated (green) genes in the trametinib group are listed.Figure 4Functional enrichment analysis of upregulated DEGs in trametinib treatment for 24h. Using the STRING online database and Cytoscape, upregulated genes in the trametinib groups were filtered into the DEG PPI network complex. (IP6K2, ISG15, IFI6, IFI27, IFITM1, MIS18A and so on are hub genes in red or orange; other linked genes in blue). Inhibition of ISG15 Enhances the Anti-Cancer Effect of Trametinib in Colon Cancer Cells {#S0003-S2004} --------------------------------------------------------------------------------------- We next tested whether combined therapeutic strategies with trametinib and ISG15 inhibition could be effective in treating colon cancer cells. As shown in [Figure 5B](#F0005){ref-type="fig"}, trametinib decreased the expression of c-Myc, a cell proliferation related protein, and improved the expression of Bax, an apoptosis inducing protein. Furthermore, trametinib increased the mRNA and protein expression of ISG15 ([Figure 5A](#F0005){ref-type="fig"} and [B](#F0005){ref-type="fig"}). Therefore, we treated cancer cells with ISG15 siRNA and trametinib and found that this combination had a synergistic effect that inhibited cell proliferation ([Figure 5C](#F0005){ref-type="fig"}, [[Figure S1D](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}) and colony formation ([Figures 5D](#F0005){ref-type="fig"}, [[S1E and F](https://www.dovepress.com/get_supplementary_file.php?f=226395.zip)]{.ul}). Next, we also examined the expression of c-Myc and Bax by Western blot following the knockdown of ISG15, and found that suppression of ISG15 had no effect on c-Myc, but promoted the upregulation of Bax ([Figure 5E](#F0005){ref-type="fig"}). Finally, we combined trametinib with si-ISG15 to investigate whether ISG15 inhibition enhances the cytotoxicity of trametinib against colon cancer at the protein levels. Our results showed that si-ISG15 synergized with trametinib to dramatically reduce the expression of ISG15 and increase the pro-apoptotic genes PARP and Bax ([Figure 5F](#F0005){ref-type="fig"}). Together, ISG15 inhibition combined with trametinib displayed synergistic inhibitory effects in colon cancer cells.Figure 5ISG15 inhibition enhances the anti-cancer effect of trametinib in colon cancer cells. (**A**) The mRNA expression of ISG15 following treatment with the indicated concentrations of trametinib (0--50 nM) in colon cancer cells for 2 d by qPCR assay (\*p\< 0.05, \*\*p\< 0.01, \*\*\*p\< 0.001). (**B**) WB for ISG15, c-Myc, Bax, pERK and ERK following treatment with the indicated concentrations of trametinib (0--50 nM) in colon cancer cells for 2 d. (**C**) Histograms shows the percentage of cell proliferation in colon cancer cells after transfection with si-ISG15 and/or 25 nM trametinib by CCK8 assays at day 3 (\*\*\*p\< 0.001). (**D**) Colony formation assays show that knockdown of ISG15 combined with 25 nM trametinib inhibited colon cancer cell proliferation at day 7. (**E**) The levels of ISG15, c-Myc and Bax following ISG15 silencing in colon cancer cells for 3 d as determined via Western blot. NC, Negative control; si-ISG15, siRNA of ISG15. (**F**) The levels of ISG15, cleaved PARP and Bax following treatment with si-ISG15 (and/or 25 nM trametinib) in RKO and HCT116 cells for 3 d. NC, Negative control; si-ISG15, siRNA of ISG15. Discussion {#S0004} ========== ISG15, a 15 KDa protein, belongs to the ubiquitin-like (UBL) superfamily[@CIT0025] which is synthesized by several cells in response to various stimuli such as type-I IFNs, virus and genotoxicity.[@CIT0026] In addition, the function of ISG15 involves in diverse cellular pathways, including translation, RNA splicing, chromatin remodeling, polymerase II transcription, cytoskeleton organization, stress responses,[@CIT0027] autophagy[@CIT0028] and metabolism.[@CIT0029] Previous studies have discovered that ISG15 expression is associated with tumor responses to chemotherapy, radiotherapy and IFN-α treatment.[@CIT0030],[@CIT0031],[@CIT0032] Recently researchers have identified ISG15-deficient patients and suggested the underlying process of ISG15 regulation, these might cast some light on the intricacy of this protein. Also, there has been voices demanding a re-evaluation of ISG15.[@CIT0033] Several investigations has demonstrated the carcinogenicity of ISG15, which played an important role in cancer-associated inflammation.[@CIT0018],[@CIT0020],[@CIT0021],[@CIT0023] ISG15 is also associated with cytoskeleton disruption, cancer cell migration and poor prognosis.[@CIT0034] Recently, ISG15 is thought to function both as an oncogene and a tumor-suppressor gene in different tumors.[@CIT0035] For example, knockdown of ISG15 expression reduced breast cancer cell motility compared to the control in cell migration assays.[@CIT0036] In another study, Jeon et al[@CIT0037] reported that ISG15, as a possible tumor suppressor, provoked anchorage-dependent cell growth after doxorubicin treatment in mouse breast cancer model. Adam R.Brown[@CIT0038] purposed a process through which those select IFN-stimulated genes, especially ISG15, may facilitate colon cancer cell survival and growth by suppressing apoptosis, which was consistent with our findings. In addition, protein ISGylation could exacerbate experimental colitis and colitis-associated cancer. Jun-Bao Fana[@CIT0039] discovered an elevation of inflammation-related cytokines as well as the p38 MAP kinase activation induced by LPS in mouse. Above all, ISG15 plays an important role in the progression of various tumors. However, the pathological or physiological functions of ISG15 and their implications for targeted therapy have not been clearly elucidated in colon cancer. In our study, immunohistochemical analysis and Western blot analysis were used to detect ISG15 expression in colon tumor tissues and pericancerous tissues. The results demonstrated that the expression of ISG15 protein was significantly higher in tumors than that in adjacent control tissues. Additionally, inhibition of ISG15 with siRNA could suppress tumor growth and migration in colon cancer. Using bioinformatics, it was also predicted that the ISG15 mRNA level of colon cancer patients was higher than that of normal patients, and the expression level of ISG15 was inversely proportional to the overall survival (OS) in colon cancer patients. Further investigation in gene expression indicated that trametinib treatment could induce the expression of ISG15. While suppression of ISG15 by siRNA could enhance the cytotoxicity of trametinib against colon cancer. In summary, we have identified ISG15 as an oncogene in colon cancer, with a mechanism that mainly dealt with the proliferation and metastasis of colon cancer cells. Notably, our data point out the possibility that inhibition of ISG15 with small molecules which do not currently exist combined with trametinib treatment would be a novel therapeutic strategy for colon cancer. Ethics Approval and Consent to Participate {#S0005} ========================================== The human colon cancer tissues as provided by the Ethics Committee of Taizhou Hospital of Zhejiang Province. Funding {#S0006} ======= This work was supported by the National Natural Science Foundation of China (Grant No.81572850). Disclosure {#S0007} ========== No potential conflict of interest was reported by the authors. [^1]: **Notes:** \*\*p\<0.01, The ISG15 expression was associated with poor histological classification. ^a^Fisher's exact test was used for statistical analyses. ^b^Pearson's χ2 test was used for statistical analyses.

Optimal patient care relies on a cohesive team of health care professionals who work synergistically, focusing on individual strengths to ensure the highest quality of patient care ([@CIT0001], [@CIT0002]). It has been shown that interprofessional collaboration enhances patient safety, decreases medical errors, and improves satisfaction among health professionals ([@CIT0002], [@CIT0003]). Skills necessary to function as part of an interprofessional team require cultivation and development during health professions training ([@CIT0004]). This has made training in interprofessional collaboration an increasingly appreciated component of medical education. Interprofessional education (IPE) occurs when two or more professions learn with, from, and about each other to improve collaboration and the quality of care ([@CIT0005]). The development and implementation of interprofessional education curriculum, however, can be a difficult task, requiring significant faculty involvement and curriculum development ([@CIT0006]). Student-led peer teaching conveys a number of benefits to the peer learner, the peer teacher, and the institution as a whole ([@CIT0007], [@CIT0008]). Peer teaching occurs when instructors and learners are at a similar stage in their education ([@CIT0009]). Students engaged in peer learning have been shown to be highly receptive to instruction from peer teachers due to the decreased power differential existing between peer teacher and peer learners ([@CIT0010]). In addition, student-led initiatives are cost-effective and can promote collegiality and socialization, leading to long-term sustainability of student-led programs ([@CIT0008]). Small-group problem-based learning (PBL) is a highly effective means of education ([@CIT0011]). During PBL sessions students are presented with a problem or clinical scenario and are then required to work together to come up with a solution with minimal input from the facilitator ([@CIT0012]). The problem-based model is designed around individual inquiry and is very effective in developing problem solving skills, independent learning, and teamwork ([@CIT0013]). This educational method presents a unique way of delivering IPE. We believe student-led small-group PBL seminars can be used to effectively deliver IPE and improve students' perceived need for interprofessional education. To measure the efficacy of IPE sessions, the Interprofessional Education Perception Scale (IEPS) tool was used ([@CIT0014]). The IEPS tool, originally developed by Leucht et al., is an 18-item questionnaire using a six-point Likert scale, designed to assess perceptions of interprofessional education. The questionnaire is then subdivided into four subscales of IPE: competence and autonomy (Subscale 1), perceived need for cooperation (Subscale 2), perception of actual cooperation (Subscale 3), and understanding of others values (Subscale 4). The IEPS was used to evaluate participants' opinions regarding IPE over 16 weeks ([Table 1](#T0001){ref-type="table"}). ###### IEPS subscale classifications ------------ --------------------------------------------- Subscale 1 Professional competence and autonomy Subscale 2 Perceived need for professional cooperation Subscale 3 Perception of interdependence Subscale 4 Willingness to share ideas ------------ --------------------------------------------- Methods {#S0002} ======= First- and second-year medical and pharmacy students at Creighton University were invited to participate in small-group PBL interprofessional seminars led by fellow second-year medical students who served as peer-teachers. The study was approved by the Institutional Review Board at Creighton University. Seminars were held weekly during the lunch hour for 16 weeks. Each seminar was led by one to two second-year medical students, serving as peer-teachers and consisted of 10--14 student learners, split evenly between medical and pharmacy students. The cases were adaptations of cases seen at the Creighton University Medical Center or cases published in peer-reviewed journals. At the beginning of each seminar the group was given a basic patient presentation. Student learners were then required to ask appropriate questions and find appropriate answers to create a differential diagnosis. Student learners were next asked to identify appropriate diagnostic testing to order. The designated test results and questions regarding history and physical were provided by the peer-teacher when appropriate. Through the course of the seminar the peer-teacher provided little to no guidance in keeping with the PBL design. During each seminar students were allowed to use any reference material they deemed necessary. Evaluation {#S0003} ========== A case--control study design was used to evaluate the efficacy of IPE seminars. Following 16 weeks of IPE seminars, student\'s attitudes toward IPE were assessed using the IEPS. Questionnaires were distributed to the entire first- and second-year medical and pharmacy student classes, including those who did not attend. Those students not attending the IPE seminars served as the control group for our study. All first- and second-year medical and pharmacy students were also given a questionnaire regarding barriers to IPE; this survey was designed by the authors of this study ([Table 2](#T0002){ref-type="table"}). ###### IEPS scaled results *n* Subscale 2 mean±SD Subscale 3 mean±SD ----------------------------------- ----- ---------------------------------------------- ---------------------------------------------- All medical students 62 94.76[a](#TF0001){ref-type="table-fn"}±6.44 75.91[b](#TF0002){ref-type="table-fn"}±10.32 All pharmacy students 35 90.24[a](#TF0001){ref-type="table-fn"}±10.00 84.29[b](#TF0002){ref-type="table-fn"}±10.05 Medical student participants 19 97.81[c](#TF0003){ref-type="table-fn"}±4.68 76.67±11.65 Medical student non-participants 43 93.45[c](#TF0003){ref-type="table-fn"}±7.02 74.76±10.04 Pharmacy student participants 10 95.37^d^±5.71 86.33±10.82 Pharmacy student non-participants 25 88.33[d](#TF0004){ref-type="table-fn"}±10.76 82.67±10.93 All medical students compared to all pharmacy students 'need for professional cooperation' *p*=0.02 all medical students compared to all pharmacy students 'perception of interdependence' *p*=0.002 medical student participants compared to medical student non-participants 'need for professional cooperation' *p*=0.006 pharmacy student participants compared to pharmacy student non-participants 'need for professional cooperation' *p*=0.02. Data from the IEPS were analyzed along four subscales as described previously ([@CIT0014]). The primary outcome was the difference in responses to the IEPS of those students who attended seminars versus those students who did not attend. The data were scaled to a score of 100 for easier interpretation. It is important to note that data can only be compared within each subscale and not between subscales. Statistical significance was determined using two-tailed *t-*test and a *p* value of 0.05. Student responses with regard to perceived barriers to interprofessional education were also recorded and served as secondary outcome data. Results {#S0004} ======= In total, 43 students (24 medical and 19 pharmacy) attended IPE seminars, of those 29 completed the IEPS (67.4% response rates). Sixty eight (43 medical and 29 pharmacy) out of 313 students who did not participate in IPE seminars completed the IEPS survey (21.7% response rate) ([Table 2](#T0002){ref-type="table"}). IEPS responses show that medical and pharmacy students who participated in seminars perceived a significantly higher need for cooperation when compared to those who did not participate (*p=*0.006 and *p=*0.02, respectively). Significance was not observed within subscale 1 or subscale 4 (data not shown). In addition, combined responses from participants and non-participants showed that pharmacy students perceived a significantly higher need for professional cooperation (*p=*0.02) and interdependence (*p=*0.002) when compared to medical students. In total, 109 (30.62%) out of 356 eligible students responded to the survey regarding current barriers to interprofessional education ([Table 3](#T0003){ref-type="table"}). The most common barrier to participation was 'I am not aware of interprofessional education programs' (61.5% of respondents). The second most common response was 'I do not have time' (52.3% of respondents). Zero respondents indicated, 'Interprofessional education is not an important part of my career'. ###### Student perceived barriers to IPE Student response \% *n* --------------------------------------------------------------------- ------ ----- I am interested in IPE but I do not have enough time to participate 52.3 57 I am not aware of currently available IPE activities 61.5 67 I have tried and all IPE sessions are full 5.5 6 I do not think IPE is important in my future 0.0 0 Discussion {#S0005} ========== The role that student leaders play in curricular development, especially pertaining to interprofessional education, can be of great value to educational institution as a whole. Current student-led initiatives in IPE have used designs such as social gatherings, conferences, lectures, and formal communications between health professions students ([@CIT0008]). Our data demonstrate that a student-led small-group PBL seminar can be used to effectively address certain IPE goals. In particular, students in our study who participated in IPE seminars demonstrated significantly greater perceived need for interprofessional cooperation when compared to those who did not attend. In addition, the most commonly cited barrier to involvement in IPE in our study was lack of awareness of IPE programs. In our case, the lack of awareness of IPE programs is due to the absence of IPE in the standard curriculum. IPE programs can be resource and time intensive because of the significant amount of coordination required to educate students in different health professions ([@CIT0015]). Peer-teachers are a significantly underutilized resource available to educators, and can be used to effectively achieve educational outcomes that may otherwise be difficult based on funds or faculty availability ([@CIT0016], [@CIT0017]). Seminars in our study were designed and led exclusively by medical students and were minimally resource intensive on faculty and school funding. Students also state that time is a major barrier to participation in interprofessional education. Minimal preseminar preparation and the relaxed nature of peer-led seminars significantly reduce the time constraint that students feel in more formal educational environments. These seminars fit nicely into a lunch hour, similar to noon-conference seminars used in many graduate medical education training programs. The use of small-group PBL sessions has been shown to be highly effective in developing problem solving skills and teamwork ([@CIT0013]). During our study it was evident that students frequently encountered problems they could not solve on their own and required the help of students from other health professions to solve such problems. This interprofessional reliance began to shape the attitudes and behaviors of students in our study and resulted in appreciation of the interprofessional collaboration required to optimally approach a patient or clinical problem. Other studies have observed similar findings using a broad range of IPE models ([@CIT0017]). There are certain limitations to this study; in particular our study was limited to pharmacy and medical students based on logistical constraints. It is yet to be proven whether these results are applicable to other health professions. Also, the voluntary nature of participation may select for those students who inherently desire growth in interprofessional collaboration, thus further studies are required to control for this variable and determine the generalizability to an entire student population. One of the most encouraging observations of this study was the vigor with which new students approached leadership roles within IPE seminars. Because of the progressive nature of medical education it is crucial that new leaders emerge who can effectively continue programs and improve upon previous iterations ([@CIT0008]). The minimal faculty involvement and complete student autonomy over IPE seminars seemed to inspire peer-learners to become peer-teachers. Conclusion {#S0006} ========== In conclusion, peer-teacher-led problem-based interprofessional seminars are effective in improving the perceptions of interprofessional collaboration among first- and second-year medical and pharmacy students. The peer-teaching model may be an effective adjunct to traditional curriculum in addressing IPE by overcoming major barriers to student involvement. Conflict of interest and funding {#S0007} ================================ The authors report no conflicts of interest.

INTRODUCTION {#s1} ============ Cytomegalovirus (CMV) is a re-emerging human herpes virus, which causes substantial morbidity and occasional mortality in immunocompromised patients \[[@CIT0001]\]. Among these, solid organ transplant recipients who maintain an allograft by receiving immunosuppressive drugs are well-known to be at risk of infection. CMV infection after kidney transplantation (KT) reportedly ranges from asymptomatic infection to symptomatic CMV syndrome and tissue-invasive disease \[[@CIT0002]\]. Data from 2 retrospective studies in Thailand revealed the prevalence of CMV seropositivity in both donors and recipients was 99 percent. Despite preexisting immunity in most patientsindicated by CMV seropositivity, they remain at moderate risk of CMV infection. The prevalence of asymptomatic CMV infection and CMV disease in KT recipients were 5%--21% and 7%, respectively. Risk factors include advanced age of the recipient and use of antithymocyte globulin (ATG) for induction or steroid-resistant rejection therapy \[[@CIT0003], [@CIT0004]\]. Patients developing CMV diseases carry a greater risk of allograft failure and death compared with those free from CMV disease \[[@CIT0003]\]. Furthermore, KT recipients who developed drug-resistant CMV infection could suffer increased morbidity from prolonged CMV DNAemia and anti-CMV therapy \[[@CIT0005]\]. Prevention of this infection remains a suitable intervention to limit unfavorable consequences. Two international guidelines from the Transplantation Society and the American Society of Transplantation Infectious Disease Community of Practice encouraged implementation of prevention strategies for CMV infection among KT recipients \[[@CIT0006], [@CIT0007]\]. Although the guideline for prevention of CMV infection in Thai KT care was developed by the Thai Transplant Society, the strategies implemented among transplant centers in Thailand remained variable \[[@CIT0008]\]. Despite the fact that a preemptive approach is recommended in CMV-seropositive KT recipients by the aforementioned guideline, impracticability and financial restrictions could limit its utilization in real-world practice. Additionally, an optimal cut-off value of plasma CMV DNA load has not been standardized among physicians. Furthermore, an investigation of real-life strategies to prevent this specific infection has never been performed, especially in resource-limited settings. Therefore, we aimed to investigate CMV prevention strategies utilized among transplant centers in Thailand. We also investigated differing perspectives in terms of CMV prevention in KT recipients between infectious disease physicians (ID) and nephrologists (NP). METHODS {#s2} ======= Questionnaire {#s3} ------------- A survey was delivered to all 31 transplant centers in Thailand during October and November 2018. One ID and 1 NP, who were directly caring for KT recipients at each transplant center, were included. The names of the transplant centers and physicians were obtained from the Thai Transplant Society in October 2018. An email with a link to the electronic survey was sent to all physicians with a reminder email if a response was not received. The study protocol was approved by the Institutional Review Board of the Faculty of Medicine, Ramathibodi Hospital, Mahidol University, Bangkok, Thailand. A questionnaire on CMV prevention strategies for KT recipients was developed using a web-based electronic survey website ([www.surveymonkey.com](http://www.surveymonkey.com)). The survey included the respondents' demographic data, such as sex, age, transplant center setting or years in KT practice, and CMV prevention strategies. CMV prevention strategies were defined according to the recently published guidelines \[[@CIT0006], [@CIT0007]\]. Prophylaxis was defined as administration of anti-CMV drugs, either intravenous (IV) ganciclovir or oral valganciclovir, for a defined period of time after KT. A preemptive approach was defined as surveillance of plasma CMV DNA load, quantified by real-time polymerase chain reaction (PCR) assay and initiation of anti-CMV drugs when the cut-off value was reached to prevent progression from asymptomatic CMV infection to disease. Plasma CMV DNA load was reported in both copies/mL and international units (IU)/mL, based on the use of a quantitative real-time PCR technique by COBAS AmpliPrep/COBAS TaqMan (TaqMan CMV Test, Roche Molecular Diagnostics, Branchburg, NJ). One copy/mL was calibrated to 0.91 IU/mL by the calibration of tests with the World Health Organization international standard \[[@CIT0009]\]. Targeted prophylaxis or preemptive approaches were defined as above, but they were focused on a defined high-risk group of patients rather than universal implementation. A hybrid approach was defined as surveillance by plasma CMV DNA quantification after cessation of a defined period of prophylaxis. A CMV-specific immunity-guided approach was defined as an intervention (prophylaxis, preemptive approach, or closed observation) guided by measurement of cell-mediated immunity, such as CMV-specific T-cell immunity assay. Statistical Analyses {#s4} -------------------- Demographic data of all physicians involved in the study were analyzed by descriptive analysis. Categorical data were described as frequencies and percentages and compared by Fisher exact test. A *P* value \< .05 was considered statistically significant. Statistical analyses were performed with the statistical software Stata version 15 (StataCorp, LLC, College Station, TX). RESULTS {#s5} ======= Demographic Data {#s6} ---------------- There were 43 respondents from 26 of the 31 (84%) transplant centers, including 26 (60%) IDs and 17 (40%) NPs. The demographic data of all physicians included in the survey are summarized in [Table 1](#T1){ref-type="table"}. Fifty-eight percent of respondents were aged 35--44 years and approximately half were males (49%). Two-thirds of the physicians had been working in a public hospital setting (63%) and encountering KT recipients for at least 2 years (74%). The demographic data between the ID and NP groups were comparable; however, there were slightly more male physicians among the IDs compared with the NP group (62% vs 29%; *P* = .06). There were also significantly more physicians from a public hospital setting in the ID group compared with the NP group (81% vs 36%; *P* \< .05). ###### Demographic Data of Infectious Disease Physicians and Nephrologists Who Participated in the Survey Characteristics Infectious Disease Physicians (n = 26) Nephrologists (n = 17) *P* value --------------------------------------------------- ---------------------------------------- ------------------------ ----------- Male, n (%) 16 (62) 5 (29) .06 Age, years, n (%)  25--34 5 (19) 9 (53) .04  35--44 18 (69) 7 (41) .11  45--54 3 (12) 1 (6) 1.00 Transplant center setting, n (%)  Public general hospital 10 (39) 2 (12) .09  Public university hospital 11 (42) 4 (24) .33  Nonprofit hospital 1 (4) 8 (47) .001  Others 4 (15) 3 (17) 1.00 Years in kidney transplant recipients care, n (%)  \<1 3 (12) 2 (12) 1.00  1--2 1 (4) 5 (29) .03  2--5 15 (57) 5 (29) .12  5--10 4 (15) 3 (18) 1.00  \>10 3 (12) 2 (12) 1.00 CMV Prevention Strategies {#s7} ------------------------- Overall responses in terms of CMV prevention strategies and the different perspectives between the ID and NP groups are shown in [Figure 1](#F1){ref-type="fig"}. Forty-one (95%) physicians agreed with a need for CMV prevention in KT recipients, although 33 physicians (77%) stated that they already utilized prevention strategies for their patients. Cytomegalovirus prevention strategies currently implemented include preemptive approaches (48%), universal prophylaxis (45%) (either by IV ganciclovir or oral valganciclovir), hybrid approaches; surveillance after prophylaxis (3%), and CMV-specific immunity-guided approach (3%). When specifically asked about the potential role of CMV prevention in Thai KT recipients, when the majority are CMV seropositive and likely to receive an allograft from a CMV seropositive donor, only 5% of physicians stated no need for prevention in this scenario. The remaining 95% reported that they preferred the preemptive approach (84%) over prophylaxis (12%). However, 81% of the former preferred targeted prophylaxis for those receiving ATG. {#F1} Among physicians choosing to implement prophylaxis, the duration of prophylaxis ranged from less than 1 to more than 6 months, with half (51%) choosing to implement prophylaxis over a 3-month course. For the preemptive approach, 65% and 93% of all physicians initiated preemptive therapy when the plasma CMV DNA load reached 2000 and 3000 copies/mL (1820 and 2730 IU/mL) or 3.3 and 3.4 log10 copies/mL (3.2and 3.4 log10 IU/mL), respectively [Figure 2](#F2){ref-type="fig"}. At a plasma CMV DNA load cut-off value of 5000 copies/mL (4550 IU/mL), almost all physicians (98%), either ID or NP, had initiated preemptive therapy for their patients. There was no difference in CMV prevention strategies between the 2 groups of physicians. A significantly greater percentage of NPs initiated preemptive therapy at a plasma CMV load of 1820 IU/mL compared with IDs (88% vs 50%; *P* = .02). {#F2} Barriers for CMV Prevention Strategies {#s8} -------------------------------------- The most common barrier to implementation of CMV prevention strategies was inaccessibility to care, including financial incompatibility resulting in restricted access to oral valganciclovir and quantitative CMV DNA testing in 67% and 12% of the physicians, respectively. Lack of logistic support for plasma CMV DNA load measurement due to impracticability in real-life practice was reported in 16% of respondents. Side effect intolerance of anti-CMV drug prophylaxis, especially bone marrow suppression from (val)ganciclovir, was reported in 5% of the physicians. Eighty-one percent felt that a guideline would enable physicians to more practically implement strategies for CMV prevention in their KT recipients. The remainder stated that greater self-education (14%) and transplant infectious disease consultation (5%) would be preferable. DISCUSSION {#s9} ========== A nationwide survey of CMV prevention practice in both IDs and NPs was conducted with a response rate of greater than 80%. We report here the first survey of CMV prevention strategies for KT recipients in a resource-limited setting. The majority of respondents agreed there was a need for CMV prevention in this immunocompromised population although the approach is currently variable among physicians. The preemptive approach is the most common strategy utilized; however, prophylaxis is more favorable in patients receiving intense immunosuppression, such as ATG. We also demonstrated common CMV DNA cut-off values for initiation of preemptive therapy for Thai KT recipients. Financial incompatibility is the main inhibitory factor in a successful intervention. However, most respondents believe that practical guideline would encourage physicians to implement CMV preventive strategies in Thai KT recipients. CMV has both direct and indirect effects in KT recipients. CMV disease, especially in those with allograft involvement, can cause significant morbidity. Allograft rejection, mortality, and opportunistic infection by infectious agents other than CMV are well-known consequences of CMV infection, which could be explained by the modulation of the immune system by CMV itself \[[@CIT0002]\]. In a resource-limited setting where the majority of adult KT recipients and donors are CMV seropositive, the risk of CMV infection after KT is considered to be moderate compared with western countries where CMV serostatus mismatches are common and pose a high risk. Data from previously cited studies in Thailand revealed prevention strategies are not universally utilized. A high rate of CMV disease in the period when quantitative CMV DNA testing is not practically available \[[@CIT0003]\]. A relatively lower rate of CMV disease and some with asymptomatic CMV DNAemia, which would likely discover more from accessibility to PCR assay in the later years. In Thailand, financial constraints remain an issue regarding access to oral valganciclovir in an outpatient setting. In our center, patients receiving ATG for induction therapy received IV ganciclovir early posttransplant while being admitted to hospital and then later switched to a preemptive approach, forming a so-called hybrid approach. We also attempted to use this strategy in patients with steroid-resistant rejection requiring ATG; however, the rate of compliance remains uncertain. In our study, we have revealed that CMV prevention strategies are generally accepted, although the real-life implementation is not fully practiced. In general, implementation of the preemptive approach and anti-CMV prophylaxis are comparable. However, upon further investigation, we also report a preference for the preemptive approach among physicians treating CMV-seropositive KT recipients. Of those treated preemptively, the optimal plasma CMV DNA cut-off value for initiation of therapy remains uncertain. Although the conversion of CMV DNA load from copies/mL to IU/mL was recommended for implementation across all transplant centers, the recent international guidelines remain vague regarding the preemptive threshold. They recommended standardizing the cut-off value based on center-specific rates of CMV infection and clinical practices. Our survey revealed the applied threshold could range from 2000 to 3000 copies/mL (1820 and 2730 IU/mL). At least 90% of physicians initiated therapy when the plasma CMV DNA load reached 3000 copies/mL (2730 IU/mL), with almost all physicians initiating therapy by 5000 copies/mL (4550 IU/mL). Martín-Gandul and colleagues conducted a retrospective study to determine a valid threshold in plasma for preemptive therapy of CMV infection in CMV-seropositive solid organ transplant, including KT recipients. They found 3983 IU/ml (2600 copies/ml) to be the optimal cut-off for initiating preemptive therapy, with a high negative predictive value of 99.6% that could almost exclude CMV disease without specific anti-CMV therapy, particularly in those with moderate risk similar to our patients. This optimal cut-off value also showed sensitivity and specificity of 89.9% and 88.9%, respectively \[[@CIT0010]\]. Our data reported a real-world plasma CMV DNA load threshold that encourages physicians to start preemptive therapy. We also found a trend of relatively early initiation of preemptive therapy in the NP group compared with the ID group. This could be explained by a greater percentage of patients being treated by NPs in a private setting where anti-CMV drug therapy is more available. However, dynamic monitoring of plasma CMV DNA load with the same clinical specimens and assays along with details of the clinical setting rather than single time-point interpretation is advised. The kinetics of CMV replication has been shown to be useful for the management of CMV infection in immunocompromised patients \[[@CIT0011], [@CIT0012]\]. Logistic support for measuring plasma CMV DNA load every week following transplantation is also not practical in our setting. A less frequent measurement, such as once every 2 weeks, in an outpatient clinic was commonly implemented in our practice. However, this strategy is not supported by Boillat Blanco et al, who revealed that less frequent monitoring may miss the opportunity to keep pace with the progression of the disease, especially in the postprophylaxis setting \[[@CIT0013]\]. Among centers utilizing prophylaxis, either IV ganciclovir or oral valganciclovir was commonly used. Duration of prophylaxis was variable with the majority of courses ranging from 1 to 6 months. Although the preemptive approach is recommended in the guidelines for CMV-seropositive KT recipients, anti-CMV prophylaxis is also acceptable, though 3 months is the recommended duration. Only in CMV serostatus mismatch settings was an extension to 6 months after transplant mentioned. When we specifically investigated the intervention for CMV-seropositive KT recipients, more physicians tended to report a preemptive approach for their patients and individually considered those with high risk for anti-CMV prophylaxis. Targeted prophylaxis has been stated in the guidelines in the setting of steroid-resistant cellular rejection that needs an additional immunosuppressant such as ATG. This practice is supported by the previously identified risk from receiving this lymphocyte-depleting agent in CMV-seropositive KT recipients \[[@CIT0003]\]. Reusing et al demonstrated some benefit of anti-CMV prophylaxis in CMV-seropositive KT recipients who received ATG in a retrospective study \[[@CIT0014]\]. More recently, Chiasakul and colleagues implemented targeted prophylaxis in those receiving a standard dose of ATG among high-risk patients such as those who underwent ABO incompatible KT with a favorable outcome \[[@CIT0004]\]. Chitasombat et al specifically conducted a cost-effectiveness analysis, revealing that few KT recipients who received ATG were prescribed oral valganciclovir as outpatients. Therefore, they developed CMV infection more frequently compared with those receiving oral valganciclovir, ultimately leading to a longer duration of hospitalization and direct and indirect costs of treatment for CMV infection compared with those without CMV reactivation \[[@CIT0015]\]. These data likely represent a burden of infectious complications among ATG-treated KT recipients, especially from CMV. More studies are required to better define the benefits of anti-CMV prophylaxis in these patient groups. In such individuals, anti-CMV prophylaxis is likely warranted, especially in those receiving immunosuppression augmented with ATG, and could be postulated for high-dose glucocorticoids. The practice setting in Thailand is likely to have an impact on decision-making in terms of prevention strategies. Although substantial portions of our respondents worked in a public hospital setting where patients rely on the national and social security budget, we found that only a small portion of physicians who reported using CMV-specific T-cell immunity to design an intervention for their patients. This intervention has been encouraged in a recent guideline, because each patient is considered to have a different state of immunity against CMV, depending on their level of risk and immunosuppression \[[@CIT0006], [@CIT0016]\]. Recent data supported measurement of patients' nonspecific and CMV-specific immunity to better stratify prevention strategies for each KT recipient \[[@CIT0017]\]. The most common barriers for establishing CMV prevention strategies in Thailand were lack of access to oral valganciclovir and quantitative CMV DNA testing due to high drug and laboratory assay costs, respectively. Based on the 2019 conversion rate of 32.5 THB to \$1 US, the cost of valganciclovir prophylaxis per 1 patient with normal glomerular filtration rate (900 mg/day) for 100 days was US \$7900, and the cost of weekly quantitative CMV DNA testing per 1 patient for 3 months was US \$936. Apart from the above mentioned, the impracticability of weekly measurement of plasma CMV DNA load was also an issue. Furthermore, a few participants reported an unacceptable side effect of bone marrow suppression from (val)ganciclovir. Our study has several limitations. First, we saw a relatively low response rate of respondents in the NP group. Second, only 1 ID and 1 NP were selected from each transplant center and these might not represent who the practice employed its entirety in that center. Lastly, other physicians who could be categorized outside of the ID and NP groups may be inadvertently excluded, even though they were potentially caring for KT recipients in some centers. Nonetheless, we have provided an overall picture of real-world CMV prevention awareness and practice in Thailand where resources are limited. This information will be helpful in highlighting the barriers faced, which must be addressed in a standardized guideline and direct efforts to implement a consistent practice for CMV prevention in Thai KT recipients. In summary, although most physicians agreed that CMV prevention was necessary for Thai KT recipients, a nationwide survey revealed a lack of uniformity in prevention strategies for CMV infection after KT. A preemptive approach is the most common intervention used, though prophylaxis is preferred in patients receiving intense immunosuppression. The financial implication of specific drugs and laboratory tests is the main barrier for CMV prevention in Thailand. A practical guideline for CMV prevention in KT recipients would be valuable in attempting to resolve these concerns. ***Author contributions. ***J.B. conceptualized, performed data curation and formal analysis, and wrote, reviewed, and edited the manuscript. A.B. performed data curation and reviewed and edited the manuscript. Sa.K. performed formal analysis and reviewed and edited the manuscript. And Su.K. reviewed and edited the manuscript. ***Financial support. *** None reported. ***Potential conflicts of interest. *** All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. All authors: No reported conflicts of interest. All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

Introduction {#s1} ============ *Toxoplasma gondii* is an obligate intracellular apicomplexan protozoan parasite that can infect all warm-blooded animals, mainly through oral and congenital infections (Tenter et al., [@B34]; Elmore et al., [@B11]; Zhou et al., [@B39]). This parasite has a significant medical and socioeconomic importance because it infects over two billion people worldwide. It is the causative agent of toxoplasmosis, an important zoonotic disease that can cause serious, even fatal health consequences in immunocompromised individuals, such as AIDS patients and organ transplant recipients (Liu et al., [@B22]; Xiao et al., [@B38]; Chemoh et al., [@B9]). Infection in immunocompetent individuals are generally asymptomatic, however the parasite persists as bradyzoites-containing cysts in brain and muscle tissues for many years. Primary infection in pregnant women can lead to fetus death, deformity, abortion, and long-term damage of the eye and central nervous system (Hill and Dubey, [@B19]). The remarkable ability of *T. gondii* to invade and colonize virtually all nucleated cells (Morisaki et al., [@B25]), and to adopt a successful intracellular lifestyle depends partly on the sequential secretion of effectors from the specialized secretory organelles micronemes, rhoptries and dense granules (Håkansson et al., [@B17]; Boothroyd and Dubremetz, [@B4]; Nadipuram et al., [@B26]). *T. gondii* tachyzoites enter the host cells through an active invasion mechanism and replicate within a membrane-bound parasitophorous vacuole (PV) inside the surrogate host cell cytoplasm (Coppens et al., [@B10]; Nam, [@B27]). After release from the host cell, newly formed parasites invade new mammalian host cells, and the replication cycle starts again. Most of the dense granule proteins (GRAs) are destined to the PV and parasitophorous vacuole membrane (PVM), and contribute to the biogenesis and maturation of the PV, and nutrient acquisition (Mercier and Cesbron-Delauw, [@B24]). Some GRA proteins have the ability to traffic to the host cytoplasm or nucleus and interfere with host cell signaling pathways. For example, *GRA6* is a polymorphic dense granule protein and activates the host transcription nuclear factor of activated T cells 4 (NFAT4) in order to manipulate host immune responses to maximize the parasite virulence in a strain-specific manner (Ma et al., [@B23]). *GRA15* is another strain-specific effector that activates NF-κB pathway and induces IL-12 secretion in type II, but not type I or III genotypes. *GRA15*-deficient type II strain cannot activate NF-κB pathway or induces IL-12 secretion, hence *GRA15*-deficient type II strains grow faster compared with wild-type strains (Rosowski et al., [@B29]). *GRA16* and *GRA24* also manipulate host gene expression and signaling pathways (Bougdour et al., [@B5]; Braun et al., [@B6]). *GRA17* mediates the transfer of small molecules between the host cell and PV, and maintains the stability of the PV (Gold et al., [@B16]). *GRA7, GRA25*, and *GRA39* are also virulence factor and can interfere with host cell signaling pathways (Alaganan et al., [@B1]; Shastri et al., [@B30]; Nadipuram et al., [@B26]). *GRA22* and *GRA41* are involved in the parasite egress (Okada et al., [@B28]; LaFavers et al., [@B21]). Despite the wealth of information regarding *T. gondii*\'s 40+ GRAs including sequence variation and expression of GRA coding genes, and their roles in the infection process, the contribution of many GRAs to the parasite growth and virulence are still unclear. The CRISPR-Cas9 system provides a novel and promising tool for editing *T. gondii* genes (Shen et al., [@B32]) and the genome (Sidik et al., [@B33]; Wang et al., [@B35]; Shen et al., [@B31]). Using CRISPR-Cas9 to target *T. gondii*\'s GRA genes may ultimately offer a new approach to achieving functional cure to toxoplasmosis. In this study, we used the CRISPR-Cas9 technique to edit 17 GRA genes, namely *GRA11, GRA12 bis, GRA13, GRA14, GRA20, GRA21, GRA28-31, GRA33-38*, and *GRA40* in the virulent *T. gondii* RH strain and examined the effects of gene loss on the parasite\'s ability to grow and exit from host cells *in vitro* and to cause death in mice. Materials and methods {#s2} ===================== Parasite and cell cultures -------------------------- Tachyzoites of *T. gondii* RH strain (type I) were maintained *in vitro* by passages in human foreskin fibroblast (HFF, ATCC, Manassas, VA, USA). HFF cells were grown in 75-cm^2^ tissue culture flasks containing Dulbecco\'s Modified Eagle medium (DMEM) supplemented with 10% fetal bovine serum (FBS) and 10 μg/ml gentamycin, at 37°C under a 5% CO~2~ atmosphere. To purify tachyzoites, infected HFF cells were lysed through a 27-gauge needle and the tachyzoites were filtered using a 5 μm pore size Millipore filter. The number of purified parasites was counted using a hemocytometer under a phase-contrast microscopy. Construction of GRA knockout *T. gondii* strains using CRISPR-Cas9 ------------------------------------------------------------------ *GRA* knockout strains were constructed by using CRISPR-Cas9 as previously described (Wang et al., [@B36]). All plasmids, primers, and gRNAs used in this study are listed in Table [S1](#SM1){ref-type="supplementary-material"}. Briefly, sgRNA of each *GRA* was engineered into pSAG1::CAS9-U6::sgUPRT by PCR using the Q5 Mutagenesis Kit (NEB). Positive plasmid was extracted with Endo-Free Plasmid DNA Mini Kit Protocols (OMEGA). The resistance cassettes (DHFR^\*^-Ts) were amplified from the plasmid pUPRT-DHFR-D and purified by agarose gel electrophoresis. About 40 μg positive plasmids and 15 μg purified DHFR^\*^-Ts amplicons were co-transfected into freshly harvested *T. gondii* RH tachyzoites by electroporation. *GRA*-deficient tachyzoites were selected with pyrimethamine and examined by PCR analysis. Stable clones were confirmed by reverse transcription PCR (RT-PCR) by comparison to WT strains. Total RNA was extracted from tachyzoites of wild type (WT) or Δ*GRA* mutant *T. gondii* strains using TRIzol (Invitrogen). Reverse transcription was performed using a PrimeScriptTM 1st Strand cDNA Synthesis Kit (TaKaRa). The central region of each target gene cDNA was amplified by RT-PCR using specific primers (see Table [S1](#SM1){ref-type="supplementary-material"} in the Supplemental Material). Assessment of parasite growth using plaque assay ------------------------------------------------ The growth rates of individual *GRA*-deficient and WT RH strains were determined in HFF cells. Cells were grown to confluence overnight in 6-well cell culture plates. The cells were then infected with tachyzoites of Δ*GRA* mutant and WT RH strains (\~200 tachyzoites/well). Cells were incubated for 3 h to allow the parasites to enter the host cells and were then washed twice with sterile 1 × phosphate-buffered saline (PBS) to remove unbound parasites, and fresh medium was added. The plates were incubated for 7 days at 37°C in 5% CO~2~ environment. Then, the culture medium was removed and infected HFF cells were fixed with 4% paraformaldehyde in PBS (pH 7.4) for 15 min at ambient temperature. Then, fixed cells were incubated with crystal violet staining solution (2% \[wt/vol\] crystal violet, pH 7.4) for 10 min at ambient temperature. The size of plaques (i.e., areas in the cell culture devoid of cells caused by the proliferating parasites) was determined using inverted microscope as previously described (Wang et al., [@B37]). This experiment was performed in triplicate and the experiments were repeated three independent times. Egress assay ------------ Asexual reproduction of *T. gondii* culminates with the egress of the newly formed tachyzoites from the surrogate host cell and subsequent parasite invasion into new host cells. Therefore, parasite egress represents a very important event which is indispensable for the parasite dissemination in the host body. In order to determine the role of the GRAs in this important process, we examined the effect of ionophore A23187 modulating Ca2^+^ homeostasis on the egress of the parasite from the host cells. Briefly, HFF cells were incubated with 10^5^ freshly harvested *T. gondii* tachyzoites in 2 ml culture medium for 3 h, followed by washing twice with DMEM medium to remove the unbound parasites. After 30--36 h of incubation, the wells were washed twice with sterile PBS and 3 μM calcium ionophore A23187 (Sigma) diluted in DMSO were added to the HFF cells. Live cell microscopy was used to monitor the timing of parasite egress from HFF cells infected with the WT strain compared with HFF cells infected with the mutant strains after addition of 3 μM calcium ionophore A23187. Mouse infection with *GRA*-deficient strains -------------------------------------------- Specific-pathogen-free (SPF) inbred female BALB/c mice (8 weeks-old) were purchased from Center of Laboratory Animals, Lanzhou Veterinary Research Institute, Lanzhou, China. Mice (5 mice/cage) were housed in a SPF environment within the animal care facility during the experiment. Each mouse was injected intraperitoneally with 200 freshly harvested tachyzoites of *GRA*-deficient or WT RH strain (10 mice per strain). The negative control mice were injected with an equal amount PBS only. All mice were monitored daily for the signs of illness and time of death. Bioinformatics analysis of *T. gondii* dense granule proteins ------------------------------------------------------------- Information on the genomic features (signal peptide, the number of exons and transmembrane domains) and time-series expression data of the *GRA* genes by the parasite cell cycle phases, parasite life cycle stages (the oocyst, tachyzoite, and bradyzoites), and the parasite genotypes were downloaded from ToxoDB (<http://ToxoDB.org>; Gajria et al., [@B14]). *GRA* gene expression data were processed using Robust Multiarray Average (RMA) algorithm of the Partek Genomics Suite package (Partek, Inc, St Louis, MO, USA). Statistical analysis -------------------- Statistical analysis for the *in vitro* and *in vivo* experiments was carried out using Graphpad Software package (GraphPad Software, La Jolla, CA). All experiments in the present study were conducted with at least 3 replicate and data were shown as means ± standard deviations (SD). Significant differences between means were determined by Student\'s *t*-test. *P*-values below 0.05 were considered significant. Results and discussion {#s3} ====================== In this study, we assessed the impact of targeted disruption of the individual 17 *GRA* genes (*GRA11, GRA12 bis, GRA13, GRA14, GRA20, GRA21, GRA28-31, GRA33-38*, and *GRA40*) on the ability of the virulent *T. gondii* RH tachyzoites to grow and exit from the cultured HFF cells. We also examined the impact of the disruption of *GRA* genes on the virulence of *T. gondii* in BALB/c mice. Generation of single Δ*GRA* knockout RH strains ----------------------------------------------- CRISPR-Cas9 technique was used to disrupt the *GRA* genes in type I RH strain. We designed RNA-guided CRISPR-Cas9 targeting *T. gondii* 17 *GRA* genes individually. The larger pyrimethamine-resist fragment (DHFR^\*^-Ts) was designed for inserting into the sgRNA-targeted coding region causing frameshift mutation for GRA proteins (Figure [1A](#F1){ref-type="fig"}). Then, single clones were identified by PCR, and small fragment (\~500 bp) was not amplified in *GRA*-deficient strains due to the insertion of the large DHFR^\*^-Ts fragment with short extension time, however it was detected in the WT strain (Figures [1B,C](#F1){ref-type="fig"}). RT-PCR was used to amplify cDNA products, reverse transcribed from mRNA, of each *GRA* gene. Our result showed that CRISPR-Cas9 abolished the expression of *GRA* genes in the transfected strains compared with the WT strains (Figures [1D,E](#F1){ref-type="fig"}), indicating that the target *GRA* genes were mutated at the Cas9 cleavage sites and that 17 Δ*GRA* mutant strains were successfully generated. {#F1} Deletion of *GRA* genes did not affect the parasite growth and egress --------------------------------------------------------------------- After confirming knockout by RT-PCR, we compared the plaque formation in HFF monolayers grown in 6-well culture plates 7 days after infection with 200 tachyzoites of wild-type and Δ*GRA* mutant strains by conventional crystal violet staining and microscopic examination. As shown in Figure [2A](#F2){ref-type="fig"}, no significant differences in the size of plaques were observed in cells infected with wild-type compared to cells infected with Δ*GRA* mutant strains (*p* = 0.1582). Next, we evaluated the role of *GRAs* in the parasite egress. The infected HFF cells were treated with 3 μM calcium ionophore A23187 and the timing of the parasite exit from the cells was monitored over 5 min (Figure [2B](#F2){ref-type="fig"}). No significant difference in the parasites egress was observed between Δ*GRA* mutants and WT strains. The 17 *GRA* KO strains and the WT strains remained within the PV after stimulation with DMSO. Calcium is a critical mediator of *T. gondii* invasion and egress processes. The calcium ionophore A23187 can stimulate the parasite to exit from the PV to the extracellular space (Arrizabalaga et al., [@B2]; Caldas et al., [@B8]). *GRA41* can regulate the timing of egress and the sensitivity to calcium (LaFavers et al., [@B21]). However, none of the 17 *GRA* mutant strains tested in the present study seem to be responsive to the action of the A23187. In agreement with our results, a previous study has shown that deletion of 15 rhoptry organelle proteins ROPs (*ROP10, ROP11, ROP15, ROP20, ROP23, ROP31, ROP32, ROP33, ROP34, ROP35, ROP36, ROP40, ROP41, ROP46*, and *ROP47*) did not suppress the parasite\'s ability to grow in HFF cells or alter its pathogenicity for BALB/c mice (Wang et al., [@B37]). This lack of effect of *GRA* deletion on the phenotype of single parasite mutants argues for possible redundancy of function for *GRAs*. {#F2} *GRA* deletion did not alter the parasite virulence in mice ----------------------------------------------------------- We tested whether deletion of *GRA* genes in *T. gondii* RH strains would cause any changes in the parasite virulence. We inoculated female BALB/c mice intraperitoneally with 200 *T. gondii* tachyzoites (wild-type or mutants) and monitored the mortality and signs of illness on a daily basis. All mice died within 7--9 days (Figure [3](#F3){ref-type="fig"}), suggesting that the 17 *GRA* genes tested in the present study might not contribute to the parasite virulence during infection. Previous studies have shown that some of dense granule proteins such as *GRA25* may regulate the production of CCL2 and CCL1 in macrophages and thus affect virulence, which was different between Type II and Type III strains (Shastri et al., [@B30]). *GRA39* is also an important virulence factor in Type II strains (Nadipuram et al., [@B26]). However, our results did not show any significant differences between WT and Δ*GRA* mutant strains. The fact that Δ*GRA* strains did not display any reduction or loss of pathogenicity in mice suggests that deletion of a single gene might not be enough to influence the highly virulent RH strain in which the knockout was performed. More research is required to elucidate the impact of individual or multiple *GRA* gene deletion on the phenotype and virulence of *T. gondii* strains. It is possible that these *GRA* genes have roles in the pathogenesis of *T. gondii*, but in *T. gondii* strains of other genotypes or in other hosts. Disruption of these GRAs in avirulent and more physiologically relevant cystogenic strains may allow the assessment of more subtle roles in the virulence, alteration in cyst formation or tissue tropism. {#F3} Sequence and expression analyses of *GRAs* in *T. gondii* --------------------------------------------------------- More than one-third of *T. gondii* mRNAs exhibit a tightly regulated expression pattern, probably driven by the diverse cell-cycle-dependent processes mediated by the parasite in the course of infection (Behnke et al., [@B3]). We mapped the transcriptomic data of the 17 *GRAs* available in ToxoDB and found that the expression profile of the majority of *GRAs* do not follow a particular cell cycle pattern, and that *GRA11* and *GRA12 bis* expression level was low (Figure [4A](#F4){ref-type="fig"}). Bioinformatics features of *GRAs*, such as the number of exons, signal peptide and transmembrane domains are summarized in Table [1](#T1){ref-type="table"}. The majority of the known *GRAs* include a signal peptide at the N terminus and a single transmembrane domain in the C-terminal part of the protein. We found that some GRAs encode multi-exons (even eight) and a small amount of *GRAs* do not have a signal peptide or a transmembrane domain. The signal peptide plays an important role in protein targeting and protein translocation in and eukaryotic cells and is considered as a feature of the GRA proteins that enter the secretory pathway (Hakimi and Bougdour, [@B18]). Most of the previously identified GRA proteins have been predicted to contain classical or non-classical signal peptide, such as *GRA1, GRA2, GRA6, GRA9, GRA16*, and *GRA21*, whereas other GRAs (*GRA10, GRA15, GRA20*, and *GRA22*) did not (Mercier and Cesbron-Delauw, [@B24]). However, a few GRAs do not seem to depend on signal peptide to enter the secretory pathway, such as *GRA5*, which is secreted into the PV as a unique soluble protein and then binds to the PVM (Gendrin et al., [@B15]). Some GRA proteins lack the transmembrane domain, typically *GRA19, GRA20*, and *GRA21*, and may interact with the PVM through protein-protein interactions (Hsiao et al., [@B20]). Other GRAs (38, 39, and 40) lack the transmembrane domains or other identifiable sequences for membrane association, suggesting they are soluble protein of the PV (Nadipuram et al., [@B26]). It is possible that GRAs that lack signal peptide or transmembrane domain perform their roles by as yet unknown mechanisms. ![The trend charts of the distinct expression profiles of *Toxoplasma gondii GRAs***. (A)** Time-series expression profile of 17 *GRA* genes of *T. gondii* RH strain by cell cycle phases of the parasite as described by Behnke et al. ([@B3]). **(B)** Transcriptomic expression profiles of 17 *GRA* genes in Type I (RH and GT1), Type II (Pru and ME49), and Type III (CTG and VEG) strains. **(C)** Transcriptomic profiles of 17 *GRA* genes related to the parasite life cycle stages (oocyst, tachyzoite and bradyzoite). Expression profile of 17 *GRA* genes of the oocysts recovered from cat feces, at 0 day (unsporulated), 4 days (4 day sporulated), and 10 days (10 day sporulated), tachyzoites grown for 2 days in HFF cells (2 day *in vitro*), bradyzoites grown in HFF cells for 4 days and 8 days (4 day *in vitro* and 8 day *in vitro*), and 21 days tissue cyst-containing bradyzoites harvested from infected mouse brains (21 day *in vivo*). Each line represents the expression value of the corresponding gene. The data were obtained from ToxoDB (36 release) and the graph was generated using GraphPad Prism version 5.0.](fcimb-08-00300-g0004){#F4} ###### Bioinformatics features of GRA proteins of *Toxoplasma gondii*. **Name** **Gene ID** **Product description** **Exons** **Phenotype value** **TMHMM[^a^](#TN1){ref-type="table-fn"}** **Predicted signal peptide** **Acute infection [^b^](#TN2){ref-type="table-fn"}** **Chronic infection [^b^](#TN2){ref-type="table-fn"}** ----------- ------------------------------ ----------------------------- ----------- --------------------- ------------------------------------------- ------------------------------ ------------------------------------------------------ -------------------------------------------------------- GRA11 [TGGT1_212410](TGGT1_212410) Dense granule protein GRA11 1 ND Yes Yes 5.97 5.21 GRA12 bis [TGGT1_275850](TGGT1_275850) Dense granule protein GRA12 3 1.31 No Yes 3.00 2.63 GRA13 [TGGT1_237880](TGGT1_237880) Hypothetical protein 1 0.76 Yes Yes 120.50 34.28 GRA14 [TGGT1_239740](TGGT1_239740) Dense granule protein GRA14 1 2.00 No Yes 116.36 53.11 GRA20 [TGGT1_200010](TGGT1_200010) Hypothetical protein 2 2.54 No No 37.97 30.74 GRA21 [TGGT1_241610](TGGT1_241610) Hypothetical protein 2 ND No Yes 18.40 13.56 GRA28 [TGGT1_231960](TGGT1_231960) Putative omega secalin 3 1.48 No Yes 33.28 3.58 GRA29 [TGGT1_269690](TGGT1_269690) Hypothetical protein 1 1.54 Yes Yes 47.08 77.60 GRA30 [TGGT1_232000](TGGT1_232000) Hypothetical protein 2 2.29 Yes Yes 33.46 44.61 GRA31 [TGGT1_220240](TGGT1_220240) Hypothetical protein 1 0.90 Yes Yes 33.92 22.57 GRA33 [TGGT1_247440](TGGT1_247440) Hypothetical protein 1 1.72 Yes Yes 48.06 45.42 GRA34 [TGGT1_203290](TGGT1_203290) Hypothetical protein 1 2.26 No Yes 64.29 248.97 GRA35 [TGGT1_226380](TGGT1_226380) Hypothetical protein 1 1.98 Yes Yes 48.74 32.29 GRA36 [TGGT1_213067](TGGT1_213067) Hypothetical protein 1 −0.21 Yes Yes 83.85 24.24 GRA37 [TGGT1_236890](TGGT1_236890) Hypothetical protein 2 2.10 Yes No 60.01 34.21 GRA38 [TGGT1_312420](TGGT1_312420) Hypothetical protein 4 −1.15 No Yes 27.17 17.82 GRA40 [TGGT1_219810](TGGT1_219810) Hypothetical protein 7 0.69 Yes No 114.88 97.16 *ND, Not determined*. *Prediction of transmembrane helices was performed using the TMHMM program version 2.0*. *Gene expression levels of fragments per kilobase of exon model per million mapped reads (FPKM) at acute (10 days post infection) and chronic infection (28 days post infection). Source: <http://www.toxodb.org/toxo/> \[accessed 26 July 2018\]*. We next analyzed the transcriptomic levels of 17 GRAs in different *T. gondii* genotypes (Type I, II, and III), and found that *GRA20* and *GRA34* were significantly different among the three genotypes (Figure [4B](#F4){ref-type="fig"}). This difference may be because these two GRAs play different roles in different strains. Levels of the transcriptomic expression of 17 *GRA* genes across *T. gondii* developmental stages are presented (Figure [4C](#F4){ref-type="fig"}). Our analysis showed that most GRAs are differentially expressed at different life cycle stages, but *GRA11* and *GRA21*. The expression of some *GRA* genes can be specifically up- or down-regulated during parasite development (Fritz et al., [@B13]). For instance, the expression of *GRA7* was significantly reduced in *in vitro* 4 day bradyzoites compared to that in the *in vitro* 2 day tachyzoites and then restored to near tachyzoite\'s levels in the *in vitro* 21 day bradyzoites (Buchholz et al., [@B7]). The expression of *GRA4, GRA6*, and *GRA8* has been reported to be reduced or even non-detectable in the bradyzoites (Ferguson, [@B12]). The different expression levels of some *GRAs* in different life cycle stages indicate that the roles played by *GRAs* may play parasite stage-dependent. Conclusion {#s4} ========== GRAs play key roles in modulating host-parasite interactions, such as parasite vacuole remodeling, nutrient uptake and manipulation of host signaling pathways (Nadipuram et al., [@B26]). In this study, 17 Δ*GRA* mutant *T. gondii* strains were successfully generated using CRISPR-Cas9 technique. The role of these 17 *GRA* genes in the pathogenicity of *T. gondii* RH strain was investigated *in vitro* and *in vivo*. We report here that no significant difference was detected between Δ*GRA* knockouts and wild-type RH strains. These findings indicate that *GRA* genes examined in this study are not absolutely essential for *T. gondii* RH virulence, suggesting that other virulence factors may be involved in these processes or that virulence of *T. gondii* is the result of a multigene effect. We also investigated the patterns of gene expression and bioinformatics features of the 17 *GRAs*, by parasite cell cycle phases, life cycle stages and genotypes. Results indicated that the expression of *GRAs* can vary across life cycle stages or genotypes of *T. gondii*. Functional analysis of these *GRAs* in other parasite strains or life cycle forms is therefore of high importance and may further elucidate the pathogenic role of *GRAs* in *T. gondii* infection. Ethics statement {#s5} ================ The study was approved by the Animal Administration and Ethics Committee of Lanzhou Veterinary Research Institute, Chinese Academy of Agricultural Science (Permit No. LVRIAEC-2017-006). All mice were handled humanely in strict accordance with the Guidelines and Animal Ethics Procedures of the People\'s Republic of China. Author contributions {#s6} ==================== X-QZ, J-LW, and HE designed the study and critically revised the manuscript. M-JB performed the experiments, analyzed data, and drafted the manuscript. Q-LL, KC, and L-BN participated in the implementation of the study. All the authors read and approved the final manuscript. Conflict of interest statement ------------------------------ 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. We thank Professor Bang Shen (Huazhong Agricultural University) for providing the pSAG1::CAS9-U6::sgUPRT and pUPRT-DHFR-D vectors. **Funding.** Project financial support was kindly provided by the International Science and Technology Cooperation Project of Gansu Province (Grant No. 17JR7WA031), by the Elite Program of Chinese Academy of Agricultural Sciences, and by the Agricultural Science and Technology Innovation Program (ASTIP) (Grant No. CAAS-ASTIP-2016-LVRI-03). Supplementary material {#s7} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fcimb.2018.00300/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Kenneth Pfarr, Universitätsklinikum Bonn, Germany [^2]: Reviewed by: Sabrina Absalon, Boston Children\'s Hospital and Harvard University, United States; Renato Augusto DaMatta, State University of Norte Fluminense, Brazil [^3]: This article was submitted to Parasite and Host, a section of the journal Frontiers in Cellular and Infection Microbiology

Introduction {#s1} ============ Sensory experience plays an essential role in establishing and refining neural organization during development. In fact, abnormal input or experiential modifications due to injury can produce severe changes in neuronal morphology including alterations in dendritic length, cell size, and can even influence the fate of a cell [@pone.0111243-LeviMontalcini1]--[@pone.0111243-Smith1]. Second-order auditory neurons in the chick brain stem have proven useful in investigating the importance of sensory experience during development [@pone.0111243-Rubel1]. Neurons in the cochlear nucleus, nucleus magnocellularis (NM) receive their sole excitatory input from the cochlea via the ipsilateral eighth nerve [@pone.0111243-Boord1], [@pone.0111243-Parks1]. Thus, unilateral cochlear ablation results in a complete loss of excitatory afferent input to the ipsilateral NM but leaves the contralateral side unaffected, allowing for within-subject comparisons of NM neurons on the intact versus the deafened side of the brain. Elimination of auditory nerve activity during a sensitive period of development leads to the death of approximately 20--30% of NM neurons on the deafened side of the brain [@pone.0111243-Born1]. The remaining neurons survive, albeit at a lower metabolic rate and with a reduction in soma size [@pone.0111243-Born1]--[@pone.0111243-Steward1]. The factors that determine which cells die and which cells survive still remain unclear; however, the chain of events that occur within NM as a consequence of deafening are well documented. Within 1 to 3 hrs following cochlear ablation, NM neurons show a threefold increase in intracellular calcium (\[Ca^2+^\]~i~) [@pone.0111243-Zirpel1], and a decrease in RNA and protein synthesis [@pone.0111243-Steward1], [@pone.0111243-Garden1], [@pone.0111243-Garden2]. At 6--12 hrs, deafened NM neurons appear to segregate into two populations: one population suffers a complete cessation of protein synthesis and eventually goes on to die, while the other population continues to synthesize proteins, albeit at a reduced level, and goes on to survive [@pone.0111243-Steward1]. The cessation of protein synthesis appears to be due to the dissociation of polyribosomes in NM neurons following cochlear ablation [@pone.0111243-Rubel2]. One way to visualize the rapid activity-dependent changes that occur in ribosomes following deafening is by using Y10B, a monoclonal antibody that recognizes a ribosomal epitope [@pone.0111243-Garden1], [@pone.0111243-Garden2], [@pone.0111243-Garden3], [@pone.0111243-Hyson1]. Changes in Y10B immunoreactivity match the changes observed in overall protein synthesis [@pone.0111243-Garden3], and these changes correspond with changes observed in ribosomes at the EM level [@pone.0111243-Garden2], [@pone.0111243-Rubel2]. Consequently, Y10B immunoreactivity provides an efficient and meaningful assay for examining changes in NM ribosomes, one of the earliest cellular changes to occur following deafferentation. Studies directed at identifying the trans-synaptic signals necessary for preventing the early changes that occur in NM neurons following cochlea removal have made use of an *in vitro* slice preparation of the chick auditory brain stem. In this condition, NM activity is eliminated bilaterally since both cochleae are destroyed. In order to mimic the case of unilateral deafening, auditory nerve fibers are electrically stimulated on one side of the slice while fibers on the opposite side of the same slice remain unstimulated. Within 1 hr, NM neurons on the stimulated side of the slice show greater protein synthesis [@pone.0111243-Hyson2] and Y10B labeling [@pone.0111243-Hyson1], [@pone.0111243-Hyson3], [@pone.0111243-Hyson4] than the neurons on the unstimulated side. Through the use of the *in vitro* slice preparation, it has been shown that glutamate\'s action on ionotropic glutamate receptors (iGluRs) is responsible for the electrical activity in postsynaptic NM neurons [@pone.0111243-Hyson3], [@pone.0111243-Nemeth1], and glutamate\'s action on metabotropic glutamate receptors (mGluRs) is responsible for providing trophic support to NM neurons [@pone.0111243-Hyson4], [@pone.0111243-Nicholas1], [@pone.0111243-Zirpel2]. Previous *in vivo* studies have also shown that mGluR activation is necessary for maintaining ribosomal antigenicity of Y10B and for the ultimate survival of NM neurons [@pone.0111243-Carzoli1]. For example, application of selective group I or group II mGluR antagonists results in NM neuronal cell death even with an intact cochlea. Although it is clear that mGluR activation is [necessary]{.ul} to preserve NM neurons in a healthy state, it is not known if mGluR activation is [sufficient]{.ul} to maintain these neurons or if other activity-dependent factors are also required. The present experiments evaluate whether or not activating mGluRs on deafferented NM neurons is sufficient for maintaining ribosomes in a healthy state during the period immediately following deafferentation. If adding mGluR activation to the deprived NM is sufficient to regulate NM neuronal ribosomes, then application of mGluR agonists should preserve ribosomal Y10B antigenicity, which is normally lost in the absence of auditory experience. Materials and Methods {#s2} ===================== Subjects {#s2a} -------- All subjects were 1 to 4 day old chicks of either sex, hatched from eggs obtained from a local supplier and reared at Florida State University. The procedures used in these experiments were approved by the Animal Care and Use Committee at Florida State University and conform to the guidelines set forth by the National Institutes of Health. All efforts were made to minimize the number of animals used and their suffering. Slice preparation {#s2b} ----------------- Subjects were anesthetized with isoflurane and decapitated. A 4 mm segment of the caudal skull containing the brain stem was removed with a razor blade, quickly submerged in room temperature artificial cerebral spinal fluid (ACSF), and was oxygenated using a 95% O~2~, and 5% CO~2~ gas mixture. ACSF consisted of (in mM) 130 NaCl, 3 KCl, 2 CaCl~2~, 2 MgCl~2~, 26 NaHCO~3~1.25 NaH~2~PO~4~ and 10 dextrose. The brain stem was rapidly dissected from the skull segment and mounted onto a custom-built stage for slicing using cyanoacrylate glue with additional support provided by a 30% gelatin compound. Using a vibrating blade microtome, a 300 µm, bilaterally symmetrical, coronal slice containing NM was obtained, transferred to a submersion-type recording chamber, and perfused (2--3 ml/min by gravity) with oxygenated ACSF. The slice was anchored in the recording chamber using nylon-strung metal "harp". Drug administration {#s2c} ------------------- Once the brain slice was placed in the chamber, a glass micropipette was lowered into the bath and positioned over the slice toward the lateral side of the IVth ventricle. Drugs were ejected onto one side of the brain slice using periodic pressure pulses (10 psi, 10--50 msec duration) through a picospritzer (General Valve Corporation, Fairfield, N.J.). The slice was oriented such that the laminar flow of ACSF through the chamber carried the drug only to NM on the outflow side of the chamber. Different groups of slices (n = 4 per group) were unilaterally treated with either vehicle, glutamate (1 mM), the general mGluR agonist, ± trans 1-amino-1,3-cyclopentanedicarboxylic acid (ACPD) (400 µM), a selective group I mGluR agonist, 3,5-dihydroxyphenylglycine (DHPG) (200 µM), or a selective group II mGluR agonist, LY354740 (200 nM) for 1 hr. Drug concentrations in the pipette were selected based on their EC~50~ and predicting an approximately10-fold reduction in concentration by the time the drug flowed to NM. A dose-response curve for glutamate was constructed using concentrations of L-glutamic acid ranging from 0.01 to 10 mM in the ejection pipette (n = 3--4 slices per group). The vehicle was dextrose-free ACSF. Confirmation of unilateral drug application {#s2d} ------------------------------------------- Preliminary tests using fast green were performed to monitor the distribution of solution released by pressure application in order to ensure drug was being administered to only one side of the slice and that minimal backflow was affecting the opposite side of the same slice. Additional confirmation was obtained by ejecting a 1% solution of a fluorescent dextran, tetramethylrhodamine (fluoro-ruby) 10,000 MW (Invitrogen), from the micropipette. Distribution of dye clearly showed that ejected substances primarily, if not exclusively, reached only one NM (see [Figure 1](#pone-0111243-g001){ref-type="fig"}). Dye was used in slices to document the application procedure and was not included for drug-treated slices, so as not to interfere with immunoreactivity. {#pone-0111243-g001} Tissue preparation {#s2e} ------------------ At the end of the 1-hr period of unilateral drug application, a notch was cut into the ventral portion of the side of the slice that received drug. The section was then post-fixed in 4% paraformaldehyde for 1 hr and cryoprotected in a 30% sucrose/phosphate-buffered saline (PBS) solution overnight. Since all comparisons were performed between opposite sides of the same tissue section, it was necessary to preserve symmetry between sides. This was accomplished by assuring that the slice was mounted flat for sectioning. A mound of tissue freezing medium was built up onto a cryostat chuck to approximately ½ inch in height. The chuck was then loaded into the cryostat and the top of the mound was sectioned off to obtain a flat surface. The 300 µm tissue slice was then placed onto a silanized microscope slide and excess fluid was removed with a kimwipe. The cryostat chuck was then placed on dry ice and a small amount of tissue freezing medium was deposited onto the flattened surface. The slide holding the brain slice was placed tissue-side-down into the fresh medium and a pellet of dry ice was placed on top of the slide to ensure rapid freezing. The microscope slide was then detached using a razor blade, leaving the brain slice embedded in the freezing medium. The mounted brain slice was allowed to equilibrate in the cryostat at −20°C for at least 30 min prior to sectioning. Using a Leica CM 1850 cryostat (Leica Microsystems Inc., Bannockburn, IL, USA), 20 µm sections were then collected from the mounted slice and were free-floated in a vial containing ice-cold PBS. Immunohistochemistry {#s2f} -------------------- Following sectioning, endogenous peroxidase activity was quenched by placing sections in 0.03% H~2~O~2~ in methanol for 20 min. After quenching, sections were rinsed 3×10 min in PBS, and were then preincubated in a 4% normal horse serum solution for 10 min. Sections were then incubated on a rotator overnight at room temperature in 1∶500 Y10B, a mouse monoclonal antibody that recognizes a ribosomal epitope. The Y10B antiserum was raised from a clone originally developed by J. Steitz and subsequently supplied to our laboratory by E. Rubel. This antibody is an established marker of changes in ribosomes [@pone.0111243-Lerner1], and has been used extensively as an indicator of the early changes resulting from deafferentation (e.g., [@pone.0111243-Garden2], [@pone.0111243-Garden3], [@pone.0111243-Hyson1], [@pone.0111243-Carzoli1], [@pone.0111243-McBride1]). Changes in antigenicity appear to match changes in overall protein synthesis and a deafferentation-induced reduction in Y10B labeling has been confirmed at the EM level [@pone.0111243-Garden2]. Control sections processed without primary antibody show no labeling. The following day, sections were washed 3×10 min in PBS then incubated for 1 hr in 1∶200 biotinylated horse anti-mouse secondary antibody. Sections were washed 3×10 min in PBS and then incubated in avidin-biotin-peroxidase complex (Vectastain ABC Kit, Vector Laboratories, Burlingame, CA, USA) for 1 hr. Following 2×10 min washes in PBS and a 10 min wash in 0.1 M phosphate buffer (PB), sections were reacted with diaminobenzidine tetrahydrochloride and 0.03% H~2~O~2~ for 15 min. Sections were mounted onto slides following 2 10-min washes in 0.1 M PB, and allowed to dry overnight. The next day, slides were cleared though graded ethanols and xylenes and then coverslipped using DPX mounting medium (Sigma-Aldrich, St. Louis, MO, USA). Densitometry {#s2g} ------------ To objectively analyze the level of immunolabeling, densitometric measurements of NM neurons were obtained using a digital image analysis program (NIH ImageJ). The staining densities of NM neurons on the drug-treated versus the untreated side of the same tissue section were compared. This within-subjects method improves statistical power. Although immunohistochemistry cannot provide quantitative differences in the amount of protein changed, it has crucial advantages over other methods (e.g., Western analysis) by selectively evaluating changes in neurons and examining if changes occur in all neurons or only a subpopulation of cells. All comparisons were between opposite sides of the same tissue section. This assures that variation in labeling intensity is attributable to the experimental treatment and not to variation in immunohistochemistry processing variables (e.g., incubation time in any of the reagents). For these analyses, the light levels and camera settings remained constant. Cells were visualized using a 40X objective and mean gray scale densities over approximately 40 NM neurons/side in a given tissue section were measured beginning with cells at the medial edge of NM and proceeding laterally. At least 3 sections from each subject were measured. The investigator analyzing the tissue remained blind to the identities of the drug-treated and the untreated sides of the section as well as to the drug group until after measurements were obtained. Differences in the density of labeling between sides of the slice that received ACSF or drug were compared using analysis of variance (ANOVA). Results {#s3} ======= Vehicle treatment to NM on one side of the slice for 1 hr had no apparent effect on Y10B immunoreactivity. Application of glutamate (1 mM) resulted in darker Y10B immunolabeling as did the general mGluR agonist, ACPD 400 µM. Relatively low concentrations of the group I mGluR agonist, DHPG (200 µM), and the potent group II mGluR agonist, LY354740 (200 nM), also resulted in darker Y10B labeling on the drug-treated side of the slice in comparison to the side of the slice that remained unexposed to drug. An example of this effect can be seen in [Figure 2A, B](#pone-0111243-g002){ref-type="fig"}. {#pone-0111243-g002} Visual impressions were confirmed through objective analyses of labeling density. The labeling densities of individual NM neurons on the drug-treated versus the untreated side of the same tissue slice were compared. A two-way mixed ANOVA was performed on the gray scale measurements using side of the section as the within-subject variable and drug as the between-subjects variable. This analysis revealed no reliable effect of drug treatment (F (4, 15) \<1), an overall effect of side (F (1, 15)  = 52.55, *p*\<0.001) and, importantly, a reliable drug treatment X side interaction (F (4, 15)  = 5.16, *p*\<0.01). Post hoc (Newman-Keuls) pair-wise comparisons revealed that all agonist-treated groups showed a reliable difference between sides (p\<0.05) whereas control-treated slices demonstrated no reliable difference between sides (see [Figure 2C](#pone-0111243-g002){ref-type="fig"}). The protective effect of glutamate is somewhat surprising given its known excitotoxic effects in other systems (for review see [@pone.0111243-Choi1]). Consequently, a dose response curve was generated over a broad range of concentrations (0.01 to 10 mM) in the application pipette. A relatively low concentration of L-glutamic acid (0.01 mM) produced no difference in Y10B labeling between the glutamate-treated and untreated side of the same slice. Treatment with 0.1 mM glutamate showed only a hint of an effect, but treatment with 1 mM glutamate replicated the first experiment and clearly yielded darker ribosomal labeling in NM neurons on the treated side of the slice compared to neurons on to the untreated side of the same slice. While the percent difference between sides following 1 mM glutamate treatment was slightly lower that that obtained in the first experiment, this difference between experiments was not statistically significant (t-test, p = .22). A high concentration (10 mM) had the opposite effect, in that the glutamate had detrimental effects on Y10B labeling on the glutamate-treated side of the slice when compared to the untreated side of the same slice. Visual impressions were confirmed through objective analyses of staining density. The staining densities of individual NM neurons on the glutamate-treated versus the untreated side of the same tissue slice were compared. A two-way mixed ANOVA was performed on the gray scale density measurements using side of brain as the within-subjects variable and glutamate concentration as the between-subjects variable. Neither main effect of side (F (1,10)  = 0.34, p\<1)) nor effect of concentration (F (3,10)  = 3.34, p = 0.063)) was statistically significant. Importantly, there was a statistically significant Side X Concentration interaction (F (3,10)  = 20.08, p\<0.0001). To further compare the effect of glutamate concentration between groups, density measurements were normalized across brains by transforming gray scale scores to percent difference ((treated-untreated)/treated \*100). A one-way ANOVA on the percent difference scores revealed a reliable dose effect (F (3,13)  = 13.3, p\<0.001). Post hoc (Newman-Keuls) pair-wise comparisons (p\<0.05) revealed that the percent difference in labeling for the 1 mM group was significantly higher than the 0.01 mM group but was only marginally different from the 0.1 mM group (p = 0.06). The 10 mM group showed a statistically reliable percent difference in labeling between sides, albeit in the opposite direction of all other groups (see [Figure 2D](#pone-0111243-g002){ref-type="fig"}). Discussion {#s4} ========== It is widely accepted that early experience is essential to normal brain development. Atypical sensory input during development can have dramatic and potentially damaging effects on the central nervous system that can persist into maturity. A number of classic experiments have demonstrated this by restricting the amount of experience an animal receives through a single sensory modality. Cell death is observed following sensory deprivation in the visual [@pone.0111243-Nucci1], somatosensory [@pone.0111243-Baldi1], olfactory [@pone.0111243-Brunjes1]--[@pone.0111243-Skeen1], vestibular [@pone.0111243-Peusner1], and auditory systems of both chicks and mammals [@pone.0111243-LeviMontalcini1], [@pone.0111243-Webster1], [@pone.0111243-Born1], [@pone.0111243-Hashisaki1], [@pone.0111243-Tierney1]. Despite this common finding across developing sensory systems, relatively little is known about the trophic mechanisms that govern such activity-dependent cell survival. The brain stem auditory system of the chick has proven to be a useful model system for examining the transneuronal signals necessary for activity-dependent cell survival. This is because deafening is relatively easy to produce in the chick. Moreover, the large cochlear nucleus cells form a relatively uniform population of neurons that undergo rapid and robust changes following deafferentation, leading to the ultimate death of a subpopulation of these cells. One early alteration that occurs following deafness in NM neurons is a change in ribosomal activity [@pone.0111243-Steward1]. This change in function corresponds with a reduction in antigenicity for Y10B, a monoclonal antibody that recognizes a ribosomal epitope [@pone.0111243-Garden1], [@pone.0111243-Garden2], [@pone.0111243-Garden3], [@pone.0111243-Hyson2], [@pone.0111243-Lerner1]. A number of studies have made use of the Y10B assay to examine the transneuronal signals necessary for the activity-dependent regulation of NM neurons. Earlier studies, have shown that electrical activity of the postsynaptic NM neuron is not sufficient to maintain NM neuronal ribosomes [@pone.0111243-Hyson1], [@pone.0111243-Hyson2], suggesting activity-dependent release of some important trophic factor from the active auditory nerve terminal. However, blockade of iGluR receptors, which are responsible for driving the electrical activity of NM neurons [@pone.0111243-Nemeth1], does not prevent the activity-dependent regulation of ribosomal antigenicity [@pone.0111243-Hyson3]. This implies that neither iGluR activation nor postsynaptic electrical activity is necessary to keep NM neurons in a healthy state. On the other hand, blockade of mGluRs does prevent this activity-dependent regulation [@pone.0111243-Hyson4], [@pone.0111243-Nicholas1]. The importance of mGluR activation has also been confirmed *in vivo* where it has been shown that blockade of mGluRs reduces ribosomal antigenicity and produces NM neuronal death even with an intact cochlea [@pone.0111243-Carzoli1]. The current set of experiments demonstrate that providing mGluR activation to NM neurons is sufficient, at least in the early stages that follow deafferentation, to maintain healthy ribosomes. Application of glutamate, a general mGluR agonist (ACPD), a selective group I mGluR agonist (DHPG), or a selective group II mGluR agonist (LY354740), effectively preserved ribosomal antigenicity in NM neurons on the treated side of the slice in comparison to those on the untreated side. Together with previous studies of this system, it appears that mGluR activation is both necessary and, in part, sufficient to provide trophic support to NM neurons, which undergo degenerative changes following the loss of afferent activation that results from deafness. One overall limitation of the studies discussed thus far is that an *in vitro* slice preparation isolates an individual brain slice from the rest of the body. This allows for the possibility that alternative sources of trophic support have also been eliminated in the slice preparation. Consequently, it is not known if mGluR activation is truly required for maintaining neuronal integrity in the intact system or if mGluR activation is only required when other forms of trophic support are no longer present. If such factors exist, however, they are released independent of auditory nerve activity, since there is no spontaneous activity in either the slice preparation or *in vivo* following cochlea ablation. Since the activity-dependent regulation of ribosomes is observed *in vitro*, any such co-factors are not circulating in the cerebrospinal fluid *in vivo*. The prospective co-factor is also unlikely to be glutamate acting at iGluRs since the present studies used selective agonists for mGluRs and previous studies have shown no effect of total iGluR blockade [@pone.0111243-Hyson3]. Consequently, if a co-factor is required, it is present in deafferented NM *in vitro* and could be any number of substances provided by glial cells or by synaptic release from neuron terminals that is independent of auditory nerve action potentials. MGluRs and neuroprotection {#s4a} -------------------------- There are various mechanisms by which the different mGluRs could provide for neuroprotection, and these mechanisms can be generally categorized as having either direct or indirect effects on the neuron. One of the direct effects by which mGluRs could be protecting NM neurons is through the control of Ca^2+^ homeostasis. Manipulations of auditory nerve input to NM neurons and *in vitro* pharmacological treatments produce changes in \[Ca^2+^\]~i~ that correlate with the Y10B assay used in the present set of experiments. For example, in the absence of activity there is a rapid rise in \[Ca^2+^\]~i~, which can be reversed by electrical stimulation of the auditory nerve or mGluR activation [@pone.0111243-Zirpel1], [@pone.0111243-Zirpel2]. Additionally, blockade of mGluRs results in a dramatic rise in \[Ca^2+^\]~i~, even if afferent activity is provided [@pone.0111243-Zirpel2]. It is possible that a change in Ca^2+^ homeostasis is a triggering event that results in the disruption of ribosomes observed in the present experiments. There are unusually high demands placed on chick cochlear nucleus neurons with respect to Ca^2+^ homeostasis. This is because NM neurons see some of the highest rates of activity in the CNS, even in the absence of acoustic stimuli [@pone.0111243-Warchol1]. This requires NM to be equipped with a variety of compensatory mechanisms to help maintain the fine balance of intracellular Ca^2+^ levels. One such mechanism could involve mGluR activation by glutamate overspill that occurs when the usual uptake systems become overwhelmed [@pone.0111243-Rubel3], [@pone.0111243-Rubel4]. In fact, this has been shown to be the case in several brain regions containing glutamatergic neurons that fire at high frequency [@pone.0111243-Gegelashvili1], [@pone.0111243-Min1]. Once activated, mGluRs are known to regulate Ca^2+^ through a number of signal transduction cascades. For example, mGluR activation of the adenylate cyclase cascade in NM can directly inhibit Ca^2+^ influx through voltage-operated calcium channels [@pone.0111243-Lachica1] via adenosine 3′,5′-cyclic monophosphate activation of protein kinase A, while activation of mGluRs that stimulate the phospholipase C cascade can indirectly regulate \[Ca^2+^\]~i~ through the generation of IP~3~ [@pone.0111243-Zirpel3], which subsequently liberates Ca^2+^ from internal stores [@pone.0111243-Zirpel4] and activates protein kinase C. In the present studies, selective agonists for group I and group II mGluRs were able to reduce the loss of antigenicity for Y10B on the side of the slice that was treated with agonist compared to the untreated side of the same slice. Additionally, the combined effect of activating both mGluR groups through use of the non-selective mGluR agonist, ACPD, appeared to be more beneficial than the activation of specific groups, suggesting that activation of both I and II subgroups of mGluRs are important for the regulation of NM ribosomes. There is a chance, however, that the differences in effect seen between drugs could simply be due to variations in the effective potency of the concentrations used and the possibility of synergistic effects was not directly evaluated. A possible indirect mechanism by which mGluR activation could lead to protection is by regulating inhibitory influences on NM neurons [@pone.0111243-Lu1]. Activation of mGluRs has been shown to suppress GABA release to NM neurons, while blockade of mGluRs reportedly increases GABA release [@pone.0111243-Lu2], [@pone.0111243-Lu3]. GABA-ergic transmission in neurons of the avian auditory brainstem is unusual in that it is depolarizing but inhibitory [@pone.0111243-Lu1], [@pone.0111243-Hyson5]. Membrane depolarizations activate voltage-gated Ca^2+^ channels and lead to Ca^2+^ influx, which could be detrimental to the cell. However, mGluR activity may effectively prevent GABA-evoked depolarizations, serving as yet another mechanism by which NM neurons buffer intracellular Ca^2+^. Finally, mGluR activation could also be exerting a neuroprotective effect by working at neighboring glial cells. Previous research in the chick auditory brainstem has demonstrated that there is rapid growth of astrocytic processes in NM following cochlea removal [@pone.0111243-Canady1]--[@pone.0111243-Rubel5]. The neuroprotective effect of mGluRs on glial cells has been documented in several systems. For example, stimulation of cortical glial cells via application of mGluR agonists has been shown to be highly neuroprotective in mixed cultures that have been exposed to toxic levels of NMDA [@pone.0111243-Bruno1]. It has also been suggested that substances released from glial cells in the hippocampus can influence local synapses [@pone.0111243-Liu1]. If this is the case in NM neurons, then it is possible that mGluR activation could promote the release of some trophic substance from astrocytes. Glutamate concentration dependence and technical considerations {#s4b} --------------------------------------------------------------- Although NM neurons are equipped with compensatory mechanisms that allow them to deal with higher levels of glutamate, the current experiments demonstrated that they are not immune to the toxic properties of glutamate. The ability of glutamate to regulate ribosomal antigenicity was contingent on concentration. Reliably darker labeling was observed when 1 mM glutamate was ejected into the media upstream of NM neurons, while a high concentration of glutamate (10 mM) had the opposite effect, yielding lighter labeling on the drug-treated side of the slice when compared to the control side of the same slice. These findings are not unexpected since it is well established that glutamate has neurotoxic properties when released in large amounts or when incompletely recycled (for review see [@pone.0111243-Choi1]). Studies looking at the role of glutamate transporters have shown that clearing glutamate from the synaptic cleft is an important regulatory control of synaptic strength and that this modulation could be traced to mGluRs located on glial cells [@pone.0111243-Turecek1]. It has also been shown that there is a proliferation of glial processes in NM following the cessation of auditory nerve activity [@pone.0111243-Canady1]. Perhaps the important aspect of mGluR activation is on glial cells, which then modulate the uptake of potentially toxic levels of glutamate. While the current experiment used a glutamate concentration of 1 mM in the ejection pipette, focal application through periodic pressure ejection allows for dilution of the drug. A disadvantage of this method of application is the uncertainty about the degree of drug dilution once expelled into the bath, which makes it impossible to know the precise drug concentration at the cell. Whole bath application would have allowed for more accurate dosimetry, but short pressure applications reduce receptor desensitization, and this method of agonist delivery more closely mimics the natural periodic activation by the endogenous transmitters. Similar to the present studies, bath superfusion of the mGluR agonist, ACPD has been shown to prevent the increase in \[Ca^2+^\]~i~ that occurs in the absence of auditory nerve stimulation, but unlike the present studies, bath superfusion of 1 mM glutamate has been shown to result in an increase in \[Ca^2+^\]~i~ [@pone.0111243-Zirpel2]. This apparent discrepancy could either be due to the additional dilution of drug in our application procedure, bringing the 1 mM pipette solution to sub-toxic levels, or it could be due to some receptor desensitization in the studies that applied drugs in the perfusate, thereby preventing the receptors from having a protective influence and disrupting Ca^2+^ homeostasis. Conclusions {#s5} =========== In summary, it is clear that activation of mGluRs protects NM neurons from early degenerative changes, both in the presence and absence of auditory nerve stimulation. Future research will have to determine the exact nature of this protective influence but there are likely multiple mediators. Activation of these receptors can modulate glutamate uptake at the auditory nerve-NM synapse, keeping glutamate at sub-toxic levels, or it could more directly protect the neuron by maintaining intracellular Ca^2+^ homeostasis, and perhaps act through a myriad of other modulatory functions that are regulated by 2nd messenger systems. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: KLC RLH. Performed the experiments: KLC. Analyzed the data: KLC RLH. Wrote the paper: KLC RLH. [^3]: Current address: Louisiana State University Health Sciences Center, Department of Cell Biology and Anatomy, New Orleans, LA, United States of America

{#sp1 .153} {#sp2 .154} {#sp3 .155} {#sp4 .156}

According to sexual conflict theory females are the choosy sex ([@zow090-B63]; [@zow090-B41]; [@zow090-B7]) and should choose males of higher quality as potential mates ([@zow090-B75], [@zow090-B76]; [@zow090-B5]; [@zow090-B31]; [@zow090-B38]; [@zow090-B32]; [@zow090-B30]). By choosing a high quality partner the female provides offspring with resources ([@zow090-B26]) or with a better genetic background ([@zow090-B23]; [@zow090-B4]). Androgen-dependent male sexual traits (STs), including behavioral characteristics, as well as immunocompetence, are theoretically assumed to be key indicators for female mate choice. The expression of androgen-dependent traits and the maintenance of a male's physical health are costly processes causing a tradeoff between competing functional systems. According to the "Immunocompetence handicap" hypothesis, when the female chooses a male, she focuses on signs of his innate health, although high levels of testosterone may negatively affect immunity ([@zow090-B19]). Theoretically, a female can base a choice on any optional trait indicating a male's quality, but the interdependence between such traits and expression of secondary sexual characters, immunocompetence, androgens, or androgen-dependent behavior is assumed. The Campbell dwarf hamster is a small, polyoestral, seasonally breeding rodent with highly pronounced sexual dimorphism, inhabiting the dry steppes and semi-deserts of Central Asia ([@zow090-B18]; [@zow090-B58]; [@zow090-B17]). Ready to mate females (of different ages) weigh from 22 g to 45 g, and males weigh from 32 g to 60 g. Males have a large specific skin gland in the middle ventral part of the body. They use a secretion of the gland to mark extensive overlapping home ranges. Information about their mating system in nature is quite scarce, so it is safe to speak only about polygamy ([@zow090-B64]; [@zow090-B72]; [@zow090-B61]; [@zow090-B60]). Receptive females attract males from a distance of up to 1 km, and may mate with several males ([@zow090-B72]; [@zow090-B60]). In captivity, adult males act aggressively toward each other from the age of 2 months, and their cohabitation in the same cage often leads to the death of one of them. Success in mating with the female can be related positively to aggressive dominance in male--male conflicts. However, this does not exclude the possibility of selective responses in the female. When a male and female share one cage, they demonstrate pronounced features of social monogamy including male participation in care of pups ([@zow090-B70], [@zow090-B71]; [@zow090-B65]). Fragmental observations in nature also signify to the close contact of a male with the female and juveniles ([@zow090-B72]; [@zow090-B59]; [@zow090-B71]). We studied experimentally the choice by receptive and sexually motivated (SM) female *Phodopus campbelli* between 2 male full-sibs that differed in the degree of expression of external STs. In addition to morphological characteristics in these males, we estimated levels of testosterone and cortisol in their blood, intensity of specific immune response to antigens, and their aggressive and sexual behavior when males competed for the female. We estimated cortisol concentrations to characterize stress level. Glucocorticoid stress hormones can play a role in sexual selection. In terms of mate choice, individuals can exhibit preferences for mates with either low baseline or peak glucocorticoid levels ([@zow090-B28]). Siblings were used in order to minimize the genetic component of variance for the female choice since it is known that the female may choose a partner according to the difference in MHC genes ([@zow090-B73]; [@zow090-B42]; [@zow090-B36]). We tested 3 hypotheses in this study: 1) The potential predictors of female mate choice (male STs, intermale aggressiveness and sexual dominance, endocrine and immune characteristics of males) will be correlated with each other. 2) A SM female will choose a partner between 2 male full-sibs divergent in the expression of ST. 3) Mate choice can be explained by a greater expression of ST, an increased level of testosterone, higher behavioral competitiveness, higher (indicator of health), or lower (reciprocal relationship with testosterone) specific immunocompetence, or by a combination of these characteristics. Materials and Methods ===================== Males ----- We obtained 18 litter with 3 or more juvenile males from 57 pairs of hamsters (aged from 6 months to a year) in 2014 and 30 litter from 120 such pairs in 2015. At 25 days we removed juvenile males from parental cages and kept them together by litter (males without females) in Ferplast plastic cages (70 × 40 × 40 cm) for up to 2 months. At 2 months the animals were photographed with a digital camera (Nikon 7000) in fixed positions with their ventrum up against a ruler for measurements of STs on the computer screen ([@zow090-B56]). We fixed animal by hand ventrum up, so that the testes became visible in scrotum. ST measurements included body mass and characteristics of male specific external morphology (both secondary and primary sexual characters): area (length × width) of mid-ventral skin gland, distance between anal and genital openings, and the testes size (average length of testes) in their external outlines from a live animal. In each litter we chose 2 males with maximal differences in body mass, area of mid-ventral gland, ano-genital distance, and testes size. From 8 to 9 AM we took a blood sample from the sublingual vein (0.3 mL) of each male. The whole procedure of sampling blood lasted no longer than 2 min, which is 2 times less than the time of glucocorticoid signal in the blood in response to handling. Samples were centrifuged at 3,000 rpm for 15 min, and the serum was separated from the hematocrite and frozen at −18°C. Then selected males were placed in individual cages 30 × 22 × 20 cm, where they lived during the experiment ([Figure 1](#zow090-F1){ref-type="fig"}). {#zow090-F1} Over 2 years 45 pairs of male sibs were selected for testing. One male died. Thus, for the statistical analysis we used the data set for 88 individuals (44 male pairs). Females ------- Young females after removal from the parental cage at the age of 25 days were kept in groups of 7--8 individuals in plastic cages "Ferplast" (70 × 40 × 40 cm). At 3--4 months the majority of sexually mature females remained in diestrus. A week before the experiment we stimulated females by placing an adult male, confined in a small box of metal mesh, in the female's cage for 1 h, which was enough time to stimulate the female's estrous cycle. We tracked the females' cycle daily (1--2 h before testing) by viewing their vaginal smears under a microscope (×20). In the tests we used virgin females 3--4 months old in the phase of transition from proestrus to estrus. Our preliminary observations showed that the female was SM only at this stage of the estrous cycle and actively looked for contact with a male. Although the female remains attractive to male during most of estrus, in the middle and at the end of estrus she opposes his attempts to mate and escapes into an area of the experimental enclosure inaccessible to the male. In Campbell's hamster a vaginal estrus, registered in smears, may not coincide with behavioral receptivity ([@zow090-B16]). Thirty minutes before the test began the female was paired in a neutral arena with an adult male for 1--2 min to ensure that she was positively motivated. All females we used were not sisters of tested males. During the test, males were able to copulate with a female. In each test (with each pair of males) we used a new female. Animals housing --------------- The animals belonged to the laboratory population of hamsters at the A.N. Severtsov Institute of Ecology and Evolution, RAS. This population is descended from hamsters taken in the 1980s from Mongolia (MPR). All hamsters were kept in a room with a constant long-day regime (14 h). Food (formula feed for rats and mice, oat, vegetables, as additives, sunflower seeds, low fat cottage cheese, boiled chicken) and water were *ad libitum*. Thin wood shavings and sawdust were used as bedding, and pieces of cardboard served for shelters. Animal care mandated that we change the bedding once every 2 weeks and check on the reproductive and health state every week. Test 1: mate choice by the female --------------------------------- For each pair of male siblings at the age of 3 months ([Figure 1](#zow090-F1){ref-type="fig"}) we selected a SM female (in transition from proestrus to estrus). Before testing we exposed this female to the odor of the litter from individual male cages for 5 min. This was done to minimize the likelihood that sexual intercourse would take place as a result of the female's accidental first entry into the compartment of 1 of the 2 males. Testing was performed in a chamber made of organic glass (Plexiglas), consisting of 3 compartments---1 central and 2 lateral of 25 × 25 × 30 cm each. Compartments were partially isolated by walls with gate-passes 10 × 10 cm on each side of the center compartment, but could be isolated completely by removable dividers---privacy screens ([Figure 2](#zow090-F2){ref-type="fig"}). {#zow090-F2} We placed a SM female in the central compartment with gates, covered by screens. After a 5 min adaptation, 2 samples of the litter from male cages were exposed to the female in Petri plates (5 cm diameter) that were mounted at the mid-center line of the compartment at a distance of 5 cm one from another. The installation of plates took 2--3 s. During the next 5 min we recorded the time spent sniffing each odor sample using 2 stopwatches. After 5 min, the Petri plates were removed. The female was left for next 5 min in the central compartment. Meanwhile, flexible nylon collars were placed on each male. We used tightening nylon cable straps, usually used to tie up wiring (cable tie 3 × 100 white). The collars were attached to 25 cm of flexible tungsten fishing leads, the free end of which was fastened on the far wall of the lateral compartment. The fastening was designed so that three-fourth of the compartment floor was available for the male, while the female had free access to come into the compartment of each male, but remained inaccessible to the male in one-fourth of its area. After a 5 min adaptation the screens that blocked passages to the side compartments were removed, allowing the female to enter freely the males' areas. Within the next 60 min one observer recorded female transitions from section to section using a computer program designed by A.V. Tchabovsky in Excel macros. We gave animals only 5 min time for familiarization or adaptation to the experimental situation due to their very high sexual motivation. Our preliminary observations showed that longer periods of adaptation could impair the result of the test and, thus, would not provide any advantage. For each female in each test, we estimated the number of transitions from compartment to compartment and the time the female spent in each compartment. The time intervals were used as a primary data for analysis ([@zow090-B68]). Simultaneously, during the entire time of the test a second observer recorded the following patterns of the males' behavior: 1) male grooming of the female, 2) series of mounts, and 3) intromissions (recognized as mounts associated with grooming of genitals; [@zow090-B53]; [@zow090-B24]). In females we recorded: 1) approaches to the male, 2) defensive postures (fall on the back), 3) vocalizations (squeals associated with defense); and 4) aggression toward the male (rushes, bites). A male starts to groom the female in the shoulder area before mounting and holds her slightly biting on the shoulder while mounting, presumably stimulating the female for copulation. However, the actions of the male can be rough. Such "aggressive grooming" can be easily identified by the defensive postures of the female and squeals. In contrast, a "non-aggressive grooming" is not associated with any expression of discontent. In parallel, we recorded the behavior of animals during the test with digital video camcorder Canon G30, and these records served as the main resource for further analysis. Afterwards the testing males were weighed and photographed in fixed positions with ventrum up against the ruler scale. If in one test the male with the higher expression of ST was tethered in the right compartment, in the next test this compartment was occupied by the male with the lower expression of ST. The choice of a male by the female was based on the amount of time the female spent with each male. We assumed that a female did make a choice if she spent ≥ 2/3 (66.7%) of the time with one male from the total time spent with males. In 75% of cases SM females spent less than 10% of the test time (Median = 8.4%, *n *= 34) in the central compartment. Although the social preference (estimated by time in the compartment) and the mating preference (sexual experience) may not be the same (for instance, in prairie voles, mating was not essential for partner preference formation; however, preferences developed more rapidly when mating occurred; [@zow090-B68]). The time the choosing individual spends with a potential partner is a common measure in studies of mate preferences ([@zow090-B74]). Test 2: encounter of males and free access to the female -------------------------------------------------------- Ten days after the end of the mate choice experiment we estimated the competitiveness of male sibs (aggressive and sexual dominance) in a series of independent tests with free access of both males to another SM female ([Figure 1](#zow090-F1){ref-type="fig"}). Males could freely contact with each other and mount the female. Males were paired in a round arena 50 cm in diameter. The arena was surrounded by an organic glass wall and was subdivided into 2 equal sections by a removable partition. A fresh sheet of filter paper was used as the substrate. After a 5 min adaptation of each male in a separate section we removed the partition wall, and during the next 5 min registered the patterns of aggressive dominance behavior (attacks, rushes, chases, fights), and defensive subordinate behavior (defensive postures on the back, escaping behavior and flights, submissive vocalization, positions under in fights). After 5 min of males' testing we placed a new non-relative SM female in the arena for 10 min and registered the aggressive and defensive behavior of males by the same scheme as before, as well as the frequency of mount series and intromissions. The behavior of the males was registered by video camcorder. We estimated the occurrence of each form of behavior within 5 s intervals of the time scale of the test. In the analyses we used number of intervals with a certain behavior during the test. Test 3: reproductive success of males ------------------------------------- Three-month old sexually mature virgin females in transition from proestrus to estrus (SM females) were introduced to males in their individual cages (1 per cage) 3 weeks after the males' immunization with sheep's red blood cells (SRBC) and 2 weeks after blood sampling ([Figure 1](#zow090-F1){ref-type="fig"}). We recorded the date of birth of the first litter and the number of young per litter. If a female did not give birth within 50 days after pairing with a male she was considered non-breeding. Male characteristics used as predictors for mate choice by the female --------------------------------------------------------------------- We used the following predictors, for female choice: 1) body weight (g), 2) mid-ventral gland area (*D*~max* *~×*~ ~d*~min~; cm^2^), 3) ano-genital distance (mm), 4) mean length of the testes (mm), 5) concentration of cortisol in the blood at 2 months old (ng/mL), 6) concentration of testosterone at the age of 2 months (ng/mL), 7) concentration of cortisol at the peak of the immune response to SRBC (ng/mL), 8) testosterone concentration at the peak of the immune response to SRBC (ng/mL), 9) intensity of the cutaneous DTH reaction to phytohemagglutinin (T-cell immunity test, mm), and 10) intensity of the immune response to SRBC (titre of antibodies in the blood). The characteristics of behavior recorded in Test 2 and used for evaluation of the males' competitiveness were as follows: 1) the aggressiveness of the male in encounter test with the sibling male (sum of 5 s intervals with aggression; first 5 min without the female), and 2) mounts on the female (sum of 5 s intervals with mounts; next 10 min with the female). Immunity: T-cell immunocompetence --------------------------------- An estate of adaptive T-cell immunity of males was estimated using a cutaneous delayed type hypersensitivity response to phytohemagglutinin (PHA-P, L8754-25mg, Sigma--Aldrich Co.), a mitogen of plant origin causing rapid T-cell recruitment and proliferation in the inflammation focus ([Figure 1](#zow090-F1){ref-type="fig"}). This assay is widely used in ecological immunological research, and a study by [@zow090-B62] confirmed its appropriateness for adaptive immunity studies. We dissolved PHA-P in PBS (E404--200TABS, 100 mL Sigma) to 2.5 mg/mL ([@zow090-B57]), and 50 mL of this solution were injected intradermally in the heel of the right hind foot using an insulin syringe, while the left hind heel was injected for the control with the same amount of PBS. The paw thickness was measured before the injection and 24 h after the injection with a soft digital caliper, and the difference between the measurements (Δ = *d*~after~−*d*~before~) was used to compare the reaction in the right and the left foot. Since the response of the control (left) foot to PBS was not pronounced and there was a highly significant difference in responses of the left and right (test) foot (*T*-test for matched pairs, *P *\< 0.001), in our further analysis we used the difference in thickness of only the paw inoculated with PHA-P. Immunity: B-cell immunocompetence --------------------------------- We injected experimental hamster males intraperitoneally with 2% suspension of SRBCs in saline and control animals (*n *= 14) with the same amount of saline (2 mL per gram of body weight). SRBCs were prewashed three times and then suspended in saline to the required concentration. The immune response was measured by hemagglutination assay ([@zow090-B67]; [@zow090-B51]) with blood serum obtained from the sublingual vein 7 days post-immunization, at the immune reaction peak (observed between days 5 and 10; [Figure 1](#zow090-F1){ref-type="fig"}). Reaction was performed with 0.5% SRBC suspension in saline. In the first well of a 96 well microbiological plate, 25 mL of experimental serum were thoroughly mixed with 75 mL of saline, and 50 mL of the resulting suspension were transferred into the next well; then the procedure was sequentially repeated. The incubation temperature was experimentally optimized as +38°C. The immune response to SRBC (serum antibody titer) was evaluated after 2 h of incubation as the number of the last well in this series of sequential saline dilutions that contained the quantity of antibodies sufficient for hemagglutination ([@zow090-B50]). As we expected, immunized animals showed a significantly stronger immune response than control ones, in whom the reaction was close to zero (Mann--Whitney *U* test: *Z *= 4.6; *P *\< 0.001, *n*~1 ~=~ ~68, *n*~2 ~=~ ~14). Hormones -------- We measured cortisol and testosterone concentrations in hamster blood serum specimens ([Figure 1](#zow090-F1){ref-type="fig"}) by enzyme linked immunosorbent assay (ELISA) using IEA cortisol and IEA TS (testosterone) commercial test systems (NPO Immunotech, Russia) according to the protocol suggested by the manufacturer. The cortisol--testosterone cross reaction for the above kits was 0.08%. Optical densities were measured at 450 nm using a Uniplan plate spectrophotometer (Russia). In rare cases of high concentrations of testosterone we diluted samples with the "Buffer for serum dilution" produced by "NPO Immunotech". Statistical analysis -------------------- We applied the principal component analysis (PCA) to all four variables describing male morphological traits in order to simplify the analysis and to reduce the dimensionality of the data. All variables were standardized, and the single principal component was extracted ([Table 1](#zow090-T1){ref-type="table"}). The component (ST expression) explained 60% of the total variance in the morphological variables and reflects the increase of body mass, the size of ventral glands, the testes size, and the anogenital distance. Table 1.Summary of principal components (PCs) analysis of 4 morphological variables describing male STsInformationPC1Eigenvalue2.38Percentage of total variance explained60.0Factor loadings:Body mass0.68Size of ventral glands0.81Testes size0.85Ano-genital distance0.74 The testosterone and cortisol levels at 60 days and after SRBC immunization did not conform to the assumptions of normality (Shapiro--Wilk's *W* test, *P *\< 0.05) and, therefore, were log-transformed to access the normal distribution. To test the possibilities of future reduction of the dimensionality in the dataset and to reveal any high correlation among variables, the correlation analysis (Pearson's correlation) was performed among male individual characteristics ([Table 2](#zow090-T2){ref-type="table"}). Male--male aggressiveness and the intensity of the male's mounting behavior in TEST 2 (conducted with the same males that participated in TEST 1 and the other female) did not conform to the assumptions of normality (Shapiro--Wilk's *W* test, *P *\< 0.05) and could not be transformed into a normal distribution. Thus, correlation statistics for these variables were assessed with Spearman's rank-order correlation. Table 2.Correlation matrix (Pearson's *r*) among all male individual characteristics (boldface type indicates significant effects; *n *=* *88)ST expressionCortisol (60 days)Cortisol (SRBC)Testosteron (60 days)Testosteron (SRBC)Immune response to SRBCImmune response to PHAST expression^[a](#tblfn1){ref-type="table-fn"}^Cortisol (60 days)^[b](#tblfn2){ref-type="table-fn"},[c](#tblfn3){ref-type="table-fn"}^**−0.25**Cortisol (SRBC)^[c](#tblfn3){ref-type="table-fn"},[d](#tblfn4){ref-type="table-fn"}^−0.060.17Testosterone (60 days)^[b](#tblfn2){ref-type="table-fn"},[c](#tblfn3){ref-type="table-fn"}^0.000.02**0.21**Testosterone (SRBC)^[c](#tblfn3){ref-type="table-fn"},[d](#tblfn4){ref-type="table-fn"}^0.18−0.01**0.400.23**Immune response to SRBC−0.04−0.130.09−0.10**0.23**Immune response to PHA0.15−0.030.14−0.12**0.26**0.16[^1][^2][^3][^4] Test 1. To determine the subset of variables that best discriminated between "preferred" and "non-preferred" males in tests with clear female choice (≥ 66.7% of time), we conducted General Discriminant Analysis (GDA; forward stepwise, *P*~to\ enter/remove~ = 0.05) ([@zow090-B25]). All male individual characteristics were included as possible predictors in the GDA; additionally, the study year was fitted in the model. To obtain a quantitative estimation of the female choice, we conducted forward stepwise general regression analysis (GRM, *P*~to\ enter~ = 0.05, *F*~to\ enter ~= 1.0) with the time that females spent with the preferred male in Test 1. Standardized differences between all individual characteristics of "preferred" males and "non-preferred" males ([Table 3](#zow090-T3){ref-type="table"}) were used as potential predictors in the model; in addition, the study year was included in the initial model as a categorical predictor. All tests were included in this analysis, *n *= 44. Table 3.Spearman's correlation coefficients for male--male aggressiveness and male--female mounting activity in Test 2 (boldface type indicates significant effect, *p* \< 0.05 *n *= 88)AggressivenessMounting activityST expression−0.130.10Cortisol (60 days)−0.100.04Cortisol (SRBC)−0.05**−0.28**Testosterone (60 days)−0.07−0.15Testosterone (SRBC)−0.10−0.18Immune response to SRBC−0.03−0.15Immune response to PHA−0.160.05Mounting activity (Test 2)−0.01 We used McNemar's test for paired nominal data to estimate the effect of male behavior on the female choice in Test 1. The forward stepwise GDA was used in order to find variables that best discriminated between aggressive and nonaggressive males (with respect to male grooming). Test 2. The Wilcoxon matched pairs test was used to estimate variation between the "preferred" and "non-preferred" males in the tests with clear female choice (*n *= 34). The differences between "preferred" and "non-preferred" males in aggressiveness and mounting behavior were normally distributed. Therefore, we performed the general regression model with differences in aggressiveness and mounting behavior as predictors and the time with the "preferred" male in TEST 1 as dependent variable (study year included as additional factor). **Test 3.** To compare the litter sizes in the preferred and non-preferred males we used Student's *T*-test for paired samples. The birth dates of litters produced by preferred and non-preferred males did not fit demands for normality. To compare the birth dates the Wilcoxon matched pairs test was used. Statistical analyses were performed using STATISTICA version 8.0 (StatSoft). All tests were 2-tailed, with a significance level of 0.05. Ethical principles in treatment of animals ------------------------------------------ In our study we followed the requirements of the "Principles of Laboratory Animal Care" (NIH Publication Vol. 25, No. 28 revised 1996; <http://grants.nih.gov/grants/guide/notice-files/not96-208.html>), of "Guidelines for the treatment of animals in behavioural research and teaching" ([@zow090-B2]) and the Federal Law of the Russian Federation. Results ======= Correlations between male traits -------------------------------- Contrary to expectation, we did not find any strong relationships among different male characteristics ([Table 2](#zow090-T2){ref-type="table"}). Namely, the correlations between ST expression and hormones, ST expression and immune status, and hormonal and immune status were low or absent. Most correlations were not significant, and the highest correlations did not exceed medium-sized effects ([@zow090-B12]; [@zow090-B29]). Thus, we did not reduce dimensionality of the dataset due to substantial information loss. Male--male aggressiveness as well as the intensity of mounting in Test 2 (conducted with the same males that participated in Test 1 and the other female) also did not correlate with most individual characteristics of the males and between themselves ([Table 3](#zow090-T3){ref-type="table"}). The number of mount series per test the tethered male made in Test 1 did not correlate with the number of mount series the male made when freely competing with a sibling for the female in Test 2 (Spearman's *r *= 0.12, *P *= 0.26, *n *= 88). Female choice between 2 tethered males (Test 1) ----------------------------------------------- In 77% of the tests (*n *= 44) females preferred 1 male and spent substantially more time with him (≥ two-third of the test duration) than with the other male. On average, females spent 82.4 ± 1.6% of the total time with the preferred male. Hence, in most tests the "preferred" and "non-preferred" males were evident. We defined tests in which females spent more than 67% of the total time with the "preferred" male as tests with a clear female choice. In 41 of 44 tests (93%) females copulated with both males. Males preferred as social partners (justified by the time a female spent with the male) also were preferred as sexual partners. For non-preferred males the total number of series of mounts per test correlated positively with the total time that the female spent in the male's compartment (*r *= 0.82, *P *= 0.0001, *n *= 34). For preferred males the total number of mount series per test did not relate to total time the female spent with the male ([Figure 3](#zow090-F3){ref-type="fig"}; *P *= 0.97, *n = *34). The preferred males demonstrated significantly higher number of mount series per test than non-preferred males \[median and limits: 79 (3--178) and 28 (0--85), respectively; Wilcoxon matched pairs test: *t *= 160, *P *\< 0.001, *n*~1 ~= *n*~2 ~=34\]. The same was true for mounts with intromissions \[median and limits for preferred male: 35 (0--95), for non-preferred males: 10 (0--67); Wilcoxon matched pairs test: *t *= 86, *P *\< 0.001, *n*~1 ~= *n*~2 ~=~ ~34\]. The high sexual activity of the preferred male may lead to exhaustion; in many cases the male falls down on his back becoming rigid up to 1 min. This was the reason for the lower number of mounts per unit time in preferred males compared with non-preferred males \[median and limits for preferred male: 1.62 (0.06--3.80), for non-preferred male: 3.29 (0--6.67); Wilcoxon matched pairs test: *t *= 95, *P *\< 0.001, *n*~1 ~= *n*~2 ~=~ ~34\]. The same was true for mounts with intromissions \[median and limits: 0.83 (0--1.92) and 0.95 (0--9.72), respectively; Wilcoxon matched pairs test: *t *= 168, *P *= 0.073, *n*~1 ~= *n*~2 ~=34\]. {#zow090-F3} We did not find a relationship between the visiting order of the males and female choice: among 34 tests with clear female choice the first-visited male became the preferred male in 19 (55.9%) cases and in the other tests females preferred the second visited male (ч^2* *^=^* *^0.47, *P *= 0.5). The stepwise GDA with the ST expression, hormonal characteristics and characteristics of the male immune status included as predictors, and the study year as an additional factor failed to discriminate between preferred and non-preferred males (in tests with clear female choice). No predictor contributed significantly in the discrimination and was selected for the model (Wilk's *π *= 1.0*, P *\> 0.1). From the initial set of predictors that included differences between the variables of preferred and non-preferred males, the stepwise GRM revealed 3 variables which significantly influenced the time that females spent with the "preferred" male. Namely, the lower values were the ST expression and the immune responses to SRBC, and the higher value was the testosterone level after SRBC immunization in the preferred males compared with the non-preferred males, the more time the females spent with the "preferred" males in the test ([Table 4](#zow090-T4){ref-type="table"}). Table 4.Summary of forward stepwise general regression model (GRM) in which the dependent variable was the time that females spent with preferred males and independent variables (predictors) were represented by differences in trait values (Δ) between preferred and non-preferred males (*p*~to\ enter~ = 0.05, *F*~to\ enter ~= 1.0, *R*^2 ^=^ ^0.32, *F *= 5.9, *P *= 0.002, all tests, boldface type indicates significant effect, *p* \< 0.05 *n *= 44)EstimatesSE*FP*Intercept39.671.3**\<0.0001**Δ ST expression−2.270.95.9**0.02**Δ Testosterone (SRBC)0.190.085.8**0.02**Δ Immune response to SRBC−1.240.47.9**0.008**Δ Immune response to PHAOut of the model0.0021.0Δ Cortisol (60 days)Out of the model1.70.2Δ Cortisol (SRBC)Out of the model0.060.8Δ Testosterone (60 days)Out of the model0.20.6Year effectOut of the model0.0080.9 Effect of the male's behavior on the female's choice (Test 1) ------------------------------------------------------------- The strongest factor that affected female choice in the tests was the male's behavior during copulations. Some males used aggressive grooming during sexual contact, and females obviously avoided such situations. From 34 tests with clear female choice, among the "preferred" males only 4 males groomed females aggressively and the other 30 did not; but among the non-preferred males there was a ratio of 20:14 aggressive to non-aggressive groomers (McNemar test for paired samples, *χ*^2* *^=^* *^9.4, *P *= 0.002). Females demonstrated significantly more defensive postures (fall on the back) per unit time with non-preferred male \[median and limits: 1.09 (0--6.96) compared with preferred males (median and limits: 0.65 (0--3.14); Wilcoxon matched pairs test: *t *= 149, *P *\< 0.011, *n*~1 ~= *n*~2 ~=34\]. They also initiated more contacts per unit time with non-preferred males than with preferred males \[median and limits respectively: 2.46 (0.28--23.32) and 1.15 (0.08--7.33); Wilcoxon matched pairs test: *t *= 2, *P *\< 0.001, *n*~1 ~= *n*~2 ~=34\]. The reason is that the higher sexual activity of the preferred male may lead to his sexual exhaustion; from time to time the male falls down on his back becoming rigid up to 1 min. The stepwise GDA model did not reveal any discrimination between males demonstrating aggressive and non-aggressive grooming (all 44 tests included). No predictor contributed significantly in the discrimination and was selected for the model (Wilk's *π *= 1.0, *P *\> 0.1). Encounter of males and free access to the female (Test 2) --------------------------------------------------------- Paired comparison of the "preferred" and "non-preferred" males from the tests with clear female choice (Test 1) did not reveal significant differences in their aggressiveness and mounting behavior in the encounter test (Wilcoxon matched pairs test: *t *= 195.5, *P *= 0.6, *n*~1 ~= *n*~2 ~=34 and *t *= 267, *P *= 0.8, *n*~1 ~= *n*~2 ~=~ ~34, respectively). GRM with aggressiveness and mounting behavior as predictors and the time with "preferred" male in TEST 1 as dependent variable (study year included as additional factor) did not reveal any significant effects (multiple *r*^2 ^=^ ^0.16, *F *= 2.5, *P *= 0.07). Reproductive success in the "preferred" and "non-preferred" males (Test 3) -------------------------------------------------------------------------- After the males that participated in Test 1 were paired with virgin females ready to mate in the male's cages, neither date of birth (Wilcoxon matched pairs test: *t *= 99.5, *P *= 0.2, *n*~1 ~= *n*~2 ~=28) nor litter size differed for the first litters between the 2 categories of males (*t*-test for paired samples: *t *= 0.98, *P *= 0.3, *n*~1 ~= *n*~2 ~=~ ~22). Discussion ========== Among hypotheses we tested, only the second one received definite support. SM female hamsters chose their male partner from a pair of tethered male siblings, and the final mate choice was not a consequence of the first visit to the male. Other hypotheses did not receive support in results of our experiment. Correlations between male traits we studied and which are assumed to affect female choice were low or absent. Females did not make choice based on a higher expression of male STs, higher level of blood testosterone, with aggressiveness in the encounter with the sibling male or with sexual dominance in a situation when males could freely compete for the female. The choice was not associated with the intensity of T-cell immune response to phitohemagglutinin (PHA). The only characteristic that influenced female choice with high statistical support was non-aggressive grooming by males during sexual contact. Females also tended to choose males with a lower expression of ST and with a lower immune response to SRBC associated with higher testosterone after immunization. We, therefore, are unable to explain our results in the framework of the widely discussed hypotheses of "good genes" ([@zow090-B23]) or immunocompetence handicap ([@zow090-B19]). We expected females to prefer males with a higher expression of ST in combination with a lower specific immune response to antigens (in case of tradeoff between the expression of androgen-dependent traits and specific immunity). Inconsistency of our results with the theory can be explained by low correlations between dependent and predictor variables, but also by the vulnerability of any inferences based on the analysis of correlations with androgens. The result of recent cross-species meta-analyses of relationships between circulating sex hormone level and immune function was not statistically significant for either testosterone or estrogen. The meta-analysis of results of experiments, presupposing control of external and internal conditions, confirmed the effect ([@zow090-B20]).The testosterone concentration in blood is very unstable in *Ceteris paribus*. Thus, it would be more productive to focus on the concentration thresholds of sensitivity of the target tissue receptors rather than routine concentrations of the hormones, although it is difficult to realize this approach in practical work ([@zow090-B69]). Effects of sex hormones on immunity may depend on body conditions, and resources availability ([@zow090-B35]; [@zow090-B52]). The leptin, which can prevent the immunosuppressive effect of testosterone, may be the reason ([@zow090-B3]). The effect of testosterone on immune function may also depend on stress levels ([@zow090-B45], but see [@zow090-B48]; [@zow090-B49] for contradictory findings). Stress hormones can be the key mediators of many condition-dependent, sexually selected traits that serve as honest signals of mate quality. It is typically thought that 2-way interactions exist between glucocorticoid stress hormones, sex steroids, and body condition. In terms of mate choice, it appears that glucocorticoid stress hormones could mediate some of the condition-dependent traits used to assess mates ([@zow090-B28]). It also needs to be taken into account that results obtained in the laboratory may not be the same as in natural conditions ([@zow090-B11]). In a superb fairy-wrens *Malurus cyaneus* testosterone treatment depresses the immune response under laboratory conditions but not under natural conditions ([@zow090-B44]). The immunocompetence handicap hypothesis proposed by [@zow090-B19] summarized what was known about the interactions between sexual signals, androgens, parasites, and the immune system. The main prediction was that there would be a trade-off between sexual displays on one hand and immune function on the other hand. At a later date most of the attention was directed to this dual effect of testosterone enhancing STs/displays, but suppressing immunity. Little attention was paid to the possible effect of testosterone on redistribution of immune cells rather than to its direct suppressive effect on the immune system ([@zow090-B10]). [@zow090-B47] revised studies that tested the suppressive role of testosterone in immunocompetence by means of meta-analysis and found this effect controversial; on average it was small and far from statistically significant. On the other hand, a meta-analysis of the effect of parasites on sexual signals revealed that experimental exposure to parasites significantly suppressed sexual signals ([@zow090-B37]). Finally, the meta-analysis of published data on the effects of experimental immune activation on testosterone showed a strong suppressive effect on testosterone. The trade-off between immunocompetence and STs/displays may primarily be generated by the effect of immune activation on testosterone, rather than the opposite effect that has received most attention ([@zow090-B9]). In addition to the above, a recent review of genetic benefits of extra-pair paternity in birds ([@zow090-B1]) found no convincing evidence for genetic benefits of the phenomenon and rejects the good genes hypothesis. Taken all together these results show that Folstad and Karter's as well as Hamilton and Zuk's ideas may not be as solid as has been assumed by many people. Although most empirical research is based on the assumption that females seek a male of the highest possible quality (in terms of the genes or resources the male can provide), manipulation of the female condition can lead to divergent female preferences ([@zow090-B46]). For instance, it was shown on zebra finches ([@zow090-B27]) and house sparrows ([@zow090-B22]) that females of lower quality prefer low-quality males. High-quality female zebra finches preferred high-quality males and the high-quality female of house sparrows did not discriminate between the quality of a male (determined by the size of the black spot on the breast). Although we kept female hamsters under standardized conditions, we were limited in the possibility to control their internal state. Females that participated in the experiment were young (3--4 months), had no sexual experience, and were in the same stage of the estrous cycle. However, we were unable to standardize strictly their age and weight. These variables introduced into the model as predictors showed no statistically significant effects. However, this does not exclude the possibility of their limited impact, as well as of the impact of unaccounted factors, on the overall variation. In a set of experimental studies the choosing females had greater reproductive success (in terms of offspring viability and survival to independence or reproductive age) when paired with preferred males ([@zow090-B14]; [@zow090-B54]; [@zow090-B8]; [@zow090-B6]). In house mice *Mus domesticus* per cent of juveniles at reproductive age was higher in females paired with the preferred male, but there was no significant difference in the number of pups born ([@zow090-B14], [@zow090-B15]). Fecundity was even higher in parental pairs with non-preferred males (*P *\< 0.1; [@zow090-B14]; [@zow090-B21]). In our study we found no difference in reproductive success between preferred and non-preferred males paired with a ready to breed new female. In test situation (Test 1) female choice was constrained by the experimental design. She was forced to choose between 2 sibling males which differed in ST expression. In nature she might not have preferred either male. We also restricted our estimation of reproductive success to the first litters the females produced after pairing with experimental males. Usually the first litters born by young females of Campbell dwarf hamsters are more variable in size and viability, and we expected the impact of male traits on breeding success to be more pronounced at first reproduction event. At the same time, we cannot exclude completely the possibility of underestimation of reproductive differences between preferred and non-preferred males. Interpreting our results we should take into consideration the species specific features of Campbell dwarf hamsters. In fact the spectrum of species for the experimental study of mate choice among rodents is still limited. Because many experiments were conducted on mice *Mus musculussensu lato*, the support of the "good genes" theory is largely associated with this species. Campbell hamsters differ from house mice in many aspects. Despite the high plasticity of the social organization of house mice (*M. musculus* s.l., [@zow090-B33]), at moderate densities of the population of synanthropic house mice (first and foremost, *M. m. domesticus*), despotic dominance of a male provides him with access to most of the group's females ([@zow090-B13]; [@zow090-B34]; [@zow090-B43]). It was the house mouse that provided good evidence for positive relationships between androgens, aggressive dominance, and expression of STs, or the negative impact of androgens on the system of acquired immunity ([@zow090-B55]; [@zow090-B40]; [@zow090-B66]; [@zow090-B39]). The negative effect of activation of specific immunity function on testosterone level has also been confirmed in the house mouse ([@zow090-B9]). Despite the pronounced sexual dimorphism in Campbell hamsters, polygyny in the conventional sense (males desperately compete for females, and many females become available for the winner male) is uncommon for the species. The mating system of Campbell hamsters in nature, apparently, has some characteristics of promiscuity. Males mate with more than 1 female, and their female partners can mate with other males. Fragmentary observations made in nature support the opinion that males do not monopolize females ([@zow090-B64]; [@zow090-B61]; [@zow090-B60]; [@zow090-B72]). It is also possible that the female focuses on other quality markers of the male, unrelated to those, which determine the advantage in direct competition between males (body mass, aggressiveness, testosterone). On the other hand, the breeding system of Campbell dwarf hamsters has the features of social monogamy ([@zow090-B70], [@zow090-B71]; [@zow090-B65]). Among hamsters reared in captivity the male always cares for the young. Deprivation among pups of male paternal care, and in particular of the opportunity to receive a secretion of his specific glands, affects the growth and survival of offspring ([@zow090-B65]). Some observations in nature also point to the participation of a male in caring of juveniles. The male was observed carrying food into the burrow with pups ([@zow090-B59]; [@zow090-B71]) and marking pups ([@zow090-B59]). In this respect it is possible that in choosing a non-aggressive male, the female focuses on the parental quality of a potential partner. This assumption suggests a separate study on parental behavior of preferred and non-preferred males; their actual reproductive success should be estimated by the number of juveniles surviving to reproductive age. Also, results of experimental studies of mate choice, including ours, should be interpreted with accuracy due to their basic restriction---females in test situations are limited in making a totally free choice. Under other equal conditions the choice made by the female could be influenced by the activity (mating or courting behavior) of males. Direct tactile and subsequent sexual contacts with both males could enhance asymmetry in preference exhibited by the female. Our finding that females preferred gentle males supports this suggestion. This situation looks realistic in the natural context (in the wild), however, it leaves questions open. We did not find a correlation between the number of mount series of tethered males and number of mount series of freely competing males. This could indicate significant impact of female activity on the net result (ultimate choice), but a complete understanding of how the mechanism of female mate choice works seems feasible only through performing a separate experiment in which the participation of males in female mate choice is limited by a lattice, making sexual intercourse impossible. To what extent the results of this and earlier experiments will coincide is a question for future research. The authors are deeply grateful to Prof. Jan A. Randall for comments and corrections to the English, to Dr A. V. Tchabovsky for valuable suggestions throughout the study and the opportunity to use his original program for registration of hamsters' behavior. We are greatly indebted to 2 anonymous reviewers for their constructive remarks, useful comments and suggestions. Funding ======= This research was supported by the Russian Foundation of Basic Research (RFBS) \[Grant Nos 14-04-00854 and 16-04-01376\]. [^1]: ^a^ The combined variable characterizing STs expression. Result of PCA of males' morphological characteristics. [^2]: ^b^ Hormone concentration in blood serum at the age of 60 days. [^3]: ^c^ Variables that were log-transformed. [^4]: ^d^ Hormone concentration in blood serum at the peak of immune response to SRBC.

Introduction ============ The prevalence of diabetes mellitus is increasing worldwide with approximately half of all persons with diabetes living in Asia \[[@B1]\]. The herb fenugreek (*Trigonella foenum-graecum* L., Fabaceae family) is used both in cooking and for the treatment of diabetes in many parts of the world, especially in China, Egypt, India and Middle Eastern countries \[[@B2]-[@B4]\]. In low-income countries, individuals with diabetes often do not have access to appropriate medications due to a lack of financial resources \[[@B5]\]. Active compounds of fenugreek included soluble fiber \[[@B6]-[@B8]\], saponins \[[@B9],[@B10]\], trigonelle \[[@B11]\], diosgenin \[[@B12]\], and 4-hydroxyisoleucine \[[@B13],[@B14]\]. Hypoglycemic activities have mainly been attributed to dietary fiber \[[@B6],[@B7]\] and saponin \[[@B9]\]. Fenugreek is a widely used herbal medicine for diabetes, but its efficacy for glycemic control remains unclear. Animal studies have shown that fenugreek seed extracts have the potential to slow enzymatic digestion of carbohydrates, reduce gastrointestinal absorption of glucose, and thus reduce post-prandial glucose levels \[[@B8]\]. In addition, fenugreek stimulated glucose uptake in peripheral tissues \[[@B15]\] and had insulinotropic properties in isolated rat pancreatic cells \[[@B16]\]. In humans, fenugreek seeds acutely reduced postprandial glucose and insulin levels \[[@B17]-[@B20]\]. In addition, several longer-term clinical trials showed reductions in fasting and post-prandial glucose levels and glycated haemoglobin (HbA1c) \[[@B9],[@B21]-[@B23]\], but some trials did not show benefit \[[@B24],[@B25]\]. Systematic reviews that have evaluated the effect of various alternative therapies for diabetes included only a few clinical trials of fenugreek \[[@B26]-[@B29]\]. We therefore conducted a systematic review and meta-analysis of the effects of fenugreek on glucose homeostasis based on a comprehensive literature search leading to the identification of a reasonably large number of trials with an evaluation of potential explanations for differences in study results. Methods ======= Data sources and searches ------------------------- To identify articles on the effect of fenugreek on glucose homeostasis we searched MEDLINE (PubMed), SCOPUS, Web of Science, BIOSIS, and Cochrane Trials Registry from inception through Nov 29, 2013 using key search terms related to fenugreek ("fenugreek", "trigonella"), an experimental study design ("trial", "clinical trial", "intervention", "therapy"), to identify potentially relevant articles. The search strategy utilized both index terms and free text to search for synonyms of trigonella, fenugreek and diabetes/healthy subjects, and was limited to human studies. Grey literature such as conference proceedings, abstracts, dissertations and technical reports was identified using the same key terms through the electronic search engines Google Scholar, SCIRUS, CINAHL, and ProQuest. No language restriction was applied. The results (titles, abstracts and citations) of electronic searches were downloaded into EndNote software (EndNote X5, 2011, Thomson Reuters, Philadelphia) and initial screening for eligibility was performed by two independent reviewers (Nithya Neelakantan, Madanagopal Narayanan). When assessment of eligibility based on the title and abstract was insufficient, the full text of the articles was obtained. The second screening of those full text articles was then independently performed by at least two reviewers (Nithya Neelakantan, Madanagopal Narayanan, Rob M van Dam). Disagreements were resolved by consensus. The kappa for the inter-reviewer reliability was 0.78. Study authors were contacted to verify results and methodological quality of retrieved articles where necessary. We used the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) statement to report our findings \[[@B30]\]. Study selection --------------- We included clinical trials that compared single herb preparations of fenugreek in any dose or form with a control intervention that was either placebo or no treatment and evaluated effects on markers of glycemia \[fasting blood glucose, 2 hr postload glucose, glycosylated hemoglobin (% HbA1c) and/or fasting serum insulin levels\]. We excluded trials that used combination preparations of fenugreek with other herbs, non-human studies, observational studies, literature reviews/editorials/letters/case reports, and articles not reporting the outcomes of interest. We also excluded trials with interventions that lasted less than 7 days. The number of articles that did not meet the eligibility criteria and the reasons for their exclusion are shown in Figure [1](#F1){ref-type="fig"}. {#F1} Data extraction and quality assessment -------------------------------------- Details of trial design, study setting, population, randomization, blinding, sample size, duration of follow-up, participant characteristics, interventions, total daily dose and outcome characteristics were independently extracted by two reviewers (Nithya Neelakantan, Madanagopal Narayanan), using a standardized data extraction form. Differences in data extraction were resolved by a third reviewer (Rob M van Dam). The quality assessment was conducted using the CONSORT statement for herbal trials \[[@B31]\] by two reviewers (Nithya Neelakantan, Madanagopal Narayanan), with disagreements resolved by consensus. From each trial, data on mean and SD for all outcomes of interest were extracted. If trials reported fasting blood glucose and 2 hr postload glucose (glucose concentrations 2 hours after the start of the oral glucose tolerance test) in units of mg/dL, this was converted to the standardized international unit \[[@B32]\] of mmol/L by multiplying the glucose values in mg/dL by 0.0555; for fasting serum insulin, we divided the serum insulin values reported in pmol/L by 6.945 and reported the results in mU/L. Parallel trials generally reported the baseline mean and standard deviation and follow-up mean and standard deviation, but not the standard deviation (SD) of change for the intervention and control groups. For parallel trials, the net changes in each outcome measure were calculated as the change in the intervention group minus the change in the control group. For crossover trials, net changes in the outcome measures were calculated as the value of the outcome measure at the end of the intervention period minus the value of the outcome measure at the end of control period. We estimated the SD of the change on the basis of reported p values for differences in means, if available \[[@B33]\]. We used the p-values cutoff if it was only reported that a p-value was below a threshold (e.g., 0.05 if p \< 0.05 was reported) leading to conservative estimates \[[@B34],[@B35]\]. If p-values were not reported, we imputed SD of the change by using a pooled correlation coefficient between baseline and final measurements from a meta-analysis of correlation coefficients from those trials reporting sufficient data. We derived correlation coefficients for individual trials according a standard formula \[[@B33]\] and we then imputed these correlations into the meta-analysis as transformed z scores (±SEs) to estimate the pooled correlation coefficient \[[@B36]\]. For HbA1c and fasting serum insulin measures, due to small number of trials, we estimated the SDs of the change assuming a conservative 0.5 correlation and performed a sensitivity analysis assuming alternative values of 0.25 or 0.75. To investigate the effect of imputed within-person correlation coefficients, we performed sensitivity analyses with a range of correlation coefficients (0.25, 0.50 and 0.75) \[[@B37]\], the pooled estimates did not change substantially. Meta-analysis ------------- The meta-analysis was performed according to the methods described by Curtin et al. \[[@B38]\]. In the combined design meta-analysis, the pooled estimate of treatment effect combining parallel and crossover trial results was the weighted sum of the separate treatment effects estimated, respectively, from parallel and crossover trials divided by the sum of the associated weights. We anticipated large differences in the fenugreek drug preparation format, active components/chemical composition, administration of supplements, and dosages as well as variation in the study population and study design. Therefore, we *a priori* decided to use a random effects model for this meta-analysis. Hence, for each outcome measure, weighted mean differences and corresponding 95% confidence intervals (CI) were calculated by using DerSimonian and Laird random-effects models. We also conducted separate meta-analyses for parallel and crossover trials for the primary outcome measures, fasting blood glucose, and 2 hr post prandial glucose. Heterogeneity in study results was tested by using the Cochran *Q* statistic (and associated p value), and was quantified by the *I*^2^ statistic. The *I*^*2*^ provides an estimate of the percentage of variation in study results that is explained by between-study heterogeneity rather than sampling error \[[@B39]\]. Potential sources of heterogeneity were investigated using *a priori* defined stratified analyses by study design (parallel or crossover), daily dose of fenugreek extract (\<5 g, 5--10 g or \>10 g), study duration (\<30 days or \> =30 days), randomization (yes or no), blinding (yes or no), baseline BMI (\<25 or \> =25 kg/m^2^), study precision (SE of the effect estimate above or below the median), geographical region (India vs. other countries) and age (above or below the median mean age of all studies). Meta-regression analyses were used to assess the significance of differences in the effects of fenugreek between strata. p-values for the overall F-test for a common mean amongst three or more groups were obtained using one way ANOVA. Publication bias was investigated by visual inspection of funnel plots and by the Egger regression test and the Begg adjusted correlation test \[[@B40]\]. The robustness of the findings of the meta-analysis to different assumptions were examined in a sensitivity analysis using both fixed and random effects models, in which the meta-analysis estimates were computed omitting one trial at a time to assess the influence of each individual trial \[[@B41]\]. All tests were two-sided and p \<0.05 was considered statistically significant. The data were analyzed by using Stata version 11 (StataCorp, College Station, Texas). Results ======= Search results -------------- We identified 161 potentially relevant articles and screened the abstracts for eligibility. The flow of trial selection is reported in Figure [1](#F1){ref-type="fig"}. We evaluated 32 full text articles in detail. Nine articles reporting 10 trials met the inclusion criteria for the meta-analysis. Of these 10 trials, one trial \[[@B42]\] reported results separately for participants with mild and severe type 2 diabetes, and thus 11 data points are presented in Table [1](#T1){ref-type="table"}. Five of these had a parallel design and six had a crossover design. ###### Characteristics of the 10 trials (11 data points) included in the meta-analysis of the effects of fenugreek seeds on glycemia **Lead author, year and country** **Population, medication** **Design** **Age, y** **Male (%)** **Sample size, N** **Duration, days** **Fenugreek, preparation** **Daily dose, g** **Control** **Outcomes** **Study quality** ------------------------------------------- ---------------------------- ------------ ------------ -------------- -------------------- -------------------- ---------------------------- ------------------- ------------- ------------------------------- ------------------- ---- ----- Bordia et al., 1997 \[[@B42]\], India Mild T2DM, NR PL NR NR 40 30 Capsule, PS 5 Placebo FBG, 2 hr glucose NR NR NR Bordia et al., 1997 \[[@B42]\], India Severe T2DM, NR PL NR NR 40 30 Capsule, PS 5 Placebo FBG, 2 hr glucose NR NR NR Chevassus et al., 2010 \[[@B43]\], France Overweight, NR PL 38.0 100 40 42 Capsule, H 1.176 Placebo FBG, FSI Yes DB 5% Gupta et al., 2001 \[[@B23]\], India T2DM, SU, BI PL 51.0 76 25 56 Capsule, H 1 Placebo FBG, 2 hr glucose, HbA1c, FSI Yes DB 4% Lu et al., 2008 \[[@B9]\], China T2DM, SU PL 54.4 55 69 84 Capsule, PS 6.3 Placebo FBG, 2 hr glucose, HbA1c Yes DB NR Alamdari et al., 2009 \[[@B17]\], Iran T2DM, Diet, OAD CO 43.1 100 12 56 PS 8 Unspecified FBG, 2 hr glucose, HbA1c Yes NR 25% Chevassus et al., 2009 \[[@B25]\], France Healthy, NR CO 22.0 100 12 14 Capsule, H 1.176 Placebo FBG, FSI Yes DB 0% Raghuram et al., 1994 \[[@B7]\], India T2DM, BI CO 46.6 NR 10 15 Chapati^a^, PS 25 Chapati FBG, 1 hr glucose, HbA1c Yes NR NR Sharma et al., 1990 \[[@B44]\], India T1DM, Insulin therapy CO 22.7 70 10 10 Chapati^a^, DPS 100 Chapati FBG, 2 hr glucose, FSI Yes NR NR Sharma et al., 1990 \[[@B22]\], India T2DM, BI/metformin CO 46.0 67 15 10 Chapati^a^, DPS 100 Chapati FBG, 2 hr glucose, FSI Yes NR NR Sharma et al., 1990 \[[@B22]\], India T2DM, NR CO 42.0 NR 5 20 Chapati^a^, DPS 100 Chapati FBG, 2 hr glucose Yes NR NR *Abbreviations*: *NR* not reported, *T1DM* type 1 diabetes mellitus, *T2DM* type 2 diabetes mellitus, *SU* sulfonylurea, *BI* biguanides, *OAD* oral antidiabetic drug, *PL* parallel, *CO* crossover trials, Daily dose, *g* fenugreek total daily dose in grams, *PS* powdered fenugreek seeds, *H* Hydro-alcoholic extract of fenugreek seeds, *DPS* Debitterized fenugreek seed powder, *FBG* fasting blood glucose, *HbA1c* % glycosylated hemoglobin, *FSI* fasting serum insulin, *RCT* randomized control trial, *DB* double blinded, ^a^Fenugreek incorporated into chapati (unleavened bread). Trial characteristics --------------------- The mean age of participants in the trials ranged from 22.0 to 54.4 years (median: 43.1 y), and the median percentage of males was 76%. Most trials included participants with type 2 diabetes treated with diet or oral anti-diabetic medication (Table [1](#T1){ref-type="table"}). One trial was conducted in persons with type 1 diabetes and two trials included overweight or non-overweight participants without diabetes. The sample size ranged from 5 to 15 participants for crossover trials and from 25 to 69 participants for parallel trials. The sample size for all trials combined was 278. The daily dose of fenugreek seed ranged from 1 g to 100 g (median: 25 g), and the study duration from 10 to 84 days (median: 30 days). Fenugreek supplements were administered as powdered fenugreek seeds, debitterized powdered fenugreek seeds, or hydro-alcoholic seed extract either in form of capsules or as an ingredient of unleavened bread. These were provided in equal doses 2 to 3 times per day. All 10 trials \[[@B7],[@B9],[@B17],[@B22],[@B23],[@B25],[@B42]-[@B44]\] (11 data points) reported fasting blood glucose, 7 trials \[[@B7],[@B9],[@B17],[@B22],[@B23],[@B42],[@B44]\] (8 data points) reported 2 hr glucose, 5 trials \[[@B22],[@B23],[@B25],[@B43],[@B44]\] reported fasting serum insulin and 3 trials \[[@B9],[@B17],[@B23]\] reported HbA1c. With regard to study quality, one trial \[[@B42]\] did not report whether groups were randomized, none of the trials reported details on the method of randomization or allocation concealment, and most trials \[[@B7],[@B17],[@B22],[@B42],[@B44]\] did not report the blinding status. Only four trials reported the percentage of drop-outs \[[@B17],[@B23],[@B25],[@B43]\]. Of which three studies \[[@B23],[@B25],[@B43]\] have reported low percentage drop-out ranging between 0 to 5% and one study \[[@B17]\] has reported 25% drop-out. These subjects were not included in the statistical analyses that were conducted for the primary studies. Most trials provided information on the concentration of components of the used fenugreek preparations. Reported components included diosgenine, saponins, trigonelline (1.4%) and 4-hydroxyisoleucine (1.5%) for the hydro-alcoholic extract \[[@B25],[@B43]\]; alkaloids, carpaine, erythricine, trigonelline, meletin, and saponins for powdered fenugreek seeds \[[@B22],[@B44]\]; and lipids (0.1%), protein (28.3%), starch (6.5%), total fiber (51.7% \[gum 19.2%\]) for debitterized fenugreek seeds \[[@B22],[@B44]\]. Effect on glucose homeostasis ----------------------------- ### Fasting blood glucose All 10 trials (11 data points) were included in the meta-analysis of fasting blood glucose. The individual trial results and the pooled estimates by trial designs are shown in Figure [2](#F2){ref-type="fig"}. Based on the overall pooled estimate, fenugreek significantly reduced blood glucose levels as compared with control treatments (pooled mean difference = -0.96 mmol/l; 95% CI: -1.52, -0.40; p = 0.001). There was large heterogeneity in study results (I^2^ = 80%; p \< 0.001). There were no significant differences in the effects of fenugreek on fasting glucose by study design, study duration, geographical region, mean age and mean BMI of the study population (Table [2](#T2){ref-type="table"}). However, the effect of fenugreek on fasting blood glucose differed significantly by diabetes status with substantial effects in persons with type 1 diabetes and type 2 diabetes, but not in persons without diabetes. There was a large variation in the dose of fenugreek used ranging from 1 g per day to 100 g per day of fenugreek seeds. The effect size differed significantly by the dose of fenugreek used suggesting no effects for the studies using low doses (\<5 g/day) and greater effects with higher doses of fenugreek. Effects of fenugreek on fasting glucose also differed by preparation method of the fenugreek supplement. Studies using debitterized fenugreek powdered showed the greatest reduction in glucose levels, but these were mostly the same trials that administered the highest dose of fenugreek. Similarly, the trials using a hydro-alcoholic extract were the same trials that used the lowest dose. Heterogeneity in effects on fasting glucose was partly explained by fenugreek dose (I^2^~residual~ = 69%; adjusted R^2^ = 47%), fenugreek preparation method (I^2^~residual~ = 38%; adjusted R^2^ = 83%), and diabetes status of the study population (I^2^~residual~ = 55%; adjusted R^2^ = 61%). {#F2} ###### Stratified meta-analyses of the effects of fenugreek on fasting blood glucose and 2 hour postload glucose according to trial and participant characteristics **Characteristics** **Fasting blood glucose** **2 hr-postload glucose** ---------------------------- --------------------------- --------------------------- ------------- --------- ------- ---- ---------------------- ------------- ------- ------ Overall 11 -0.96 (-1.52, -0.40) 80 (65, 89) \<0.001   8 -2.19 (-3.19, -1.19) 71 (40, 86) 0.001   Study design                       Parallel 5 -0.78 (-1.93, 0.37) 86 (69, 94) \<0.001 0.48 4 -1.71 (-2.73, -0.70) 59 (0, 86) 0.07 0.33   Crossover 6 -1.20 (-2.03, -0.38) 77 (49, 90) 0.01   4 -3.32 (-5.90, -0.75) 82 (54, 93) 0.001 Region                       India 7 -1.43 (-2.26, -0.60) 48 (0, 78) 0.07 0.21 6 -2.60 (-4.06, -1.13) 73 (39, 88) 0.002 0.50   Others 4 -0.55 (-1.25, 0.16) 89 (75, 95) \<0.001   2 -1.59 (-3.08, -0.10) 74 (0, 94)   Study precision^a^                       Below median 5 -0.54 (-1.15, 0.07) 86 (69, 94) \<0.001 0.10 4 -1.55 (-2.52, -0.57) 67 (2, 89) 0.03 0.11   Above median 6 -1.68 (-2.62, -0.74) 43 (0, 77) 0.12   4 -3.42 (-5.56, -1.28) 67 (4, 89) 0.03   Study population^b^                       Healthy 2 0.05 (-0.51, 0.61) 82 (23, 96) 0.02   \- \- \- \-     T1DM 1 -4.20 (-6.73, -1.67) \- \- 0.01 1 -3.20 (-6.92, 0.52) \- \- 0.79   T2DM 8 -1.21 (-1.69, -0.73) 26 (0, 67) 0.22   7 -2.14 (-3.19, -1.09) 74 (45, 88) 0.001   Randomization                       Yes 9 -0.94 (-1.56, -0.32) 83 (69, 91) \<0.001 0.83 6 -2.54 (-3.87, -1.21) 73 (37, 88) 0.003 0.49   Unknown 2 -1.11 (-2.10, -0.12) 0 (0, 100) 0.49   2 -1.58 (-3.61, 0.45) 81 (17, 95) 0.02   Blinding status                       Double-blinded 4 -0.44 (-1.22, 0.34) 88 (72, 95) \<0.001 0.15 2 -1.95 (-3.02, -0.88) 16 (0, 55) 0.28 0.74   Unknown 7 -1.37 (-2.08, -0.67) 47 (0, 78) 0.08   6 -2.43 (-3.82, -1.04) 78 (50, 90) 0.000   Daily dose (grams)^b^                       \<5 3 0.02 (-0.46, 0.50) 63 (0, 89) 0.07   1 -1.34 (-2.83, 0.14) \- \-     5-10 4 -1.27 (-1.80, -0.74) 21 (0, 88) 0.28 0.01 4 -1.55 (-2.52, -0.57) 67 (2, 89) 0.04 0.02   \>10 4 -2.07 (-3.58, -0.55) 68 (9, 89) 0.02   3 -4.42 (-5.96, -2.89) 8 (0, 35) 0.58   Preparation^b^                       Powdered seed 5 -1.12 (-1.63, -0.61) 29 (0, 72) 0.23   4 -1.55 (-2.52, -0.57) 67 (2, 89) 0.03     Hydro-alcoholic extract 3 0.02 (-0.46, 0.50) 63 (0, 89) 0.07 0.002 1 -1.34 (-2.83, 0.14) \- \- 0.05   Debitterized seed powder 3 -2.68 (-3.82, -1.54) 0 (0, 90) 0.42   3 -4.42 (-5.96, -2.89) 0 (0, 90) 0.58   Study duration (days)                       \<30 5 -1.48 (-2.64, -0.32) 80 (52, 91) 0.001 0.39 3 -4.42 (-5.96, -2.89) 8 (0, 35) 0.58 0.02   \> = 30 6 -0.78 (-1.68, 0.11) 83 (65, 92) \<0.001   5 -1.49 (-2.28, -0.70) 55 (0, 84) 0.06   Mean Age (years)                       \<43.1 4 -0.73 (-1.61, 0.16) 86 (67, 94) \<0.001   2 -4.14 (-5.80, -2.47) 0 (0, 100) 0.58     \> = 43.1 5 -1.11 (-1.76, -0.47) 45 (0, 80) 0.12 0.60 4 -1.90 (-3.19, -0.61) 67 (4, 89) 0.03 0.21 Mean BMI (kg/m^2^)                       \<25 3 -1.62 (-3.21, -0.02) 93 (82, 97) \<0.001 0.29 2 -2.52 (-3.75, -1.30) 0 (0, 100) 0.71 0.17   \> = 25 3 -0.19 (-1.14, 0.76) 74 (12, 92) 0.02   2 -1.01 (-1.72, -0.30) 0 (0, 100) 0.62   *Abbreviations*: *BMI* body mass index, *CI* confidence interval, *P*~*hetr*~ p value for heterogeneity, *P*~*EM*~ p value for effect modification. ^a^The cutoffs are based on the median of standard error of the effect size for fasting blood glucose and 2hr glucose respectively; For mean age and study duration, the cutoffs are based on the median values of 11 data points. ^b^P~EM~-value for the overall F-test for a common mean amongst the three groups (i.e., study population, and fenugreek drug preparation format), the overall p-value for daily dose was obtained by modeling this as a continuous variable in meta-regression analysis. The funnel plot for effects of fenugreek on fasting blood glucose by study precision appeared to be asymmetrical (Additional file [1](#S1){ref-type="supplementary-material"}: Figure S1, Begg test, p = 0.10) and the Egger test was significant (p = 0.03) suggesting potential publication bias. However, these tests are based on detecting an association between study precision (lower SE of effect estimates) and effect size. Less precise studies also tended to use a greater dose of fenugreek (r = 0.51 between dose and SE of the effect estimates) and we could therefore not distinguish between potential publication bias and the dose of fenugreek used. ### 2 hr glucose Seven trials (8 data points) reported effects of fenugreek on 2 hr glucose values. Forest plots of 2 hr glucose effects in individual trials and the pooled analyses are shown in Figure [3](#F3){ref-type="fig"}. Meta-analysis of the trials yielded a pooled estimate for the effect of fenugreek on 2 hr glucose of -2.19 mmol/l (95% CI: -3.19, -1.19, p \<0.001). There was large heterogeneity in study results (I^2^ = 71%; p = 0.001). Stratified analyses of effects of fenugreek on 2 hr glucose according to study characteristics are presented in Table [2](#T2){ref-type="table"}. As observed for fasting glucose, a higher dose was associated with greater effects on 2 hr glucose concentrations. We also observed stronger effects for trials with study duration less than 30 days than for trials with a longer duration and for trials that administered debitterized fenugreek seed powder than for trials using other fenugreek preparations. However, the trials of short duration that used debitterized fenugreek powder were the same three trials that administered a substantially higher dose of fenugreek than the other trials. Heterogeneity in effects of fenugreek on 2 hr glucose was partly explained by fenugreek dose (I^2^~residual~ = 42%; adjusted R^2^ = 72%) and fenugreek preparation method (I^2^~residual~ = 50%; adjusted R^2^ = 61%). The funnel plot for 2 hr glucose data appeared to be asymmetrical (Additional file [1](#S1){ref-type="supplementary-material"}: Figure S2, Begg test, p = 0.05) and the Egger test (p = 0.03) also provided evidence for publication bias. Again, less precise studies also tended to use a greater dose of fenugreek (r = 0.54 between dose and SE of the effect estimates) and we could not distinguish between potential publication bias and the dose of fenugreek used. {#F3} ### Other outcome measures (HbA1c and fasting serum insulin) Three trials (2 parallel and 1 crossover trial) reported the effects of fenugreek on HbA1c. Fenugreek significantly reduced HbA1c values as compared with control treatment (pooled mean difference -0.85%; 95% CI: -1.49%, -0.22%, p = 0.009) (Additional file [1](#S1){ref-type="supplementary-material"}: Figure S3) without significant heterogeneity in study results (I^2^ = 0%; p = 0.78). Five trials reported fasting serum insulin concentrations. However, fasting insulin may have been affected by exogenous insulin use in persons with type 1 diabetes and fasting insulin values had a strongly skewed distribution in persons with type 2 diabetes. Therefore, we only included two trials \[[@B25],[@B43]\] in persons without diabetes in our meta-analysis of the effects of fenugreek on fasting insulin. The pooled effect of fenugreek on fasting serum insulin was not statistically significant (pooled mean difference = -1.42 mU/L; 95% CI: -3.04, 0.19 mU/L; p = 0.08) (Additional file [1](#S1){ref-type="supplementary-material"}: Figure S4). Heterogeneity for effects on fasting insulin was substantial, but not statistically significant (I^2^ = 62%; p = 0.10). Subgroup analyses were not performed for HbA1c and fasting serum insulin because of the limited number of trials for these outcomes. Sensitivity analyses -------------------- We conducted a sensitivity analysis excluding the trial that did not report randomization status. This exclusion had little effect on the pooled effect on fasting blood glucose (-0.94; 95% CI: -1.56, -0.32) or 2 hr glucose (-2.54; 95% CI: -3.87, -1.21). We also conducted sensitivity analyses for the effects of fenugreek on fasting blood glucose and 2 hr glucose concentrations omitting one study at a time. None of the individual trials dramatically influenced pooled effect estimates, which ranged from -0.78 mmol/l (95% CI: -1.31, -0.25) to -1.20 (95% CI: -1.85 to -0.56) for fasting blood glucose and from -1.79 mmol/l (95% CI: -2.69, -0.90) to -2.52 (95% CI: -3.64 to -1.39) for 2 hr glucose. We also conducted sensitivity analyses with simultaneous exclusion of trials conducted by the same researchers. The three trials reported by Sharma et al. \[[@B22],[@B44]\] had a much larger dose of fenugreek than the other trials and also the lowest precision. In a sensitivity analysis that excluded these trials, the pooled effect on fasting blood glucose (-0.62 mmol/l, 95% CI: -1.14, -0.10) and 2 h glucose (-1.49 mmol/l, 95% CI: -2.28, -0.70) remained statistically significant. We also conducted a sensitivity analysis after excluding the Chevassus trials (\[[@B25],[@B43]\]) that were the only trials in persons without diabetes, used a low dose, and had the largest precision. After exclusion of these studies, the pooled effect estimate for fasting glucose was larger (-1.35 mmol/L; 95% CI: -1.92, -0.74). Adverse effects --------------- Chevassus et al. reported 2 cases of specific urine smell and 1 case of abdominal pain in one trial \[[@B25]\] and 4 cases of mild gastrointestinal symptoms, and 1 case of specific urine and sweat smell in both the treatment group and the control group in their other trial \[[@B43]\]. Of the 12 participants in the treatment group in the study by Gupta et al. \[[@B23]\], 5 developed dyspepsia and mild abdominal distention for the first few days of therapy. This subsided on continuation of the therapy. No renal or hepatic side effects were reported and there were no withdrawals due to the side effects. Lu et al. \[[@B9]\] reported that 2 out of 46 participants of the treated group suffered from stomach discomfort and nausea, and one from diarrhea during the treatment period. These symptoms disappeared after 2 day drug withdrawal without special treatment. No adverse reaction was found after the treatment resumed and all participants finished the trial. Discussion ========== In our meta-analysis of 10 clinical trials, intake of fenugreek seeds resulted in a significant reduction in fasting blood glucose, 2 hr glucose, and HbA1c. However, we observed substantial heterogeneity in study results. Differences in the diabetes status of participants and the large variation in dose of fenugreek seed extract used and type of preparation appeared to be contributors to variation in study results. No major harmful side effects of fenugreek were reported in all included studies. We only found a significant reduction in glucose parameters for trials that administered medium to high doses (≥5 g) of fenugreek seed powder and not for trials that administered low doses (\< 2 g) of hydro-alcoholic extracts. Medium to high doses (range: 5--25 g) of fenugreek seed powder also lowered postprandial glucose levels in acute studies \[[@B6],[@B18],[@B19],[@B21],[@B45],[@B46]\]. Lower doses, as used in three of the trials in our meta-analyses, were not evaluated in acute studies of fenugreek. The mechanisms by which fenugreek may lower blood glucose levels have not been well established in humans. Acute hypoglycemic effects of fenugreek seeds and its extract have been evaluated in individuals with and without diabetes \[[@B18],[@B22],[@B44]\]. Whole fenugreek raw seeds, extracted seed powder, cooked seeds (25 g) and gum isolate of seeds (5 g) decreased postprandial glucose levels, whereas degummed seeds (25 g) showed little effect \[[@B18]\]. These findings suggest that acute effects of fenugreek seeds are mainly due to the gum fraction, but do not exclude a longer term effect of other fenugreek components on glycemia. Animal studies also indicate that the soluble fiber fraction of fenugreek seeds reduces the rate of enzymatic digestion and the absorption of glucose from the gastrointestinal tract \[[@B8]\]. However, data from other studies suggest an effect of other fenugreek components on glucose homeostasis. In diabetic rats, trigonelline ingestion increased insulin sensitivity and reduced blood glucose levels \[[@B47]\]. In addition, a novel amino acid derivative extracted from fenugreek seeds, 4-hydroxyisoleucine, stimulated glucose-dependent insulin release in isolated rat and human pancreatic islet cells \[[@B14]\]. In a trial of acute effects in healthy volunteers, trigonelline reduced the early glucose response during an OGTT \[[@B48]\]. The only previous meta-analysis of the effects of fenugreek on glycemia included only two clinical trials as compared with 10 in the current meta-analysis \[[@B26]\]. Strengths of our study included the comprehensive literature search leading to the identification of a reasonably large number of trials and a detailed analysis of potential sources of heterogeneity in study results. Our study also has several limitations that need to be considered in the interpretation of the results. First, the quality of the included trials was generally poor. None of the trials reported the methods of randomization or allocation concealment, and only a few trials provided information on blinding status and drop-out rates. In addition, with some exceptions \[[@B9]\] it was unclear whether other diabetes medication remained constant during the trial. Most of the included crossover trials did not test the carryover effect or report a washout period. However, we did not find a difference in results between parallel and cross-over trials suggesting that carryover effects did not substantially affect the results. Second, tests for publication bias suggested that such bias may have been present. Tests for publication bias are based on detecting differences in effect sizes by study precision with a greater effect size for less precise (\'smaller') studies suggesting the presence of publication bias. In our meta-analysis less precise studies were also more likely to use larger doses of fenugreek. Differences in dose are thus a possible alternative explanation for the observed \'small study effect', but we were unable to distinguish between the effects of dose and publication bias on effect sizes. Finally, we only found a significant effect on glycemia for powdered fenugreek seeds and our findings do not apply to other forms of fenugreek and may differ for other strains as a result of natural variation in active ingredients. Our systematic review and meta-analysis suggest that fenugreek seeds may contribute to better glycemic control in persons with diabetes mellitus with a similar magnitude of effect as intensive lifestyle \[[@B49]\] or other pharmaceutical treatment added to standard treatment \[[@B50]\]. Fenugreek is widely available at low cost and generally accepted in resource poor countries such as India and China where a large proportion of persons with diabetes in the world reside. Therefore, fenugreek may be a promising complementary option for the clinical management of diabetes. The previously reported lipid lowering effect of fenugreek may be an additional benefit \[[@B19],[@B44],[@B51]\]. However, given the limited quality of the included trials and potential for publication bias, a larger double blind randomized trial should be conducted according to rigorous standards for herbal interventions \[[@B31]\] with an appropriate randomization procedure, an adequate method of allocation concealment and transparent reporting of these methods. The fenugreek herbal product must be standardized and tested for the composition and can be administered in the form of capsules with a recommended dose of at least 5 g per day. In order to provide more conclusive evidence on the benefit of fenugreek for glucose homeostasis, a trial in at least 100 (50 subjects in each of the study arms) persons with diabetes is warranted. The duration should preferably be at least three months to be able to evaluate effects on HbA1c levels and given the longer duration a parallel trial appears most appropriate. Competing interest ================== The authors declare that they have no competing interest. Authors' contributions ====================== NN, MN and RMvD contributed to the conception and design of the study. NN and MN conducted the literature search and data extraction. NN performed the statistical analyses. NN and MN drafted the manuscript. RMvD supervised the study. NN, RJD and RMvD contributed to the interpretation of data and critically revised the manuscript for important intellectual content. All authors gave final approval. NN and RMvD are the guarantors of this work and, as such, had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. Supplementary Material ====================== ###### Additional file 1: Figure S1 Funnel plot for effect of fenugreek on fasting blood glucose. The solid line represents the pooled effect estimate expressed as the weighted mean difference and the dashed lines represent pseudo-95% confidence limits. **Figure S2.** Funnel plot for effect of fenugreek on 2 hour postload glucose. The solid line represents the pooled effect estimate expressed as the weighted mean difference and the dashed lines represent pseudo-95% confidence limits. **Figure S3.** Forest plot of the effect of fenugreek on HbA1c. The effects in individual trials are depicted as open squares with 95% confidence intervals (CIs). The pooled estimate with 95% CI is depicted as an open diamond. **Figure S4.** Forest plot of the effect of fenugreek on fasting serum insulin. The effects in individual trials are depicted as open squares with 95% confidence intervals (CIs). Pooled estimate with 95% CI is depicted as an open diamond. ###### Click here for file Acknowledgements ================ This work was supported by Saw Swee Hock School of Public Health, National University of Singapore. Funding ======= Saw Swee Hock School of Public Health, National University of Singapore.

Introduction {#sec001} ============ Oropharyngeal candidiasis (OPC) occurs in a diverse group of patients. Risk factors for OPC include the use of dentures, corticosteroid inhalers, cigarettes, broad-spectrum antibiotics, and immunosuppressive and chemotherapeutic agents. Patients with HIV, diabetes, and iatrogenic or autoimmune-induced dry mouth are also at substantial risk for OPC. This infection is caused primarily by *Candida albicans*, a ubiquitous polymorphic fungus that is part of the normal microbiota of the gastrointestinal and reproductive tracts of healthy individuals. In order to persistently colonize the oropharynx, *C*. *albicans* must adhere to the epithelial cell lining of the oral mucosa while avoiding being killed by host antimicrobial factors. OPC develops when local host defenses are weakened, permitting the fungus to invade and damage oral epithelial cells. The epithelial cells respond to fungal infection by secreting antimicrobial peptides that directly kill the fungus and by releasing pro-inflammatory cytokines that recruit neutrophils to the focus of infection, where they can kill *C*. *albicans* and limit the extent of epithelial cell damage \[[@ppat.1006056.ref001],[@ppat.1006056.ref002]\]. In this Pearl, we summarize recent advances in our knowledge of the pathogenesis of OPC, focusing on fungal-epithelial interactions. *C*. *albicans* Invades, Damages, and Stimulates Oral Epithelial Cells {#sec002} ====================================================================== Although OPC is a superficial fungal infection, it is characterized by invasion of the epithelial cell lining of the oropharynx \[[@ppat.1006056.ref003]\]. *C*. *albicans* can invade into oral epithelial cells by two distinct mechanisms, induced endocytosis and active penetration ([Fig 1](#ppat.1006056.g001){ref-type="fig"}). During induced endocytosis, *C*. *albicans* hyphae express the Als3 \[[@ppat.1006056.ref004]\] and Ssa1 \[[@ppat.1006056.ref005]\] invasins, which bind to E-cadherin and a heterodimer composed of the epidermal growth factor receptor (EGFR) and HER2 on the epithelial cell surface \[[@ppat.1006056.ref006]\] ([Fig 1](#ppat.1006056.g001){ref-type="fig"}). These binding events trigger the clathrin-dependent endocytosis machinery and induce epithelial cells to produce pseudopods that engulf the fungus and pull it into the cell. Treatment of mice with a small molecule inhibitor of EGFR/HER2 reduces the severity of OPC, suggesting that induced endocytosis contributes to the pathogenesis of this infection \[[@ppat.1006056.ref006]\]. Additional epithelial cell signaling pathways that are required for maximal endocytosis of *C*. *albicans* include the platelet-derived growth factor BB (PDGF BB) and neural precursor-cell-expressed developmentally down-regulated protein 9 (NEDD9) pathways. Activation of these signaling pathways requires *C*. *albicans* hyphal formation and expression of the Als3 invasin \[[@ppat.1006056.ref007]\]. The relationship among these pathways and those activated by E-cadherin and EGFR/HER2 has not yet been determined. {#ppat.1006056.g001} Active penetration occurs when elongating *C*. *albicans* hyphae physically push their way into the epithelial cell ([Fig 1](#ppat.1006056.g001){ref-type="fig"}) \[[@ppat.1006056.ref008]\]. While it is likely that both induced endocytosis and active penetration occur during OPC, it has been difficult to determine the role of active penetration in the pathogenesis of OPC because most invasins are hyphal specific. Thus, *C*. *albicans* mutants that do not form hyphae also fail to express invasins and, consequently, are defective in both active penetration and induced endocytosis. *C*. *albicans* can also penetrate epithelial cell barriers via a paracellular route by secreting lytic enzymes such as members of the secreted aspartyl proteinase (SAP) family. These proteases degrade E-cadherin and other inter-epithelial cell junctional proteins, enabling the organism to penetrate between epithelial cells \[[@ppat.1006056.ref009]\] ([Fig 1](#ppat.1006056.g001){ref-type="fig"}). Recently, it has been found that *C*. *albicans* infection also stimulates the activity of epithelial cell calpain, a cysteine protease that degrades E-cadherin. Moreover, calpain activity is dramatically enhanced in epithelial cells that are co-infected with *C*. *albicans* and *Streptococcus oralis* \[[@ppat.1006056.ref010]\]. When *C*. *albicans* invades epithelial cells, it damages them. Epithelial cells in turn respond to *C*. *albicans* by secreting proinflammatory cytokines and antimicrobial peptides. Recently, it was discovered that *C*. *albicans*-induced epithelial cell damage is mainly caused by candidalysin, a toxin released by the fungus ([Fig 1](#ppat.1006056.g001){ref-type="fig"}). Candidalysin is encoded by the hyphal-specific gene *ECE1*. The protein is cleaved by the Kex2 protease into eight peptides, of which only one permeabilizes epithelial cell membranes, induces epithelial cell damage, and stimulates epithelial cell cytokine secretion. A *C*. *albicans ece1*Δ/Δ deletion mutant is defective in damaging and stimulating oral epithelial cells in vitro. Furthermore, it has a greatly attenuated virulence in the mouse model of OPC \[[@ppat.1006056.ref011]\]. Thus, candidalysin and epithelial cell damage play a central role in the pathogenesis of OPC. Although the presence of candidalysin is required to induce a maximal epithelial cell response to *C*. *albicans*, binding of yeast, yeast lysates, and Als3 to epithelial cells is sufficient to induce some epithelial cell responses even in the absence of this toxin. For example, contact with *C*. *albicans* yeast cells, which do not secrete candidalysin, stimulates phosphorylation of the Akt serine/threonine kinase within 5 minutes \[[@ppat.1006056.ref012]\]. In addition, lysates of yeast cells induce secretion of IL-8, possibly by interacting with intercellular adhesion molecule-1 (ICAM-1) \[[@ppat.1006056.ref013]\]. Finally, *Saccharomyces cerevisiae* cells expressing the Als3 invasin stimulate the phosphorylation of EGFR and HER2 within 20 minutes \[[@ppat.1006056.ref006]\]. These data suggest that oral epithelial cells sense and respond to contact with *C*. *albicans*, and that this initial response is subsequently amplified by candidalysin. While EGFR/HER2 and E-cadherin interact with *C*. *albicans* hyphae, the epithelial cell receptor(s) that is activated by yeast-phase *C*. *albicans* is currently unknown. Interplay between Epithelial Cell-Derived Antimicrobial Peptides and *C*. *albicans* {#sec003} ==================================================================================== Host defense peptides (HDPs) represent one of the first lines of defense against invading microbes. Epithelial cells release multiple types of HDPs, including β-defensins and cathelicidins, that either kill or inhibit the growth of *C*. *albicans*. Different HDPs have different mechanisms of action, and their targets include the fungal cell membrane and mitochondria \[[@ppat.1006056.ref014]\]. HDPs that are highly expressed during the early stage of OPC in mice include murine β-defensin 1 (mBD1), β-defensin 3 (the homolog of human β-defensin 2), and the alarmins S100A8 and S100A9 \[[@ppat.1006056.ref015]\]. The importance of epithelial cell--derived HDPs in the host defense against OPC is demonstrated by finding that mBD1-deficient mice develop severe OPC, even in the absence of immunosuppression \[[@ppat.1006056.ref016]\]. By contrast, immunocompetent wild-type mice are resistant to OPC, and when orally inoculated with *C*. *albicans*, clear the infection within 2--3 days. Mice that are deficient in either IL-17 receptor A (IL-17RA) or IL-17RC fail to up-regulate HDPs in the oral epithelium in response to *C*. *albicans* infection and are highly susceptible to OPC \[[@ppat.1006056.ref001],[@ppat.1006056.ref015]\]. Recently, it was shown that mice with oral epithelial cell specific deletion of IL-17RA have reduced expression of *Defb3* (encoding β-defensin 3) in the oral mucosa and increased susceptibility to OPC. This increased susceptibility is phenocopied by deletion of *Defb3* \[[@ppat.1006056.ref017]\]. Finally, people with mutations in signal transducer and activator of transcription 3 (STAT3) fail to mount a Th17 response, have significantly reduced salivary β-defensin 2 and histatins, and suffer from chronic mucocutaneous candidiasis \[[@ppat.1006056.ref018]\]. Collectively, these data indicate that the response of oral epithelial cells to *C*. *albicans*, especially the production of HDPs, contributes to the host defense against OPC. When *C*. *albicans* colonizes the oral mucosa of healthy individuals and causes OPC in patients, it must be able to withstand HDPs. The fungus has evolved multiple different mechanisms that enable it to evade the deleterious effects of HDPs and persist in the oral cavity \[[@ppat.1006056.ref014]\]. These mechanisms include shedding of the Msb2 mucin, which binds to and inactivates HDPs, and secretion of aspartyl proteases, which break down HDPs \[[@ppat.1006056.ref019],[@ppat.1006056.ref020],[@ppat.1006056.ref021]\] ([Fig 1](#ppat.1006056.g001){ref-type="fig"}). *C*. *albicans* also expresses the Flu1 efflux pump that reduces intracellular HDP levels \[[@ppat.1006056.ref022]\]. Exposure to HDPs induces a stress response in *C*. *albicans*, and susceptibility to HDPs is governed in part by the signaling pathways that are activated in response to these peptides. Activation of the high-osmolarity-glycerol (HOG) pathway enables *C*. *albicans* to resist histatin 5 and human β-defensins 2 and 3 \[[@ppat.1006056.ref023],[@ppat.1006056.ref024]\]. The Ssd1 RNA-binding protein acts in part through the Bcr1 transcription factor to mediate resistance to human β-defensin 2 \[[@ppat.1006056.ref025],[@ppat.1006056.ref026]\]. Conversely, activation of the Cek1 mitogen-activated protein kinase (MAPK) enhances surface exposure of β-1,3-glucans, elevating the uptake of histatin 5 and increasing sensitivity to this HDP \[[@ppat.1006056.ref027]\]. Co-infection with *C*. *albicans* and Other Microorganisms Influences Virulence during OPC {#sec004} ========================================================================================== *C*. *albicans* is frequently isolated from the oral cavity in conjunction with other microbial pathogens. In a study of AIDS patients with OPC \[[@ppat.1006056.ref028]\], 107 out of 1,106 episodes of infection (9.7%) were caused by more than one species of *Candida*. Among the subjects with multispecies OPC, *C*. *albicans* plus *C*. *glabrata* (68.2%) was the most common co-infection, followed by *C*. *albicans* plus *C*. *tropicalis* (15.0%) and *C*. *albicans* plus *C*. *glabrata* plus *C*. *tropicalis* (8.4%). These epidemiological data suggest that *C*. *glabrata* has unique characteristics that facilitate co-infection with *C*. *albicans*. Silva et al. \[[@ppat.1006056.ref029]\] reported that *C*. *glabrata* is unable to invade reconstituted human oral epithelium in vitro and causes minimal epithelial cell damage. A potential explanation for this finding is that *C*. *glabrata* grows only as yeast in vivo and does not form hyphae at foci of infection. However, when oral epithelial cells are infected with a mixture of *C*. *glabrata* and *C*. *albicans*, the *C*. *glabrata* cells bind to *C*. *albicans* hyphae and are carried deeper into the epithelium, where they cause extensive epithelial cell damage ([Fig 1](#ppat.1006056.g001){ref-type="fig"}). It was recently reported that *C*. *glabrata* alone is not able to cause disease in a mouse model of OPC. However, when mice are inoculated with a mixture of *C*. *albicans* and *C*. *glabrata*, they develop more severe disease than mice infected with *C*. *albicans* alone \[[@ppat.1006056.ref030]\]. This synergistic infection requires the expression of the Als1 and Als3 adhesins/invasins by the *C*. *albicans* hyphae. Multiple *C*. *glabrata* adhesins, including Epa8, Epa19, Awp2, Awp7, and CAGL0F00181, mediate adherence to *C*. *albicans* hyphae; whether they interact specifically with *C*. *albicans* Als1 and/or Als3 remains to be determined \[[@ppat.1006056.ref030]\]. Collectively, these results provide an explanation for the frequent co-isolation of *C*. *glabrata* and *C*. *albicans* in patients with OPC. Although *C*. *tropicalis* can occasionally cause OPC in conjunction with *C*. *albicans*, the mechanistic basis for this co-infection is currently unknown. Co-infection with *C*. *albicans* can also augment infection with *S*. *aureus*. When mice are immunosuppressed with cortisone acetate and then orally inoculated with *S*. *aureus*, the bacteria persist in the oral cavity in relatively low numbers and do not cause detectable disease \[[@ppat.1006056.ref031]\]. However, when mice with OPC due to *C*. *albicans* are orally inoculated with *S*. *aureus*, the bacteria adhere to the fungal hyphae, which transport the bacteria along with them as they invade into the superficial oral epithelium. Not only does the presence of *C*. *albicans* result in higher levels of *S*. *aureus* in the oral cavity but it also enables *S*. *aureus* to invade the oral mucosa and cause a fatal hematogenously disseminated infection. This pathologic interaction between *S*. *aureus* and *C*. *albicans* is dependent on Als3; even though an *als3*Δ/Δ null mutant is still able to cause OPC, *S*. *aureus* co-infection does not result in a disseminated staphylococcal infection \[[@ppat.1006056.ref031]\]. Although OPC is not known to be associated with disseminated staphylococcal infection in patients, it is possible that the interaction of *S*. *aureus* with *C*. *albicans* at other anatomic sites, such as the gastrointestinal tract or peritoneal cavity \[[@ppat.1006056.ref032]\], might facilitate the development of *S*. *aureus* bacteremia. *C*. *albicans* is also known to interact with common oral commensal bacteria. In vitro studies demonstrate that *Streptococcus gordonii* and *C*. *albicans* co-adhere during mixed-species biofilm formation \[[@ppat.1006056.ref033]\]. This interaction is mediated by the *S*. *gordonii* cell wall-anchored proteins SspA and SspB and the *C*. *albicans* surface proteins Als3, Eap1, and Hwp1. The presence of *S*. *gordonii* enhances *C*. *albicans* hyphal development and increases biofilm mass \[[@ppat.1006056.ref033],[@ppat.1006056.ref034],[@ppat.1006056.ref035]\]; whether it also augments *C*. *albicans* virulence during OPC is currently unknown. Co-infection with *Streptococcus mutans* also increases *C*. *albicans* biofilm formation in vitro. However, *Galleria mellonella* larvae infected with both organisms have prolonged survival compared to larvae infected with *C*. *albicans* alone. This reduction in virulence is likely due to inhibition of fungal hyphal formation by the bacteria \[[@ppat.1006056.ref036]\]. By contrast, *Streptococcus oralis* both augments *C*. *albicans* biofilm formation in vitro and increases virulence during experimental OPC in mice, as manifested by increased size of the oral lesions and greater fungal dissemination to distal organs \[[@ppat.1006056.ref037],[@ppat.1006056.ref038]\]. These data demonstrate that the interaction of *C*. *albicans* with the oral microbiota has the potential to significantly influence the pathogenesis of OPC. Conclusions and Perspective {#sec005} =========================== The oral epithelium is a critical component of the host defense against OPC. However, *C*. *albicans* has evolved multiple strategies to breach the epithelial cell barrier, withstand HDPs, and cause a superficial infection. In mouse models of infection, other microbial pathogens can interact with *C*. *albicans* and alter virulence. This pathogenic interaction is undoubtedly relevant to human infection in the case of *C*. *glabrata* and streptococci because these organisms are commonly isolated in conjunction with *C*. *albicans*. Whether mucosal infection with *C*. *albicans* predisposes patients to develop invasive staphylococcal infections is currently unknown but is clearly worthy of investigation. We would like to thank the students and faculty of the molecular mycology course "Current Approaches to Fungal Pathogenesis," who stimulated us to write this Pearl. [^1]: SGF is a co-founder of and shareholder in NovaDigm Therapeutics, Inc.

{#sp1 .26} {#sp2 .27} {#sp3 .28}

Key question {#Sec1} ============ Does hypo-attenuated leaflet thrombosis occur in bioprosthetic mitral valve? Key finding {#Sec2} =========== Hypo-attenuated leaflet thrombosis was observed in patients who underwent bioprosthetic mitral valve replacement. Take-home message {#Sec3} ================= Although rare, hypo-attenuated leaflet thrombosis of mitral valve occurs and must be further characterized. Introduction {#Sec4} ============ A series of reduced aortic-valve leaflet motion and a radiographic finding of hypoattenuating opacities in bioprosthetic aortic valve was first described in 2015 \[[@CR1]\], bringing attention to this previously unrecognized entity. Using 4D volume-rendered computed tomography (CT) scans, this was identified in 40% of the evaluated cohort, occurring in both transcatheter and surgically implanted aortic valve prostheses. Recent analysis of two large registries reported 13% prevalence of this entity, and reported a higher prevalence in valves implanted via transcatheter aortic valve replacement (TAVR) compared to those implanted via surgical aortic valve replacement (SAVR) \[[@CR2]\]. Importantly, its previous perception as a benign finding was questioned by the demonstration of increased incidences of cerebrovascular events associated with subclinical leaflet thrombosis in this series \[[@CR2]\]. Hypo-attenuated leaflet thickening (HALT) of bioprosthetic valve is a hallmark finding of subclinical leaflet thrombosis \[[@CR3]\]. Its characteristics have been described fairly extensively in aortic valves, but its occurrence and characteristics are unknown in bioprosthetic mitral valves (bMV). Mitral valve prosthesis may be at higher risk of leaflet thrombosis for its exposure to lower flow compared to that of prostheses in the aortic position. In addition, possible impact of anticoagulation strategy on HALT would be of interest in the context of current variability in anticoagulation following mitral valve replacement \[[@CR4]\] despite the guideline recommendation \[[@CR5]\]. This study describes the characteristics and prevalence of HALT in surgically-implanted bMV in a single-center series. Materials and methods {#Sec5} ===================== Patient selection {#Sec6} ----------------- A cross-sectional study of 175 consecutive patients who underwent mitral valve replacements with bMV at Toyohashi Heart Center, Japan, between 2007 and 2017 was conducted. Fifty-six patients died during follow-up. Thirty-four patients were excluded for history of chronic kidney disease, due to an elevated risk of further renal injury from contrasted CT scan. Eighty-five surviving patients without chronic kidney disease were contacted between June and October 2017 to undergo contrasted multi-dimensional CT scan, of whom 53 agreed to undergo the scan between June and October 2017 to evaluate the present of HALT in bMV (Additional file [1](#MOESM1){ref-type="media"}: Figure S1). Toyohashi Heart Center Institutional Review Board approved this study, and individual patient consent was obtained at the time of CT scanning. Anticoagulant and antiplatelet therapy use were ascertained at the time of hospital discharge from index hospitalization and at the time of CT scanning. At our center, patients undergoing MVR with bMV are discharged on warfarin and 81 mg aspirin unless contraindicated, and anticoagulation is continued for up to 3 months, after which anticoagulation is discontinued unless there are other indications to continue. CT scanning and HALT definition {#Sec7} ------------------------------- All patients underwent cardiac-gated contrasted CT scan using a 256-slice CT scanner (Brilliance iCT; Philips Medical Systems, Eindhoven, the Netherlands) at the Toyohashi Heart Center. Cardiac gating was performed with prospective electrocardiogram triggering (75% of R-R interval) to obtain a slice thickness of 2.5 mm. The scan was performed between the tracheal bifurcation and the diaphragm with the following parameters: collimation width, 32 × 0.625 mm; rotation time, 330 ms/revolution; tube voltage, 120 kV; and maximum effective tube current, 412 mA. Image reconstruction was gated prospectively to 30--45% of R-R interval. CT images were reconstructed across the entire cardiac cycle using a cardiac standard filter with a slice thickness of 2.5 mm. CT datasets were transferred to an offline workstation (Intelli Space Portal; Philips Medical Systems) for image analysis. The images were evaluated by an attending imaging cardiologist (T.S) and an attending cardiac surgeon and consensus was obtained in all findings of HALT. HALT was defined as the presence of a low-density area on bMV during systole on image with the most closure of the valve, which parallels the definition of HALT in aortic valve bioprosthesis \[[@CR3]\]. Transthoracic echocardiogram was obtained at the time of CT scan. Data were collected on patient demographics, comorbidity, presenting symptoms, operative details, postoperative survival and stroke, echocardiographic and CT scan findings, and use of anticoagulant or antiplatelet therapy at the time of index hospital discharge and at the time of CT scans. CT characteristics and echocardiographic findings including ejection fraction (EF), mitral regurgitation (MR), mean mitral pressure gradient (PG), and mitral valve area (MVA) are reported. The prevalence of HALT was calculated by the number of patients with HALT divided by the 53 patients who underwent CT scan as a denominator. All analysis was conducted using SAS version 9.4 (SAS Institute, Cary NC). Results {#Sec8} ======= Of the 53 patients who underwent CT evaluation, 3 patients (5.7%) were found to have a HALT on bMV, each affecting different leaflets. The mean time from index MVR to CT scan was 3.4 ± 0.8 years in HALT cohort and 3.4 ± 2.7 years in non-HALT cohort. Table [1](#Tab1){ref-type="table"} summarizes patient characteristics of the entire cohort. Oral anticoagulant and antiplatelet therapy use at the time of hospital discharge and at the time of CT scan are summarized in Table [2](#Tab2){ref-type="table"}. Among the entire cohort of 53 patients, 74% were discharged on the dual therapy of single antiplatelet agent (SAPT) and warfarin, and additional 21% were on warfarin only. At the time of CT scan, 43% were on warfarin only, and 26% were on the dual therapy of SAPT and warfarin. One patient who was found to have HALT was on warfarin with the INR in the therapeutic range at the time of CT scan. Table 1Baseline characteristics of patients with CT scanVariables***N*** = 53Age (years)71.6 ± 8.9Male26 (49.0)Diabetes12 (22.6)COPD0 (0.0)Creatinine (mg/dL)1.02 ± 0.95Chronic kidney disease0 (0.0)NYHA functional class I or II42 (79.2)NYHA functional class III or IV11 (20.8)Previous PMI0 (0.0)Previous CABG1 (1.9)Previous valve surgery4 (7.5)Operative variables Emergency3 (5.7) Urgent0 (0.0) Elective50 (94.3)Concomitant operations MAZE34 (64.2) CABG9 (17.0) AVR14 (26.4) TAP29 (54.7)Cross-clamp time (minutes)105.7 ± 31.2Perfusion time (minutes)154.4 ± 37.5Operation time (minutes)271.5 ± 54.0Valve size (mm) 2515 (28.3) 2727 (50.9) 2910 (18.9) 311 (1.9)Valve type Carpentier-Edwards Pericardial Bioprosthesis12 (22.6) Magna Ease16 (30.2) Mosaic15 (28.3) St. Jude Medical Epic10 (18.9) Stroke within 30 days of operation2 (3.8)Data are displayed as mean ± SD or n (%). *AVR* Aortic valve replacement, *CABG* Coronary artery bypass grafting, *COPD* Chronic obstructive pulmonary disease, *NYHA* New York Heart Association, *PMI* Pacemaker implantation, *TAP* Tricuspid annuloplastyTable 2Oral anticoagulant and antiplatelet therapy use at the time of hospital discharge and at the time of CT scanMedicationsAt discharge (N = 53)At CT (N = 53)SAPT3 (5.7)13 (24.5)DAPT0 (0.0)2 (3.8)SAPT + warfarin39 (73.6)14 (26.4)SAPT + DOAC0 (0.0)0 (0.0)Warfarin11 (20.8)23 (43.4)DOAC0 (0.0)1 (1.9)Unknown0 (0.0)0 (0.0)Data are displayed as n (%). *CT* Computed tomography, *SAPT* Single antiplatelet therapy, *DAPT* Dual antiplatelet therapy, *DOAC* Direct oral anticoagulant All three patients were asymptomatic at the time of CT scan. Table [3](#Tab3){ref-type="table"} summarizes pertinent finding in each patient who was found to have HALT. Figure [1](#Fig1){ref-type="fig"} displays CT findings of HALT in the three patients. First patient was 3.6 years out of mitral valve replacement (MVR) with 27 mm Magna Ease bMV (Carpentier-Edwards Lifescience, Irvine CA). CT image demonstrated a HALT in one leaflet on the periphery near the sewing ring (Fig. [1](#Fig1){ref-type="fig"}). Echocardiogram demonstrated EF of 45%, trace MR with mean PG of 8 mmHg, and MVA of 2.24 cm^2^ at the time of CT. Second patient was 2.3 years out of MVR. The mitral bioprosthesis was 27 mm Mosaic bMV (Medtronic, Minneapolis, MN). CT image demonstrated a HALT in a posteromedial single leaflet (Fig. [2](#Fig2){ref-type="fig"}). Echocardiogram demonstrated EF of 61%, mild MR with PG of 5 mmHg with MVA of 2.1cm^2^. Third patient was 4 years out of MVR and concomitant AVR with 27 mm Magna Ease bMV. CT showed a HALT extending to almost the entire leaflet (Fig. [3](#Fig3){ref-type="fig"}). Echocardiogram demonstrated EF of 64%, mild MR with PG of 2.7 mmHg and MVA of 2.76 cm^2^. Table 3Summary of three patients with HALTAge (years)/SexPreoperative diagnosisProcedureAntithrombotic therapy at dischargeAt the timing of detectionDegree of leaflet immobility in TTE (concordance with CT)HALT area (mm^**2**^)Clinical symptomMax mean PG (mmHg) /MR165/MalePMRMVR, CABGCoumadinAspirin 3.6 yearsModerate40.8None8/Trivial274/FemaleMSRedo-MVR (previous ASD repair)Coumadin and aspirinCoumadin 2.3 yearsMild40.3None5/Mild372/MaleAS, MRAVR, MVR, MAZEAspirinNone 4.1 yearsModerate48.7None2.7/Mild*AS* Aortic stenosis, *ASD* Atrial septal defect, *AVR* Aortic valve replacement, *CABG* Coronary artery bypass graft, *CT* Computed tomography, *MR* Mitral regurgitation, *HALT* Hypo-attenuated leaflet thickening, *MS* Mitral stenosis, *MVR* Mitral valve replacement, *PG* Pressure gradient, *PMR* Papillary muscle rupture, *TTE* Transthoracic echocardiogramFig. 165-year-old male. 3.6 years out of mitral valve replacement. Long-axis and short-axis view. CT image demonstrated a HALT in one leaflet on the periphery near the sewing ring. All panels were obtained on the same dayFig. 274-year-old female. 2.3 years out of mitral valve replacement. CT image demonstrated a HALT in a posteromedial single leafletFig. 372-year-old male. 4 years out of mitral and aortic valve replacement. CT showed a HALT extending to almost the entire leaflet First patient had stable angina related to failed saphenous vein graft, 38 months postoperatively and underwent cardiac-gated CT scan. This CT scan also demonstrated HALT. He underwent percutaneous coronary intervention (PCI) for the vein graft with resolution of symptoms. Dual antiplatelet therapy was initiated following the PCI. Repeat CT scan was obtained 12 months after the event to reevaluate coronary and mitral valve, and demonstrated complete resolution of the thrombus (Fig. [4](#Fig4){ref-type="fig"}). His mean transvalvuar pressure decreased to 5 mmHg from 8 mmHg at the time of first CT. Fig. 4**a** initial CT in 65-year-old male. **b** Repeat CT scan was obtained 12 months after prasugrel and aspirin, and demonstrated complete resolution of the thrombus Discussion {#Sec9} ========== In this single-center cross-sectional study, the prevalence of HALT in bMV was low at 5.7%. This number should be interpreted in the context of the anticoagulant use pattern, which was routine anticoagulation with warfarin along with 81 mg aspirin for at least 3 months following the index operation. Surprisingly, one patient was on warfarin with the INR in the therapeutic range at the time of the CT scan, in which HALT was identified. Possible explanations are numerous, and include the possibility that HALT is a point in the spectrum of structural valve degeneration, in which the etiology is not necessarily a thrombus formation \[[@CR6]\]. All three patients with HALT presented without valve-related symptoms and echocardiographic abnormality was identified in all three patients. The extent of hypoattenuation varied from those only occupying minimal portion of the leaflet to the one extending to almost the entire leaflet. The temporal progression and potential reversal with anticoagulation remains topic of further investigation. This study offers the first insight into this previously suspected but poorly elucidated entity. Its association with the longevity of prosthetic valve is of interest as well. In the aortic valve domain, majority of patients with HALT detected by multi-detector CT (MDCT) appears to have normal echocardiographic findings. In a series of 140 patients who underwent TAVR with Edwards Sapien XT, MDCT detected HALT in 5 patients (4%), although 4 of them were asymptomatic with normal echocardiographic findings \[[@CR7]\]. Similar pattern was expected in HALT in bMV, and this prompted us to obtain both MDCT and echocardiographic imaging in our cohort. Indeed, of the three patients with HALT detected on MDCT, echocardiographic findings were abnormal in all three patients. The incidence of HALT in bioprosthetic aortic valve varies across studies but most resides in the range between 4 and 12% \[[@CR2], [@CR7]--[@CR9]\]. The incidence of 5.7% in our bMV cohort falls within this range. The original report by Makkar and colleagues is an outlier with the prevalence of 40% at 30 days in their series of 55 patients who underwent either SAVR or TAVR \[[@CR1]\]. The vast majority of the patients in Makkar's study was only on either dual or single antiplatelet therapy, which may partly explain the discrepancy in the prevalence. However, it should also be noted that in a series of 249 Sapien S3 TAVR patients who were only on dual or single antiplatelet therapy, the prevalence of HALT was 10% \[[@CR9]\]. Importantly, this study also reported that CT finding of HALT resolved almost completely following full anticoagulation \[[@CR9]\], corroborating its hypothesized thrombotic etiology. Unfortunately, repeat CT scans following the institution of anticoagulation in patients with HALT were not obtained in our cohort. Structural valve degeneration (SVD) and valve longevity remains active area of investigation with limited long-term data compared to the surgical aortic valve prosthesis. A standardized definition of SVD was proposed in 2018 to help facilitate this \[[@CR10]\]. HALT and leaflet thrombosis are suspected to play a role in accelerating SVD, possibly via triggering inflammation and subsequent fibrocalcific remodeling of valve leaflets \[[@CR6]\]. Subclinical leaflet thrombosis in bioprosthetic aortic valve appears to be more common in TAVR valves compared surgically placed valves \[[@CR2]\], with balloon-expandable valve being associated with highest risk of overt valve thrombosis in TAVR \[[@CR11]\]. In vitro study has suggested that the difference among valves in thrombogenicity is due to the flow properties \[[@CR12]\]. The trans-mitral flow property differs significantly from that of trans-aortic flow, which may produce a differential effect towards the thrombogenicity. The prevalence of HALT in our series was relatively low compared to previously reported prevalence in aortic valve bioprosthesis, although this is difficult to interpret in the context of limited sample size. In the aortic valve domain, it is hypothesized that HALT exist along the spectrum of SVD in that HALT represents the earliest phase of SVD, with a more severe form manifesting in reduced leaflet motion and hemodynamically overt valve thrombosis, eventually leading to premature SVD \[[@CR6], [@CR10]\]. Extrapolating from the excellent long-term durability of surgically implanted aortic valves \[[@CR13]\] and a differential in the risk of HALT between SAVR and TAVR valves, the sequence of events likely occurs more commonly in TAVR valves compared to SAVR valves \[[@CR2], [@CR14]\]. In our series, all three patients with HALT in bMV had hemodynamically occult HALT with mean PG in a normal range with normal leaflet motions on echocardiogram. Importantly, the cumulative incidence of HALT increases as the duration of follow-up increase. In a series of 70 patients following TAVR, only 1 was found to have HALT at the time of discharge, 7 at 6-month follow-up, and 10 at 1-year \[[@CR15]\]. Our cohort had a mean follow up time of 3.4 years since the index valve replacement, which yielded the HALT prevalence of 3 among 53 patients. Longer follow up would provide insights on temporal changes in the risk of HALT in bMV, although obtaining serial contrasted CT scans poses logistical challenges. The finding of HALT has been associated with increased risk of transient ischemic attack (TIA) or stroke \[[@CR2]\]. In the series of 890 patients who underwent surgical or transcatheter AVR, those with HALT had stroke or TIA rate of 7.9 per 100 person-years whereas those without HALT had the rate of 2.4 per 100 person-years, and this difference was statistically significant \[[@CR2]\]. Interestingly, the presence of reduced leaflet motion alone was not associated with a different risk of cerebrovascular event. In our series, none of the three patients with HALT had neurologic symptoms, and brain imaging was deferred. Evaluation of the incidence of cerebrovascular event associated with bMV at a larger scale is in need. Current data are insufficient to infer whether routine screening for this entity should be recommended. Additionally, the role of lipid lowering agent is debated for it relationship with slowing of aortic valve degeneration. Only the first patient was on a lipid lowering agent at the time of presentation and its significance remain unclear. Reversal of HALT and presumably the protective effect of anticoagulation are evident from previous observational studies. It also appears that anticoagulation with direct oral anticoagulant or warfarin is effective towards HALT while antiplatelet therapy alone may not be^2^. An analysis from the national Society of Thoracic Surgeons Adult Cardiac Surgery database indicated that the use of anticoagulation following bioprosthetic mitral valve replacement varies significantly across centers and surgeons \[[@CR4]\], despite the class IIa recommendation according to the American Heart Association/American College of Cardiology guideline \[[@CR16], [@CR17]\]. In addition, the higher prevalence of atrial fibrillation in the population with mitral valve disease and subsequent need for anticoagulation may modulate the risk of HALT in bMV. This association is also currently unknown and warrants further investigation. Limitations {#Sec10} ----------- Aside from limitations associated with single-center retrospective design of this study, the small sample size precluded application of longitudinal models to account for the time-dependent nature of HALT risk. Therefore, we conducted cross-sectional analysis of prevalence. Conclusions {#Sec11} =========== In this first series of HALT in surgically-implanted bMV, the prevalence of HALT evaluated via CT was low at 5.7%. All 3 patients found to have bMV HALT were asymptomatic. One patient presented with HALT while on therapeutic oral anticoagulation. Further study is needed to characterize the clinical significance of this finding in the mitral position. Supplementary information ========================= {#Sec12} **Additional file 1: Figure S1.** CONSORT-style diagram. Figure shows starting cohort and subsequent exclusions and reasons for exclusion to reach the final cohort. **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= **Supplementary information** accompanies this paper at 10.1186/s13019-020-01120-3. None. Study design, manuscript draft, critical revision, analysis (SH MM, YG); critical revision, supervision (TS MY); data collection and critical revision (YF). The authors read and approved the final manuscript. None. Supporting data are not available due to the sensitive nature of the data. This study was approved by the Toyohashi Heart Center Institutional Review Board and individual consent was waived (THC-00048356). All authors reviewed and approved the final version of the manuscript and consented for publication. None.

1. Introduction {#sec1} =============== Electrophysiology in audiology is an objective tool to check the integrity of the auditory system. Auditory evoked potentials are electrophysiological tests, which give information about a number of events happening in the peripheral and central nervous systems that are generally related to the sensory pathway ([@bib53], [@bib9]). These sound-related evoked potentials are categorized as endogenous and exogenous potentials ([@bib21]). The exogenous potentials are mainly recorded by external event related dimensions of the stimulus. The endogenous potentials are responses which are due to internal events such as perception and cognition ([@bib47], [@bib40]). Studies have considered the possibility of studying auditory discrimination using a technique referred to as event-related potentials ([@bib6]; [@bib7]). Mismatch negativity (MMN) is an event-related potential that has been extensively studied by researchers to assess the pre-attentive auditory discrimination capability and storage of regularities in features of stimulus ([@bib41]). Pre-attentive processing is the unintentional gathering of information from the environment. First, all gathered information is pre-attentively processed. Then, our brain sieves and processes the prime information. The important information is selected for further analysis by attentive processing ([@bib1]). Our auditory system has an imperative role in gathering sound information for pre-attentive processing. At the point where auditory stimulus or sound waves hits the tympanic membrane, it transmits the message to the auditory cortex by means of auditory nerve for pre-attentive processing. The proficiency to appropriately filter information from pre-attentive auditory processing to attentive auditory processing is crucial for normal development of speech perception ([@bib52]). According to [@bib19], pre-attentive process uses the Gesalt laws of organization which says temporal proximity, physical similarity and good continuity is required to group the sound, which improves speech perception in quiet as well as in noise. For acoustic pre-attentive auditory processing, the temporal cortex is the primary site of activation, but research additionally demonstrated the association of frontal cortex as well ([@bib11], [@bib17]). Studies also suggest that perception of minute variation in complex musical patterns triggers the right ventromedial prefrontal cortex ([@bib11]). [@bib36] showed MMN as an endogenous potential with a negative component elicited by any discriminable change in regular auditory stimuli. MMN is usually obtained by presenting a train of repetitive homogenous tones at a rate of approximately one tone per second. It is occasionally interspersed with a tone that differs physically ([@bib10]). [@bib35] described MMN as "an electric brain response, a negative component of the event-related potential (ERP), elicited by any discriminable change (deviant) in some repetitive aspect of auditory stimulation (standard), usually peaking at around 100--200 ms from onset". MMN seems to depict a neuronal representation of the difference perceived between the auditory stimuli. Thus, MMN is well advised as an objective tool to check auditory discrimination skills at pre-attentive level. In which case, it could also be of clinical importance as speech perception, by its nature, depends on neuronal responses to changes in stimulus ([@bib20]). Music demands cognition, which requires specific and appropriate timing of many actions, such as perceiving the exact interval and control of pitch which are otherwise not involved in language. Enhanced auditory perception in musicians is likely to result from auditory perceptual learning during years of training, practice and experience. The musician\'s brain is presumed to be a good and appropriate model to investigate neuroplastic changes ([@bib33]). Professional musicians have fine-tuned auditory skills which are achieved by aural training that they receive during their musical training. It is considered as an important component of their vocational formation ([@bib13]). A study done by [@bib54] assessed pre-attentive auditory discrimination skills in amateur musicians and non-musicians. They reported significantly larger MMN in amateur musicians compared to non-musicians. Another study by [@bib3] reporting a strong advantage for musicians in accompanying behavioral task of detecting the deviants while attending to the stimuli for all pattern lengths showed that long-term musical training differentially affects the memory capacity. [@bib28] investigated MMN in non-musicians native speaker of a quality language, Finnish, in which duration is a phonemically contrastive cue, with French musicians compared to French non-musicians. They reported that pre-attentive and attentive duration processing of duration deviants was enhanced in Finn non-musicians and French musicians compared to French non-musicians. They also observed that MMN in French musicians was significantly larger compared to Finns and French non-musicians. Along a similar line, [@bib22] investigated neuronal representation of vowels and temporally manipulated CV syllables among string players and non-musicians with MMN odd ball paradigm. They showed that musicians are not only advantaged in the pre-attentive encoding of temporal cues but also in processing vowels. Previous literature from western countries have investigated MMN in western classical musicians and reported an enhanced pre-attentive auditory discrimination skill in musicians ([@bib54], [@bib3], [@bib28], [@bib46], [@bib22]). There is some basic mechanistic difference between Western and Indian classical music in terms of pitch structure and temporal patterning. Some basic elements of Indian music i.e. taala (rhythmic pattern), shruti (relative musical pitch), raaga (melody) and swara (the musical sound of a single note) are rarely found in western classical music. These features are difficult to perceive for western listeners without special training. In the case of vocal singers, control of pitch is important and is done by biomechanical and aerodynamic systems. Investigators agree that the ability to produce a precise pitch is very important for the professional vocal musician. Literature shows that accurate pitch control mainly depends on auditory perceptual monitoring, proprioceptive feedback of the laryngeal system and phonatory reflex systems ([@bib16], [@bib34]). The Recent literature reported enhanced auditory skills through different behavioral tests in Indian classical musicians ([@bib48], [@bib31], [@bib32], [@bib24], [@bib49], [@bib50]). It is interesting to know the effect of Indian classical vocal music training and practice on pre-attentive auditory discrimination skills in musicians through an electrophysiological test like MMN. There is a lack of literature regarding pre-attentive auditory discrimination skills in Indian classical vocal musicians. Hence, there is a need to compare pre-attentive auditory discrimination skills in Indian classical vocal musicians with non-musicians. 2. Materials and methods {#sec2} ======================== 2.1. Participants {#sec2.1} ----------------- Two groups of participants (the experimental and control group) were involved in the study. The experimental group consisted of 25 female right handed Indian classical vocal musicians with a mean age of 24.52 ± 2.6 years (age range 18--30 years). According to the inclusion criteria, only those with a minimum of 10 years of experience were taken. The participants of the experimental group in this study had an average experience of 12.3 years in Indian classical vocal music. All of them had started musical training after the age of 8 years. They practiced music for 19.2 ± 9.3 h per week regularly. As the control group, 25 female age-matched right-handed participants (age range = 18--30 years, mean = 24.8 ± 2.2 years) were included. None of them had any kind of formal training in music. The reason of taking only female participants was availability of more female participants in experimental as well as control group. 2.2. Participant selection criteria {#sec2.2} ----------------------------------- The participants selected for the study had hearing threshold within the normal limit as defined by AC and BC thresholds less that 15dBHL from 250 Hz to 8000 Hz and from 250 Hz to 4000 Hz respectively. They also had normal middle ear function as revealed by tympanometry and reflexometry. Otological problems were ruled out by otological evaluation with the help of a qualified otolaryngologist. Auditory Brainstem Responses for site of lesion were recorded to rule out any neurological problem in the subjects. An informed written consent was taken from all participants before involving them in the study. 2.3. Testing environment {#sec2.3} ------------------------ All behavioral as well as electrophysiological tests were carried out in a sound-treated room. The permissible noise levels were as per the guidelines in ANSI S3.1 (1999). Laboratory room was well lit and air-conditioned for the tranquility of the investigators as well as the subjects. 2.4. Instrumentation {#sec2.4} -------------------- For pure tone audiometry, a calibrated dual channel clinical diagnostic audiometer (Orbitor-922) was used for all participants. For tympanometry and reflexometry, a calibrated GSI-Tympstar Immittance meter was used for all participants. Mismatch Negativity was recorded on all participants using Intelligent Hearing System with smart EP. 2.5. Procedure {#sec2.5} -------------- The Modified version of Hughson and Westlake\'s procedure given by [@bib5] was used for pure-tone audiometry across octave frequencies from 250 Hz to 8000 Hz for air conduction. Octave frequencies from 500 to 4000 Hz were tested for bone conduction. To carry out tympanometry, a 226 Hz probe tone was used, whereas 500 Hz, 1000 Hz, 2000 Hz, and 4000 Hz stimuli were used for ipsilateral and contralateral reflex. Previous literature reported difficulty in identifying MMN at an individual level ([@bib26], [@bib30], [@bib45]). [@bib8] reported that MMN identification rate was too low (29%) to allow reliability to be evaluated. To obtain a clear and distinguishable MMN for statistical analysis, a larger difference between the frequent and infrequent stimulus was considered in the current study. The stimuli taken were 1000 Hz and 1100 Hz, where 1000 Hz served as the frequent stimulus and 1100 Hz served as the infrequent stimulus. The reason for taking larger difference between frequent and infrequent stimuli was to elicit a distinct waveform of MMN, as previous literature reported distinct MMN as the discrimination became easier between frequent and infrequent stimuli. They also reported as the discrimination became easier, MMN was earlier in latency and greater in amplitude ([@bib56], [@bib57]). In the present study the total duration of both stimuli was kept constant at 200 ms with 30 ms rise-fall time and a plateau of 140 ms. The Aux Viewer program was used for the preparation of stimulus. The wave file was then converted to stimulus file for AEPs using the software "Stimconv" provided by Intelligent Hearing System. Vertical montage with 'Fz' as the non-inverting electrode referenced to the nape of the neck was used to record MMN. The ground electrode was placed on the lower forehead. Eye blink responses were also recorded by another channel. Those sweeps with large eye blink artifact were not taken for averaging. The 1000 Hz and 1100 Hz pure tone stimuli were given in the odd ball paradigm in which the probability of frequent stimulus (1000 Hz) was 80% and that of infrequent stimulus was 20% at 70 dB nHL. The stimuli were presented at a repetition rate of 1.1/second in rarefaction polarity. To get MMN, the responses of −50 to 500 ms (with reference to stimulus onset) were averaged for 150 sweeps (20%) of infrequent stimulus and correspondingly 600 sweeps of frequent stimuli to maintain the 80%/20% frequent and infrequent stimulus ratio. The response was amplified to 50,000 times. The filter setting used was 0.1--30 Hz. Stimuli were presented binaurally. The participants were seated in a relaxed and comfortable position in order to avoid muscular artifacts and were made to watch a silent movie in order to promote passive listening. All the participants were asked not to pay attention to the auditory stimuli. Disc electrodes were placed on the cleaned skin surface of the targeted electrode sites. Absolute impedance was less than or equal to 5 kΩ and inter-electrode impedance was less than or equal to 2 kΩ while recording MMN. Apart from recording MMN in the conventional paradigm for each stimulus pair, LLRs (Long Latency Responses) were also recorded for the infrequent stimulus for 150 presentations, keeping the same recording parameters as for MMN. 2.6. Response analysis {#sec2.6} ---------------------- Conventional MMN recording was obtained in the odd ball paradigm which consisted of waveforms for the frequent and infrequent stimulus. This was followed by a second recording which was the conventional LLR for the infrequent stimulus at the rate of 1.1/second, averaged for 150 sweeps. The LLRs obtained for the infrequent stimulus were later used to analyze MMN by comparing it with the infrequent stimulus waveforms of the conventional odd ball paradigm. This paradigm was adopted to rule out any chances of error marking in MMN parameters due to the difference in LLRs elicited by the two stimuli of the odd ball paradigm and also to reduce the N1 affect ([@bib29]). MMN was located in the difference wave to obtain its onset, peak and offset latency. Similarly, peak amplitude and the area under the curve were also considered for measurement in MMN response for all participants. Onset latency was the time in millisecond at which the negativity started in the subtracted waveform. Offset latency was the time in milliseconds at which the negativity reached the baseline activity in the subtracted waveform. Peak latency was the time in millisecond at which negativity reached its peak in the subtracted waveform. Peak amplitude was the maximum amplitude of the peak of the negativity with respect to the baseline and area under the curve was the area under the negativity trough, derived from multiplying the peak amplitude with MMN duration. 2.7. Waveform analysis {#sec2.7} ---------------------- Visual detection was used for recognition of the MMN response. The criteria defined MMN as the first negative broad peak in the latency range of 100--300 ms, i.e. the N~1-~P~2~ or P~2-~N~2~ complex of LLRs. The first negativity should have the amplitude of more than −0.3 μV and a positive peak should follow the negative peak. If any extra negativity occurred in the P~1~ area, it was ignored by investigators. 2.8. Statistical analysis {#sec2.8} ------------------------- Descriptive statistics was done to find out mean and standard deviation (SD) for all the measures of MMN, i.e. onset latency, offset latency, peak latency, peak amplitude and area under the curve. To reduce the chance of type 1 error, MANOVA was used to compare between Indian classical vocal musicians and non-musicians for each measure of MMN. 3. Results {#sec3} ========== To inspect the data collected from Indian classical vocal musicians and non-musicians, descriptive statistics and MANOVA was done. Out of 25 musicians and 25 non-musicians, MMN was present only in 17 (68%) musicians and 16 (64%) non-musicians. Hence, further statistical analysis was done only for these subjects. The various measures of MMN i.e. onset latency, offset latency, peak latency, peak amplitude and area under the curve, were noted down from the MMN waveform through visual inspection for individual subjects. Sample waveform of MMN in musicians and non-musicians are represented in [Fig. 1](#fig1){ref-type="fig"}, [Fig. 2](#fig2){ref-type="fig"}, respectively. Descriptive statistics was done to find out mean and standard deviation (SD) for all the parameters of MMN (onset latency, offset latency, peak latency, peak amplitude and area under the curve) for the 17 Indian classical vocal musicians and 16 non-musicians ([Table 1](#tbl1){ref-type="table"}). Shapiro Wilk test was used to check the normal distribution of collected data from musicians and non-musicians. Based on the result of normality test, MANOVA was used to check any significant difference between musicians and non-musicians for each measure of MMN. From [Table 1](#tbl1){ref-type="table"}, the standard deviation for onset and peak latency was less (better) for musicians in comparison to non-musicians. The mean values of onset and peak latency for musicians were also less (better) in comparison to non-musicians. However, the mean value of offset latencies was similar between the two groups. [Fig. 3](#fig3){ref-type="fig"} shows an error bar graph for onset, offset and peak latency in musicians and non-musicians.Fig. 1A sample waveform of mismatch negativity in Indian classical vocal musicians along with the response measures.Fig. 2A sample waveform of mismatch negativity in non-musicians along with the response measures.Fig. 3Error bar graph of onset latency, offset latency and peak latency for Indian classical vocal musicians and non-musicians.Table 1Mean and standard deviation (SD) of onset latency, offset latency and peak latency for the Indian classical vocal musicians and non-musicians.ParametersOnset Latency (ms)Offset Latency (ms)Peak Latency (ms)MeanSDMeanSDMeanSDNon-musicians172.3735.66266.2538.48217.4339.62Musicians155.2323.26268.4334.36204.6423.98 MANOVA was carried out to compare differences between Indian classical vocal musicians and non-musicians for onset latency, offset latency and peak latency. Results of MANOVA showed marginally significant difference for onset latency \[F (1, 31) = 3.57; p = 0.06; ƞ^2^ = 0.103\], whereas no significant difference was observed for offset latency \[F (1, 31) = 0.00; p = 0.98; ƞ^2^ = 0.00\] and peak latency \[F (1, 31) = 1.01; p = 0.32; ƞ^2^ = 0.032\] between musicians and non-musicians in spite of higher mean observed for peak latency in musicians. Descriptive statistics was done to find out mean and standard deviation (SD) of the area under curve and peak amplitude for the Indian classical vocal musicians and non-musicians. From [Table 2](#tbl2){ref-type="table"}, the mean peak amplitude and area under the curve were higher (better) for musicians in comparison to non-musicians. However, standard deviation (SD) was less for non-musicians in comparison to Indian classical vocal musicians ([Table 2](#tbl2){ref-type="table"}). [Fig. 4](#fig4){ref-type="fig"}, [Fig. 5](#fig5){ref-type="fig"} show error bar graphs for peak amplitude and area under the curve in musicians and non-musicians respectively.Fig. 4Error bar graph of peak amplitude for the Indian classical vocal musicians and non-musicians.Fig. 5Error bar graph of the area under the curve for the Indian classical vocal musicians and non-musicians.Table 2Mean and standard deviation (SD) of peak amplitude and area under the curve for the Indian classical vocal musicians and non-musicians.ParametersPeak Amplitude(μv)Area under curve (μVμsec)GroupsMeanSDMeanSDNon-Musicians2.780.80131.7539.68Musicians4.061.32218.65110.36 MANOVA was carried out to compare peak amplitude and area under the curve between Indian classical vocal musicians and non-musicians. Results revealed statistically significant difference for peak amplitude \[F (1, 31) = 11.32; p = 0.00; ƞ^2^ = 0.267\] and area under curve \[F (1, 31) = 7.64; p = 0.00; ƞ^2^ = 0.198\] between Indian classical musicians and non-musicians. 4. Discussion {#sec4} ============= Out of 25 Indian classical vocal Musicians and 25 non-musicians, MMN was present only in 17 Indian classical vocal musicians and 16 non-musicians. In our study, absence of MMN in some of the subjects may be due to their inability to follow instruction of 'passive listening' during recording of MMN. So, data from those subjects on whom MMN was absent were excluded. MMN was studied by [@bib18] on professional violinists and non-musicians. The results showed that a distinct MMN was evoked in professional violinists but MMN was absent in non-musicians. Previous studies have also reported MMN to be robust at the group level, but identification of MMN can be difficult at an individual level ([@bib26], [@bib30], [@bib45]). [@bib8] also reported that MMN identification rate was too low (29%) to allow reliability to be evaluated. A study by [@bib51] also showed that MMN was present only in 66% of the normal hearing population. 4.1. Findings in onset, offset and peak latency of MMN {#sec4.1} ------------------------------------------------------ The present study showed a marginally significant difference in onset latency between Indian classical vocal musicians and non-musicians, and no significant difference in offset and peak latency between Indian classical musicians in comparison to non-musicians. The present outcomes are in consonance with previous literature ([@bib38]). In addition, there are studies done on different populations that have obtained similar outcomes ([@bib27], [@bib15], [@bib51]). However, there are a few studies that are not in agreement with the present findings ([@bib39], [@bib14]). [@bib38] compared MMN in formally trained instrumental musicians and age-matched non-musicians using harmonic tones. The result showed no significant difference in latency of MMN between instrumental musicians and non-musicians. Since there are not many researchers who have explored in the area of music, studies with MMN done in different populations are considered for the support of the present study. [@bib27] measured MMN in individuals with cochlear implants and reported that MMN latencies to frequency deviance did not show any changes over time. Similarly, [@bib15] compared MMN in children with and without stuttering. The results showed no significant difference in peak latency of MMN between the two groups. The finding of the current study is in contrast with the finding of a study done by [@bib39]. They assessed MMN in trained musicians and non-musicians. The results showed that musicians had shorter (better) MMN latencies to frequency changes in pure tones than non-musicians. In both groups, as the frequency difference between standard and deviant stimuli increased, MMN latency decreased (better). They also observed that mismatch negativity latencies for harmonic tone and speech syllable were significantly lesser (better) for musicians when compared to non-musicians. [@bib2] reported that latency measures were not reliable in MMN. They suggested that amplitude measures were more reliable than latency and most studies in this area have reported only amplitude measures. [@bib49], [@bib50] also showed no significant difference in latency measures across gross and fine differences between auditory stimuli. This result was attributed to poor reliability and high variability in latency measures of MMN. A similar study can be replicated on a large number of subjects to validate the findings. 4.2. Findings in peak amplitude and area under curve of MMN {#sec4.2} ----------------------------------------------------------- The results of the present study showed that peak amplitude and area under the curve were significantly higher (better) in Indian classical vocal musicians compared to non-musicians. This indicates enhanced pre-attentive auditory discrimination skills when it is measured in terms of peak amplitude and area under the curve. The present study\'s outcome is well supported by other researchers ([@bib54], [@bib39], [@bib3], [@bib28], [@bib22], [@bib12], [@bib46]). However, the findings of the current study are in contrast to a few studies ([@bib55]) as reported in the literature. [@bib54] recorded MMN with changes in acoustic features (gap, duration, frequency, location and intensity) and abstract features (interval size and melodic contour) as stimulus in non-musicians and amateur band musicians. The results showed that musicians had a larger MMN amplitude (better) and a greater area under the curve (better) as compared to non-musicians fora location change. Whereas, no statistically significant group differences were observed in response to other feature changes or in abstract-feature in mismatch negativity. This study shows that even amateur musicians have neural sound processing advantage when compared with non-musicians. [@bib28] investigated pre-attentive skills in musicians and non-musicians using MMN. The results revealed that mismatch negativity peak amplitude was significantly larger (better) in musicians compared to non-musicians for frequency deviants. A similar study was done by [@bib22] investigating MMN in musicians and non-musicians using vowels and temporally manipulated consonant-vowel syllables as stimuli. They found that musicians were not only advantaged in the pre- attentive encoding of temporal speech cues than non-musicians, but most notably also in processing vowels. [@bib12] recorded event-related brain potential responses in musicians and non-musicians to discrepancies of rhythm between pairs of unfamiliar melodies based on western classical rules. They noticed that musicians were able to detect rhythm deviations significantly better than non-musicians. [@bib46] recorded MMN for changes in melody, rhythm, musical key, timbre, tuning and timing in musically trained children. When compared to non-trained children, the musically trained children showed a significantly larger amplitude in MMN for all changes in stimuli. Therefore, it can be inferred that musical training helps in enhancing auditory discrimination for musically central sound dimensions in pre-adolescence. A similar study was done by [@bib39] using mismatch negativity on trained musicians. In this study, they reported that amplitude was significantly higher (better) for musicians with a pure tone as stimulus, but there was no significant difference seen in terms of the amplitude of MMN elicited by harmonic tones and speech syllables. They suggested that "musicians may have been slower to detect pure tones because they perceived this audible stimulus energy as irrelevant sensory stimuli. However, once the stimuli were detected, musicians automatically discriminated changes in pure tone frequency earlier than nonmusicians without an increase in response amplitude, suggesting more efficient acoustic processing". They also showed that the size of frequency deviance significantly affected the neural response, i.e. with increase in difference between frequent and infrequent stimuli, the MMN latency decreased and amplitude increased. The current study is in contrast to the study by [@bib55]. MMN was recorded in professional musicians in their study. They were presented with frequent standard sounds and rare deviant sounds at 0.8%, 2% and 4% higher in frequency. They reported no significant difference in peak amplitude between musicians and non-musicians when MMN was recorded in reading condition. They attributed these results to musical expertise that could have exerted its effects at merely attentive level of processing but not at the pre-attentive level. Similar to most previous studies ([@bib3], [@bib28], [@bib22], [@bib12], [@bib46]), the present study also shows that amplitude and area under the curve measure of the MMN have a significant effect from Indian classical musical training. 5. Clinical implication of the study {#sec5} ==================================== The present study shows that Indian classical vocal musical training have enhanced pre-attentive auditory discrimination skills in Indian classical musicians. Earlier studies have reported poor pre-attentive auditory processing in several clinical populations, i.e. central auditory processing disorders ([@bib37]), dyslexia ([@bib23]), Parkinson\'s disease ([@bib43]), Alzheimer\'s disease ([@bib43]), schizophrenia ([@bib44]), developmental language disorders ([@bib2]) and cochlear implant ([@bib25]). Indian classical musical training can be used to enhance pre-attentive auditory discrimination skills in these clinical populations. Earlier literature shows enhanced speech perception in quiet and noise in musicians compared to non-musicians ([@bib42]). Auditory scene analysis is defined as the internal process of segregating and subsequent grouping of auditory system for better speech perception ([@bib4]). Auditory scene analysis is based on the assumption that pre-attentive process uses the Gesalt laws of organization which says temporal proximity, physical similarity and good continuity is required to group the sound, which improves speech perception in quiet as well as in noise ([@bib19]). Enhanced pre-attentive skills in musicians is reported by many researchers ([@bib28], [@bib22], [@bib12], [@bib46]). Therefore, it can be hypothesized that musical training can be used for enhancement of pre-attentive auditory discrimination skills in these populations and may result in improvement in speech perception. 6. Conclusion {#sec6} ============= The current study shows enhanced peak amplitude and area under the curve of MMN in Indian classical vocal musicians compared to non-musicians. This indicates better pre-attentive auditory discrimination skills in Indian classical vocal musicians compared to non-musicians. It can also be stated that Indian classical vocal musical training has an effect on pre-attentive auditory discrimination skills in musicians, leading to higher peak amplitude and area under the curve. It can also be hypothesized that musical training (Indian classical) can be used to improve pre-attentive auditory discrimination skills in clinical populations including those with central auditory processing disorders, learning disability, Parkinson\'s disease, schizophrenia, Alzheimer\'s disease, children with cochlear implant and developmental language disorders. We want to acknowledge Director and HOD of Audiology of All India Institute of Speech and Hearing, Mysuru-6, Karnataka, India. We want to acknowledge Ms. Madhuri Sharma, Research Officer at AIISH for her valuable input during preparation of the manuscript. We also want to acknowledge participants of this study. Peer review under responsibility of PLA General Hospital Department of Otolaryngology Head and Neck Surgery. [^1]: Tel.: +91 9886833741.

In the wild, feral cats typically eat small mammals, reptiles, birds and insects. It is often not possible to mimic natural feeding behaviours of feral cats. Extruded diets have been the traditional alternative fed to domestic cats. Commercial chicken-based extruded diets (EXT) have complex diet formulations, including protein, fat, carbohydrate, fibre, vitamin and mineral ingredients. Association of American Feed Control Officials (AAFCO)^(^[@ref1]^)^ recommend minimum concentrations of 26 % crude protein (DM basis) and 9 % fat DM basis. If these minimum concentrations are targeted by formulators, commercial feline EXT may contain up to 50--55 % carbohydrates. Plantinga *et al.*^(^[@ref2]^)^ estimated the diet of feral cats expressed on a DM basis would contain 63 % crude protein (CP), 23 % fat and 2·8 % nitrogen-free extract (i.e. digestible carbohydrate). Given their carnivorous nature, it has been hypothesised that the protein:carbohydrate ratio of feline diets is important for feline health (i.e. obesity, feline diabetes and gut microbiota)^(^[@ref3]^--^[@ref6]^)^. As lower bacterial diversity and great shifts in commensal bacteria are often present in inflammatory bowel diseases, it suggests that a balanced gut microbiota is important for maintaining host health^(^[@ref7]^--^[@ref9]^)^. Several studies have examined the impact of dietary alterations on faecal microbial populations in cats^(^[@ref3]^--^[@ref5]^,^[@ref10]^--^[@ref13]^)^; however, very few have examined the microbial population of cats fed a 'wild-type' diet^(^[@ref12]^,^[@ref13]^)^. Commercially available whole prey may be more similar to the feral cat diet. For example, commercially available 1--3-d-old chicks (CHI) are approximately 72--76 % CP, 16--20 % fat and \<5 % nitrogen-free extract^(^[@ref14]^,^[@ref15]^)^. Previous studies have shown that extruded and whole-prey diets differ in digestibility as well as macronutrient composition^(^[@ref14]^--^[@ref17]^)^, and this may alter the fermentable substrates that are available to the gastrointestinal microbiota for fermentation^(^[@ref18]^,^[@ref19]^)^. The objective of the present study was to compare the faecal microbiota of cats fed an EXT chicken-based diet to those fed commercially available whole CHI. Experimental methods {#sec1} ==================== Study design {#sec1-1} ------------ The animal protocol was approved by the University of Illinois Animal Care and Use Committee. Faecal samples were collected from neutered male domestic cats (mean age = 5·7 years; body condition score 4·5--5·5 of 9). A completely randomised design was utilised to test the impacts of two dietary treatments ([Table 1](#tab01){ref-type="table"}): (1) EXT (*n* 3 cats; P & G Petcare); and (2) raw CHI (*n* 5 cats; Rodent Pro). The raw chicks were frozen (−20°C) upon arrival, and thawed in the refrigerator for 24 h prior to feeding. Fresh water was available *ad libitum*. A computer was used to randomly allot cats to treatment. Cats were adapted to diets for 10 d, prior to fresh faecal collection (\<15 min from defection). Faecal samples were stored at −80°C until DNA extraction. Table 1.Chemical composition of CHI and an EXT fed to domestic cats[\*](#tfn1_1){ref-type="table-fn"}ItemCHIEXT†DM (%)24·295·4Organic matter (% DM)91·192·8Crude protein (% DM)71·438·9Acid-hydrolysed fat (% DM)20·014·4Gross energy (kcal/g DM)5·95·2[^1][^2] Sample analysis {#sec1-2} --------------- Faecal bacterial DNA was isolated according to procedures described previously^(^[@ref20]^)^ using the MO BIO PowerSoil™ Kit (MO BIO Laboratories). Amplification of a 600 bp sequence of the V4--V6 variable regions of the 16S rRNA gene was done using barcoded primers as previously described^(^[@ref21]^)^. PCR amplicons were further purified utilising AMPure XP beads (Beckman-Coulter Inc.). Amplicons were combined in equimolar ratios to create a DNA pool that was used for pyrosequencing. DNA quality of amplicon pools was assessed before pyrosequencing using a 2100 Bioanalyzer (Agilent Technologies). Pyrosequencing was performed at the W. M. Keck Center for Biotechnology at the University of Illinois utilising a 454 Genome Sequencer and FLX titanium reagents (Roche Applied Science). Data analysis {#sec1-3} ------------- High-quality (quality value \>25) sequence data derived from the sequencing process was processed using a proprietary analysis pipeline ([www.mrdnalab.com](www.mrdnalab.com)) and as described previously^(^[@ref22]^)^. Statistical analysis {#sec1-4} -------------------- Sequence percentages at each taxonomic level were analysed using the Mixed models procedure of SAS (version 9.3; SAS Institute). The fixed effect of diet was tested. Means were separated for treatments using a Fisher-protected least significant difference with Tukey\'s adjustment. Results are reported as least-squares means with *P* ≤ 0·05 defined as significant and *P* ≤ 0·10 as trends for treatment effects. Results {#sec2} ======= Regardless of dietary treatment, Firmicutes (62--88 % of all sequences) was the predominant bacterial phylum in cat faeces (data not shown). Fusobacteria (0·2--17 % of all sequences), Proteobacteria (2--16 % of all sequences), Actinobaceria (1·4--18 % of all sequences), Tenericutes (1·4--9 % of all sequences) and Bacteroidetes (0--3 % of all sequences) also were predominant phyla present (data not shown). The proportion of Bacteroidetes was greater (*P* = 0·03) in faeces of cats fed EXT (1·6 % of all sequences) than those fed CHI (0·2 % of all sequences; data not shown). Proportions of genera, however, depended on dietary treatment ([Table 2](#tab02){ref-type="table"}). The predominant genera in faeces of cats fed CHI were *Clostridium* (11--25 % of sequences), *Blautia* (4--19 % of sequences), unidentified Lachnospiraceae (14--16 % of sequences), *Peptococcus* (7--13 % of sequences), *Fusobacterium* (4--13 % of sequences), *Ruminococcus* (2--9 % of sequences) and *Collinsella* (2--8 % of sequences). The predominant genera in faeces of cats fed EXT were *Megamonas* (2--28 % of sequences), *Megasphaera* (0·01--26 % of sequences), *Blautia* (10--16 % of sequences), *Collinsella* (1--16 % of sequences), *Lactobacillus* (0·2--14 % of sequences), *Clostridium* (8--12 % of sequences) and unidentified Lachnospiraceae (4--7 % of sequences). Table 2.Predominant bacterial genera (expressed as percentage of sequences) in faeces of domestic cats fed CHI (*n* 5) or EXT (*n* 3)[\*](#tfn2_1){ref-type="table-fn"}PhylumFamilyGenusCHIEXT[sem]{.smallcaps}*P* valueActinobacteriaBifidobacteriaceae*Bifidobacterium*ND†0·3----Bifidobacteriaceae*Collinsella*4·96·917·00·94Coriobacteriaceae*Slackia*0·1\<0·1\<0·10·06Unidentified genera0·10·50·10·08FirmicutesAcidaminococcaeae*Phascolarctobacterium*0·61·30·20·09Clostridiaceae*Clostridium*16·310·52·30·12Enterococcaeae*Enterococcus*1·00·30·30·13Erysipelotrichaceae*Allobaculum*0·80·10·20·06Unidentified genera0·7\<0·10·30·22Eubacteriaceae*Eubacterium*5·82·51·10·09Lachnospiraceae*Blautia*9·712·32·70·51*Coprococcus*2·30·61·00·30*Psuedobutyrivibrio*4·01·00·80·04*Roseburia*1·20·30·50·25Unidentified genera15·25·30·5\<0·01Lactobacilliaceae*Lactobacillus*ND7·6----Oscillospiraceae*Oscillibacter*0·3\<0·10·10·09Peptococcaeae*Peptococcus*9·23·21·0\<0·01Peptostreptococcaeae*Peptostreptococcus*0·7\<0·10·50·43Ruminococcaceae*Anaerotruncus*2·31·30·40·13*Faecalibacterium*0·21·00·20·02*Ruminococcus*4·32·21·20·28Unidentified genera2·41·10·40·06Veillonellaceae*Megamonas*\<0·111·94·10·09FusobacteriaFusobacteriaceae*Fusobacterium*8·75·915·10·83ProteobacteriaCampylobacteraceae*Campylobacter*2·40·62·20·59Enterobacteriaceae*Shigella*2·70·71·20·29Succinivibrionaceae*Anaerobiospirillum*0·81·20·60·64*Succinivibrio*\<0·11·20·2\<0·01[^3][^4] Four genera were present in a majority of samples (*n* 5) for cats fed CHI (*Holdemania* (three of five; 0--0·5 % of sequences), *Escherichia* (four of five; 0--0·2 % of sequences), *Marvinbyantia* (five of five; 0·01--0·1 % of sequences) and *Acetanaerobacterium* (three of five; 0--0·02 % of sequences)), but were not detected in the samples for cats fed EXT (data not shown). Five genera were present in a majority of samples (*n* 3) for cats fed EXT (*Megasphaera* (three of three; 0·01--26 % of sequences), *Lactobacillus* (three of three; 0·2--14 % of sequences), *Prevotella* (three of three; 0·1--2·6 % of sequences), *Subdoligranulum* (two of three; 0--1·4 % of sequences) and *Bifidobacterium* (two of three; 0--0·8 % of sequences), but were not detected in the samples for cats fed CHI (data not shown). Cats fed CHI had greater (*P* \< 0·05) *Psuedobutyrivibrio*, unidentified Lachnospiraceae and *Peptococcus* populations and tended to have greater (*P* \< 0·10) *Slackia*, *Allobaculum*, *Eubacterium*, *Oscillibacter* and unidentified Ruminococcaceae populations. In contrast, cats fed CHI had lower (*P* \< 0·05) *Faecalibacterium* and *Succinivibrio* populations and tended to have lower (*P* \< 0·10) *Phascolarctobacterium*, *Megamonas* and unidentified Coriobacteriaceae populations. Discussion {#sec3} ========== We identified a significant shift in the faecal bacteria of cats fed CHI *v.* EXT. As these diets contained ingredient and nutrient differences, differences in proportions of bacterial populations can only be attributed to the treatments as a whole. To our knowledge, most of the studies investigating the effects of diet on bacterial composition utilising next-generation sequencing in cats have examined the effects of commercial dry^(^[@ref5]^,^[@ref10]^,^[@ref11]^)^ and canned diets^(^[@ref3]^,^[@ref4]^)^. Only preliminary data for the differences in bacterial composition between a raw meat and kibbled diets fed to dogs have been reported^(^[@ref23]^,^[@ref24]^)^. No data have been reported for a whole-prey diet type in either cats or dogs; however, in a companion paper, we also present the microbial populations of cats fed whole and ground chicks and the effects of clinically confirmed symptomatic salmonellosis^(^[@ref13]^)^. Bermingham *et al.*^(^[@ref4]^)^ reported increased faecal proportions of *Lactobacillus* (32 *v.* 0·1 % of sequences) and *Megasphaera* (23 *v.* \<0·1 % of sequences) in the faecal microbiota of cats fed a commercial dry diet (i.e. lower protein, higher nitrogen-free extract; CP = 33 %, DM basis; fat = 11 %, DM basis) compared with those maintained on a commercial wet diet (i.e. higher protein, lower nitrogen-free extract; CP = 42 %, DM basis; fat = 42 %, DM basis). Hooda *et al.*^(^[@ref5]^)^ reported increased faecal proportions of *Megasphaera* (18--33 *v.* \<0·1--0·1 % of sequences), *Subdoligranulum* (2--6 *v.* 0·1--0·3 % of sequences) and *Bifidobacterium* (12--21 *v.* \< 0·1--0·1 % of sequences) in kittens fed a moderate protein-moderate carbohydrate diet (CP = 34 %, DM basis; fat = 19 %, DM basis) compared with those fed a high-protein, low-carbohydrate diet (CP = 53 %, DM basis; fat = 24 %, DM basis). Beloshapka *et al.*^(^[@ref23]^)^ reported increased faecal proportions of *Faecalibacterium* (10 *v.* 0·3 % of sequences), *Lactobacillus* (9 *v.* \< 0·1 % of sequences) and *Prevotella* (9 *v.* 0·2 % sequences) for dogs fed an EXT compared with raw-meat-based diets. Although these studies also reported other differences not observed herein, those listed here are similar to our data, and indicated that the protein:carbohydrate ratio may impact these genera. However, both Bermingham *et al.*^(^[@ref4]^)^ and Hooda *et al.*^(^[@ref5]^)^ reported decreased faecal proportions of *Faecalibacterium* (\<0·1 *v.* 0·5 % of sequence, and 0·1--2 *v.* 5--7 % of sequences, respectively) for the lower-protein *v.* higher-protein diet, which are contrary to our results and those reported by Beloshapka *et al.*^(^[@ref23]^)^. Another aspect of diet that can impact microbial populations is dietary fibre. Although it has been recognised that animal tissues provide substrate for fermentation (i.e. animal fibre^(^[@ref18]^,^[@ref19]^)^), their role in gut health has not been fully elucidated, and little is known about their impacts on microbial populations. The EXT diet tested herein included multiple ingredients that would contribute to the dietary fibre fraction, including beet pulp and fructooligosaccharides, which likely contributed to the differences in microbial populations. *Lactobacillus*, *Bifidobacterium* and *Faecalibacterium* species are generally considered beneficial bacteria and are often targeted with dietary fibre and prebiotic inclusions. Middelbos *et al.*^(^[@ref25]^)^ reported increased faecal proportions of *Faecalibacterium* (30 *v.* 9 % of sequences) in dogs fed a diet containing 7·5 % beet pulp fibre (total dietary fibre = 4·5 %, DM basis) compared with 0 % supplemental fibre (total dietary fibre = 1·4 %, DM basis). Fructooligosaccharides are rapidly fermented and serve as a source of soluble, prebiotic fibre^(^[@ref26]^,^[@ref27]^)^. Several studies in cats and dogs have reported that fructooligosaccharides exert a prebiotic effect in the colon, stimulating the growth of *Bifidobacterium* spp., *Lactobacillus* spp. or both^(^[@ref10]^,^[@ref28]^)^. These studies are consistent with the results reported herein. The present study had limitations, including our sampling protocol, number of animals and the nature of the diets tested. First, baseline samples were not collected before dietary treatments were administered. Thus, while differences due to diet were identified, we were unable to identify microbiome shifts from baseline. Second, given the low number of animals studied, our statistical power was low and our ability to translate the data to larger cat populations was limited. Finally, because the diets were greatly different in terms of nutrient composition and physical form, microbiome differences could not be attributed to any single factor or nutrient, but only the entire diet as a whole. To conclude, there is growing evidence that the proportions of gastrointestinal microbes are altered in some disease states, including cancer, gastrointestinal diseases and metabolic diseases^(^[@ref29]^--^[@ref31]^)^. Given the potential of diet to modulate microbial populations, diet therapies may play a role in their treatment. However, it is unclear if this dysbiosis is causative or symptomatic of these disease states. Additionally, there is little data available regarding microbial populations, dysbiosis and disease states for cats^(^[@ref8]^,^[@ref13]^)^. The present study has highlighted some interesting differences in gastrointestinal microbes of cats eating extruded *v.* raw diets. More research is needed, however, to determine the long-term impacts of the alterations in the proportions of faecal microbial populations and the health of domestic cats. The authors have no conflicts of interest to declare. This research received no specific grant from any funding agency in the public, commercial or not-for-profit sectors. In regards to authorship, K. R. K. and K. S. S. contributed to formulation of research, study design and study execution. K. R. K. and S. E. D. contributed to data analysis. K. R. K. and K. S. S. contributed to manuscript writing. As stated in the Experimental Methods section, the animal protocol for this experiment was approved by the University of Illinois Animal Care and Use Committee. This paper was published as part of the WALTHAM International Nutritional Sciences Symposium Proceedings 2013, publication of which was supported by an unrestricted educational grant from Mars Incorporated. The papers included in these proceedings were invited by the Guest Editor and have undergone the standard journal formal review process. They may be cited. [^1]: CHI, 1--3-d-old chicks (Rodent Pro); EXT, extruded chicken-based diet (P & G Petcare). [^2]: †Ingredient composition of EXT as reported by manufacturer: chicken, chicken by-product meal, maize meal, maize grits, dried beet pulp, poultry by-product meal, natural flavour, dried egg product, brewers dried yeast, sodium bisulphate, potassium chloride, fructooligosaccharides, animal fat (preserved with mixed tocopherols, a source of vitamin E), fish oil (preserved with mixed tocopherols, a source of vitamin E), DL-methionine, choline chloride, calcium carbonate, vitamins (vitamin E supplement, niacin, ascorbic acid, vitamin A acetate, calcium pantothenate, biotin, thiamine mononitrate (source of vitamin B~1~), pyridoxine hydrochloride (source of vitamin B~6~), vitamin B~12~ supplement, riboflavin supplement (source of vitamin B~2~), inositol, vitamin D~3~ supplement and folic acid), taurine, minerals (zinc oxide, manganese sulphate, copper sulphate, potassium iodide and cobalt carbonate) and rosemary extract. [^3]: CHI, 1--3-d-old chicks (My Pet Carnivore); EXT, extruded chicken-based diet (P&G Petcare). [^4]: †ND, not detected.

INTRODUCTION {#s1} ============ Lung cancer is the leading cause of cancer-related deaths worldwide, with non-small cell lung cancer (NSCLC) accounting for approximately 85% of all cases \[[@R1]\]. Most NSCLC patients are diagnosed at an advanced stage and have a 5-year survival rate of less than 20% \[[@R1], [@R2]\] because of their advanced stage diagnoses \[[@R3]\]. Lack of early diagnosis markers and high potential for the invasion ability of NSCLC are challenging for NSCLC therapy. Hence, the molecular mechanisms involved in the development and progression of NSCLC must be investigated. Long noncoding RNAs (lncRNAs), defined as a class of noncoding RNA with a length of more than 200 nucleotides, have critical roles in the gene expression regulation \[[@R4]\], epigenetic control \[[@R5]\], chromatin structure \[[@R6], [@R7]\], development process, genomic imprinting, and pluripotency of embryonic stem cells \[[@R6], [@R8], [@R9]\]. In addition, dysregulation of lncRNAs has been reported to play a vital role in the carcinogenesis, disease progression, and metastasis of human cancers \[[@R6], [@R7], [@R10]--[@R12]\]. Some lncRNAs such as H19, HOTAIR, ANRIL, MALAT1, and SCAL1 \[[@R13]--[@R15]\] have been reported to be associated with the development and progression of lung cancers. However, the roles of lncRNAs in NSCLC development and metastasis remain largely unknown. Hence, the identification of lung cancer-associated lncRNAs and the investigation of their molecular and biological functions in lung cancers are vital. Myocardial infarction-associated transcript (MIAT) is one of the noncoding RNAs first identified as an lncRNA in 2006 \[[@R16]\]. MIAT is involved in various cellular processes, including myocardial infarction \[[@R16], [@R17]\], microvascular dysfunction \[[@R18]\], paranoid schizophrenia \[[@R19]\], nuclear body formation \[[@R20]\], and neurogenic commitment \[[@R21]\]. Because MIAT physically interacts with SF1 splicing factor, it is supposed to be involved in RNA splicing and regulating gene expression \[[@R22]\]. Recent studies have demonstrated that MIAT constitutes a loop with Oct4 in malignant mature B cells and is essential for cell survival \[[@R23]\]. MIAT is also upregulated and interacts with the polycomb in neuroendocrine prostate cancer to participate in tumorigenesis \[[@R24]\]. However, the expression pattern, biological function, and underlying mechanism of MIAT in NSCLC are still unclear. In the present study, we investigated the potential mechanisms of MIAT in NSCLC progression. We observed that MIAT was upregulated and played a role in the advanced pathological stage. Moreover, our data revealed that MIAT could interact with MLL and epigenetically activate MMP9 to facilitate cell proliferation, migration, and invasion in NSCLC. RESULTS {#s2} ======= MIAT expression was upregulated and correlated with advanced tumor stage {#s2_1} ------------------------------------------------------------------------ To explore whether MIAT played a role in carcinogenesis, we first profiled the expression of MIAT in 60 pairs of NSCLC tissues (30 paired of adenocarcinoma and 30 paired of squamous) and paired adjacent non-tumor tissues. The qPCR data indicated that the expression level of MIAT in tumor tissues was significantly higher than that in the corresponding non-tumor tissues (mean dCT of tumor vs. normal tissue: 2.95 vs. 3.71, *p* = 0.0014; Figure [1A](#F1){ref-type="fig"}). Furthermore, we analysed the association between *MIAT* gene expression and the clinical stage of NSCLC and the state of metastasis. *MIAT* upregulation in the tumor tissues was associated with an advanced stage (stages III, IV, *n* = 24, *p* = 0.001) but not early stage cancer (stages I and II, *n* = 36, *p* = 0.09; Figure [1B](#F1){ref-type="fig"}). Next we tested the MIAT expression in NSCLC cell lines, including A549, H1299, H460, and H520. Among these cell lines, MIAT was relative higher expressed in A549 and H1299 (Figure [1C](#F1){ref-type="fig"}); thus, we chose A549 and H1299 cells to perform the following experiments. Moreover, to investigate the clinical significances of MIAT, we evaluated the correlation between MIAT level and clinicopathological factors. Results revealed that MIAT levels were correlated with tumor size (*p* = 0.0035), TMN stage (*p* = 0.001), and lymph node metastasis (*p* = 0.0185) in NSCLC. Nevertheless, MIAT levels were not associated with age (*p* = 1.000) or gender (*p* = 0.0581) (Table [1](#T1){ref-type="table"}). These results indicated that upregulated expression of MIAT might play a role in NSCLC tumorigenesis. {#F1} ###### MIAT expression and clinicopathological factors in NSCLC patients (*n* = 60) Parameter *N* Relative MIAT expression *p*-value -------------------------------- ----- -------------------------- ----------- -------- Age (year) 1.000  ≤ 65 34 20 14  \> 65 26 16 10 Gender  Male 37 26 11 0.0581  Female 23 10 13 Tumor size (maximum diameters)  ≤ 3 cm 32 25 7 0.0035  \> 3 cm 28 11 17 Lymph node metastasis  N1 26 11 15 0.0185  N0 34 25 9 TMN stage  I--II 36 28 8 0.0010  III--IV 24 8 16 *P* values when expression levels were compared using Fisher\'s exact test. Knockdown of MIAT impaired lung cancer cells proliferation and cell cycles arrest *in vitro* {#s2_2} -------------------------------------------------------------------------------------------- Because the overexpression of MIAT was significantly associated with progression in NSCLC patients, we further modulated MIAT expression to examine whether MIAT regulated the proliferation of A549 and H1299 cells. A cell counting assay revealed that cell growth rate of A549 and H1299 were dose-dependently inhibited with siMIAT compared with the control (Figure [2A](#F2){ref-type="fig"}). Colony formation assay data also revealed that clonogenic survival were inhibited in si-MIAT-treated A549 and H1299 cells (Figure [2B](#F2){ref-type="fig"}). To further examine whether the effect of MIAT on proliferation reflected cell cycle arrest, cell cycle progression was analysed using flow cytometry analysis. The results indicate that MIAT knockdown retarded the G1/S transition in si-MIAT A549 and H1299 cells (Figure [2C](#F2){ref-type="fig"}). We then performed Western blot and found that knockdown of MIAT would decrease the expressions of cyclin D3 and cdk2 in A549 and H1299 cells (Figure [2D](#F2){ref-type="fig"}). These data indicated that MIAT could promote the proliferation phenotype of NSCLC cells. {#F2} MIAT silencing impaired cell migration and invasion *in vitro* {#s2_3} -------------------------------------------------------------- Next, we explored the efficiency of MIAT on migration and invasion in A549 and H1299 cells. The wound healing scratch assay revealed that the ratio of the recovered region were decreased in MIAT knockdown A549 and H1299 cells compared with the control (Figure [3A](#F3){ref-type="fig"}). Furthermore, a matrigel transwell assay demonstrated that decreasing MIAT expression could dose-dependently impede the invasion abilities of A549 and H1299 cells (Figure [3B](#F3){ref-type="fig"}) compared with control. These results indicate that knockdown of MIAT expression retarded cell migration and invasion motility in NSCLC cells. {#F3} MIAT silencing impaired A549 cell migration and invasion *in vivo* {#s2_4} ------------------------------------------------------------------ To validate the oncogenic efficiency of MIAT *in vivo*, A549 cells stably transfected with shMIAT or scramble were subcutaneously inoculated into the left flank (stable shMIAT cells) and right flank (scramble cells) of BALB/c athymic male nude mouse, respectively (*n* = 6). After 24 days, the tumors formed in the shMIAT group were substantially smaller than those in the scramble group (Figure [4A](#F4){ref-type="fig"} and [4B](#F4){ref-type="fig"}). Moreover, the tumor weight at the end of the experiment was lower in the shMIAT group (0.466 ± 0.021 g) compared with that in the scramble group (0.333 ± 0.051 g) (Figure [4C](#F4){ref-type="fig"}). QPCR analysis confirmed that the MIAT levels were lower in shMIAT tumor tissues than in scramble tumor tissues (Figure [4D](#F4){ref-type="fig"}). These findings indicate that *MIAT* knockdown would decelerate tumor growth *in vivo* by repressing cell proliferation and migration. {#F4} MIAT knockdown suppressed MMPs expression {#s2_5} ----------------------------------------- Because the MIAT expression level is correlated with advanced stage and affects cell invasion, we further explored whether MIAT regulated epithelial--mesenchymal transition (EMT) or matrix metallopeptidases (MMPs) expression. The qRT-PCR results revealed that MIAT knockdown has no effect on EMT inducers such as TWIST1, SANI1, and ZEB1; however, MIAT knockdown significantly reduced MMP9 expression (*p* = 0.02) and has a trend of decreasing MMP2 (*p* = 0.07; Figure [5A](#F5){ref-type="fig"}) in A549 and H1299 cells. We then detected the activities of MMP2 and MMP9 through gelatin zymography. The data revealed that MIAT knockdown could reduce MMP9 expression but has no effects on MMP2 in A549 and H1299 cells (Figure [5B](#F5){ref-type="fig"}). We also observed that MMP9 but not MMP2 was overexpressed in tumor parts (mean dCT of tumor vs. normal tissue: 1.30 vs. 3.43, *p* \< 0.0001; Figure [5C](#F5){ref-type="fig"}), and that increased MMP9 in early and advanced stages in 60 paired NSCLC tissues was correlated with MIAT (Figure [5D](#F5){ref-type="fig"}). These results indicated that MMP9 might be a downstream gene that was regulated by MIAT to affect NSCLC migration and invasion. {#F5} MIAT was associated with MLL and epigenetically regulates MMP9 activity {#s2_6} ----------------------------------------------------------------------- To investigate the potential mechanism of MIAT in regulating MMP9 in NSCLC cells, we first analysed the distribution of MIAT in cells. The cell fractional data revealed that MIAT was distributed in both the cytoplasm and nucleus; however, the ratio of MIAT in the nucleus was higher than that in the cytoplasm (Figure [6A](#F6){ref-type="fig"}). Previous studies have reported that lncRNAs could function in cooperation with chromatin-modifying enzymes to promote epigenetic activation or silencing of gene expression. Therefore, we performed the RIP assay to examine whether a physical interaction is present between histone modifiers and MIAT. The results revealed that MIAT could interact with H3K4 methyltransferase MLLs (activator complex) but not EZH2 (enhancer of zeste homolog 2 in polycomb group repressive complex) in A549 and H1299 cells (Figure [6B](#F6){ref-type="fig"}). We also performed an immunostaining combined with RNA-FISH experiment and observed that MIAT could colocalize with MLL in the nucleus (Figure [6C](#F6){ref-type="fig"}). To elucidate whether MLL could regulate MMP9 activity, we knockdown MLL expression in A549 and H1299 cells. inthen transfected 50 nM of siMLL into A549 and H1299 to elucidate whether MLL could affect the expression levels of MMP9. The data showed that the expressions of MLL and MMP9 were been silenced (Figure [6D](#F6){ref-type="fig"}). Therefore, to investigate whether MLL is involved in the regulation of MIAT on MMP9, we first designed three sets of primers of the MMP9 promoter region and found that MLL was bound to the sequences within 1634 bp upstream of the transcript start site (Figure [6E](#F6){ref-type="fig"}). To further address whether MIAT regulated MMP9 through MLL enrichment, we performed ChIP in the A549 and H1299 cell lines with MIAT knockdown. The results revealed that MIAT silencing reduced the binding activity of MLL and the status of H3K4me3 with the MMP9 promoters in A549 and H1299 cells (Figure [6F](#F6){ref-type="fig"} and [6G](#F6){ref-type="fig"}). These results indicated that MIAT interacted with MLL in the nucleus and was essential for MLL to regulate MMP9 activity through epigenetic regulation. {#F6} DISCUSSION {#s3} ========== Many studies have reported that lncRNA dysregulation is associated with pathological and physiological processes in different human diseases. The expression levels of lncRNA are also associated with cancer development and progression, including NSCLC \[[@R25]--[@R27]\]. For example, lncRNA BANCR functions as a tumor suppressor in NSCLC \[[@R28]\], whereas HOTAIR and MALAT1 promote oncogenic functions in NSCLC \[[@R29], [@R30]\]. However, the roles of lncRNAs in NSCLC tumorigenesis are still unknown. MIAT was first identified as a candidate gene for myocardial infarction \[[@R16]\]; it is abundantly expressed in the nervous system \[[@R31]\] and retinal tissue \[[@R32]\]. Recent studies have reported that MIAT participates in chronic lymphocytic leukaemias progression \[[@R23]\] and prostate cancer formation \[[@R24]\]; however, the underlying mechanism of MIAT in tumorigenesis remains unclear. We speculated that MIAT might be involved in NSCLC progression. In the NSCLC cohort, we observed that MIAT was upregulated in lung cancer tissues compared with the non-tumor tissues. In addition, MIAT expression level was significantly different in the advanced tumor stage. Because treatment failure and the poor prognosis of lung cancer are due to high metastasis and invasion, we further identified the biological role of MIAT in tumor progression. A loss-of-function assay demonstrated that MIAT silencing impaired cell proliferation, migration, and invasion *in vitro* and inhibited tumor formation *in vivo*. To further elucidate the molecular mechanism through which MIAT contributes to invasion and metastasis in NSCLC, we investigated potential target proteins involved in cell motility and matrix invasion. During tumor metastasis, cancer cells lose their polarity and intercellular adhesions and then get invasive characteristics of mesenchymal cells through EMTs \[[@R33]\]. MMPs could degrade the proteins of the extracellular matrix and basement membrane to promote tumor metastasis \[[@R34]\]. Therefore, we determined whether any interaction was present between EMTs, MMPs, and MIAT. MIAT silencing had no effects on EMT transcriptional factors (TWIST1, SNAI1, and ZEB1), but repressed MMP2 and MMP9 expressions in A549 cells. We further determined the expression levels of MMP2 and MMP9 in our NSCLC cohort; the results revealed that the expression level of MMP9 but not MMP2 was correlated with MIAT to express highly in tumor tissue and advanced stage (Figure [1A, 1B](#F1){ref-type="fig"} and Figure [5C, 5D](#F5){ref-type="fig"}). MMP9 is a member of the MMP family. MMP9 is overexpressed in colorectal and lung tumor and is related to metastasis and invasion in the breast, upper urinary tract, and oral squamous cell tumor \[[@R35]--[@R38]\]. Our findings demonstrated that MIAT-mediated NSCLC cell migration, invasion, and metastasis suppression through MMP9. Although several recent studies have indicated that MMP9 expression could also be regulated by some lncRNAs, such as BCRYN1 and MVIH \[[@R39], [@R40]\], the regulation mechanism remains unclear. Generally, lncRNAs regulate cancer cell phenotypes by interacting with specific RNA binding proteins and resulting in gene activation or repression through chromosome reprogramming, DNA methylation, RNA decay, and histone protein modification \[[@R41], [@R42]\]. Our RNA-IP experiment demonstrated that MIAT could interact with the histone methyltransferase MLL. MLL is a histone methyltransferase that mediates histone H3 lysine 4 and can catalyze H3K4 monomethylation, dimethylation, and trimethylation \[[@R43]\]. Abnormal methylation of H3K4 frequently occurs in cancer progression. A recent study also revealed that decreased H3K4me3 modification at the MMP9 promoter reduces MMP9 expression and inhibits tumor cell proliferation \[[@R44]\]. Moreover, lncRNAs regulated gene expression by mediating histone modifiers to the promoter of target genes \[[@R45]--[@R47]\]. Therefore, we further validated MIAT and MLL on the MMP9 promoter and found that MLL could bind on the MMP9 promoter and that MIAT silencing reduced the binding of MLL and H3K4me3 level with the MMP9 promoter. The hypothesis of the mechanism of MIAT on NSCLC is presented in Figure [7](#F7){ref-type="fig"}. These results indicated that MIAT might play crucial roles in the MLL-mediated activation of oncogenes in NSCLC. {#F7} In conclusion, our results clarified that MIAT was upregulated in NSCLC tumor tissues and was correlated with tumor advanced stage. MIAT knockdown inhibited cell proliferation, migration, and invasion in A549 and H1299 cells and inhibited tumorigenesis *in vivo.* Furthermore, MIAT-mediated oncogenic effects are partially due to the epigenetic silencing of MMP9 through the direct binding of MIAT with MLL. These findings indicate a role of MIAT-dependent histone H3K4 methylation in MMP9 transactivation and lung carcinogenesis and reinforce the notion that targeting the lung cancer epigenome may yield novel therapeutic solutions. MATERIALS AND METHODS {#s4} ===================== Patients and tissue samples {#s4_1} --------------------------- Sixty paired NSCLC tissues (30 paired of adenocarcinoma and 30 paired of squamous) and adjacent non-tumor tissues from patients (37 males, 23 females, mean age 63.0, SD ± 12.6) who received surgical resection of NSCLC between 2006 and 2014 were obtained from the Bio-Bank of China Medical University Hospital (CMUH) after approval from CMUH\'s Institutional Research Ethics Committee (CMUH103-REC2-140), according to the Declaration of Helsinki guideline. None of the patients had received chemotherapy or radiotherapy prior to surgery. All surgical specimens were snap-frozen and stored in liquid nitrogen immediately after resection until total RNA extraction. All tumor and paired non-tumor tissues were confirmed by experienced pathologists, and the pathological stage, grade, and nodal status of the tissues were provided. Clinical and pathological characteristics were also collected for each patient. Informed written consent was obtained from all patients in this study. Cell culture and transfection {#s4_2} ----------------------------- A549 and H1299 cell lines were respectively cultured in DMEM and RPMI supplemented with 10% fetal bovine serum (10% FBS, Gibco), 100 U/ml penicillin, and 100 mg/ml streptomycin in humidified air at 37°C with 5% CO~2~. A549 and H1299 cells were transfected with various dosages of siRNA or negative control by using RNAimax Lipofectamine (Invitrogen, USA) according to the manufacturer\'s instructions. The siRNA oligonucleotides were synthesized by MDBio, Inc. The siRNA sequences of MIAT were as follows: 5′-ACUUCUUCGUAUGUUCGGCTT-3′ and a negative control (a scrambled matched %GC oligonucleotide). The siRNA sequences of MLL were as follows: 5′-GCUCUUUCCUAUUGGAUAUTT-3′ and a negative control (a scrambled matched %GC oligonucleotide). Subcellular fractionation, total RNA extraction, and qRT-PCR analysis {#s4_3} --------------------------------------------------------------------- The separation of the nuclear and cytosolic fractions of the A549 and H1299 cell lines was performed according to the protocol of the PARIS Kit (Life Technologies, Carlsbad, CA, USA) Total RNA was extracted from tissues or cultured cells with TRIzol reagent (Life Technologies, Scotland, UK, USA) according to the supplier\'s instructions. Two micrograms of RNA was reverse-transcribed into cDNA by using a high-capacity cDNA reverse transcriptase kit (Thermo Fisher Scientific-Applied Biosystems, Waltham, MA, USA). Quantitative PCR was performed using the Taqman assay and GAPDH mRNA was employed as an endogenous control for mRNA. Relative expression levels of the target genes were calculated as ratios normalized against GAPDH. The quantification of gene expression was performed by using the 2−ΔΔCt method. All primers were designed and synthesized by Genomics BioSci & Tech, Taipei, Taiwan. The following primer sequences were used: *MMP2*-forward: ccccaaaacggacaaagag, reverse: cttcagcacaaacaggttgc; *MMP9*-forward: cgcagacatcgtcatccagt, reverse: cgcagacatcgtcatccagt; *TWIST1*-forward: ggcatcactatggactttctctatt, reverse: ggccagtttgatcccagtatt; *SANI1-*forward: aggatctccaggctcgaaag, reverse: tcggatgtgcatcttgagg; and *ZEB1-*forward: aactgctgggaggatgacac, reverse: tcctgcttcatctgcctga. Cell proliferation assay {#s4_4} ------------------------ Cell proliferation was assessed using a trypan blue exclusion assay. Cells were seeded in 60-mm culture dishes at a density of 1 × 10^5^ cells/dish and incubated for 24 h or 48 h. After incubation, cell number was determined using a trypan blue exclusion test with trypan blue (0.4%) purchased from Sigma Chemical Co. Colony formation assay {#s4_5} ---------------------- A total of 10,000 control and siMIAT cells were placed in a 6-well plate and maintained in media containing 10% FBS; the medium was replaced every 3 days. After 14 days, cells were fixed with methanol and stained with 0.1% crystal violet. Viable colonies were manually counted. For each treatment group, wells were measured in triplicate. Cell cycle analysis {#s4_6} ------------------- Transfected A549 and H1299 cells were collected and fixed in 70% ethanol at −20°C overnight. Fixed cells were washed once with PBS and then labelled with propidium iodide (Sigma-Aldrich) in the presence of RNase A (Sigma-Aldrich) and Triton X-100 for 30 min in the dark. Cells were run on a FACSCanto flow cytometer (Becton-Dickinson, FL, NJ, USA). The percentages of the cells within each phase of the cell cycle were analysed using the ModFit LT program. Western blot {#s4_7} ------------ Whole cell extracts were prepared from A549 and H1299 cells by adding RIPA lysis buffer (150 mM NaCl, 0.1% SDS, 0.5% sodium deoxycholate, 1% NP-40) (Sigma) with complete protease inhibitor cocktails (Sigma). Equal quantities of total protein samples were separated on 10% SDS-PAGE gels and transferred to PVDF membranes. Blots were incubated with primary antibody against Cyclin D3, cdk2 (Abcam) and β-actin (GeneTex) overnight at 4°C. After secondary antibody incubation, the electrochemiluminescence (ECL) kit (EMD Millipore, St. Charles, MO) was used to visualize protein signals. β-actin was used as internal control. Wound healing scratch and transwell assays *in vitro* {#s4_8} ----------------------------------------------------- A wound healing assay was used to assess the ability of cell migration, and appropriate A549 and H1299 cells were seeded into 24-well plates at a density that reached 95%--100% confluence as a monolayer. The monolayer was gently scratched across the centre of the well with a 200-μl plastic tip. The rate of closure was assessed through imaging with an inverted microscope (DMi1; Leica, Wetzlar, Germany). The migration movement throughout the wound area was examined and calculated using the free software 'TScratch'. For the invasion assay, 24-well transwell chambers with 8 mM pore size polycarbonate membranes were used. Approximately 1 × 10^5^ control or siMIAT cells were seeded into the upper chamber of the insert. After culturing the cells in the upper chamber for 24 h, they were carefully removed, and cells adhering to the underside of the membrane were stained with 0.1% crystal violet solution. The numbers of cells were counted under an inverted microscope (DMi1; Leica, Wetzlar, Germany). For each experimental group, the assay was performed in triplicate. *In vivo* tumor formation {#s4_9} ------------------------- BALB/c athymic nude mice (male, 4--6 weeks old) were purchased from National Laboratory Animal Breeding and Research Center, Taiwan. To establish a lung cancer xenograft model, 1 × 10^7^ scramble or shMIAT-A549 cells were suspended in 100 ml PBS and inoculated subcutaneously into the flanks of six nude mice (left: shMIAT; right: scramble). The following shRNA sequence was used for MIAT knockdown: 5′-GATCCCCGGACA GAGAATGCAAATAATTCAAGAG ATTATTTGCATTC TCTGTCCTTTTTA-3′. The tumor size was calculated by measuring length (L) and width (W) with callipers every 3 days. The tumor volumes were calculated using the formula (L × W^2^)/2. All animal experiments were performed in accordance with the guidelines set by the Institutional Animal Care and Use Committee (IACUC) of China Medical University (CMU). All animals were housed in the Laboratory Animal Center of CMU under a 12 h light/dark (08:00/20:00) cycle with free access to food and water. The mice were sacrificed using CO~2~, and the tissues were subsequently harvested. All breeding and subsequent use of animals in this study, including sacrifice, was approved by the IACUC of CMU. The IACUC approval number was 102-203-N. Gelatin zymography assay {#s4_10} ------------------------ Gelatin zymography was performed to determine the activity of MMP2 and MMP9. In brief, the protein in control or siMIAT cell medium were separated in 10% SDS-PAGE containing 3 mg/ml gelatin at 4°C. PAGE was then incubated at 37°C with incubation buffer (50 mM Tris-HCl pH 7.6, 10 mM CaCl~2~.2H~2~O, 50 mM NaCl) for 24 h. Gelatinolytic activities appeared as clear bands after the cells were stained with 0.25% Coomassie brilliant blue R-250. RNA immunoprecipitation (RIP) {#s4_11} ----------------------------- RNA immunoprecipitation (RIP) was performed using ChIP-IT (Active Motif, Carlsbad, CA, USA) according to the manufacturer\'s instructions. In brief, endogenous MLL and EZH2 complexes from the whole-cell extract were pulled down using anti-MLL1 (EMD Millipore, St. Charles, MO) and anti-EZH2 (Cell Signaling Technology, Danvers, MA, USA) antibody-coated beads. The beads were washed with wash buffer and eluted with elution buffer. The eluted samples were incubated with 0.5 mg/ml protease K to remove proteins. The isolate from the IP product was further analysed using qRT-PCR. The primers for detecting MIAT expression were as follows: forward: ctggagagggaggcatctaa and reverse: aactcatccccacccacac. Immunofluorescence combined with RNA-FISH {#s4_12} ----------------------------------------- Simultaneous protein and mRNA detection using immunofluorescence-combined single-molecule RNA fluorescence *in situ* hybridization (FISH) was performed as described previously \[[@R48]\]. The RNA-FISH probe primer were as follows: forward: tgactccctgaagatctcatcc and reverse: tgctaggaagctgttccagac. The PCR product of MIAT RNA was purified and labelled using Label IT Cy^®^3 nucleic acid labelling kit (Mirus Bio Corp., Madison, WI, USA). In brief, A549 cells were plated on glass coverslips in 6-well culture plates at a density of 10,000 cells/well. Cells were fixed for 10 min in 4% formaldehyde (Thermo Scientific, Rockford, IL, USA) in 1× RNase-free PBS at room temperature. Next, specimens were blocked and permeabilised for 60 min at room temperature in blocking buffer (1× RNase-free PBS, 1% acetylated BSA, 0.3% Triton X-100, and 2 mM vanadyl ribonucleoside complexes). Blocked specimens were incubated with antibodies diluted in blocking buffer. MLL protein was stained with anti-MLL1 (EMD Millipore, St. Charles, MO, USA) and Alexa Fluor 488-conjugated donkey anti-mouse antibody (Jackson ImmunoResearch, West Grove, PA, USA). Incubations with primary antibodies were performed overnight in the dark at 4°C and those with secondary antibodies for 90 min in the dark at room temperature in a humidifying chamber. The RNA-FISH probe were 2192 bps PCR product of MIAT RNA was purified and labelled by Label IT Cy^®^3 nucleic acid labelling kit (Mirus Bio Corp., Madison, WI). After postfixation (10 min in 4% methanol-free formaldehyde in 1× RNase-free PBS at room temperature), the RNA-FISH procedure was performed as described above. Finally, all samples were mounted onto slides in the Vectashield Mounting Medium with DAPI (Vector Laboratories, Burlingame, CA, USA), sealed with nail polish, and imaged using a Leica DMI6000B (AF7000 version) inverted widefield fluorescence microscope (Leica Microsystems, Wetzlar, Germany). Chromatin immunoprecipitation (ChIP) {#s4_13} ------------------------------------ DNA ChIP assay was performed using ChIP-IT (Active Motif, Carlsbad, CA, USA) according to the manufacturer\'s instructions by using Anti-MLL1 (EMD Millipore, St. Charles, MO, USA), anti-EZH2 (Cell Signaling Technology) antibodies, and IgG. Three sets of primers were designed to amplify the MMP9 promoter region. P1 (−1563 to −1634)-forward: ggagatttggctgcatgg, reverse: gcaggatatgggggaaaataat; P2 (−893 to −966)-forward: cctagcagagcccattcctt, reverse: ccctgacagccttctttgac; and P3 (−207 to −288)-forward: cagtccacccttgtgctctt, reverse: ctaggtgtttgcccacctct. Statistical analysis {#s4_14} -------------------- All experimental data from three independent experiments were analyzed by GraphPad Prism version 5 (GraphPad Software Inc., La Jolla, California, USA) and results were expressed as mean ± SD (standard deviation, SD). The association between relative MIAT RNA expression levels and clinical parameters (age, gender, tumor size, lymph node metastasis and TNM stage) was analyzed using Fisher\'s exact test. Student *t*-test was conducted to analyze the *in vitro* and *in vivo* assays. *P* \< 0.05 was considered to indicate a statistically significant difference. This work was supported by the grant to the Ministry of Science and Technology (NSC 99-2320-B-039-038-MY3, NSC 105-2811-B-039-038, MOST 106-2632-E-468-002, MOST 106-2221-E-468-018), National Health Research Institutes (NHRI-EX103-10326BI), Ministry of Health and Welfare (MOHW106-TDU-B-212-113004), China Medical University Hospital (DMR-104-101) and Asia University (ASIA-104-CMUH-22, ASIA-105-CMUH-15). **CONFLICTS OF INTEREST** The authors declare that there are no conflicts of interest.

Introduction ============ Brucellosis is the collective name of a group of zoonotic diseases afflicting a wide range of domestic and wild mammals ([@B75]; [@B77]). In domestic livestock brucellosis is manifested mostly as abortions and infertility, and contact with infected animals and consumption of unpasteurized dairy products are the sources of human brucellosis, an incapacitating condition that requires prolonged antibiotic treatment ([@B78]). Eradicated in a handful of countries, brucellosis is endemic or even increasing in many areas of the world ([@B33]; [@B18]; [@B39]). This disease is caused by facultative intracellular parasites of the genus *Brucella*. Taxonomically placed in the α-2 *Proteobacteria* ([@B51]), the brucellae are close to plant pathogens and endosymbionts such as *Agrobacterium, Sinorhizobium*, and *Rhizobium* and to soil bacteria such as *Ochrobactrum*, the latter including some opportunistic pathogens, and comparative analyses suggest that soil bacteria of this group are endowed with properties that represent a first scaffold on which an intracellular life style develops ([@B70]; [@B50]; [@B5]). The brucellae owe their pathogenicity mainly to their ability to multiply within dendritic cells, macrophages, and a variety of other cells. Due to their ability to control intracellular trafficking and be barely detected by innate immunity, these bacteria are able to reach a safe intracellular niche before an effective immune response is mounted, and to multiply extensively ([@B25]; [@B4]). A mechanism used by *Brucella* to scape from the host immune response is the interference with the toll-like receptor (TLR) signaling pathway by the injection of active effectors such as BtpA and BtpB through the Type IV secretion system T4SS. Both effector proteins contain a TIR domain that interferes with TLR signaling by directly interacting with MyD88 ([@B12]; [@B61], [@B60]; [@B11]) and contribute to the control of dendritic cell (DC) activation during infection. Moreover, *Brucella* has modified outer membrane (OM) components in order to reduce the pathogen-associated molecular patterns (PAMP) of the cell envelope. In Gram-negative bacteria, these PAMP are created by the conserved composition of the OM lipopolysaccharide (LPS) and the free lipids on which the topology of the OM also depends. However, in addition to free-lipid species present in most Gram-negative bacteria (i.e., cardiolipin, phosphatidylglycerol, and phosphatidylethanolamine), *Brucella* also possesses phosphatidylcholine and amino lipids. Phosphatidylcholine is a eukaryotic-type phospholipid required for *Brucella* full virulence ([@B13]; [@B15]). Among the amino lipids, only the ornithine lipids (OL) have been investigated which unlike their counterparts in *Bordetella*, do not trigger the release of IL-6 or TNF-α by macrophages, possibly on account of their longer acyl chains that reduce the OL PAMP ([@B55]). Concerning the LPS, most bacteria carry C1 and C4′ glucosamine disaccharides with C12 and C14 acyl and acyl-oxyacyl chains. This highly amphipathic structure, named lipid A, is adjacent to additional negatively charged groups of the core oligosaccharide, namely the heptose phosphates and 2-keto-3-deoxyoctulosonate carboxyl groups ([@B36]; [@B52]). This lipid A-core PAMP is so efficiently detected by the innate immunity system that some pathogens partially conceal it by removing phosphate groups or substituting them with arabinosamine and/or ethanolamine, or by hydroxylating the acyl chains ([@B67]; [@B43]; [@B48]; [@B54]; [@B44]; [@B69]). In contrast, *Brucella* lipid A is a diaminoglucose disaccharide amide-linked to long (C16, C18) and very long (C28--C30) acyl chains ([@B70]; [@B31]; [@B21]). Furthermore, negative charges in lipid A phosphates and 2-keto-3-deoxyoctulosonate are counterbalanced by four glucosamine units present in the core ([@B38]; [@B21]). As illustrated by the unusually reduced endotoxicity of the *Brucella* LPS this structure is defectively detected by the innate immune response ([@B40]; [@B47]; [@B14]). It remains unknown, however, whether *Brucella* LPS undergoes post-synthetic modifications that have been described for other bacteria that could alter its PAMP potential and contribution to virulence. In this work, we investigated in *Brucella* the role of gene homologs to phosphatases, phospho-ethanolamine (pEtN) transferases, and acyl hydroxylases (**Figure [1](#F1){ref-type="fig"}**) that have been shown in other Gram-negative pathogens to act on LPS and to contribute to overcoming innate immunity defenses. {#F1} Materials and Methods {#s1} ===================== Bacterial Strains and Growth Conditions --------------------------------------- The bacterial strains and plasmids used in this study are listed in Supplementary Table [S1](#SM1){ref-type="supplementary-material"}. Bacteria were routinely grown in standard tryptic soy broth or agar either plain or supplemented with kanamycin at 50 μg/ml, or/and nalidixic at 5 or 25 μg/ml or/and 5% sucrose. All strains were stored in skim milk at -80°C. DNA Manipulations ----------------- Genomic sequences were obtained from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database^[1](#fn01){ref-type="fn"}^. Searches for DNA and protein homologies were carried out using the National Center for Biotechnology Information (NCBI^[2](#fn02){ref-type="fn"}^) and the European Molecular Biology Laboratory (EMBL) -- European Bioinformatics Institute server^[3](#fn03){ref-type="fn"}^. Primers were synthesized by Sigma-Genosys (Haverhill, United Kingdom). DNA sequencing was performed by the "Servicio de Secuenciación del Centro de Investigación Médica Aplicada" (Pamplona, Spain). Restriction--modification enzymes were used under the conditions recommended by the manufacturer. Plasmid and chromosomal DNA were extracted with Qiaprep Spin Miniprep (Qiagen) and Ultraclean Microbial DNA Isolation Kits (Mo Bio Laboratories), respectively. When needed, DNA was purified from agarose gels using the Qiack Gel Extraction Kit (Qiagen). Mutagenesis ----------- To obtain *BmeΔlptA, BaΔlpxE*, and *BmiΔolsC* in-frame deletion mutants, directed mutagenesis by overlapping PCR were performed using genomic DNA as template and pJQK ([@B63]) as the suicide vector. The corresponding gene was deleted using allelic exchange by double recombination as previously described ([@B15]). For the construction of the *BmeΔlptA* mutant, we first generated two PCR fragments: oligonucleotides *lptA-*F1 (5′-GAACGCGAGACTATGGAAAC-3′) and *lptA-*R2 (5′-TGGTGAACGCCAGAAGATAGA-3′) were used to amplify a 400-bp fragment including codons 1--26 of *BmelptA* ORF, as well as 324 bp upstream of the *BmelptA* start codon, and oligonucleotides *lptA-*F3 (5′-TCTATCTTCTGGCGTTCACCGCACGACAATCTCTTC-3′) and *lptA*-R4 (5′-AATATTCCATGGCGCATTTC-3′) were used to amplify a 472-bp fragment including codons 506--544 of the *lptA* ORF and 353-bp downstream of the *lptA* stop codon. Both fragments were ligated by overlapping PCR using oligonucleotides *lptA*-F1 and *lptA*-R4 for amplification, and the complementary regions between *lptA*-R2 and *lptA*-F3 for overlapping. The resulting fragment, containing the *lptA* deleted allele, was cloned into pCR2.1 (Invitrogen, Barcelona, Spain), sequenced to ensure maintenance of the reading frame, and subcloned into the *Bam*HI and the *Xba*I sites of the suicide plasmid pJQK. The resulting mutator plasmid (pRCI-32) was introduced in *B. melitensis 16M* by conjugation using the *Escherchia coli* S.17 strain ([@B64]). For the construction of the *BaΔlpxE* mutant, we first generated two PCR fragments: oligonucleotides *lpxE*-F1 (5′-CGCGTGTGCCATAGGTATATT-3′) and *lpxE-*R2 (5′-TATAGGCAGGGCGCAGAA-3′) were used to amplify a 482-bp fragment including codons 1--29 of *lpxE* ORF, as well as 394 bp upstream of the *lpxE-*1 start codon, and oligonucleotides *lpxE-*F3 (5′-TTCTGCGCCCTGCCTATAGATTCGTTTCCGCATGGT-3′) and *lpxE-*R4 (5′-CCAATACAC CCGTCATGAGA-3′) were used to amplify a 577-bp fragment including codons 226--255 of the *lpxE* ORF and 488-bp downstream of the *lpxE* stop codon. Both fragments were ligated by overlapping PCR using oligonucleotides *lpxE-*F1 and *lpxE-*R4 for amplification, and the complementary regions between *lpxE*-R2 and *lpxE*-F3 for overlapping. The resulting fragment, containing the *lpxE* deleted allele, was cloned into pCR2.1 (Invitrogen, Barcelona, Spain), sequenced to ensure maintenance of the reading frame, and subcloned into the *Bam*HI and the *Xba*I sites of the suicide plasmid pJQK ([@B63]). The resulting mutator plasmid (pRCI-36) was introduced in *B. abortus* 2308 by conjugation using the *E. coli* S.17 strain ([@B64]). For the construction of the *BmiΔolsC* mutant, we first generated two PCR fragments: oligonucleotides *olsC-*F1 (5′-TGCTGGATCGTATTCGTCTG-3′) and *olsC-*R2 (5′-GCCATAAGCCGATGGAACTA-3′) were used to amplify a 334-bp fragment including codons 1--15 of *olsC* ORF, as well as 289 bp upstream of the *olsC* start codon, and oligonucleotides *olsC-*F3 (5′-TAGTTCCATCGGCTTATGGCAGGAGGGGCTAGACAACCAC-3′) and *olsC*-R4 (5′-AACCAGCGACAGGGTAAGC-3′) were used to amplify a 320-bp fragment including codons 286--313 of the *olsC* ORF and 237-bp downstream of the *olsC* stop codon. Both fragments were ligated by overlapping PCR using oligonucleotides *olsC*-F1 and *olsC*-R4 for amplification, and the complementary regions between *olsC*-R2 and *olsC*-F3 for overlapping. The resulting fragment, containing the *lptA* deleted allele, was cloned into pCR2.1 (Invitrogen, Barcelona, Spain), sequenced to ensure maintenance of the reading frame, and subcloned into the *Bam*HI and the *Xba*I sites of the suicide plasmid pJQK ([@B63]). The resulting mutator plasmid (pRCI-65) was introduced in *B. microti* CM445 by conjugation using the *E. coli* S.17 strain ([@B64]). Deletion of each gene was checked with oligonucleotides *gene*-F1 and *gene*-R4 and internal primers hybridizing in the non-deleted regions. Complementation of Deleted Genes -------------------------------- For pBME*lpxE* and pBME*lptA* construction we took advantage of the *Brucella* ORFeome constructed with the Gateway cloning Technology (Invitrogen) ([@B17]). The clones carrying Bme*lpxE* or Bme*lptA* were extracted and the DNA containing the corresponding ORF was subcloned in plasmid pRH001 ([@B28]) to produce pBME*lpxE* and pBME*lptA*. For pBMI*olsC, olsC* was amplified using genomic DNA of Bmi-parental as DNA template. The primers used were *olsC*-F6 (5′-GCTTTCCGAACAAGCACTGA-3′) and *olsC*-R7 (5′-GCCTCCCTTCACCGGTTATT-3′). The resulting PCR product, containing the ORF from 342 bp upstream to 84 bp downstream, was then cloned into pCR2.1 TOPO (Invitrogen) plasmid by "TA cloning" (Life Technologies). The resulting plasmid was sequenced to ensure that the gene was correctly cloned. Then, the gene was subcloned into the *BamHI* and the *XbaI* sites of the replicative plasmid pBBR1 MCS ([@B37]) pBME*lpxE*, pBME*lptA*, and pBMI*olsC* were introduced into *Brucella* by conjugation using *E. coli* S.17-1 strain and the conjugants harboring corresponding plasmid were selected by plating onto TSA-Nal-Cm plates. Sensitivity to Cationic Peptides -------------------------------- Exponentially growing bacteria were adjusted to an optical density equivalent to one of the McFarland scale and the minimal inhibitory concentrations (MICs) of polymyxin B were determined by the *e*-test method on Müller--Hinton agar (Izasa) or by the serial dilution method in a similar broth. LPS Preparation --------------- Lipopolysaccharide was obtained by methanol precipitation of the phenol phase of a phenol--water extract ([@B42]). This fraction \[10 mg/ml in 175 mM NaCl, 0.05% NaN~3~, 0.1 M Tris--HCl (pH 7.0)\] was then purified by digestion with nucleases \[50 μg/ml each of DNase-II type V and RNase-A (Sigma, St. Louis, MO, United States), 30 min at 37°C\] and three times with proteinase K (50 μg/ml, 3 h at 55°C), and ultracentrifuged (6 h, 100,000 × *g*) ([@B1]). Free lipids (OLs and phospholipids) were then removed by a fourfold extraction with chloroform--methanol \[2:1 (vol/vol)\] ([@B70]). Infections in Mice ------------------ Seven-week-old female BALB/c mice (Charles River, Elbeuf, France) were kept in cages with water and food ad libitum and accommodated under biosafety containment conditions 2 weeks before the start of the experiments. To prepare inocula, tryptic soy agar (TSA) grown bacteria were harvested and suspended in 10 mM phosphate buffered saline (pH 6.85), and 0.1 ml/mouse containing approximately 5 × 10^4^ colony forming units (CFU) for *B. melitensis* or *B. abortus* and 1 × 10^4^ CFU for *B. microti* was administered intraperitoneally. The exact doses assessed retrospectively by plating dilutions of the inocula. Number of CFU in spleens was determined at diferent time after inoculation. For this, the spleens were aseptically removed and individually weighed and homogenized in 9 volumes of PBS. Serial 10-fold dilutions of each homogenate were performed and each dilution was plated by triplicate. Plates were incubated at 37°C for 5 days. At several points during the infection process, the identity of the spleen isolates was confirmed by PCR. The individual data were normalized by logarithmic transformation, and the mean log CFU/spleen values and the standard deviations (*n* = 5) were calculated. Intracellular Multiplication Assays ----------------------------------- Bone marrow cells were isolated from femurs of 7--8-week-old C57Bl/6 female and differentiated into dendritic cells \[bone-marrow derived dendritic cells (BMDCs)\] as described by [@B30]. Infections were performed by centrifuging the bacteria onto the differentiated cells (400 x *g* for 10 min at 4°C; bacteria:cells ratio of 30:1 followed by incubation at 37°C for 30 min under a 5% CO~2~ atmosphere). BMDCs were gently washed with medium to remove extracellular bacteria before incubating in medium supplemented with 50 μg/ml gentamicin for 1 h to kill extracellular bacteria. Thereafter, the antibiotic concentration was decreased to 10 μg/ml. To monitor *Brucella* intracellular survival at different time-points post-infection, BMDC were lysed with 0.1% (vol/vol) Triton X-100 in H~2~O and serial dilutions of lysates were plated onto TSA plates to enumerate the CFU. Flow Cytometry -------------- To assess activation and maturation, BMDC were analyzed for surface expression of classical maturation markers at 24 h post-treatment with the different *Brucella* strains and derived mutants. Cells were labeled with fluorochrome-conjugated antibodies specific for mouse CD11c:APC-Cy7 (clone N418), IA-IE:PE (MHC class II clone M5/114.15.2) (PE), CD86:FITC (Clone GL-1), CD40:APC (clone 3/23), and CD80:PE-Cy5 (clone 16-10A1), all from BioLegend. Labeled cells were then subjected to multi-color cytometry using a LSR II UV (Becton Dickinson) and the data analyzed using FlowJo Software by first gating on the CD11c^+^ population (100,000 events) prior to quantifying expression of receptors. Cells were stimulated with *E. coli* LPS (055:B5) as a positive control. Lipid A Extraction ------------------ Five milligrams of LPS was hydrolyzed in 5 ml 1% acetic acid by sonication, heating to 100°C for 30 min, and cooling to room temperature. Concentrated HCl was added to the mixture until the pH was 1--2. The solution was converted to a two-phase acidic Bligh--Dyer mixture by adding 5.6 ml of chloroform and 5.6 ml of methanol. Phases were mixed by inverting the tubes and separated by centrifugation at 4000 × *g* for 20 min. The lower phases containing lipid A were collected, washed two times with water, and dried under a stream of nitrogen. Extraction was repeated, and the lower phases (11.2 ml) were combined and neutralized with a drop of pyridine. Samples were evaporated to dryness under a stream of nitrogen. Mass Spectrometry ----------------- Mass spectrometra were acquired on a Bruker Autoflex^®^Speed TOF/TOF Mass Spectrometer (Bruker Daltonics Inc.) in negative reflective mode with delayed extraction. The ion-accelerating voltage was set at 20 kV. Each spectrum was an average of 300 shots. A peptide calibration standard (Bruker Daltonics Inc.) was used to calibrate the Matrix Assisted Laser Desorption/Ionization Time-of-Flight (MALDI-TOF), and lipid A extracted from *E. coli* strain MG1655 grown in LB medium at 37°C. Extraction and Analysis of Envelope Lipids ------------------------------------------ The free-lipid fraction was extracted as described by [@B7], and analyzed on a silica gel 60 high-performance thin layer chromatography (HPLC) plates (Merck, Darmstadt, Germany). Chromatography was performed either monodimensionally with chloroform--methanol--water \[14:6:1 (volume)\] or bidimensionally with chloroform--methanol--water \[14:6:1 (volume)\] first and chloroform--methanol--acetic acid \[13:5:2 (volume)\] in the second dimension ([@B74]). Plates were developed with 0.2% ninhydrin in acetone at 180°C or 15% sulfuric acid in ethanol at 180°C. Results ======= The *Brucella lptA* Orthologs Encode a Lipid A Phosphate-Ethanolamine Transferase --------------------------------------------------------------------------------- A genomic search in the KEGG database revealed that all *Brucella* spp. carry an ORF (BMEI0118 in *B. melitensis*) homologous to *Neisseria meningitidis lptA*, a pEtN transferase that modifies lipid A ([@B16]). Strikingly, in *B. abortus* but not in other *Brucella* spp., all genomic sequences available at KEGG show a deletion of a thymine in position 774 that should result in a truncated protein lacking the amino acids related to the enzymatic activity ([@B53]; **Figure [1](#F1){ref-type="fig"}** and Supplementary Figure [S1](#SM4){ref-type="supplementary-material"} and Supplementary Table [S2](#SM2){ref-type="supplementary-material"}). In addition to LptA, two other pEtN transferases have been identified in *N. meningitidis*: Lpt-3 and Lpt-6, which, respectively, modify the LPS core at the third and sixth position of heptose II ([@B45]; [@B76]). By multiple sequence alignment, the *B. melitensis* putative pEtN transferase showed highest homology with *Neisseria* LptA and also displayed the LptA membrane-associated domains not present in Lpt-3 and Lpt-6 (ORFs NMB1638, NMB2010, and NMA0408, respectively). Accordingly, it can be predicted that ORF BMEI0118 (henceforth BME*lptA*) encodes a pEtN transferase that acts on lipid A, a hypothesis fully consistent with the absence of heptose in the *Brucella* LPS core ([@B31]; [@B21]). To test this hypothesis, we constructed a *B. melitensis* non-polar mutant (BmeΔ*lptA*) lacking the LptA enzymatic domain (amino acids 26--506), which as expected maintained a smooth (S) phenotype (negative crystal violet test and positive coagglutination with anti-S-LPS antibodies). As a consequence of the increased positive charge of the amino group, pEtN has been shown to decrease binding of the polycationic lipopeptide polymyxin B to LPS, and to increase resistance to this antibiotic in a variety of bacteria ([@B54]; [@B69]; [@B29]). In keeping with this possibility, the BmeΔ*lptA* mutant was more sensitive to polymyxin B than the parental strain *B. melitensis* 16M (Bme-parental) (**Figure [2A](#F2){ref-type="fig"}**). In contrast, and consistent with the frame-shift in its *lptA* homolog, *B. abortus* 2308 (Ba-parental) displayed polymyxin B sensitivity similar to that of BmeΔ*lptA*. Moreover, complementation of BmeΔ*lptA* with the multi-copy plasmid pBME*lptA* or its introduction into *B. abortus* 2308 leads to restoration of polymyxin B resistance in BmeΔ*lptA* or an increase up to *B. melitensis* level in *B. abortus* (**Figure [2A](#F2){ref-type="fig"}**). As expected both constructs kept the S type features (negative crystal violet test and positive coagglutination with anti-S-LPS antibodies) of the parental strains. *N. gonorrhoeae* shows increased resistance to the action of complement in non-immune serum that is dependent on lipid A-linked pEtN ([@B43]). Testing for a similar contribution here, we found that BmeΔ*lptA* was more sensitive than either the parental strain or the complemented mutant (25% vs. no decrease in viability after 3 h of incubation in normal sheep serum) relevant given that *B. melitensis* is characteristically resistant to killing by normal serum. {#F2} By MALDI-TOF analysis, the lipid A of Bme-parental was found to contain four main clusters of ions (A, B, C, and D in **Figure [2B](#F2){ref-type="fig"}**). BmeΔ*lptA* lipid A was qualitatively identical to Bme-parental with respect to groups A, B, and C but clearly differed in group D (**Figure [2B](#F2){ref-type="fig"}** and Supplementary Table [S3](#SM3){ref-type="supplementary-material"}). In group D, the 2191 m/z″ species of Ba-parental was consistent with the isotopic mass of a molecule (C~120~H~232~N~4~O~25~P~2~) formed by a hexaacylated and bisphosphorylated diaminoglucose disaccharide carrying the hydroxylated long and very long chain acyl groups characteristic of *Brucella* ([@B70]; [@B20]). According to this interpretation, the signal(s) at 2112 m/z (mass of - H~2~PO~3~ o - HPO~3~ -, 80.9 - 79.9) could correspond to a monophosphorylated (C~120~H~232~N~4~O~25~P) 2191 m/z″ equivalent. Substitution of this monophosphorylated form with pEtN (^+^H~3~NCH~2~-CH~2~- HPO~3~ mass 125) should account for signal m/z 2237, in keeping with the fact that m/z 2237 did not appear in the spectrum of the lipids A from either BmeΔ*lptA* or Ba-parental (**Figure [2B](#F2){ref-type="fig"}**). Although a clear cut demonstration requires direct analyses of the enzymatic analyses of LptA, these results and the homologies with LptA of other bacteria are consistent with the hypothesis that LptA acts as a pEtN transferase in *B. melitensis* and lacks functionality in *B. abortus*. It is remarkable that pEtN activity was detected for only a fraction (D) of lipid A species. This could be explained by a preferential activity of the enzyme for higher MW lipid A molecules. The *Brucella lpxE* Orthologs Encode a Phosphatase Involved in the Remodeling of the OM --------------------------------------------------------------------------------------- As described above, MALDI-TOF analyses showed the presence of molecular species with a mass compatible with monophosphorylated lipid A. Since lipid A synthesis produces C1 and C4′ bisphosphorylated disaccharide backbones ([@B57]), a possible explanation could be its dephosphorylation by a phosphatase such as LpxE, an inner membrane enzyme that in the phylogenetic neighbor *Rhizobium leguminosarum* removes the lipid A phosphate at C1 ([@B58]). A search in KEGG showed that all *Brucella* spp. carry an ORF homologous to *R. leguminosarum lpxE* (Supplementary Table [S2](#SM2){ref-type="supplementary-material"}). However, the start codon in the *B. melitensis* 16M homolog (BMEI1212) is annotated to a position different from that determined for other brucellae (Supplementary Table [S2](#SM2){ref-type="supplementary-material"}), including other *B. melitensis* strains. Thus, whereas the *B. abortus* homolog (BAB1_0671) is predicted to encode a protein of 255 amino acids, the *B. melitensis* one could encode a protein of either 235 or 255 amino acids (**Figure [1](#F1){ref-type="fig"}**). Both proteins conserve the consensus sequence of the lipid phosphatase superfamily \[KX6RP-(X12--54)-PSGH-(X31--54)-SRX5HX3D\] ([@B66]) which is also present in LpxE from *R*. *leguminosarum, Sinorhizobium meliloti*, and *Agrobacterium tumefaciens* ([@B34]). Although BAB1_0671 and BMEI1212 code for proteins that contain the three motifs conserved in the LpxF phosphatase from *Francisella*, they lack two amino acids of the central motif, NCSFX2G, which seems LpxF specific ([@B72], [@B73]). Thus, the *Brucella* proteins were named BALpxE and BMELpxE. To study whether BALpxE actually acts as a lipid A phosphatase, we constructed a non-polar mutant (BaΔ*lpxE*) and tested it against polymyxin B, since the permanence of a phosphate group in an OM molecule should increase sensitivity to this antibiotic. Mutant BaΔ*lpxE* was eight times more sensitive than the parental strain (MIC 0.2 and 1.6 μg/ml, respectively). Moreover, when we introduced a plasmid containing the BME*lpxE* ortholog into BaΔ*lpxE*, the resistance to polymyxin B was restored (MIC 1.6 μg/ml). Although final confirmation of this interpretation would require to assay the enzymatic activity of the protein, these results are consistent with the predicted role of *lpxE* as a phosphatase and its functionality in both *B. abortus* and *B. melitensis* 16M, a strain where the annotation of the start codon was a source of ambiguity. By MALDI-TOF analysis, the Ba-parental lipid A spectrum showed three of the four predominant clusters of ions (A, B, and C) found in *B. melitensis* (**Figure [2B](#F2){ref-type="fig"}** and Supplementary Table [S3](#SM3){ref-type="supplementary-material"}). Cluster A (m/z 2173) was consistent with an hexaacylated bisphoshorylated diaminoglucose disaccharide (C~120~H~232~N~4~O~24~P~2~) and the signal at 2093 m/z, which differed in the mass of one phosphate group (i.e., 80), was consistent with the cognate monophosphorylated lipid A (C~120~H~232~N~4~O~25~P) (A-Pi, **Figure [2](#F2){ref-type="fig"}**). Other signals differing in a mass of 14 or 28 units should result from the heterogeneity in acyl chain length that is typical of lipid A. The B and C clusters also contained signals differing in 80 mass units that could correspond to bis- and mono-phosphorylated species. The mass spectrum of BaΔ*lpxE* lipid A (not shown) did not differ significantly from that of Ba-parental, and again showed acyl chain heterogeneity in the A, B, C clusters, as well as the -80 m/z signals indicative of mono- and bisphoshorylated lipid A species. As mutation of lpxE is concomitant with an increase in polymyxin B sensitivity, it is tempting to speculate that LpxE directly or indirectly modulates *Brucella* cell envelope by removing an accessible phosphate group from a substrate different from lipid A. Further studies need to be performed to clarify the role of LpxE. The *Brucella lpxO* Orthologs Encode an Acyl Hydroxylase Acting on Ornithine Lipids ----------------------------------------------------------------------------------- The genomes of all *Brucella* species available at KEGG contain an ORF homologous to *Salmonella lpxO* ([@B23]), which encodes an enzyme hydroxylating the 3′-secondary acyl chain of lipid A. In all *Brucella* spp. except *B. microti* and *B. vulpis* this ORF presents a frame-shift leading to a truncated protein that lacks the consensus of the aspartyl/asparaginyl β-hydroxylases family to which LpxO belongs (**Figure [1](#F1){ref-type="fig"}** and Supplementary Table [S2](#SM2){ref-type="supplementary-material"}). These characteristics are consistent with chemical studies that previously failed to observe S2 hydroxylated fatty acids in *B. abortus* lipid A ([@B70]). Moreover, a *lpxO* homolog is present in *Ochrobactrum anthropi* where S2 hydroxylated fatty acids were also not observed in the lipid A ([@B70]), indicating that a role similar to that of *Salmonella* LpxO is unlikely. Thus, the *lpxO* homologs present in these *B. microti* and *O. anthropi* could be acting on a free lipid and, in fact, it has been reported that the corresponding *R. tropici* homolog is a β-hydroxylase acting on OLs ([@B71]). If this were the case in *O. anthropi* and the brucellae, the end product \[a hydroxylated OL (OH--OL)\] of the pathway described previously in members of the *Rhizobiaceae* (**Figure [3A](#F3){ref-type="fig"}**) should be observed in *O. anthropi* and *B. microti* (and *B. vulpis*) but not in other *Brucella* spp. ![The *Brucella lpxO* orthologs encode an acyl hydroxylase acting on ornithine lipids. **(A)** Pathway of synthesis of ornithine lipids in α-2 *Proteobacteria* (adapted from [@B22]); the ORFs of *B. abortus* and *B. microti* are indicated, whereas *B. microti, B. vulpis*, and *O. anthropi* contain an intact *olsC* acyl hydroxylase gene, *B. abortus* and other *Brucella* spp. carry a frame-shift in the *olsC* homolog. **(B)** Lipid profile of *B. abortus* wild-type (Ba-parental) and *B. microti* wild-type (Bmi-parental) showing the absence or presence, respectively, of OH--OL. **(C)** Amino lipid profile of *B. abortus* wild-type (Ba-parental), *B. microti* wild-type (Bmi-parental), *B. microti* deleted in *olsC* (BmiΔ*olsC*), the cognate reconstituted mutant (BmiΔ*olsC*pOlsC), and *B. abortus* wild-type carrying a plasmid with the *B. microti olsC* gene (Ba-parentalpOlsC).](fmicb-08-02657-g003){#F3} To investigate these hypotheses, we compared the free lipids of *B. abortus, B. melitensis, B. suis, B. ovis, B. microti*, and *O. anthropi*. As can be seen in **Figure [3B](#F3){ref-type="fig"}**, *B. microti* but not *B. abortus* produced an amino lipid with the migration pattern predicted for OH-OL ([@B71]), and results similar to those of *B. microti* were obtained for *O. anthropi* but not for the other *Brucella* spp. tested (not shown). These observations support the interpretation that *O. anthropi* and *B. microti* LpxO are OL hydroxylases and are fully consistent with the aforementioned genomic and chemical evidence. Accordingly, *Brucella lpxO* should be named *olsC.* To confirm this, we examined the amino lipids of a non-polar *olsC* mutant in *B. microti* (BmiΔ*olsC*). As predicted, this mutant did not synthesize OH--OL and complementation with a plasmid containing *B. microti olsC* restored the wild-type phenotype (**Figure [3C](#F3){ref-type="fig"}**). Furthermore, introducing this plasmid or a plasmid carrying *O. anthropi olsC* into *B. abortus* resulted in the synthesis of OH--OL (**Figure [3C](#F3){ref-type="fig"}** and Supplementary Figure [S2](#SM5){ref-type="supplementary-material"}). No difference in polymyxin sensitivity was observed in these constructs or the mutant BmiΔolsC when compared to the corresponding parental strains. LptA, LpxE, and OlsC Are Not Required for *Brucella* Virulence in Laboratory Models ----------------------------------------------------------------------------------- *Brucella abortus, B. melitensis*, and *B. suis* have been shown to multiply in murine and human monocyte-derived dendritic cells while interfering with their activation and maturation and reducing both antigen presentation and an effective adaptive response ([@B6]; [@B47]; [@B14]; [@B26]; [@B56]). To assess whether LptA, LpxE, and OL β-hydroxylase (OlsC) were involved, we compared parental and mutant strains of *B. melitensis, B. abortus*, and *B. microti* in mouse BMDCs. As shown in **Figure [4](#F4){ref-type="fig"}**, the kinetics of multiplication of the mutants and wild-type strains were similar. We also performed a phenotypic characterization of MHC II and co-stimulatory receptors CD86 and CD80 (**Figure [5](#F5){ref-type="fig"}**). In agreement with previous studies, these analyses showed that activation and maturation was only partially induced in BMDC infected with *B. melitensis* and *B. abortus* ([@B47]). In addition, a similar partial-activation profile was evident both for *B. microti*, for which no previous studies exist in infected BMDC, and all of the tested mutants obtained for each of the three *Brucella* spp. {#F4} {#F5} The mouse model has been widely used for testing *Brucella* virulence ([@B27]). In this model, the LptA and LpxE mutants and the parental strains behaved identically (**Figure [6](#F6){ref-type="fig"}** upper panels). Deletion of *olsC* in *B. microti* did not alter the CFU/spleen profile produced by this species which is characterized by a lower lethal dose in mice as well as a faster clearance from mouse spleens ([@B32]; **Figure [6](#F6){ref-type="fig"}**, lower left panel). Moreover, when we tested whether the expression of *B. microti olsC* in *B. abortus* could affect virulence, we found no differences between the *B. microti olsC*-carrying and the wild-type *B. abortus* strains (**Figure [6](#F6){ref-type="fig"}**, lower right panel). {#F6} Discussion ========== In this work we investigated three *Brucella* ORFs that according to homologies with genes of known function in other pathogens could modify the lipid A and contribute to further altering the LPS PAMP of representative *Brucella* species. The results show that, whereas *Brucella* LptA modifies the lipid A, this is not the case for *lpxE* and *lpxO* (redesignated *olsC*), the former encoding a putative phosphatase acting on an unidentified OM molecule and the latter for an enzyme with OlsC activity. Our data strongly suggest that *B. melitensis* LptA is involved in the addition of pEtN to lipid A, homologous proteins carrying out this function are not uncommon in Gram-negative pathogens and modulate the properties of lipid A. In *Salmonella* Typhimurium, *Shigella flexneri, E. coli, Vibrio cholerae, Helicobacter pylori, Haemophilus ducreyi, N. gonorrhoeae*, and *N. meningitidis* pEtN reduces the binding of cationic bactericidal peptides by balancing the negative charge of lipid A ([@B54]; [@B69]). Conversely, pEtN promotes binding to *N. gonorrhoeae* lipid A of factors that downregulate the complement cascade and thwart building of the membrane-attack complex and opsonophagocytosis ([@B43]). *N. meningitidis* pEtN also promotes adhesion of non-encapsulated bacteria to endothelial cells ([@B67]). Indeed, properties that parallel some of those observed for the above-listed pathogens can also be attributed to the pEtN transferase counterpart in *Brucella*. An intact *lptA* was related to polymyxin B resistance in *B. melitensis* and the introduction of *B. melitensis lptA* into *B. abortus* increases polymyxin B resistance to the level of *B. melitensis*, suggesting that LptA function is severally impaired in *B. abortus*. This is in agreement with the presence of a frame-shift in *B. abortus lptA* encompassing the consensus sequence, which makes likely that it codes for a protein with no or residual enzymatic activity. Previous analyses are contradictory with regard to the presence ([@B10]) or absence ([@B51]) of ethanolamine in *B. abortus* lipid A but the materials analyzed differ in methods of extraction and presence of *B. abortus* lipid A markers, such as very long chain fatty acids (VCLFA). Although further chemical and enzymatic analyses are necessary for a definite conclusion, our results strongly suggest that, if present, pEtN is in much less amounts in *B. abortus* than in *B. melitensis* lipid A. It is also worth noting that such genetic and phenotypic differences in the lipid A of *B. abortus* and *B. melitensis* could relate to differences in biological properties. The LPS of *B. abortus* and *B. melitensis* is a poor activator of the complement cascade, and this property has been traced to the core and lipid A structure ([@B49]; [@B14]; [@B21]). Since *B. abortus* is less resistant than *B. melitensis* to normal serum ([@B24]), it is tempting to suggest that, like in *N. gonorrhoeae, B. melitensis* pEtN could sequester regulatory elements enhancing complement resistance in this species. Concerning LpxE, phosphatases acting on lipid A have at least been shown in *Francisella tularensis, H. pylori, Porphyromonas gingivalis*, and *Capnocytophaga canimorsus*, bacteria where lipid A dephosphorylation is involved both in resistance to bactericidal peptides and the reduction of TLR-4-dependent recognition ([@B54]). Although these properties are displayed by the LPS of *B. abortus* and *B. melitensis* ([@B46]; [@B41]; [@B14]), our results do not support a role for BALpxE as a lipid A phosphatase. This is consistent with genomic analysis showing that, whereas in bacteria where LpxE acts on lipid A the gene is located together with *lptA* in an operon ([@B68]; [@B59]), *Brucella lpxE* is instead located upstream of three sequences annotated as pseudogenes and downstream, but in the opposite direction, of a cystathionine beta-lyase. On the basis of the data shown here, the origin of monophosphoryl lipid A in *Brucella* remains to be explained. Further, we believe it unlikely to be an artifact resulting from the hydrolytic steps used to obtain lipid A and instead favor the hypothesis of the existence of an as yet unidentified lipid A phosphatase. LpxE belongs to the type 2 family of phosphatases that can act on lipid A but also on phosphatidylglycerol phosphate, phosphatidic acid, sphingosine phosphate, and lysophosphatidic acid ([@B9]; [@B62]). Significantly, LpxE from *Agrobacterium*, although predicted to be a lipid A phosphatase, dephosphorylates phosphatidyl glycerophosphate ([@B35]) to generate phosphatidylglycerol, a cell envelope phospholipid. Indeed, a hypothetical phosphatidyl glycerophosphate phosphatase activity of *Brucella* LpxE could account for both the polymyxin B sensitivity of the mutated bacteria and the unaltered mass spectra of the lipid A of the mutant. Such a modification of a phospholipid could be meaningful by itself on account of the LpxE-dependent bactericidal peptide resistance but there are other possibilities. In some bacteria (i.e., *Rhizobium*) phosphatidylglycerol is a precursor for the synthesis of amino lipids such as lysyl-phosphatidylglycerol. This synthesis is induced by acid pH and brings about resistance to daptomycin and polymyxin B ([@B65]; [@B19]; [@B2]). Interestingly, whereas the BaΔ*lpxE* mutant is impaired for growth at pH 6, the parental *B. abortus* becomes more resistant to cationic peptides (L. Palacios-Chaves and R. Conde-Álvarez, Unpublished observations). These observations suggest the existence in *Brucella* of pH-dependent envelope modifications that require a functional LpxE. Research is in progress to elucidate the mechanisms behind the increased resistance at acid pH and the implication regarding a role for LpxE. In *S.* Typhimurium, *Pseudomonas aeruginosa, Bordetella bronchispetica, Legionella pneumophila*, and *Klebsiella pneumonia*, LpxO is a Fe2+/α-ketoglutarate-dependent dioxygenase that catalyzes the hydroxylation of the 3′-secondary acyl chain of lipid A. LpxO has been implicated indirectly in stress responses at the envelope level ([@B54]) and, in *K. pneumoniae*, it has been shown to be relevant *in vivo* by increasing bactericidal peptide resistance and reducing the inflammatory responses ([@B44]). However, as discussed above, previous chemical analysis ([@B70]) of lipid A and the evidence presented here indicate that the *Brucella lpxO* homolog is not a lipid A hydroxylase but rather an OlsC whose mutation, in contrast with LpxO, does not result in increased sensitivity to polymyxin B. This absence of an effect on polycation resistance is in keeping with both the lack of activity on lipid A and the fact that OL do no play a major role in resistance to polycationic bactericidal peptides in *B. abortus* ([@B55]). At the same time, it would also appear to rule out, the involvement of this protein in the metabolism of succinate in *B. microti* as has been previously suggested ([@B3]). Previous data showing *lptA, lpxE*, and *lpxO* to be involved in modulating the properties of the OM in a way that in some cases confers *in vitro* resistance to innate immunity bactericidal peptides, complement, and cytokine responses ([@B54]) have been drawn upon as evidence for a role in virulence. However, to the best of our knowledge, a role *in vivo* has thus far been shown only for *lpxO* from *K. pneumoniae* ([@B44]). Moreover, contrasting results have been obtained with mutants both showing bactericidal peptide sensitivity *in vitro* and no phenotype *in vivo* have been reported for at least *H. ducreyi* ([@B69]) and may reflect the complexities of the infection processes and/or the inadequacies of the currently available *in vivo* models. Despite their effect on the envelope, our results show that *Brucella lptA, lpxE*, or *olsC* do not play a role in the ability of *Brucella* to replicate in BMDC and do not modulate the activation and maturation profile in these cells. Similarly, the mouse model did not reveal any effect on its ability to colonize and multiply in the spleen. However, further experimental work in the natural hosts and alternative routes of infection might provide evidence on the role in virulence of these genes. The fact that *lptA* and *olsC* are not functional in all *Brucella* spp. must therefore be considered in the context of the models used. While the absence of a functional *lptA* in *B. abortus* suggests that the gene is not essential for the virulence of this species we cannot conclude it to be totally irrelevant. Differences between *B. melitensis* and *B. abortus* related to *lptA* could explain the higher invasiveness of the former species noted by early researchers in studies carried out in guinea pigs, animals that are highly susceptible to brucellosis ([@B8]). This possibility together with the presence of intact *lptA* and *olsC* in *Ochrobactrum* and *B. microti* is also compatible with the hypothesis that they represent ancestral characters that are liable to be lost in the absence of a selective pressure during the intracellular life cycle or, in the case of *lptA*, that is no longer present in the ruminant host species (i.e., cattle) to which *B. abortus* is characteristically associated. Ethics Statement ================ Female BALB/c mice (Charles River, France) were kept in cages with water and food *ad libitum* under P3 biosafety conditions in the facilities of "Centro de Investigación Médica Aplicada" (registration code ES31 2010000132) 2 weeks before and during the experiments. The procedures were in accordance with the current European (directive 86/609/EEC) and Spanish (RD 53/2013) legislations, supervised by the Animal Welfare Committee of the University of Navarra, and authorized by the "Gobierno de Navarra" \[CEEA045/12 and E36-14 (045-12E1)\]. Author Contributions ==================== IM, MI, J-PG, JB, and RC-Á conceived the study. RC-Á, LP-C, YG-R, MB-V, MS-B, BA-A, EM-G, AZ-R, MdM, TLB, SH, M-JG, MV-G, and VA-G carried out the experimental work. IM, MI, and RC-Á wrote the paper. All authors participated in the presentation and discussion of results. Conflict of Interest Statement ============================== 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. **Funding.** This research was supported by the Institute for Tropical Health funders (Obra Social la CAIXA, Fundaciones Caja Navarra and Roviralta, PROFAND, Ubesol, ACUNSA, and Artai) and grants MINECO (AGL2014-58795-C4-1-R, Bru-Epidia 291815-FP7/ERANET/ANIHWA), Aragón Government (Consolidated Group A14), and Marie Curie Career Integration Grant U-KARE (PCIG13-GA-2013-618162). TLB is the recipient of a Ph.D. Fellowship funded by the Department for Employment and Learning (Northern Ireland, United Kingdom). The authors thank A. Delgado-López for excellent technical assistance in the extraction and purification of LPS. <http://www.genome.jp/kegg/> <http://www.ncbi.nlm.nih.gov/> <http://www.ebi.ac.uk/> Supplementary Material ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fmicb.2017.02657/full#supplementary-material> ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. [^1]: Edited by: *Axel Cloeckaert, Institut National de la Recherche Agronomique (INRA), France* [^2]: Reviewed by: *Diego J. Comerci, Instituto de Investigaciones Biotecnológicas (IIB-INTECH), Argentina; Roy Martin Roop II, East Carolina University, United States* [^3]: This article was submitted to Infectious Diseases, a section of the journal Frontiers in Microbiology

Ischemia-reperfusion injury represents a pathological condition characterized by an initial undersupply of blood to an area or organ followed by a restoration of perfusion and concomitant reoxygenation (= reperfusion). Ischemia typically occurs in the presence of embolism or thrombosis but can also be triggered by surgery and transplantation. Anyway, the disturbance in perfusion results in a severe imbalance between metabolic supply and demand, subsequently causing tissue hypoxia \[[@B1]\]. Notably, these initial changes cause time-dependent molecular and structural alterations. In this context, it is also important to consider that all tissues and organs are susceptible to ischemia, but susceptibility to an ischemic insult differs between organ systems. Whereas the brain can endure ischemia only a few minutes, other tissues (e.g., muscle) are able to withstand ischemia for a long time without signs of irreversible damage. Interestingly, restoration of blood flow and reoxygenation is commonly associated with an exacerbation of tissue injury and a profound inflammatory response ("reperfusion injury") \[[@B1], [@B2]\]. Ischemia-reperfusion injury contributes to pathology in a wide range of conditions. For example, myocardial ischemia followed by reperfusion typically manifests in microvascular dysfunction, death of myocytes, and myocardial stunning or dysfunction. Ischemia-reperfusion injury (IRI) of the lung, for example, following transplantation, is characterized by nonspecific alveolar damage, edema formation, and hypoxemia. The clinical spectrum of pulmonary IRI may range from mild hypoxemia to acute respiratory distress syndrome. In contrast to other organs, the brain is particularly susceptible to ischemia and irreversible neuronal damage already occurs after only 5 minutes of complete ischemia \[[@B3]\]. For brain ischemia, as occurring in the setting of stroke, reestablishing reperfusion seems to be only beneficial, if carried out within a short time period after the onset of ischemia. Reperfusion of ischemic stroke seems to be very critical, as patients may suffer from cerebral reperfusion injury manifesting in fatal cerebral edema formation and intracranial hemorrhage. IRI of the kidney may occur in the setting of transplantation and cardiac arrest and during cardiac surgery. Here it is important to note that renal injury is usually associated with a high morbidity and mortality. The cortical-medullary region is the most susceptible region to tubular injury, inflammation, and vascular alterations. Generally, IRI of a single organ causes the release of different proinflammatory mediators, which may subsequently induce inflammation in other organs, thereby potentially contributing to multiple organ dysfunction or even failure \[[@B4]\]. Different pathological processes contribute to tissue injury secondary to ischemia-reperfusion. During ischemia, limited oxygen availability leads to an impaired endothelial cell barrier function with a concomitant increase in vascular permeability and leakage due to decreases of intracellular cAMP levels caused by a reduced adenylate cyclase activity \[[@B1]\]. Furthermore, ischemia-reperfusion induces cell death due to apoptosis, necrosis, and autophagy \[[@B5]\]. During the ischemic period, alterations in the transcriptional control of gene expression likewise occur. Another mechanism implicated in the pathophysiology of injury during ischemia is the inhibition of oxygen-sensing prolyl hydroxylase (PHD) enzymes, because they require oxygen as a cofactor. Hypoxia-triggered inhibition of PHD enzymes induces the posttranslational activation of hypoxia and inflammatory signaling cascades, which in turn regulate the stability of the transcription factors, hypoxia-inducible factor (HIF) and nuclear factor-*κ*B (NF-*κ*B) \[[@B2]\]. Reperfusion of ischemic tissue activates a complex inflammatory response without the involvement of pathogenic triggers, a phenomenon also referred to as sterile inflammation. During the initiation of this inflammatory response, endogenous molecules act as alarmins or danger-associated molecular patterns (DAMPs) \[[@B6]\]. The inflammatory process is stimulated through self-antigens, which are functional components of intact cells but become stimulators of innate immunity when released from injured or dying cells \[[@B6]\]. In 1996, Weiser et al. discovered and described a novel mechanism for reperfusion injury that involves antibody deposition and activation of the complement system leading to an acute inflammatory response \[[@B7]\]. One decade later, the concept of innate autoimmunity was introduced, which is based on the discovery that circulating natural antibodies recognize self-antigens and elicit an acute inflammatory response involving the complement system \[[@B8]\]. Although ischemia-reperfusion is typically established in a sterile environment, activation of innate and adaptive immune responses occurs and contributes to injury, including activation of pattern-recognition receptors such as TLRs and inflammatory cell trafficking into the injured organ \[[@B9]\]. During this inflammatory process, the coagulation system is also activated, because the innate immune system and coagulation system are highly interconnected \[[@B10]\]. As ischemia-reperfusion injury is a common clinical problem and is associated with relevant complications, it is important to identify therapeutic approaches which prevent or at least mitigate ischemia-reperfusion-induced organ injury and organ dysfunction. This special issue is devoted to the modulation of ischemia-reperfusion injury by different measures and contains eight original papers addressing this clinically relevant topic. These papers are accompanied by two review articles dealing with the effects of anesthetics on ischemia-reperfusion injury. Papers from B. U. Togrul et al., D. Dohman et al., and Y. Demirci et al. are focusing on ischemia-reperfusion injury of the liver. In two of these three papers, different therapeutic interventions on hepatic ischemia-reperfusion injury are evaluated, whereas the third paper is a retrospective study in which the authors investigated the efficacy and safety of intermittent portal triad clamping with low central venous pressure during liver resection. In this context, it has been reported that remote ischemic preconditioning and therapeutic interventions can reduce liver damage after inducing ischemia-reperfusion injury. The studies by S. C. Karahan et al., B. Michèle et al., S. C. Karahan et al., D. Dohman et al., and G. Altun et al. elucidate the effects of different anesthetic techniques and drugs on ischemia-reperfusion injury. These eight papers are entitled as follows:*"The effects of remote ischemic preconditioning and N-acetylcysteine with remote ischemic preconditioning in rat hepatic ischemia-reperfusion injury model"* by B. U. Togrul et al.,*"The effects of spinal, inhalation, and total intravenous anesthetic techniques on ischemia-reperfusion injury in arthroscopic knee surgery"* by S. C. Karahan et al.,*"Efficacy and safety of hepatectomy performed with intermittent portal triad clamping with low central venous pressure\"* by D. Dohman et al.,*"Adalimumab ameliorates abdominal aorta cross clamping induced liver injury in rats"* by Y. Demirci et al.,*"Evidence for the use of isoflurane as a replacement for chloral hydrate anesthesia in experimental stroke: an ethical issue"* by B. Michèle et al.,*"The effect of dexmedetomidine on oxidative stress during pneumoperitoneum"* by S. C. Karahan et al.,*"The comparison of the effects of sevoflurane inhalation anesthesia and intravenous propofol anesthesia on oxidative stress in one lung ventilation"* by D. Dohman et al., and*"Role of ethyl pyruvate on systemic inflammatory response and lung injury in an experimental model of ruptured abdominal aortic aneurysm"* by G. Altun et al. *Alexander Zarbock* *Ahmet Eroglu* *Engin Erturk* *Can Ince* *Martin Westphal*

About 1 in every 10 adults worldwide is overweight or obese[@b1] and obesity is a risk factor for morbidity and mortality from cardiovascular diseases, diabetes, certain cancers, and musculoskeletal disorders. Despite strong interest in research addressing obesity, safe and effective pharmacological options for the prevention and treatment of this condition remain elusive[@b2][@b3]. Currently, most drug-discovery efforts are based on *in vitro* assays with candidate targets, but *in vitro* assays do not reconstitute the complexity of whole organisms. This is particularly relevant for drugs modulating feeding behavior and metabolic homeostasis, since both arise from complex interactions within and between the central nervous system, the digestive tract, and fat-storage organs[@b4][@b5][@b6], which cannot be modeled *in vitro*. An alternative to *in vitro-*based screens is phenotype-based whole organism screens[@b7][@b8][@b9][@b10]. Whole organism drug screens provide several advantages over *in vitro* assays. Active drugs are by definition bio-available and potential toxicity can be evaluated at early project stages. Further, an effective drug need not act through a well-validated target, but can have novel or complex mechanisms of action. However, whole organism drugs screens can be costly and time-consuming. These disadvantages can be partially overcome by using model organisms that can be raised cost-effectively in large quantities, like the vinegar fly *Drosophila melanogaster*. Flies and vertebrates share many metabolic functions, molecular machinery, and analogous organ systems that control nutrient uptake, storage, and metabolism[@b11][@b12][@b13][@b14][@b15]. Like humans, flies regulate circulating sugar levels according to food availability, and store excess energy in the form of glycogen and lipids. These reserves are mobilized during periods of energy consumption[@b12][@b16][@b17]. As seen in many animals, fasted flies increase food foraging and intake[@b18]. Two master metabolic regulators in vertebrates, insulin and leptin, have functional homologues in the fly[@b13][@b14]. In addition, an unbalanced diet can trigger a type-2 diabetes-like insulin resistance and obesity phenotypes in the fly[@b19]. Here we report the development of a high-throughput drug-screen for *Drosophila* larval feeding. We identify the serotonin (5-hydroxytryptamine or 5-HT) receptor antagonist metitepine as a potent anorectic drug and show that all five fly 5-HT receptors are inhibited by this drug. Despite its broad spectrum antagonism of *Drosophila* 5-HT receptors, metitepine requires only receptor 5-HT2A for its *in vivo* anti-feeding activity. Our results highlight the potential of *Drosophila* as a tool for pharmacological study of feeding behavior and provide a powerful method for drug discovery and target identification. Results ======= High-throughput feeding assay for *Drosophila* larvae ----------------------------------------------------- To screen for drugs that modify food intake in whole animals, we developed a high-throughput assay that allowed us to monitor ingestion of fluorescent liquid food by *Drosophila* first instar larvae in 96-well plates read by a plate reader ([Fig. 1a](#f1){ref-type="fig"}). Larvae were dispensed into plates and fed liquid food consisting of sugar and yeast extract and supplemented with fluorescein for visualization. After washing uningested fluorescent food from the wells, we quantified the fluorescein ingested by the larvae that was visible in the digestive tract ([Fig. 1b](#f1){ref-type="fig"}). To evaluate the dynamic range of our assay, we carried out control experiments to either decrease or increase food intake in the larvae. When the animals were cold-paralyzed while feeding on fluorescent food, ingestion was reduced ([Fig. 1c](#f1){ref-type="fig"}). When larvae were selectively fasted for protein by removing yeast extract from the liquid food before being exposed to fluorescent food, they showed a post-fasting rebound in which they ingested more standard liquid food than control animals continuously fed standard liquid food ([Fig. 1d](#f1){ref-type="fig"}). The dynamic range of feeding suppression was more than two times greater than feeding enhancement in these experiments, perhaps because larvae are continual feeders and may be ingesting at a near maximal rate during basal conditions[@b20]. We next tested a known feeding mutant in our assay by comparing the fluorescent signal accumulated in three different wild-type strains (*w*^1118^, Canton-S, and Oregon-R) and a *klumpfuss*^09036^ mutant (*klu*, ref. [@b20]). *klu* encodes a transcription factor necessary for proper expression of the neuropeptide hugin, whose activity is required for normal feeding behavior in *Drosophila*[@b20]. All three wild-type strains ingested significantly more than the feeding mutant ([Fig. 1e](#f1){ref-type="fig"}). In addition, we confirmed previous reports[@b21] that high concentrations of dietary amino acids suppressed food intake ([Fig. 1f](#f1){ref-type="fig"}). Drug screen for modulators of food intake ----------------------------------------- After validating our feeding assay, we screened for small molecules that modulated food intake. In a pilot screen of 415 compounds tested individually at 20 μg/ml, we identified one compound, cycloheximide, which inhibited feeding (data not shown). This established a hit rate of 0.24%. To improve the throughput of the subsequent primary screen, we used pools of 3 to 4 compounds per well (see Methods for details). 3630 small molecules were tested in the primary screen ([Fig. 2a](#f2){ref-type="fig"}). The compounds were obtained primarily from annotated chemical libraries, such that each compound had at least one known cellular target (see Methods for details). The average signal of all the drug-treated wells was less than 3% different from the average of solvent-treated wells, indicating that the drugs did not cause generalized toxicity ([Fig. 2b](#f2){ref-type="fig"}). 279 and 114 compounds were identified as candidate anorectic (feeding suppressant) and orexigenic (feeding stimulant) drugs, respectively, by the criterion that they differed from solvent controls by more than one standard deviation ([Fig. 2a](#f2){ref-type="fig"}). The anorectic compounds caused an average decrease in fluorescent signal of 27%, while the orexigenic compounds caused an average increase of 18% ([Fig. 2b](#f2){ref-type="fig"}). The 393 candidate small molecules were re-tested individually in the secondary screen ([Fig. 2a](#f2){ref-type="fig"}). Of the anorectic and orexigenic candidates, 32 and 10 compounds, respectively, were reconfirmed as hits, defined as differing from solvent controls by more than one standard deviation ([Fig. 2a, c, d](#f2){ref-type="fig"}, [Supplementary Table 1](#s1){ref-type="supplementary-material"}). We searched for reported molecular targets for each of the 42 hits of the secondary screen using a drug-discovery database (<https://www.collaborativedrug.com/>). Two known insect anti-feedants, gedunin and plumbagin[@b22][@b23], were among these compounds ([Fig. 2d](#f2){ref-type="fig"}), confirming the efficacy of our screen. We chose 14 compounds for verification of a dose-response curve ([Supplementary Table 1](#s1){ref-type="supplementary-material"}), based on their annotation as drugs that target neuromodulators, cell signaling, and/or neuronal activity. From these, only metitepine, a non-selective antagonist of 5-HT receptors, and reserpine, an inhibitor of the vesicular mono-amine transporter (VMAT), showed reliable dose-dependent responses and were selected for further characterization. Reserpine was subsequently discarded because it dramatically reduced larval locomotion at concentrations as low as 10 μM (data not shown). In light of these results, we concluded that the effects of reserpine on feeding were secondary to a general effect on locomotion and muscular tone, as confirmed by the sluggish phenotype of the dVMAT mutant larvae[@b24]. Metitepine decreased food accumulation by more than one standard deviation when tested in combination with other drugs, during the primary screen ([Fig. 2e, f](#f2){ref-type="fig"}), or alone, in the secondary screen ([Fig. 2g](#f2){ref-type="fig"}). Dose-response experiments indicated that the threshold concentration for metitepine efficacy was 10 μM, with increasing effects at 50 and 100 μM ([Fig. 2h](#f2){ref-type="fig"}). Metitepine decreases feeding persistently, reversibly, and specifically ----------------------------------------------------------------------- To ask if metitepine decreases larval feeding on conventional fly food, we tested the effect of the drug on larvae fed a standard laboratory cornmeal-agar-molasses diet supplemented with the dye bromophenol blue. We quantified the amount of food ingested by measuring the optical density (O.D.) of the gut in individual larvae. Larvae treated with the drug accumulated less solid food in their digestive tract as reflected by a lower O.D. ([Fig. 3a](#f3){ref-type="fig"}). The effect of metitepine on solid food accumulation was dose-dependent ([Fig. 3b](#f3){ref-type="fig"}) with a threshold efficacy dose of 50 μM. To rule out the possibility that metitepine was causing a general locomotor defect in larvae, we measured their crawling speed when fed solvent or metitepine. Metitepine-treated animals crawled at the same speed as solvent treated controls ([Fig. 3c](#f3){ref-type="fig"}). Food accumulation in the digestive tract is a function of both ingestion and excretion. To establish a direct link between metitepine treatment and food intake, we measured mouth-hook contraction rates of larvae that were treated with either solvent or 100 μM metitepine. Mouth-hook contraction is the motor behavior associated with food intake in larvae. Animals treated with the drug displayed decreased mouth-hook contraction rates, confirming that metitepine had a direct effect on food intake ([Fig. 3d](#f3){ref-type="fig"}). Notably, since mouth-hook contractions were measured after drug treatment (see Methods), this result also suggests that metitepine is not required to be present in food to induce a decrease in ingestion. To further explore the time course of metitepine action on feeding behavior, we measured food intake at various time points after metitepine treatment. When tested immediately after exposure, drug-treated larvae showed reduced food accumulation ([Fig. 3e](#f3){ref-type="fig"}, 0-h recovery). The anorectic effect of metitepine lasted for 2 hours but was not evident 4 hours after treatment ([Fig. 3e](#f3){ref-type="fig"}). At 24 hours after treatment, metitepine-treated larvae ate significantly more, suggesting that larvae were rebounding from drug-induced fasting ([Fig. 3e](#f3){ref-type="fig"}). Metitepine is a non-selective antagonist of *Drosophila* 5-HT receptors ----------------------------------------------------------------------- Metitepine is a broad spectrum antagonist of vertebrate 5-HT receptors. To investigate the pharmacology of this drug on *Drosophila* 5-HT receptors, we expressed each in mammalian tissue culture cells and carried out calcium imaging experiments. Four 5-HT receptors have been previously identified and cloned: *5-HT1A* and *5-HT1B* (ref. [@b25]), *5-HT2* (ref. [@b26]), and *5-HT7* (ref. [@b27]). Of these, 5-HT1A, 5-HT1B, and 5-HT7 were previously expressed in heterologous systems, shown to respond to 5-HT, and to activate different intracellular effector systems[@b25][@b27][@b28]. In binding-competition assays 5-HT2 was shown to bind to both 5-HT and metitepine[@b26]. A fifth putative receptor (CG42796) was annotated as a 5-HT receptor based on homology[@b29]. We cloned CG42796 and propose that it be named 5-HT2B, since its closest homologue is 5-HT2 (ref. [@b29]). We suggest that the gene formerly known as 5-HT2 be denoted as 5-HT2A. This revised nomenclature for 5-HT2A and 5-HT2B is used throughout the manuscript. We expressed all five 5-HT receptors in HEK-293T cells and monitored their activity by measuring intracellular calcium concentrations. All of the receptors induced a dose-dependent response to 5-HT ([Fig. 4a](#f4){ref-type="fig"}). From the dose-responses curves of each receptor we calculated a half effective concentration (EC~50~, [Fig. 4b](#f4){ref-type="fig"}). The most sensitive receptor was 5-HT7 (EC~50~ = 12 nM) and the least sensitive receptor was 5-HT1A (EC~50~ = 1 μM). We next established stimulus conditions in which we applied two pulses of 5-HT without desensitizing any of the receptors ([Fig. 4c](#f4){ref-type="fig"}). Under these conditions, 100 μM metitepine dramatically suppressed or completely abolished responses to 5-HT in all 5-HT receptors ([Fig. 4d](#f4){ref-type="fig"}). Control experiments confirmed that metitepine did not kill the cells because ATP, a ligand for endogenous purinergic receptors, activated all cells after metitepine treatment ([Fig. 4d](#f4){ref-type="fig"}). Metitepine had an effective inhibitory concentration (IC~50~) in the μM range, ranging from 2 μM for 5-HT2B to 58 μM for 5-HT1B ([Fig. 4e](#f4){ref-type="fig"}). These pharmacological experiments confirmed that all five candidate 5-HT receptors in the *Drosophila* genome respond to serotonin. However, since they all showed sensitivity to metitepine, further genetic experiments were required to identify the molecular target of the anorectic drug *in vivo*. 5-HT2A is required for the anorectic actions of metitepine and for normal feeding behavior ------------------------------------------------------------------------------------------ We reasoned that if metitepine were acting through a specific 5-HT receptor, mutating that gene would render larvae resistant to the drug. We generated or obtained null mutants for each one of the five *Drosophila* 5-HT receptors and asked if they were sensitive to the anorectic effects of 100 μM metitepine. All mutants except *5-HT2A*^Gal4^ were sensitive to metitepine and showed a decrease in feeding similar to that seen in wild-type strains ([Fig. 5a](#f5){ref-type="fig"}). In contrast, 5-HT2A receptor mutants were resistant to the effects of this drug. To confirm this observation, we tested additional 5-HT2A mutant alleles. *5-HT2A*^e01363^ and *5-HT2A*^PL00052^ are different pBac transposon insertions in the *5-HT2A* locus. We tested each as homozygous insertions and in heteroallelic combinations with the original *5-HT2A*^Gal4^ mutant. *5-HT2A*^e01363^ (*5-HT2A*^e^ in [Fig. 5b](#f5){ref-type="fig"}) interferes with proper splicing of the transcript, but is not a null[@b30]. Consistent with this, *5-HT2A*^e^ was sensitive to the effects of metitepine when tested as a homozygote ([Fig. 5b](#f5){ref-type="fig"}). However, when *5-HT2A*^e^ was tested in combination with the original *5-HT2A*^Gal4^, the heteroallelic mutant combination *5-HT2A*^Gal4/e^ was insensitive to metitepine ([Fig. 5b](#f5){ref-type="fig"}). *5-HT2A*^PL00052^ (*5-HT2A*^PL^ in [Fig. 5b](#f5){ref-type="fig"}) has dramatically reduced levels of *5-HT2A* mRNA[@b31]. Both *5-HT2A*^PL^ homozygous mutants and the heteroallelic mutant combination *5-HT2A*^Gal4/PL^ were insensitive to metitepine ([Fig. 5b](#f5){ref-type="fig"}). Heterozygous *5-HT2A*^Gal4^ larvae showed normal sensitivity to the drug ([Fig. 5b](#f5){ref-type="fig"}). If metitepine suppresses feeding by blocking the activation of 5-HT2A, mutating the gene should result in larvae that eat less. Indeed, *5-HT2A*^Gal4/e^ mutants ate less than wild-type larvae ([Fig. 6a](#f6){ref-type="fig"}). To further confirm the role of *5-HT2A* in larval feeding behavior, we knocked down *5-HT2A* by conditional expression of a *5-HT2A* RNAi with a pan-neuronal GeneSwitch system[@b32]. Larvae in which neuronal expression of 5-HT2A-RNAi was induced ate less than larvae in which the RNAi was not induced or in control larvae in which GFP was induced ([Fig. 6b](#f6){ref-type="fig"}). Thus, genetic knock-down of 5-*HT2A* in a time frame similar to the action of metitepine was sufficient to phenocopy the drug-induced feeding phenotype. Discussion ========== Serotonin is involved in regulating appetite, food intake, and metabolic homeostasis in organisms ranging from *C. elegans* to humans. In *C. elegans*, serotonin activates overall feeding by activating two separate neural pathways that respectively control pharyngeal pumping and isthmus peristalsis[@b33]. In *Drosophila*, serotonin has been shown to play a trophic role during embryonic development in the establishment of neuronal innervation to the gut[@b34]. In adult female flies, serotonin controls the postmating dietary switch to protein-rich food[@b35]. We show here that the 5-HT receptor antagonist metitepine reduces food intake in *Drosophila* larvae and that the drug acts selectively through the 5-HT2A receptor. Interestingly, metitepine was previously identified in a different small molecule screen as a compound that extends lifespan in *C. elegans*[@b8]. There is a known connection between dietary restriction and lifespan across several organism including primates[@b36]. We have not tested the effect of metitepine on *Drosophila* lifespan, but this would be of interest in future studies on this drug. In mammals, the role of serotonin in controlling appetite and body-weight is complex[@b37]. It appears that the level of brain serotonin signaling has an inverse relationship with food intake: when brain serotonin signaling is increased, food intake is reduced, and vice versa. For example, serotonin re-uptake inhibitors reduce food intake[@b2][@b37]. Part of the complexity of the serotonin system that controls metabolic homeostasis in mammals might arise from the fact mammals have 14 5-HT receptor subtypes. This raises the possibility of serotonin having divergent effects on food intake and body weight depending on the receptor subtype activated. *Drosophila*, with only five receptor subtypes, offers a simplified model in which to study the core mechanisms by which serotonin regulates food intake and coordinates metabolic homeostasis. Here we established that *Drosophila* larvae can be used to screen for drugs that modulate food intake. Few model organisms allow the combination of large-scale drug screens with genetic screens to identify bioactive small molecules and their *in vivo* targets[@b2]. *Drosophila* is an appealing model to study appetite control because 60% of functional human genes have orthologues in the fly[@b38] and *Drosophila* has a specialized tissue for fat storage that controls metabolic homeostasis using a leptin-like signaling mechanism[@b13]. Future work can apply these methods to larger scale screens of novel compounds to identify new pathways regulating feeding behavior. Methods ======= Fly stocks ---------- Flies were maintained on conventional cornmeal-agar-molasses medium and, unless otherwise stated, under a natural light-dark cycle, at room temperature. *klumpfuss*^09036^ (stock \#11733), *5-HT1A*^Δ5Kb^ (stock \#27640), *5-HT1B*^MB08181^ (stock \#24240), *5-HT2A*^PL00052^ (stock \#19367), *5-HT2A*^RNAi^ (stock \#31882) and *Df(3R)tll-e* (stock \#5415) were obtained from the Bloomington Stock Center. *5-HT2A*^e01363^ was obtained from the Harvard Exelixis Collection. *5-HT1A*^Gal4^, *5-HT2A*^Gal4^, *5-HT2B*^Gal4^, and *5-HT7*^Gal4^ were generated by J.H and Y.R. by replacing the first coding exon of each gene by Gal4, and will be reported elsewhere (Huang and Rao, in preparation). Embryo collection ----------------- For embryo collection, grape-juice 2%-agar plates were used. Flies were allowed to lay eggs for 24 hours at 25°C. Eggs were further incubated at 18°C for another 24 hours. Egg laying and embryo development were both performed at 70% humidity and in a 12 hour light: 12 hour dark cycle. First instar larva collection ----------------------------- Previously hatched larvae were removed from the embryo-collection plates under a stream of water. Plates were further incubated for 2 hours at 25°C to allow new larvae to hatch. ### For primary and secondary screens Newly hatched larvae were collected in a cell strainer with a gentle stream of distilled water, rinsed and re-suspended in liquid food (100 g/l yeast extract, 100 g/l glucose, 7.5% sucrose, 0.15% nipagin, 6.25 μg/ml cholesterol). ### For individual larval assays Newly hatched larvae were collected with a brush, and transferred to a vial with conventional medium or to a 96-well plate with liquid food. To facilitate penetration of the larvae into the solid food, incisions were made on the surface of the medium. Larvae were incubated at 25°C and 70% humidity. Liquid food feeding assay ------------------------- Seventy-five larvae were dispensed into each well a filter-bottom 96-well plate (Millipore, Part. No. MSRLN04) using a COPAS Select worm sorter (Union Biometrica). Once loading was completed, the liquid content of the plates was filtered away and replaced with 100 μl of liquid food. The plates were incubated for 16--18 hours at 25°C and 70% humidity. Thereafter, the food was replaced with food that contained 0.3% fluorescein (sodium salt, Sigma Cat. No. 6377). Before the fluorescent signal was quantified the plates were washed 4× with 300 μl double-distilled water, 10× with 0.05% PBT, 6× with 400 mM lysine and 2× with 100 mM Na-citrate, 100 mM NaCl, pH2 (citrate buffer). The larvae were kept in 50 μl of citrate buffer for quantification or imaging capture. The fluorescent signal from the plate was acquired in a 5 × 5 circular grid using an EnVision Plate Reader (Perkin Elmer). The total fluorescent value of each well was calculated by adding the data points. Small-molecule screen --------------------- ### Primary screen The small molecules were obtained from the LOPAC1280, Prestwick, GreenPharma, and MicroSource Spectrum libraries, and were provided by the High-Throughput Screening Resource Center (HTSRC) of The Rockefeller University. A total of 3688 compounds, representing 3630 unique structures, were screened. Small molecules were pooled at 10 μg/ml each such that each compound was represented twice in two independent mixtures. Only those compounds that showed the required effect two times were chosen for confirmation in the secondary screen. For the primary screen, sixteen 384-well plates with a different small-molecule in each well were mixed using a 4 × 4 grid into eight 384-well destination plates. Since we screened 3688 small molecules, and the 384-well plates contain 352 usable wells each (32 wells in each plate are reserved for solvent controls), a total of 1944 wells in the source plates were empty (16 × 352 − 3688 = 1944). While the majority of mixtures contained 4 compounds, about a third contained only 3. To apply the drugs to the larvae loaded into the 96-well plates, each 384-well plate quadrant was treated as an independent 96-well plate. The cut-off for hit-identification was arbitrarily set to one standard deviation above or below the solvent-treated wells for anorectic and orexigenic compounds, respectively. The compounds were applied to the larvae in 96-well plates in the liquid food together with fluorescein for 16--18 hours. In each 96-well plate, 80 compound-mixtures were tested and 16 wells were treated only with solvent as control (1.6% DMSO). Each plate was tested in duplicate. ### Secondary screen Compounds were screened individually at 40 μg/ml. Each hit was tested in two rounds, in duplicate. Each screening plate contained 8 solvent-treated control wells. Solid food feeding assay ------------------------ Zero-to-two hour old larvae were transferred to conventional solid food containing either solvent or 100 μM metitepine and 0.05% bromophenol blue, and allowed to feed for 17 hours at 25°C and 70% humidity. Larvae were recovered from the food with a brush, rinsed in PBS, and photograph under a SMZ1500 dissecting scope (Nikon). Images were captured with a DS-2Mv digital camera (Nikon) and a NIS-Elements F acquisition program. Larvae were immobilized in a cold plate set to 4°C. Reflective light was adjusted to make the background of each picture 12% gray. Images of individual larvae were analyzed in MetaMorph (Molecular Devices) to calculate the optical density of the digestive tract. Optical density values of drug-treated larvae were analyzed always in parallel with same day solvent-treated larvae. Mouth-hook contractions ----------------------- Zero-to-two hour old larvae were transferred to conventional solid food containing either solvent or 100 μM metitepine, and allowed to feed for 17 hours at 25°C and 70% humidity. Larvae were recovered from the food with a brush, rinsed in PBS and transfer to a 2% yeast suspension at room temperature. After a one-minute acclimation, a one-minute movie was recorded at 6.25 frames per second (fps) using a Nikon SMZ1500 dissecting scope, a Rolera-RX camera (Q Imaging), and Q Capture 6.0 Software (Q Imaging). Movies were manually scored to quantify mouth-hook contraction rate. Larval locomotion ----------------- Zero-to-two hour old larvae were transferred to conventional solid food containing either solvent or 100 μM metitepine and 0.05% bromophenol blue, and allowed to feed for 17 hours at 25°C and 70% humidity. Bromophenol blue facilitated tracking of the animals since it enhanced contrast. Larvae were recovered from the food with a brush, rinsed in PBS, and transferred to a 3% agar plate at room temperature. After a one-minute acclimation, a one-minute movie was recorded at 6.25 fps using a Nikon SMZ1500 dissecting scope, a Rolera-RX camera (Q Imaging), and Q Capture 6.0 Software (Q Imaging). Movies were analyzed in EthoVision XT 8.0 (Noldus) to calculate linear velocity. Cloning *Drosophila* 5-HT receptors ----------------------------------- 5-HT1A, 5-HT1B and 5-HT2A receptors were PCR-cloned from genomic DNA extracted from transgenic flies expressing full-length cDNAs under regulation of UAS promoter (5-HT1A: Bloomington stock \#27630; 5-HT1B: Bloomington stock \# 27632; 5-HT2A (formerly 5-HT2): Bloomington stock \#24504). The 5-HT2B receptor cDNA (GenBank accession \#KC852205) was PCR-cloned from whole adult fly cDNA prepared using poly-A primers and SuperScript III Reverse Transcriptase (Invitrogen) according to the manufacturer\'s instructions. 5-HT7 receptor was PCR-cloned from *w*^1118^ genomic DNA. High-fidelity KOD polymerase (Novagen) was used for all cloning amplifications and the full sequences of the amplicons were verified. The following primers were used in the cloning reactions: 5-HT1A: Forward: 5′-ATGGCGCACGAGACCAGC-3′ Reverse: 5′-CTAGAGCTTCCCGCTGCGG-3′ 5-HT1B: Forward: 5′-ATGCTGAAAACTGTGACAACAGC-3′ Reverse: 5′-TCAAATTTTCGCACTGCG-3′ 5-HT2A: Forward: 5′-ATGGAGATGCAAAGCTACTCTG-3′ Reverse: 5′-TCACCGTTTGCAGTTGCACTTG-3′ 5-HT2B: Forward: 5′-ATGGAAGAGGATGTGTATGCCT-3′ Reverse: 5′-TTATCTGCTCGGTCGCCA-3′ 5-HT7: Forward: 5′-ATGGCTTTATCTGGACAGGACT-3′ Reverse: 5′-CTAGAGAAAGCTCTCCCTCGC-3′ A 5′-GCCACC-3′ vertebrate Kozak sequence was added upstream of every forward primer. The amplicons were cloned into the XhoI-NotI sites in the pME18ST vertebrate expression vector[@b39]. Expression of *Drosophila* 5-HT receptors in HEK-293T cells ----------------------------------------------------------- HEK-293T cells were seeded on glass-bottom 35-mm Petri dishes (MatTek Corporation, Part No. P35GC-1.5-10-C) and allowed to reach \~70% confluence. Cells were transiently transfected with 2 μg of each receptor-expressing plasmid and Gα~15~-expressing plasmid using Lipofectamine 2000 (Invitrogen) according to the manufacturer\'s instructions. Gα~15~ is a promiscuous G protein that couples activation of a wide variety of G-protein coupled receptor to release of calcium from intracellular stores[@b40]. Cells were kept for 24--32 hours at 37°C and 5% CO~2~ before imaging. To monitor intracellular Ca^2+^ concentrations, transfected cells were loaded for 20 minutes with 2 μM Fura2-AM. Imaging was carried out in a saline solution containing 140 mM NaCl, 5.6 mM KCl, 2 mM CaCl~2~, 2 mM MgCl~2~, 1.25 mM KH~2~PO~4~, 2 mM Na-pyruvate, 0.17% Glucose, 5 mM HEPES (pH 7.4 with NaOH). The Petri dish was placed on an Eclipse TE 2000-U inverted microscope (Nikon) equipped with a Retiga Exi Fast Camera (Q Imaging) and excited with a Lambda DG-4 xenon lamp illumination system (Sutter Instruments). Images were acquired at 1.43 fps. Ligands and drugs were dissolved freshly every day at the indicated concentrations in saline solution and superfused directly into the Petri dish with a peristaltic pump (Miniplus 3, Gilson). Ca^2+^ fluctuations were recorded with MetaFluor software (version 7.1.2.0; Molecular Devices). To determine 5-HT sensitivity for each receptor, several pulses of increasing concentration of 5-HT were applied to the same cells. Pulses were delivered 2 minutes after full recovery of the previous pulse to Ca^2+^ resting levels. 5-HT dose-response curves were calculated normalizing the responses elicited by each concentration tested, by the maximal response elicited by 5-HT in those cells: ΔF/ΔF~MAX~. For metitepine inhibition, a pair pulse protocol was used. Each set of cells was stimulated twice; first they were exposed to 5-HT alone (ΔF~0~), then they were exposed to 5-HT and a given concentration of metitepine (ΔF~Met~). The concentration of 5-HT used in these experiments was closed to the calculated EC~50~ for each receptor and within 95% confidence intervals. The degree of metitepine inhibition was calculated by the following equation: ΔF~Met~/ΔF~0~. Data analysis and curve-fitting was done in Prism 5 (GraphPad). Drug preparation ---------------- ### Primary and secondary screens Drugs were obtained from the HTSRC of The Rockefeller University pre-dissolved in DMSO at 2.5 mg/ml. ### Dose-response curves, in vitro pharmacology, and behavioral experiments Stock solutions of metitepine (Sigma) were prepared in DMSO (100 mM or 500 mM). Stock solutions of 5-HT and ATP (both from Sigma) were prepared in water (50 mM), kept frozen at −20°C, and aliquots thawed only once. 5-HT1B mutant generation ------------------------ Homozygous males carrying the Minos MB05181 transposable insertion inserted in the sixth intron of 5-HT1B were crossed to virgin females of this genotype: Sna^Sco^/SM6a, p(w\[+mC\] = hslLMiT) (Bloomington stock \#24613). The heat-shock scheme to induce expression of the Minos transposase, marker selection, and the establishment of excision lines were carried out as originally described[@b41]. Ninety-four independent excision events were analyzed by PCR with the following primers: Forward primer: 5′-CTGCGCTCCTTCTTCAGC-3′ Reverse primer: 5′-CGTAATTGCCGCCATTATACTC-3′ One imprecise excision that removed a 1344 bp fragment from genomic DNA, thus producing a 1002 bp PCR product instead of the wild-type 2346 bp product, was selected for further characterization. The resulting allele was named *5-HT1B*^ΔIII-V^, since the deleted exons encode transmembrane segments III-V. The breakpoints of the sequence are: AAAAAGGATGTAGAGGAATAGAATA //deletion// CTCCAAAAATAATATTTATACAATA A precise excision was also identified in this screen by virtue of producing a 2346 bp PCR fragment, diagnostic of a clean deletion of the Minos element. Expression of *5-HT2A-*RNAi --------------------------- Zero-to-two hour old transgenic larvae carrying Elav-GeneSwitch and either UAS-5-HT2A-RNAi (Bloomington stock \#31882) or UAS-mCD8-GFP were transferred to fluorescent liquid food containing 160 μg/ml of RU-486 (induced) (Sigma Cat. No. M8046) or 1% ethanol (uninduced) for 17 hours. Data presentation ----------------- Fluorescence and optical density data were always normalized to controls ran in parallel, according to the equation: Unless otherwise noted, data are presented as mean ± s.e.m. Statistical analysis -------------------- Statistical analysis was conducted as indicated in the figure legends using Prism (GraphPad). Every single set of data was tested for normality (Shapiro-Wilk test) and equal variance (F test or Bartlett\'s test). If those criteria were met, parametric comparisons were performed (*t* test or ANOVA). Otherwise, Mann Whitney test or Kruskal-Wallis test were used. *Post hoc* test (Dunnett\'s or Dunn\'s) are listed for each condition examined. Significance is as described in the figure legends with \**p* \< 0.05; \*\**p* \< 0.01; \*\*\**p* \< 0.001. Author Contributions ==================== G.G. and L.B.V. conceived the project. G.G. carried out all experiments and analyzed the data. S.C. provided technical help for the drug screen and the mouth-hook contraction experiments. J.H. and Y.R. made the Gal4 knock-in *Drosophila* mutant lines. G.G. and L.B.V. wrote the manuscript and prepared the figures. Supplementary Material {#s1} ====================== ###### Supplementary Information Supplementary Table 1 Dragana Rogulja, Vanessa Ruta, and members of the Vosshall Laboratory provided helpful comments on the manuscript. We thank Yi-Chen Hsieh for technical help in setting up the feeding assay and Fraser Glickman and members of The Rockefeller University HTSRC for advice in implementing the drug screen. G.G. was a Pew Latin American Fellow in the Biomedical Sciences. This work was funded by National Natural Science Foundation of China (Young Scientists Fund 31000547 to J.H.), Chinese Ministry of Science and Technology (973 Program 2010CB833900 to Y.R.), and The Klarman Family Foundation Grants Program in Eating Disorders Research to L.B.V. L.B.V. is an investigator of the Howard Hughes Medical Institute. {#f1} {#f2} {#f3} {#f4} {#f5} {#f6}

Introduction ============ A giant retinal tear (GRT) is defined as a full-thickness break in the retina equal to or larger than 3 clock hours associated with vitreous detachment.[@b1-opth-12-2053] Risk factors for developing a GRT include trauma, high myopia, Marfan syndrome, Stickler's syndrome or other hereditary vitreoretinopathies, and extensive lattice degeneration.[@b2-opth-12-2053],[@b3-opth-12-2053] Additional etiologies for GRT include iatrogenic and idiopathic causes. The frequent rolled posterior edge of the retinal flap renders repair of retinal detachment (RD) due to GRTs technically challenging. Attempts to improve treatment through the 1970s and 1980s involved prone positioning for fluid-gas exchange, rotating head movements ("steamrolling") to unfold the tear, and manipulation of the retinal flap using intraocular balloons, retinal microincarceration, tissue adhesives, sodium hyaluronate, and retinal tacks, screws, and sutures.[@b4-opth-12-2053],[@b5-opth-12-2053] The advent of perfluorocarbon liquids changed the way that these cases were approached and allowed for higher surgical success rates as it made the surgical technique easier.[@b6-opth-12-2053] Currently, there is still debate about whether adjunctive procedures such as scleral buckling, lensectomy, or silicone oil injection are recommended. Thus, the optimal surgical approach for these challenging cases remains controversial. The current study represents an update to a previously reported noncomparative case series of patients with GRT-associated RDs undergoing primary management at a university referral center.[@b7-opth-12-2053] Materials and methods ===================== In accordance with the guidelines from the Declaration of Helsinki, this study was approved by the Institutional Review Board at the University of Miami, Miller School of Medicine, as a retrospective noncomparative case series of patients undergoing primary RD repair for GRT at the Bascom Palmer Eye Institute between January 2011 and August 2017. The Institutional Review Board at the University of Miami did not require patient consent for retrospective review of medical records, and all patient data were deidentified upon review. Patients with a history of retinopathy of prematurity or retinal tear of less than 3 clock hours were excluded from this study. Patients were identified using the ICD-10 codes for GRTs (H33.031, H33.032, H33.033, H33.039), and their charts were reviewed to verify coding information. A total of 427 patients were initially reviewed, yielding 80 eyes of 79 patients that fulfilled the abovementioned inclusion and exclusion criteria. Data collected included demographic information, recorded etiologies, and surgical techniques. Outcome variables that were studied included best-corrected visual acuity (BCVA), occurrence of single operation retinal reattachment, and occurrence of reoperation. Data were collected postoperatively at 1 week, 1 month, 3 months, 6 months, 1 year, and last follow-up date. The surgical technique was at the discretion of the individual surgeon since there was no defined surgical protocol for this study. The data on proliferative vitreoretinopathy (PVR) were not included in this study as there were inconsistencies in the medical records and lack of standardization in its classification among the surgeons. Current data were compared with data from a previously published article from the same institution from 2005 to 2010.[@b7-opth-12-2053] Surgical variables were analyzed using the chi-squared test and Student's *t*-test. All statistical analyses were performed using SPSS version 24.0 (IBM Corporation, Armonk, NY, USA). Results ======= This study comprises 80 eyes of 79 patients who had an initial presentation of GRT associated with RD. The study included 62 (78%) male subjects. The mean age at initial presentation was 48.1±16 years, ranging from 4 to 71 years. The mean follow-up interval was 10.3±9 months. Right eyes were involved in 42/80 eyes (53%). Principal associations for the development of GRTs included prior history of blunt trauma (23 eyes, 18%), high myopia ≥6 diopters (33 eyes, 26%), lattice degeneration (14 eyes, 11%), and prior RD in the fellow eye (12 eyes, 10%). Other less common associations of GRTs included Marfan syndrome (two eyes, 2%) and Stickler's syndrome (one eye, 1%). The size of the GRT was \<180° in 60/80 eyes (75%) ([Table 1](#t1-opth-12-2053){ref-type="table"}). The majority (68% of eyes) of the GRTs were superior and temporal. The associated RD was macular involving in 65%. The majority of eyes (76%) were initially phakic, while 21% were pseudophakic, and 2.5% were aphakic. Primary scleral buckle (SB) was performed in 3/80 eyes (4%), with two patients receiving the \#41 band and one patient receiving the \#240 band ([Table 2](#t2-opth-12-2053){ref-type="table"}). These eyes were buckled because they were inferior GRTs without rolled edges, and 2 of the 3 eyes had shallow detachments. The majority of eyes (61/80, 76%) were treated with a combined pars plana vitrectomy (PPV) and SB procedure, utilizing a combination of the \#41, \#42, and \#240 bands. All PPVs were performed with either 23 or 25 gauge instrumentation. PPV alone was performed in 16 eyes (20%). In cases of PPV, endolaser was applied in a confluent fashion for one or two rows posterior to the GRT. In the three cases with SB only, cryopexy was used. Lensectomy was performed in 9/61 phakic eyes (15%). Perfluorocarbon liquid was used in 60/77 eyes (78%) undergoing PPV. Among eyes treated with PPV, silicone oil was used in 54/77 eyes (70%), perfluoropropane (C~3~F~8~) was used in 22/77 eyes (29%), and sulfur hexafluoride (SF~6~) was used in 1/77 eyes (1%). SB was more likely to be placed in phakic eyes compared with pseudophakic/aphakic eyes (52/61 eyes, 85% vs 12/19 eyes, 63%; *P*\<0.036). Recurrent RD occurred in 1/3 eyes (33%) that received only a primary SB. However, eyes that received an initial SB (with or without PPV) had the same recurrent RD rate compared with cases managed without SB (10/64 eyes, 16% vs 1/16 eyes, 6%; *P*=0.33). Eyes associated with trauma had similar rates of recurrent RD compared with those not associated with trauma (*P*=0.55). For internal tamponade, there was no difference in rates of primary success between silicone oil (87%) and C~3~F~8~ (91%, *P*=0.6), but the one eye treated initially with SF~6~ developed a recurrent RD. Perfluorocarbon liquid was used equally (and frequently) in GRTs involving both \<180° and ≥180°: 47 eyes (78%) vs 16 eyes (80%), respectively ([Table 3](#t3-opth-12-2053){ref-type="table"}). Silicone oil was more frequently used as tamponade in eyes with GRTs ≥180° compared with C~3~F~8~ (14/20, 70% vs 6/20, 30%; *P*=0.01) and also was more commonly used for GRTs \<180° (40/60 eyes, 67% vs 16/60, 27%; *P*\<0.001). GRTs ≥180° were not less likely to be attached at final follow-up and were not more likely to require further surgery for a recurrent RD compared with GRTs \<180° ([Table 3](#t3-opth-12-2053){ref-type="table"}). The percentage with final BCVA ≥20/400 was not different in those presenting with GRTs ≥180° compared with GRTs \<180°. Rate of lensectomy was not different between phakic eyes with GRTs ≥180° compared with GRTs \<180° ([Table 3](#t3-opth-12-2053){ref-type="table"}). The retina was reattached at the last recorded follow-up examination in 76/80 eyes (95%). Recurrent RD after the primary repair occurred in 11/80 eyes (14%). The silicone oil had been removed in 24/54 eyes (44%) with silicone oil tamponade at last follow-up ([Figure 1](#f1-opth-12-2053){ref-type="fig"}). Of these eyes, 5/24 (21%) had recurrent detachment, and 1/54 eyes (2%) required a silicone oil exchange. Retinal redetachment occurred in 4/11 eyes (36%) after silicone oil was removed, with all redetachments occurring an average of 3 months after initial repair. When comparing with GRT size, 5/14 (36%) of the GRTs ≥180° had oil removed compared with 18/40 (45%) of the GRTs \<180° (*P*=0.55). Preoperative visual acuity was ≥20/400 in 43 eyes (54%). At 6 months, only one eye (1%) had BCVA ≥20/40 and 16 eyes (20%) had BCVA ≥20/400. At 1 year, four eyes (5%) had BCVA ≥20/40 and 21 eyes (26%) had BCVA ≥20/400. At last follow-up examination, 15 eyes (19%) had BCVA of ≥20/40 and 54 eyes (68%) had ≥20/400. Compared with visual acuity at initial presentation, 51 eyes (64%) had stable or improvement in BCVA at final follow-up. There was no difference in mean final BCVA between eyes with macula-off vs macula-on RDs (*P*=0.27). No difference was seen in mean final BCVA between eyes with silicone oil tamponade vs C~3~F~8~ (*P*=0.40). The surgical management and results from the earlier case series of 79 eyes were similar to that of this series in terms of SB use, silicone oil use, final reattachment rate, and visual acuity outcomes ([Table 4](#t4-opth-12-2053){ref-type="table"}).[@b7-opth-12-2053] The final reattachment rate in the current series was 95% compared with 92% in the previous series. In the current series, fewer lensectomies were performed in phakic eyes (9/61 eyes, 15%) at the time of initial surgery compared with the earlier series (23/56 eyes, 41%; *P*=0.001). Discussion ========== The management of GRT-associated RDs has been substantially more successful with modern surgical techniques. In addition to favorable anatomic outcomes, postoperative BCVA in this study has demonstrated improvement compared with previous series.[@b8-opth-12-2053],[@b9-opth-12-2053] The visual success of eyes in this study was encouraging, given that the majority (65%) already had macular detachment. Although another study found gas tamponade associated with better final BCVA in fovea-sparing RDs compared with silicone oil, the current series did not identify any difference in final BCVA between the two groups.[@b10-opth-12-2053] Surgical outcomes for RDs associated with GRTs have improved during the past 20 years.[@b3-opth-12-2053],[@b7-opth-12-2053]--[@b9-opth-12-2053],[@b11-opth-12-2053] The historic reattachment rate has been as low as 11%, whereas the rates of single surgery reattachment (86%) and final reattachment (95%) rates in the current study are comparable with recently published studies and reflect the value of the vitreoretinal tools now available.[@b3-opth-12-2053],[@b6-opth-12-2053]--[@b9-opth-12-2053],[@b11-opth-12-2053],[@b12-opth-12-2053] Perfluorocarbon liquids seem to have offered the most benefit in this improvement, but outcomes have also improved concurrent with the inception of wide-field viewing systems, high-speed cutters, and small gauge vitrectomy. There are still a variety of surgical technique variations advocated, but there is virtual unanimity that PPV and perfluorocarbon liquids are preferred management components.[@b13-opth-12-2053] The current study documents perfluorocarbon liquid use in 78% of cases, regardless of GRT extent, contrasting with the earlier study using it more frequently in larger GRTs (statistical significance not reported).[@b7-opth-12-2053] This current series also found that reoperations did not differ regardless of GRT size. Although the extent of the GRT may impact the technical difficulty of the surgery, it did not seem to determine surgical management or anatomic success. Silicone oil was most commonly used for internal tamponade and had a high anatomic success rate similar to that of C~3~ F~8~, consistent with previous studies.[@b11-opth-12-2053],[@b14-opth-12-2053],[@b15-opth-12-2053] In one report with the predominant use of silicone oil rather than gas, the superiority of silicone oil tamponade could not be confirmed.[@b16-opth-12-2053] While it is uncommon to use SF~6~, and this was used in a tiny subset in the current study, this study was consistent with other studies that have associated SF~6~ with a lower success rate.[@b11-opth-12-2053],[@b14-opth-12-2053] Scleral buckling in GRT surgery has been controversial, as there is a general tendency away from SB use in current RD repair approaches. Scleral indentation from a buckle can distort the shape of the globe and might potentially increase the risk of GRT slippage.[@b17-opth-12-2053] On the contrary, an encircling buckle may provide vitreous base support and, particularly in the case of anterior PVR, may reduce the risk of redetachment by preemptively neutralizing subsequent traction.[@b3-opth-12-2053],[@b8-opth-12-2053] Given that there was a higher proportion of phakic eyes that received an SB, it is possible that the indentation from the buckle would allow for easier shaving of the vitreous base in phakic eyes. The high rate of SB usage in the current study was similar to that of many other studies and might reflect a bias of the investigators.[@b7-opth-12-2053],[@b18-opth-12-2053] Goezinne et al compared the addition of SB procedure with PPV in patients undergoing retinal reattachment surgery for GRTs and found that the presence of an encircling SB was a positive predictive factor for retinal reattachment.[@b19-opth-12-2053] In contrast, other retrospective studies (as well as the current study) found that the rate of redetachment was similar in the cases of PPV alone compared with a combined PPV/SB approach.[@b7-opth-12-2053],[@b11-opth-12-2053] In a 2018 reported series regarding the use of 25-gauge vitrectomy for GRT, neither the extent of GRT nor the use of scleral buckling influenced anatomic or visual outcomes.[@b16-opth-12-2053] The rationale for lensectomy is to allow access for more effective vitreous base shaving, especially since progressive postoperative nuclear sclerosis with the vitrectomy is inevitable. In the earlier study from the same institution, 41% of phakic patients underwent pars plana lensectomy with initial reattachment surgery, while only 15% of phakic patients in the current series did so ([Table 4](#t4-opth-12-2053){ref-type="table"}).[@b7-opth-12-2053] While it has been reported that GRT eyes without PVR usually achieve surgical success without undergoing lensectomies, it is possible that the presence of PVR would make lensectomies more likely.[@b20-opth-12-2053] Larger GRTs were not more likely to be managed with lensectomies in this series, and while other studies have not compared rate of lensectomy with GRT size, the decision to perform lensectomy is multifactorial and subject to the surgeon's bias. The limitations of this study include its retrospective nature, nonstandard management protocol among several surgeons, and variable follow-up. The individual surgeon determined the approach, and therefore, it is difficult to draw definitive conclusions on techniques. In spite of these limitations, the current study represents a contemporary surgical case series of patients with GRTs. Conclusion ========== GRTs are still managed with variable surgical techniques even within the same practice. No techniques have been established objectively, but there is a consensus on the use of perfluorocarbon liquids, while many other maneuvers are clinical impressions gleaned from individual surgeons' experiences. The current series affirms the high success rates that accompany a high prevalence of initial PPV, perfluorocarbon liquids use, common combined PPV/SB, and either silicone oil or C~3~F~8~ that seem to have yielded reasonable results consistent with contemporary reports and experience.[@b7-opth-12-2053],[@b11-opth-12-2053],[@b21-opth-12-2053] Supported in part by the NIH core grant and an unrestricted grant from Research to Prevent Blindness, New York, NY. **Disclosure** The authors report no conflicts of interest in this work. {#f1-opth-12-2053} ###### Initial anatomic characteristics of patients undergoing surgery for GRT-related retinal detachment N (%) --------------------------------------- ----------- Extent of GRT  \<180° 60 (75)  ≥180° 20 (25) Average size of GRT, clock hours (SD) 4.5 (1.8) Macula status, n (%)  On 28 (35)  Off 52 (65) Lens status, n (%)  Phakic 61 (76)  Pseudophakic 17 (21)  Aphakic 2 (2.5) **Abbreviation:** GRT, giant retinal tear. ###### Intraoperative data for patients with giant retinal tear-related retinal detachments Intraoperative parameters N (%) ---------------------------- --------- Primary SB 3 (3.8) Primary PPV 16 (20) Combined PPV/SB 61 (76) Intraocular tamponade  1) SF~6~ 1 (1.3)  2) C~3~F~8~ 22 (28)  3) Silicone oil 54 (68) Perfluorocarbon liquid use 60 (78) **Abbreviations:** C~3~F~8~, perfluoropropane; PPV, pars plana vitrectomy; SB, scleral buckle; SF~6~, sulfur hexafluoride. ###### Perfluorocarbon liquid use, lensectomy, anatomic, and visual outcomes by GRT extent Extent of GRT \<180° N=60 (%) ≥180° N=20 (%) *P*-value -------------------------------------------- ----------------- ---------------- ----------- Perfluorocarbon liquid use 47 (78) 16 (80) 0.88 Lensectomy 4/14 (29) 5/47 (11) 0.10 Not reattached at last follow-up 2 (3.3) 2 (10) 0.24 Operation for recurrent retinal detachment 7 (12) 4 (20) 0.35 Preoperative VA (≥20/400) 32 (53) 11 (55) 0.90 Final VA (≥20/400) 43 (72) 12 (60) 0.34 **Abbreviations:** GRT, giant retinal tear; VA, visual acuity. ###### Surgical management of giant retinal tear-associated retinal detachments and anatomic outcomes over 13 years Intraoperative/postoperative parameters 2005--2010[a](#tfn4-opth-12-2053){ref-type="table-fn"} N=79 (%) 2011--2017 N=80 (%) 2005--2017 combined N=159 (%) ------------------------------------------------------------------------- ----------------------------------------------------------------- --------------------- ------------------------------- Scleral buckle/pars plana vitrectomy 67/79 (85) 61/80 (76) 128/159 (81) Perfluorocarbon liquid use 56/79 (71) 60/80 (75) 116/159 (73) Lensectomy (in phakic eyes)[b](#tfn5-opth-12-2053){ref-type="table-fn"} 23/56 (41) 9/61 (15) 32/117 (27) Silicone oil[c](#tfn6-opth-12-2053){ref-type="table-fn"} 45/79 (57) 54/80 (68) 99/159 (62) Recurrent retinal detachment 14/79 (18) 11/80 (14) 25/159 (16) Final reattachment 73/79 (92) 76/80 (95) 149/159 (94) VA ≥20/40 17/79 (22) 15/80 (19) 32/159 (20) VA ≥20/400 58/79 (73) 54/80 (68) 112/159 (70) **Notes:** Data from Gonzalez et al.[@b7-opth-12-2053] Lensectomy used during the initial surgical procedure. Silicone oil used during the initial surgical procedure. **Abbreviation:** VA, visual acuity.

1. Introduction {#S1} =============== Water is essential for sustaining all life forms and access to clean and safe drinking water is a basic human need. Latest estimates of the Joint Monitoring Programme for Water Supply and Sanitation state that the world is on track to meet the Millennium Development Goals target for drinking water of "halving by 2015 the proportion of people without sustainable access to safe water and basic sanitation" \[[@R1]\] \[[@R2]\]. But this progress has been uneven. An estimated 768 million people did not use an improved source for drinking-water in 2011 and 185 million still depended on surface water (from lakes, rivers, dams, or unprotected dug wells or springs) for their daily drinking-water needs \[[@R3]\] \[[@R4]\]. India has only 3% of the world's fresh water with 20% of its population. Although, 92% - 96% of the urban population could access improved sources of water by 2010, 4% - 8% continues to use unimproved sources \[[@R4]\]-\[[@R6]\]. The World Bank estimates that, in spite of the improvement in provision for drinking water, 21% of all communicable diseases in India are related to unsafe water \[[@R7]\], with diarrhoea alone causing more than one hundred thousand deaths annually \[[@R8]\]. Groundwater constitutes 85% of the source of drinking water in India \[[@R9]\] \[[@R10]\] and none of the major Indian cities have a continuous water supply \[[@R11]\]. The Ministry of Urban Development, Government of India commissioned a study in 1999 to assess the status of water supply, sanitation and solid waste management in 305 selected Metropolitan, Class I and Class II cities and towns (of population more than 100,000). This study found that the coverage by the formal water supply was 94%. Paradoxically, coverage did not ensure quantity, quality, duration of supply or the mode of provision. In spite of 100 percent coverage in twenty-two sampled urban centres, regular daily supply was not ensured due to acute water shortage. Six sampled urban centres did not have individual house service connections. In most of the sampled cities, the duration of water supply ranged between 1 and 6 hours daily. The study also found that 43% of the sampled urban centres depended entirely on surface water, 34% depended entirely on groundwater while 22% used both surface and groundwater sources \[[@R12]\] \[[@R13]\]. The city of Ahmedabad located on the banks of a seasonal river---Sabarmati, has begun to receive water from the Narmada canal of the Sardar Sarovar project since the year 2000. This now supplements the underground water from French wells and Bore wells. This development has slowed the groundwater table depletion, but groundwater extraction is still widespread, especially in areas not provided by municipal water supply \[[@R14]\]. We investigated the drinking water surveillance parameters in one vulnerable ward of Ahmedabad. The WHO defines drinking water surveillance as the "the continuous and vigilant public health assessment and review of the safety and acceptability of drinking-water supplies". It must include an assessment of quality, quantity, accessibility, affordability and continuity of the drinking-water supply \[[@R15]\]. 2. Methods {#S2} ========== Our selection of this study ward was based on the municipal health officials' categorisation of this particular ward as the most vulnerable in the city. It is situated within the old city limits with more than 85% of its population residing in slums. The data collection for the study took place in the summer of 2012. A questionnaire was developed based on the interviews we conducted with the municipal water supply officials. This was pilot tested in the neighbouring areas before being administered at the study site. The researchers made several visits to the study ward to understand the distribution of the household clusters and water supply across the ward. We used a mixed methodology of: 2.1. Interviews with Key Informants {#S3} ----------------------------------- We interviewed field level sanitary and health workers and supervisors posted at the ward office and health centre. We also interviewed water supply personnel (engineers and operators) at the main water storage and distribution facility located in our study ward and personnel at the Water Testing Laboratory. 2.2. Secondary Analysis and Mapping of Field Test Reports {#S4} --------------------------------------------------------- Field testing for water samples by the sanitary inspector is recorded on a standard format in a register including the date, source of water supply, site of sample collection and the result. We obtained this list of locations and results of water sample field tests in the study ward for a six month period from October 2011 to March 2012. These were analysed and mapped using ArcGIS software. 2.3. A Short Survey {#S5} ------------------- We obtained lists of population settlements from field health supervisor and used this to survey households in different types of tenements across the ward. The list divided the ward population of 65,403 into two groups; 86% of them living in 11,072 households (HHs) grouped into 47 slum clusters and 14% living in 4344 households (HHs) grouped into 42 housing society clusters. We sorted these cluster types separately in ascending order of population sizes. We sampled every alternate slum cluster from 47 slum clusters and every fifth housing society from 42 housing society clusters. (Slum clusters were oversampled because we were more interested in the poorer population of the ward.) We surveyed two HHs from every sampled cluster, one close to and one away from the municipal water line on the main road which supplies each cluster. This approach enabled us to assess the reach of the municipal water to the entire slum or society cluster and check for any disparities in supply between the proximal and distal households. The survey questionnaire was administered to a member of the household, the female head being the first preference. We finally surveyed 42 HHs out of 11,072 HHs in 21 slum clusters and 14 HHs from 4344 HHs in 8 housing society clusters. In two societies we sampled only one HH each because all the apartments were grouped around the water connection. Therefore the total tally was 56 households from 29 locations ([Figure 1](#F1){ref-type="fig"}). The data was analysed in MS-Excel and EpiInfo7. The data was stored securely on a hard disk and no personally identifiable information was collected. All participants were assured of confidentiality of their responses and a formal verbal consent was taken before starting the interview. Ethical clearance for the study was sought from the Institutional Ethics Committee of the Public Health Foundation of India. Support letters were also arranged from the Ahmedabad Municipal Corporation. 3. Results and Discussion {#S6} ========================= Our interviews revealed that Ahmedabad city receives approximately 1 billion litres of water per day, which averages out to nearly 180 litres per head. Supply is maintained for 2 hours every morning throughout the year (6 to 8 am) and during summers, for an additional one hour in the evenings. Water supply is not metered anywhere in Ahmedabad and therefore the actual quantity of water consumed by residents is not recorded. The Central Water Testing laboratory tests at least 200 to 250 routine water samples (3 to 4 from each of 64 wards) each day. Additional samples are processed based on public complaints and during focal outbreaks of waterborne diseases. These samples are subjected to the following tests---turbidity, alkalinity, hardness, ${Ca}^{2 +},{Mg}^{2 +},{Cl}^{-},{SO}_{4}^{2 -}$, total dissolved solids, before being tested for faecal indicators (total coliform and fecal *E. coli*). The Sanitary Inspector of every ward sends two water samples to the Central Water Testing laboratory and carries out 10 Orthotoludene field tests for residual chlorine from across the ward each day. The water samples for the field tests as well as the laboratory tests are collected during morning supply hours. There was no map of the existing water supply network and water sampling was not related to the physical organisation of the distribution system. Sampling is based on an arbitrary pattern of collecting samples from select slum clusters, municipal schools, public offices, etc. which has been set in practice over the years and on complaints from residents. Analysis and mapping of the records of 217 field tested water samples in the study ward over a six month period from October 2011 to March 2012 showed that 1) all samples had been reported fit; 2) most slum clusters were sampled repeatedly; 3) but none of the housing society clusters were covered by the sampling process ([Figure 2](#F2){ref-type="fig"}). The water supply department personnel of the ward examined the list of population settlements we had obtained from field health supervisor and marked out the most probable source of water (municipal supply, bore well or both) and the probable purification method used in each of the settlements. They estimated that of the 42 housing society clusters, only two supplement their municipal supply with bore-well water and one completely depends on bore well water for all of their water needs, and 12 of the 42 housing society clusters could afford small home-based Reverse Osmosis systems. Our survey showed this to be an underestimate. 3.1. Accessibility and Continuity {#S7} --------------------------------- Our site visit revealed that, except for a tiny cluster of 250 HHs, 3.2% of the ward population which drew water from common stand posts, municipal water supply reached into the smallest and poorest of homes. Our survey of half of all the slum clusters found that all slum households depended on municipal water supply. Of the 8 housing society clusters, two tapped into groundwater while the rest depended on municipal supply. 3.2. Quantity {#S8} ------------- A visual estimation of the volume of stored water showed all HHs stored an average of 35 to 50 litres of drinking water and 600 to 1300 litres for other purposes ([Table 1](#T1){ref-type="table"}). Of the 42 slum HHs, 16% (7 HHs) used either mechanical or motorised pumps during water supply hours to get additional water. Two of the 8 housing society clusters used motorised pumps to draw ground water into their storage tanks. The proportion of HHs that were dissatisfied with the quantity and/or pressure of water during summer was 45% in the slums (19 out of 42 HHs), and 50% (7 out of 14 HHs) in the society clusters. This dissatisfaction reduced to 26% in the slums HHs (11 out of 42 HHs) and 21% (3 out of 14 HHs) society HHs during the rest of the year ([Table 1](#T1){ref-type="table"}). 3.3. Quality {#S9} ------------ All slum and 80% of housing society households directly collected municipal water for drinking after simple sieving. The rest used Reverse Osmosis systems to purify water stored in overhead tanks for drinking. This city's supply of 1 billion litres of water per day averages to 180 litres per head. This falls within the range of 100 to 200 litres per capita, estimated by WHO as the amount likely to be collected when there is optimal access to water. At this level of water usage, public health risk due to poor hygiene is expected to be very low \[[@R15]\]. Our survey data bears out that the WHO water standards were met in the study ward. The water supply in Ahmedabad is unmetered and water tax is included in annual property tax bills. Thus, water is affordable in the city, the bottleneck being the restricted supply hours and poor pressure of water. Citizens, depending on their affluence level, circumvent these shortcomings using one or more methods; by storing large quantities of water, by applying hand or motorized pumps directly on the municipal water line or by tapping into groundwater sources. In our study ward, we found that poorer ward residents depended completely on the municipal water for both quality and quantity on a day to day basis. Each day, they separately collect and store water for drinking and for other purposes. Having larger families, and unable to afford storage facilities, they spent considerable time collecting and storing water. The richer households are able to ensure quality and quantity of water they require by investing in permanent storage solutions like cement or plastic tanks. They have the option of either collecting the direct municipal supply water or use various types of water purifiers for drinking. Interestingly, in these richer housing societies, filtered and chlorinated municipal water is collected in underground storage tanks. Here, it is mixed with groundwater and thus rendered non-potable. This mixture is pumped to overhead tanks, and some of this now non-potable water is processed through Reverse Osmosis systems or other commercial water purifiers in individual households, for drinking. The engineers responsible for water supply of the ward were unaware and unable to estimate how much piped water is consumed in the study ward and the extent of groundwater usage. They surmised that since this ward was upstream, there wouldn't be any water shortage in this area. The groundwater of Ahmedabad is known to have high levels of iron, fluoride, chloride and nitrates (\[[@R16]\] \[[@R17]\]). The Municipal Corporation as well as richer housing societies routinely resort to supplementing their water supply with ground water. The present drinking water surveillance system of the municipal corporation is designed for a twenty-four hour piped distribution of treated surface water. In reality, the water supply is neither 24 hours nor exclusively treated surface water. Therefore, as suggested by WHO \[[@R15]\], the water sampling process in the city needs to include periodic surveys of water actually consumed by the citizens. The Central Water Testing Laboratory needs to expand its laboratory capacity to include more refined tests for microbial and chemical contamination. The WHO drinking water quality guidelines recommend that for every 50,000 population in a city with more than five million populations, faecal indicator testing in distribution systems should be carried out on a minimum of 12 samples with an additional 600 samples every year. By these standards, if the entire city had had an interconnected water supply network with positive pressure in the system 24 hours of the day, the city would require to test 2040 samples every year \[[@R15]\]. However, the Central Water Testing Laboratory tests a minimum of 224 water samples for faecal indicators every day and 67,200 samples every year. The water samples are subjected to only basic physical and chemical analysis. Capacity to test heavy metals (lead, cadmium, mercury, arsenic, and nitrates), trihalothanes and fluorides and advanced microbial tests for bacteriophages and spores are presently not available. 3.4. Strengths and Weakness of Our Study {#S10} ---------------------------------------- We studied only a single ward of the city, therefore, the generalizability of these findings for the city are limited. Also the ward is located close to the water supply source for the city so the researchers suspect a better water availability in the study ward than the rest of the city. The investigators did not look for the existence of water vendors and thus cannot comment on purchase of water (and its financial implications) in times of non-supply of municipal water. 4. Conclusion {#S11} ============= In conclusion, the water sampling strategy for the city of Ahmedabad needs to include periodic surveys of water actually consumed by the community and the testing strategy needs to incorporate tests for viruses, heavy metals, fluorides, trihalothanes and surface contaminants at appropriate intervals of time and distance in piped water, using maps of the water distribution systems. There is a need for consultations with experts to adjust the numbers and sophistication of tests to appropriately reflect available scientific knowledge about Drinking Water Quality Surveillance. Also, a practice of Sanitation Risk Scoring by Sanitary Inspectors of the city needs to be developed and implemented. Although there are official references to reducing groundwater exploitation \[[@R17]\], guidelines need to be developed for tapping into underground aquifers in the country. The authors gratefully acknowledge Mr. Ashish Upadhya and, Mr. Ajit Rajiva for their contributions in developing the map. We thank the officials from the Municipal Corporation for extending us all the cooperation. We thank Dr. V. S. Saravanan from the University of Bonn for his valuable suggestions. **Funding** This work is supported by a Wellcome Trust Capacity Strengthening Strategic Award to the Public Health Foundation of India and a consortium of UK universities. {#F1} {#F2} ###### Salient findings from the households' survey of the ward. --------------------------------------------------------------------------------------------------------------------------------------------------- Slum clusters Society clusters ---------------------------------------------------------------------------------------- ------------------- -------------------------------------- No. of slums/societies 21 8 (+1 refused) No. of HHs 42 14 Persons per HH 6 5 **Source of drinking water** Municipal supply 6 municipal supply and\ 2 supplemented with private borewell **Quantity** Water stored (lt/HH) Drinking 41 28 Cleaning 822 1320 Per capita presence of drinking water (litres) 7 6 Per capita presence of other water (litres) 143 286 HHs complaining of low pressure and/or less quantity supply in summer rest of the year 19 (45%) 11 (26%) 7 (50%) 3 (15%) HHs using pumps 7 (16.67%) 2 (14.3%) **Drinking water practices** Municipal water used directly for drinking 42 (100%) 11 (75% ) Earthenware pots used for storing drinking water 42 (100%) 13 (92.9%) ---------------------------------------------------------------------------------------------------------------------------------------------------

Abedi E, Pourmohammadi K, Abbasi S. Dual‐frequency ultrasound for ultrasonic‐assisted esterification. Food Sci Nutr. 2019;7:2613--2624. 10.1002/fsn3.1115 1. INTRODUCTION {#fsn31115-sec-0001} =============== Starch is an accessible, thickening, and texturizing stabilizer, and of paramount importance when it comes to augmenting the overall quality of the products, reducing costs, and facilitating the processing procedure (Berski et al., [2011](#fsn31115-bib-0009){ref-type="ref"}; Choi & Kerr, [2003](#fsn31115-bib-0013){ref-type="ref"}); however, due to the high affinity toward retrogradation, its application is restricted in certain food products (Kaur, Ariffin, Bhat, & Karim, [2012](#fsn31115-bib-0027){ref-type="ref"}; Kaur, Sandhu, & Lim, [2010](#fsn31115-bib-0029){ref-type="ref"}). In frozen foods and desserts containing starch, resistance to syneresis of water, resistance to freeze and thaw, and reduction in retrogradation are important issues. In this regard, native starch was modified to overcome this limitation. By introducing the functional groups into the starch molecules, chemically modified starches, such as oxidized, hydroxypropylated, and acetylated starches, were employed. When starch is modified, its gelatinization, pasting, and retrogradation characteristics undergo different changes (Kaur et al., [2012](#fsn31115-bib-0027){ref-type="ref"}, [2010](#fsn31115-bib-0029){ref-type="ref"}; Singh, Kaur, & McCarthy, [2007](#fsn31115-bib-0044){ref-type="ref"}). Acetylated starch (AC) was introduced with CH~3~CO group which caused a cleavage in the hydrogen bonds inside the starch chain and resulted in amphiphilic properties (Hong, Chen, Zeng, & Han, [2016](#fsn31115-bib-0025){ref-type="ref"}). AC with a low degree of substitution (DS) is used as an emulsifier, coating, and thickening agent, which is stable and resistant to retrogradation (Chi et al., [2008](#fsn31115-bib-0012){ref-type="ref"}; Singh, Chawla, & Singh, [2004](#fsn31115-bib-0045){ref-type="ref"}). Moreover, in such kinds of modified starch, solubility, swelling power, viscosity, hardness, adhesiveness, cohesiveness, and translucency of the gels are increased, while the initial gelatinization temperature is reduced (González & Perez, [2002](#fsn31115-bib-0018){ref-type="ref"}). The replacement of the modifier groups may boost the free movement of the starch chains within the amorphous regions in the granule, owing to the disruptions taking place among the inter‐ and intramolecular hydrogen bonds. The weakened internal bond structure in the starch granules, due to the derivatized modifier groups, enhances the freeze--thaw stability, reduces the gelatinization temperature, generates high levels of peak viscosity, and leads to starch paste clarity (Han, [2010](#fsn31115-bib-0021){ref-type="ref"}; Han, Lee, Lim, & Lim, [2005](#fsn31115-bib-0022){ref-type="ref"}). Ultrasonic treatments can accelerate chemical reaction and increase yields. Under sonication, the surface area of starch granules increases due to the formation of channels and/or holes on the surface and inside the granules (Huang, Li, & Fu, [2007](#fsn31115-bib-0026){ref-type="ref"}). Several modifications have been performed upon ultrasonication, including acetylation of dioscorea starch following ultrasound treatment (Zhang, Zuo, Wu, Wang, & Gao, [2012](#fsn31115-bib-0055){ref-type="ref"}), octenyl succinate starch (Chen, Huang, Fu, & Luo, [2014](#fsn31115-bib-0011){ref-type="ref"}), carboxymethyl starch (Gao et al., [2011](#fsn31115-bib-0017){ref-type="ref"}; Shi & Hu, [2013](#fsn31115-bib-0042){ref-type="ref"}), and octenylsuccinates of carboxymethyl starch (Čížová, Sroková, Sasinková, Malovíková, & Ebringerová, [2008](#fsn31115-bib-0014){ref-type="ref"}). However, there is no information about the acetylation of starch under various frequencies of ultrasonication. The aim of this study was to optimize the AC wheat starch in various parameters such as acetic anhydride concentration, pH, and temperature under frequencies (25, 40, and 25 + 40 kHz) of bath ultrasound; the optimization was further investigated on the amount of modification (molar substitution) in the modified starch. Based on the minimal reagent consumption, the optimized samples were selected and their physicochemical properties (solubility, water absorption), X‐ray diffraction, scanning electron microscopy (SEM), Rapid Visco Analyzer (RVA), differential scanning calorimetry (DSC), and Texture Profile Analyzer (TPA) were investigated at three frequencies (25, 40, and 25 + 40 kHz). The freeze and thaw stability of AC wheat starch was further compared with that made with native wheat starch. 2. MATERIALS AND METHODS {#fsn31115-sec-0002} ======================== 2.1. Materials {#fsn31115-sec-0003} -------------- Native wheat starch was obtained from Fars Glucosin Company, Shiraz, Iran. The moisture, fat, ash, and protein of native (N) and AC wheat starch prepared under ultrasonication were specified along with the standard methods of AACC, [2000](#fsn31115-bib-0001){ref-type="ref"}. Amylose content (%) was calculated by iodine method (Pourmohammadi, Abedi, Hashemi, & Torri, [2018](#fsn31115-bib-0038){ref-type="ref"}). 2.2. Acetylation of wheat starch {#fsn31115-sec-0004} -------------------------------- We added 100 g of starch to 224 ml of distilled water and stirred for 1 hr at 24°C. The pH was set to 6.5, 8, and 9.5 with NaOH solution (1 M), and acetic anhydride (4, 6 and 8%) was slowly added to the suspension at various pHs. Whole acetylation reaction was performed at 10 min. The suspensions were sonicated at an amplitude of 24% (with an input power of up to 400 W and working frequencies of 25, 40, and 25 + 40 kHz) at 45, 60, and 75°C. The time duration of sonication in each frequency was 5 min (25 kHz: 5 min, 40 kHz: 5 min, and 25 + 40 kHz: 5 min). pH was then maintained at 4.5 with HCl solution (1 M) for 5 min. All the experiments were performed in bakers placed at known positions in ultrasonic bath (Pacisa SA) equipped with a temperature controlling system. The ultrasonic bath with internal dimensions of 300 × 150 × 150 mm^3^consisted of a rectangular tank containing four transducers at the bottom. The suspensions were completely washed, lyophilized, and milled using a laboratory mill (AlexanderWerk, Model WEL82), and manually sieved to obtain particle sizes \< 40 μm (Wojeicchowski, Siqueira, Lacerda, Schnitzler, & Demiate, [2018](#fsn31115-bib-0054){ref-type="ref"}). 2.3. Acetyl percentage {#fsn31115-sec-0005} ---------------------- The titrimetric method of Mbougueng, Tenin, Scher, and Tchiégang ([2012](#fsn31115-bib-0035){ref-type="ref"}) was employed in order to specify the acetyl group percentage (AC%) and (DS). AC (5.0 g), 50 ml of distilled water, and 24 ml of 0.45 M NaOH were added in a 240‐mL flask and mixed for 30 min at room temperature. The surplus alkali was back‐titrated using 0.2 M HCl and phenolphthalein as an indicator. The reaction mixture was put to stand for 2 hr, where the alkali was removed from the titrated sample. The native starch was also used as a blank sample. Initially, the acetyl (%) and DS were calculated according to Equations ([1](#fsn31115-disp-0001){ref-type="disp-formula"}) and ([2](#fsn31115-disp-0002){ref-type="disp-formula"}), respectively:$$\%{\mspace{6mu}\text{Acetylation}} = \lbrack\left( {V_{\text{B}} - V_{\text{S}}} \right) \times N_{\text{HCl}} \times 0.043 \times 100\rbrack/W$$ *V* ~B~ and *V* ~S~ are the volumes of the control and modified samples (0.2 M HCl basis), respectively. *N* ~HCl~is the normality of the consumed acid. *W* is the weight of the sample.$$\text{DS} = (162 \times {\text{Acetyl}\mspace{6mu}}\%)/\lbrack 4300 - (42 \times {\text{Acetyl}\mspace{6mu}}\%)\rbrack$$ 2.4. X‐ray diffraction {#fsn31115-sec-0006} ---------------------- To obtain a relative humidity of 75%, all starch samples were primarily placed, for 5 days, in a relative humidity box containing supersaturated NaCl at room temperature. X‐ray diffraction pattern of the samples was determined by use of an X‐ray diffractometer (Model D8 Advance, Germany) (Li et al., [2014](#fsn31115-bib-0032){ref-type="ref"}). Utilizing the instrument software (EVA, Version 9.0), the degree of starch crystallinity was specified through dividing the area under the peaks by the total curve area. 2.5. Solubility and water absorption {#fsn31115-sec-0007} ------------------------------------ The method of Li et al. ([2014](#fsn31115-bib-0032){ref-type="ref"}) was used to determine water swelling and solubility of the samples with a slight modification. Native and modified wheat starches (1 g) and distilled water (30 ml) were added into a centrifuge tube, and, using a vortex mixer set at high speed, they were shaken at room temperature for 5 min. Each tube was incubated in water bath for 30 min at 95°C followed by centrifugation for 15 min at 3000 g. The weights of the pellet and dried supernatants (at 120°C for about 2 hr) were further obtained. The water solubility and swelling values of the native and modified samples were calculated using Equations ([3](#fsn31115-disp-0003){ref-type="disp-formula"}) and ([4](#fsn31115-disp-0004){ref-type="disp-formula"}), respectively.$$\text{Water\ solubility}\left( \% \right) = \text{weight\ of\ dissolved\ solids\ in\ supernatant/weight\ of\ dry\ sample\ solids\ in\ the\ original\ sample} \times 100$$ $${\text{Water\ swelling}\mspace{6mu}}\left( \% \right) = \left( {\text{weight\ of\ sediment} \times 100} \right)/\left( {{\mspace{6mu}\text{weight\ of\ dry\ sample\ solids}} \times \left( {100 - \text{solubility}} \right)} \right)$$ 2.6. Pasting properties {#fsn31115-sec-0008} ----------------------- Using the method described by Pourmohammadi et al. ([2018](#fsn31115-bib-0038){ref-type="ref"}), the pasting properties of the native and modified wheat starches were determined through RVA (Newport Scientific Pty. Ltd) interfaced with a personal computer. 2.7. Thermal properties {#fsn31115-sec-0009} ----------------------- The gelatinization properties of the native and modified wheat starches were analyzed using a DSC instrument (OIT‐5000; Sanaf) equipped with STAR software making using of Pourmohammadi et al. ([2018](#fsn31115-bib-0038){ref-type="ref"}) method. The device was calibrated with indium and mercury, and an empty pan was used as a reference. Native and modified wheat starches (10 g) were weighed into the standard aluminum pan. Distilled water (30 ml) was added, and the pans were sealed and equilibrated for 24 hr at room temperature. After that, the starch slurry was gelatinized in the DSC instrument, by heating at 6°C/min, from 10 to 160°C. Following heat treatment, the samples were cooled to 25°C and then removed from the DSC. The changes in enthalpy (ΔH in J/g of dry starch), onset temperature (*T* ~o~), peak temperature (*T* ~p~), and conclusion temperature (*T* ~c~) for gelatinization were obtained from the exotherm DSC curves. 2.8. Microstructure determination {#fsn31115-sec-0010} --------------------------------- The microstructure of the modified starch was specified using SEM (Model Leica Cambridge) by use of the method proposed by Pourmohammadi et al. ([2018](#fsn31115-bib-0038){ref-type="ref"}). 2.9. Textural analysis {#fsn31115-sec-0011} ---------------------- Textural properties of starch gels were specified using a Texture Profile Analyzer (TA Plus; Stable Micro System). To prepare the paste of gelatinized starch, 10 g of starch was added to 100 ml of distilled water, where the slurry remained in water bath (95°C for 30 min). The hot paste was collected and transferred into a cylindrical plastic container of 10 mm diameter and 10 mm height, and kept at 4°C for 24 hr. The gels were then removed from the container and tested for their textural properties. Utilizing a cylindrical aluminum plunger with a diameter of 10 mm, the gels were packed at a test speed of 2 mm/s, pretest and post‐test speed of 5 mm/s, time interval of 10 s, and strain deformation of 24%. The recorded force--time plots were analyzed for the following: hardness (N), springiness (Length 2/ Length 1), cohesiveness, the ratio of the areas of the two resistance peaks (A2/A1), and gumminess according to Abedi, Majzoobi, Farahnaky, Pourmohammadi, and Mahmoudi ([2018](#fsn31115-bib-0002){ref-type="ref"}). Data were analyzed three times, and the results were averaged. 2.10. Freeze--thaw stability {#fsn31115-sec-0012} ---------------------------- With slight modifications in the methods described by Lawal ([2010](#fsn31115-bib-0030){ref-type="ref"}), we determined the syneresis of the starch gels belonging to the native and modified starches during cold and frozen storage. A starch suspension (5% dry basis, w/w) was heated at 95°C under constant agitation for 1 hr. 20 g of noodle strands belonging to N and AC starch was centrifuged at 2000 g for 10 min to remove free water. The free water (supernatant) was decanted, and the tubes containing starch paste were exposed to freeze and thaw cycles, followed by centrifugation at 3,500 g for 30 min. The freezing process was done at −18°C for 24 hr, and the melting process was carried out at 30°C for 4 hr. The two processes were repeated for 8 cycles, in each of which the isolated water was determined. The weight of water was taken, and the amount of syneresis was calculated as the percentage of water separated as Equation ([5](#fsn31115-disp-0005){ref-type="disp-formula"}):$$\text{Syneresis}\left( \% \right) = \text{Water\ separated}\left( g \right){\times 100)/\text{Total\ weight\ of\ sample}}\left( g \right)$$ 2.11. Statistical analysis {#fsn31115-sec-0013} -------------------------- So as to specify the regression coefficients, the Design‐Expert (version 6.0.5) methodology (Abedi, Sahari, Barzegar, & Azizi, [2015](#fsn31115-bib-0003){ref-type="ref"}) and statistical software packages SAS 9.1 (SAS Institute) were made. A Box--Behnken design from the response surface was used to study the effect of three different factors on the acetylation degrees of wheat starch. Factors associated with acetylation were acetic anhydride (4, 6, and 8%), pH (6.5, 8, and 9.5), and temperature (45, 60, and 75°C). The experimental data were fitted in accordance with Equation ([6](#fsn31115-disp-0006){ref-type="disp-formula"}) as a second‐order polynomial equation, including the linear and interaction effects of each factor:$$Y = \beta_{0} + \sum\limits_{i = 1}^{k}{\beta_{i}X_{i}} + \sum\limits_{i = 1}^{k}{\beta_{\mathit{ii}}X_{i}^{2}} + \sum\limits_{\begin{array}{l} {i = 1} \\ {i < j} \\ \end{array}}^{k - 1}{\sum\limits_{j = 2}^{k}{\beta_{\mathit{ij}}X_{i}}X_{j}}$$ where *Y* is the predicted response, *X~i~* and *X~j~* are independent factors, b0 is the offset term, bi is the ith linear coefficient, bii is the ith quadratic coefficient, and bij is the ijth interaction coefficient. All analyses were obtained in triplicates and reported as mean values. 3. RESULTS AND DISCUSSION {#fsn31115-sec-0014} ========================= 3.1. Molar substitution of AC wheat starch {#fsn31115-sec-0015} ------------------------------------------ Concerning the AC wheat starch at sonication frequencies of 25, 40, and 25 + 40 kHz, optimization was done in order to investigate the effect of acetic anhydride content, pH, temperature, and ultrasonic frequency on the amount of modification or molar substitution of the modified starch (Figure [1](#fsn31115-fig-0001){ref-type="fig"}).$$\text{Degree\ of\ acetylation}\left( \% \right)\text{at\ frequency\ 25\ kHz} = + 1.48 + 0.37 \times A + 0.18 \times B + 0.017 \times C + 0.17 \times A \times B + 0.16 \times A \times C + 0.043 \times B \times C - 0.26 \times A^{2} - 0.37 \times B2 - 0.50 \times C^{2}$$ $$\text{Degree\ of\ acetylation}\left( \% \right)\text{at\ frequency\ 40\ kHz} = + 1.02 + 0.32 \times A + 0.13 \times B + 0.13 \times C + 0.15 \times A \times B + 0.15 \times A \times C + 0.12 \times B \times C - 0.12 \times A^{2} - 0.29 \times B^{2} - 0.36 \times C^{2}$$ $$\text{Degree\ of\ acetylation}\left( \% \right){\text{at\ frequency}\mspace{6mu}}25 + 40\,\text{kHz} = + 1.52 + 0.35 \times A + 0.12 \times B - 0.14 \times C + 0.093 \times A \times B + 0.080 \times A \times C + 0.001 \times B \times C - 0.21 \times A^{2} - 0.33 \times B^{2} - 0.63 \times C^{2}$$ {#fsn31115-fig-0001} Optimized samples were selected based on minimum reagent consumption (acetic anhydride with achieving to maximum modification). According to response surface methodology (RSM) analysis, the optimized AC wheat starch was acetic anhydride (5%), pH 8.5, and temperature 60°C with 1.22%, 1.37%, and 1.02% acetylation for 25 kHz, 25 + 40 kHz, and 40 kHz, respectively. As a result of Equations ([(6)](#fsn31115-disp-0006){ref-type="disp-formula"}, [(7)](#fsn31115-disp-0007){ref-type="disp-formula"}, [(8)](#fsn31115-disp-0008){ref-type="disp-formula"}), the effect of A (acetic anhydride content) \> B (pH) \> C (reaction temperature) is on route to the maximum acetylation. The negative effect of temperature at dual frequency (25 + 40 kHz) was more than 25 and 40 kHz; however, the effect of temperature at 25 kHz frequency was the same as 40 kHz. The addition--elimination is the main mechanism for acetylation reaction, in which hydroxyl groups are converted to acetyl. The hydroxyl groups presented different levels of reactivity. The most reactive group was the hydroxyl of Carbon 6, followed by those of Carbons 2 and 3 (Wojeicchowski et al., [2018](#fsn31115-bib-0054){ref-type="ref"}). The C* = *O bond of the acetyl group was substituted in all amorphous areas and at the outer lamellae of crystalline sections due to the poor ability of acetic anhydride to penetrate granule structures (Singh et al., [2007](#fsn31115-bib-0044){ref-type="ref"}). The differences in the DS of modified starch are related to several parameters, such as starch source, amylose‐to‐amylopectin ratio, degree of crystallinity, molecular weight (MW) of amylose and amylopectin, granular size, presence of lipids, and reaction medium factors, such as pH, temperature, reaction time, mixing uniformity, and reagent content during the modification of the starches (Ayucitra, [2012](#fsn31115-bib-0006){ref-type="ref"}; González & Perez, [2002](#fsn31115-bib-0018){ref-type="ref"}; Halal et al., [2015](#fsn31115-bib-0019){ref-type="ref"}; Salcedo Mendoza, Hernández RuyDíaz, & Fernández Quintero, [2016](#fsn31115-bib-0040){ref-type="ref"}; Singh et al., [2007](#fsn31115-bib-0044){ref-type="ref"}; Sodhi & Singh, [2005](#fsn31115-bib-0047){ref-type="ref"}; Wang & Wang, [2002](#fsn31115-bib-0052){ref-type="ref"}). 3.2. Effect of ultrasonication {#fsn31115-sec-0016} ------------------------------ Under sonification, DS was increased since the solubility and swelling of granules were improved as a result of sufficient contact with reagents, which enhanced the homogeneity of the reactants. Collapse of cavitation bubbles upon ultrasonic treatment is capable of mechanically damaging and rupturing the starch granules and breaking the chains of polymers due to high‐pressure gradients, high local velocities of liquid layers, shearing force, and microstreaming. According to Amini, Razavi, and Mortazavi ([2015](#fsn31115-bib-0005){ref-type="ref"}) and Zhu ([2015](#fsn31115-bib-0057){ref-type="ref"}), ultrasound creating fracture, pore, and crack on glanular surfaces may enhance the surface area of starch, facilitate the penetration of esterifying agents (acetyl) into the granular structure, and accelerate the chemical reactions. The frequency of ultrasound is a key factor for enhancing cavitational bubble due to certain reasons (Abedi, Sahari, & Hashemi, [2017](#fsn31115-bib-0004){ref-type="ref"}): (a) Cavitation yield decreases with the increase in frequency, which is attributed to the scattering, attenuation, and shortening of the acoustic cycle at high frequencies (Abedi et al., [2017](#fsn31115-bib-0004){ref-type="ref"}). (b) Instable cavitation at low frequencies causes the bubble to collapse very quickly and violently, resulting in more rapid agitation and mass transfer (Abedi et al., [2017](#fsn31115-bib-0004){ref-type="ref"}). 3.3. Effect of temperature {#fsn31115-sec-0017} -------------------------- Temperature had a progressive effect on the DS of the modified starch, which is attributed to the swelling of starch granules, diffusion of the esterifying agents, mobility of reactant molecules for reaction with amorphous regions, destruction of the crystalline regions of the starch granules, and conversion to amorphous regions (Han et al., [2012](#fsn31115-bib-0020){ref-type="ref"}). This phenomenon can simplify the acceptability of granules in acetic anhydride and propylene glycol percolation. Temperature is an important factor, yet a very high temperature is not conducive to the absorption of the esterifying agent (acetyl) by the starch. This is because of the exothermic esterification reaction of starch (Han et al., [2012](#fsn31115-bib-0020){ref-type="ref"}; Singh et al., [2007](#fsn31115-bib-0044){ref-type="ref"}) and the facilitation of the gelatinization layer, which inhibits the contact of reagents with starch molecules (Han et al., [2012](#fsn31115-bib-0020){ref-type="ref"}; Singh et al., [2007](#fsn31115-bib-0044){ref-type="ref"}). 3.4. Effect of pH {#fsn31115-sec-0018} ----------------- DS increased following the addition of NaOH to the reaction medium, when the pH was set to 6.5, 8, and 9.5. Alkaline pH was able to enhance the DS owing to the disruption in the hydrogen bonds between molecules in amorphous and crystalline regions, and the facilitation of the percolation and penetration of the esterifying groups into the starch granular structure (Halal et al., [2015](#fsn31115-bib-0019){ref-type="ref"}; Han et al., [2012](#fsn31115-bib-0020){ref-type="ref"}). Although pH plays a major factor in the DS improvement of acetylation, very high pHs can diminish DS due to the hydrolyzation of acetic anhydride and the inhibition effect of gelatinized layer on the efficient contact between starch and acetic anhydride (Halal et al., [2015](#fsn31115-bib-0019){ref-type="ref"}; Han et al., [2012](#fsn31115-bib-0020){ref-type="ref"}). Various botanical starches have been acetylated with different degrees of acetylation, such as 0.104%--0.184% for maize starch (Singh et al., [2004](#fsn31115-bib-0045){ref-type="ref"}), 0.081 for waxy maize starch (Wang & Wang, [2002](#fsn31115-bib-0052){ref-type="ref"}), 0.087%--0.118% (Sodhi & Singh, [2005](#fsn31115-bib-0047){ref-type="ref"}) for rice starch, 0.041%, 0.059, and 0.076 for tapioca starch (Babic et al., [2007](#fsn31115-bib-0007){ref-type="ref"}), 1.85, 0.85, and 2.79% for corn starch (Chi et al., [2008](#fsn31115-bib-0012){ref-type="ref"}), 0.11 and 0.05 with 6 and 8% acetic acid consumption for oat starch (Mirmoghtadaie, Kadivar, & Shahedi, [2009](#fsn31115-bib-0036){ref-type="ref"}), and 0.9 and 2.7% for barely starch (Bello‐Pérez, Agama‐Acevedo, Zamudio‐Flores, Mendez‐Montealvo, & Rodriguez‐Ambriz, [2010](#fsn31115-bib-0008){ref-type="ref"}), 0.08%--0.21% for corn starch (Ayucitra, [2012](#fsn31115-bib-0006){ref-type="ref"}). Potato, oat, wheat, maize, and rice starches presented significant differences in their DS when acetylated under similar reaction conditions (Mirmoghtadaie et al., [2009](#fsn31115-bib-0036){ref-type="ref"}; Singh et al., [2007](#fsn31115-bib-0044){ref-type="ref"}). The high molar substitution acetylation in this study was interpreted by reaction medium under ultrasonic process, which affects the amylose regions. The extent of modification by ultrasonication in the crystallinity of starch granules depends on experimental conditions and the type of starch (Zhu, [2015](#fsn31115-bib-0057){ref-type="ref"}), and requires the introduction of acetyl group throughout the amylose or amylopectin sections. According to Singh et al. ([2004](#fsn31115-bib-0045){ref-type="ref"}), with low amylose content (low amylose/amylopectin ratio) a very high DS of acetylation can be observed. Surface of granule is another major factor affecting DS. Furthermore, González and Perez ([2002](#fsn31115-bib-0018){ref-type="ref"}) interpreted that the low degree of substitution in rice starch might be due to the lack of enough large inner channels or granular surface pores which can facilitate the access of acetic anhydride to the interior of the granule. 3.5. The crystalline structure {#fsn31115-sec-0019} ------------------------------ The X‐ray diffraction pattern of native and modified acetylated wheat starches is presented in Figure [2](#fsn31115-fig-0002){ref-type="fig"}. X‐ray diffraction analysis estimates the crystalline structure and the amount of crystal existing in the starch granules. The structure is related to amylopectin double helix. The results showed that the X‐ray pattern of wheat starch was the same as other cereals (A pattern). The native starch had sharp diffraction peaks at 15°, 17°, 18°, 23° (2θ), indicating the typical A pattern of cereal starch (Zobel, Young, & Rocca, [1988](#fsn31115-bib-0058){ref-type="ref"}). Following the esterification process, the degree of crystallinity was significantly reduced (*p* \< 0.05) from *N* (36.89 ± 0.62) to AC at 25 kHz (17.80 ± 0.38), 40 kHz (22.43 ± 0.42), and 25 + 40 kHz (13.57 ± 0.41) frequencies of wheat starches. The X‐ray diffraction showed that with acetylation, the structures of native starch were destroyed, and new structures of esterified starches were formed. Crystalline regions are the main factors preventing the loss of granular structure during reaction with reagents, and preserving the integrity of granular structure (Sha et al., [2012](#fsn31115-bib-0041){ref-type="ref"}; Singh et al., [2007](#fsn31115-bib-0044){ref-type="ref"}). Moreover, it is to be noted that starch granules in amorphous regions are more susceptible to reagents during the modification process, while crystalline regions remain intact. Acetylation under ultrasonic conditions created fractures and cracks on granular surfaces, thereby facilitating the penetration of acetic anhydride for reaction with amorphous (granule composed of amylose) and crystalline regions. The amylose content of *N* (26.30 ± 0.26) was reduced to 12.68 ± 0.12, 18.93 ± 0.24, and 9.23 ± 0.36 following acetylation at ultrasound frequencies 25, 40, and 25 + 40 kHz, respectively. This means that amorphous regions reacted with acetic anhydride, a finding which is in accordance with Diop, Li, Xie, and Shi ([2011](#fsn31115-bib-0016){ref-type="ref"}); Simsek, Ovando‐Martínez, Whitney, and Bello‐Pérez ([2012](#fsn31115-bib-0043){ref-type="ref"}); Mbougueng et al. ([2012](#fsn31115-bib-0035){ref-type="ref"}); Wani, Sogi, and Gill ([2012](#fsn31115-bib-0053){ref-type="ref"}). {#fsn31115-fig-0002} 3.6. Solubility and water absorption {#fsn31115-sec-0020} ------------------------------------ Figure [3](#fsn31115-fig-0003){ref-type="fig"} illustrates the water absorption (WA) and solubility of native and acetylated wheat starch at different temperatures. The solubility and WA of N and AC wheat starch at various frequencies differed significantly (*p* \< 0.05) at all measured temperatures. The solubility and water absorption of AC wheat starch at frequencies 25, 40, and 25 + 40 kHz were 7.37--36.77 and 6.30--23.98, 6.57--34.49, and 5.83--20.94, and 8--40.77 and 7.03--27.60, respectively, which is much higher than N wheat starch (2.28--9.03 and 2.28--12.72 at temperatures ranging between 30 and 90°C, respectively). Similar increasing trends in solubility and WA due to acetylation have been reported for potato and corn starches (Singh et al., [2004](#fsn31115-bib-0045){ref-type="ref"}), waxy and normal maize (Liu, Ramsden, & Corke, [1999](#fsn31115-bib-0033){ref-type="ref"}), small and large wheat starch (Van Hung & Morita, [2005a](#fsn31115-bib-0050){ref-type="ref"}), tapioca (Babic et al., [2007](#fsn31115-bib-0007){ref-type="ref"}), normal and waxy rice starch (Liu et al., [1999](#fsn31115-bib-0033){ref-type="ref"}), rice starch (González & Perez, [2002](#fsn31115-bib-0018){ref-type="ref"}; Sodhi & Singh, [2005](#fsn31115-bib-0047){ref-type="ref"}), barely (Bello‐Pérez et al., [2010](#fsn31115-bib-0008){ref-type="ref"}; Halal et al., [2015](#fsn31115-bib-0019){ref-type="ref"}), sweet potato (Lee & Yoo, [2009](#fsn31115-bib-0031){ref-type="ref"}), oat (Mirmoghtadaie et al., [2009](#fsn31115-bib-0036){ref-type="ref"}), and corn (Diop et al., [2011](#fsn31115-bib-0016){ref-type="ref"}; Han et al., [2012](#fsn31115-bib-0020){ref-type="ref"}). Further reported is the increasing pattern of swelling and solubility properties of various starch sources under ultrasonic processes (Chan, Bhat, & Karim, [2010](#fsn31115-bib-0010){ref-type="ref"}; Herceg et al., [2010](#fsn31115-bib-0023){ref-type="ref"}; Luo et al., [2008](#fsn31115-bib-0034){ref-type="ref"}; Montalbo‐Lomboy et al., [2010](#fsn31115-bib-0037){ref-type="ref"}; Režek Jambrak et al., [2010](#fsn31115-bib-0039){ref-type="ref"}; Sujka & Jamroz, [2013](#fsn31115-bib-0048){ref-type="ref"}; Zheng et al., [2013](#fsn31115-bib-0056){ref-type="ref"}). Such increase is probably due to the destruction, disorganization, and reduction in the crystallinity degree of starch granules. {#fsn31115-fig-0003} The acetyl group disorganized the starch components upon acetylation due to the following: (a) disruption of hydrogen bonds in the starch granules and facilitation of water access to the amorphous region (Babic et al., [2007](#fsn31115-bib-0007){ref-type="ref"}; González & Perez, [2002](#fsn31115-bib-0018){ref-type="ref"}; Liu et al., [1999](#fsn31115-bib-0033){ref-type="ref"}), (b) repulsion force between starch molecules as well as starch chain, (c) partial depolymerization principally in the amylopectin, reducing the average molecular weight (MW) (Bello‐Pérez et al., [2010](#fsn31115-bib-0008){ref-type="ref"}; González & Perez, [2002](#fsn31115-bib-0018){ref-type="ref"}), (d) loss of granule crystallinity based on the X‐ray diffraction pattern (Bello‐Pérez et al., [2010](#fsn31115-bib-0008){ref-type="ref"}; Van Hung & Morita, [2005a](#fsn31115-bib-0050){ref-type="ref"}, [2005b](#fsn31115-bib-0051){ref-type="ref"}; Liu et al., [1999](#fsn31115-bib-0033){ref-type="ref"}), and (e) reduced interactions between the starch chains due to the introduction of acetyl groups, solubilizing and releasing the amylose to the exterior of the swollen starch granule (Salcedo Mendoza et al., [2016](#fsn31115-bib-0040){ref-type="ref"}). According to Figure [3](#fsn31115-fig-0003){ref-type="fig"}, temperature had a progressing effect on solubility and swelling of N and AC starch. The increase in the temperature of the medium induces the thermodynamic activity of starch molecules; moreover, a high granular mobility can increase the water penetration in the amorphous regions of starch granules and destroy the crystalline regions as well (Mirmoghtadaie et al., [2009](#fsn31115-bib-0036){ref-type="ref"}). 3.7. The pasting properties {#fsn31115-sec-0021} --------------------------- Figure [4](#fsn31115-fig-0004){ref-type="fig"} and Table [1](#fsn31115-tbl-0001){ref-type="table"} present the RVA profile of N and AC starch suspensions (5%, w/w) at various frequencies (25, 40, and 25 + 40 kHz) during heating from 30 to 90°C. The peak viscosity is indicated by the maximum swelling of the majority of starch granules and their subsequent collapse (Van Hung & Morita, [2005a](#fsn31115-bib-0050){ref-type="ref"}, b). The peak and final viscosity of native starch (3,757 ± 48 and 7,034 ± 34) were found to be significantly (*p* \< 0.05) lower than AC starches at frequencies 25 (7,188 ± 32 and 8,851 ± 31), 40 (5,846 ± 65 and 7,648 ± 66), and 25 + 40 kHz (9,423 ± 112 and 11,174 ± 131). The AC at frequency 25 + 40 kHz (50.9 ± 1.2) had lower pasting temperatures compared with 25 kHz (58.3 ± 1.4), 40 kHz (64.4 ± 1.2), and *N* (71.6 ± 1.6). These findings are in accordance with the results of Li et al. ([2014](#fsn31115-bib-0032){ref-type="ref"}) where the starch with higher amylose content reduced the gelatinization temperatures due to less crystalline and more amorphous regions (Li et al., [2014](#fsn31115-bib-0032){ref-type="ref"}). These results are in accordance with certain other research (González & Perez, [2002](#fsn31115-bib-0018){ref-type="ref"}; Van Hung & Morita, [2005a](#fsn31115-bib-0050){ref-type="ref"}, b; Liu et al., [1999](#fsn31115-bib-0033){ref-type="ref"}; Salcedo Mendoza et al., [2016](#fsn31115-bib-0040){ref-type="ref"}). In these studies, the modification of starch by acetylation increased the peak viscosity due to (a) the decrease in the strength of associative intermolecular forces in the amorphous regions of the granules, (b) lower degrees of crystallinity in AC compared with N wheat starches, (c) increase in solubility and swelling power (Figures [3](#fsn31115-fig-0003){ref-type="fig"} and [4](#fsn31115-fig-0004){ref-type="fig"}), (d) the solubilizing amylose released outside the swollen starch granule (Han, [2010](#fsn31115-bib-0021){ref-type="ref"}; Salcedo Mendoza et al., [2016](#fsn31115-bib-0040){ref-type="ref"}), and (e) improvement in the hydrophilic capacity of the reorganized structures (Han, [2010](#fsn31115-bib-0021){ref-type="ref"}; Salcedo Mendoza et al., [2016](#fsn31115-bib-0040){ref-type="ref"}). On the other hand, compared with the native starches, the AC wheat starches presented better stability during heating. Reassociation of amylose molecules occur when leached out from swollen starch granule, thus providing a higher resistance to thermal or mechanical forces (Colussi et al., [2015](#fsn31115-bib-0015){ref-type="ref"}). {#fsn31115-fig-0004} ###### Pasting profile (RVA), thermal (DSC), and textural (TPA) analysis of native (N), acetylated (AC), wheat (dry basis) starches   RVA parameters Thermal parameters TPA parameters ------------------------------- ---------------- -------------------- ---------------- ------------- ------------- --------------- --------------- ---------------- --------------- --------------- ---------------- -------------- ---------------- --------------- *N* wheat starch 71.6 ± 1.6^a^ 3,758±48^d^ 7,034±34^d^ 1,214±84^d^ 3,576±38^a^ 56.8 ± 0.8^a^ 62.0 ± 1.4^a^ 67.4 ± 1.3^a^ 11.8 ± 0.8^a^ 1,420 ± 34^a^ 0.93 ± 0.05^a^ −234 ± 13^a^ 1.18 ± 0.09^a^ 1,321 ± 85^a^ AC wheat starch (25 kHz) 58.3 ± 1.4^c^ 7,188±32^b^ 8,851±31^b^ 3,102±18^b^ 1663±53^d^ 48.2 ± 1.0^c^ 53.4 ± 1.3^c^ 59.4 ± 1.84^c^ 6.4 ± 0.8^c^ 634 ± 28^c^ 0.87 ± 0.05^a^ −312 ± 38^c^ 0.71 ± 0.04^c^ 852 ± 34^c^ AC wheat starch (40 kHz) 64.4 ± 1.2^b^ 5,846±65^c^ 7,648±66^c^ 2,492±16^c^ 1802±67^b^ 51.0 ± 0.8^b^ 57.6 ± 1.2^b^ 62.1 ± 1.4^b^ 9.5 ± 0.4^b^ 896 ± 48^b^ 0.91 ± 0.03^a^ −281 ± 24^b^ 0.95 ± 0.08^b^ 936 ± 52^b^ AC wheat starch (25 + 40 kHz) 50.9 ± 1.2^d^ 9,423±112^a^ 11,174±131^a^ 4,011±24^a^ 1753±53^c^ 44.9 ± 1.4^d^ 50.8 ± 1.1^d^ 54.6 ± 1.8^d^ 5.8 ± 0.8^d^ 574 ± 31^d^ 0.86 ± 0.09^a^ −327 ± 30^d^ 0.62 ± 0.07^d^ 616 ± 43^d^ Different letters in each column show significant statistical difference between the values (*p* \< 0.05). John Wiley & Sons, Ltd 3.8. Thermal behavior {#fsn31115-sec-0022} --------------------- Table [1](#fsn31115-tbl-0001){ref-type="table"} summarizes the DSC gelatinization parameters including transition temperatures (To, Tp, and Tc) and enthalpy of gelatinization (ΔH gel) related to N and AC (under sonication at frequencies 25, 40, and 25 + 40 kHz) wheat starches. It was observed that these values were lower in AC at frequency 25 + 40 kHz compared with AC at frequency 25 and 40 kHz and N starches. Modification induced a major reduction from 56.8 ± 0.8°C for N wheat starch to 48.2 ± 1.04°C, 51.0 ± 0.8°C, and 44.9 ± 1.4 for AC wheat starches at frequencies 25, 40, and 25 + 40 kHz, respectively. Among the starches, N starch presented the highest ΔH~gel~ value (11.8 ± 0.8 J/g), followed by AC wheat starches at frequency 25 kHz (6.4 ± 0.8), 40 kHz (9.5 ± 0.4), and 25 + 40 kHz (5.8 ± 0.8). Singh et al. ([2004](#fsn31115-bib-0045){ref-type="ref"}), Babic et al. ([2007](#fsn31115-bib-0007){ref-type="ref"}), and Lee and Yoo ([2009](#fsn31115-bib-0031){ref-type="ref"}) also reported a statistically significant (*p* \< 0.05) decrease in the gelatinization temperature of corn, potato, tapioca, and sweet potato starches following acetylation. The starch gelatinization is controlled by two factors: (a) amylopectin molecular structures such as molar mass, integrity and order of crystallites, polydispersity, and amylopectin chain length with a degree of polymerization (DP) ranging between 5 and 12, and (b) granule structures including granule size, amylose/amylopectin ratio, crystalline zone relationship, and the quality and amount of amorphous crystallites (Tester & Morrison, [1990](#fsn31115-bib-0049){ref-type="ref"}). The enthalpy value attributed to the ordered double‐helix loss was more than the crystallinity loss. The difference in enthalpy value between the N and modified wheat starch showed that the acetylation of wheat starch with a high substitution degree generated disorganization as is shown by X‐ray, SEM, and RVA experiments. The reduction in the gelatinization parameters of acetylated starches was due to the insertion of acetyl and propyl oxide groups into the starch molecules, particularly into the amorphous areas. This further reduced the integrity of the amorphous and crystalline sections of starch granules and disrupted the inter‐ and intramolecular hydrogen bonds, destabilizing the granular structure (Babic et al., [2007](#fsn31115-bib-0007){ref-type="ref"}). ΔH indicates an overall measure of crystallinity and loss in the molecular order. The esterifying agents reduce ΔH due to (a) the change in the amylopectin double helices, (b) diminished crystallinity, (c) increase in starch chain flexibility, and (d) reduction in gelatinization temperature (Halal et al., [2015](#fsn31115-bib-0019){ref-type="ref"}). These findings are consistent with Halal et al. ([2015](#fsn31115-bib-0019){ref-type="ref"}); Bello‐Pérez et al. ([2010](#fsn31115-bib-0008){ref-type="ref"}); Mirmoghtadaie et al. ([2009](#fsn31115-bib-0036){ref-type="ref"}); Kaur, Singh, and Singh ([2004](#fsn31115-bib-0028){ref-type="ref"}). 3.9. The effects of modification on starch morphology {#fsn31115-sec-0023} ----------------------------------------------------- The morphological patterns of N and modified starches are shown in Figure [5](#fsn31115-fig-0005){ref-type="fig"} where A and B granules of wheat starch can be clearly observed. In AC wheat starches at dual frequency (25 + 40 kHz), the surfaces of all large granules were more damaged after modification, compared with those of the small granules. Meanwhile, at frequency 25 and 40 kHz of wheat starch, small granules were more affected than large granules. Starch granules indicated a partial disintegration of granules with a slight gelatinization when esterification reactions were performed in the presence of catalyst (NaOH solution; Halal et al., [2015](#fsn31115-bib-0019){ref-type="ref"}). According to Hirsch and Kokini ([2002](#fsn31115-bib-0024){ref-type="ref"}), the cross‐linked bonds formed by acetic anhydride mostly occur on the surface of the starch granules. {#fsn31115-fig-0005} Modification by ultrasonication influenced the granule morphology through forming deep grooves in the central core zone, cracks on the starch granules, and blister‐like and doughnut‐like shapes, which is in line with the SEM morphology of Kaur et al. ([2004](#fsn31115-bib-0028){ref-type="ref"}). The major impact of acetylation on the surface of the granules is due to the reaction of acetic anhydride with starch granules under ultrasonic processes. Ultrasonic‐induced cracks, pores, and fractures on granular surfaces facilitate the penetration of chemical reagents into structure granules. According to the results of Bello‐Pérez et al. ([2010](#fsn31115-bib-0008){ref-type="ref"}), esterification of starch generated higher granule fusion, which is attributed to the increase in hydrogen bonding due to the insertion of hydrophilic groups into the starch molecules. Accordingly, coalescing starch molecules induce the fusion of starch granules (Bello‐Pérez et al., [2010](#fsn31115-bib-0008){ref-type="ref"}), which is in line with the SEM morphology in the present study (Figure [5](#fsn31115-fig-0005){ref-type="fig"}). The SEM results are also consistent with Van Hung and Morita ([2005a](#fsn31115-bib-0050){ref-type="ref"}, b). 3.10. Gel texture {#fsn31115-sec-0024} ----------------- Table [1](#fsn31115-tbl-0001){ref-type="table"} demonstrates the texture profiles of N and AC wheat starches under various frequency ultrasounds. The order of the hardness of AC wheat starch was as follows: 25 + 40 kHz \< 25 kHz \< 40 kHz. Acetylation was able to reduce the hardness of the gels due to (a) the introduction of large and hydrophilic acetyl groups which aided ultrasonication via increasing the space between the starch chain molecules and reducing retrogradation, (b) slight depolymerization of the starch molecules by the acetic anhydride under sonication (although more amylose molecules are reassociated, the depolymerization can weaken the gel), and (c) acetyl group preventing the formation of amylose gel network and the reassociation of amylose and amylopectin molecules (reducing retrogradation). There were no significant differences (*p* \< 0.05) between N and AC wheat starches as far as cohesiveness is concerned; however, their gumminess was reduced. Adhesiveness is the attraction between the gel and an external surface. Adhesiveness of N wheat starch gel was significantly (*p* \< 0.05) higher than AC wheat starch gels, which is in line with Liu et al. ([1999](#fsn31115-bib-0033){ref-type="ref"}), Choi and Kerr ([2003](#fsn31115-bib-0013){ref-type="ref"}), Babic et al. ([2007](#fsn31115-bib-0007){ref-type="ref"}), and Colussi et al. ([2015](#fsn31115-bib-0015){ref-type="ref"}). 3.11. Freeze--thaw stability {#fsn31115-sec-0025} ---------------------------- The purpose of the present research was to stabilize the starch gels from syneresis during freezing and thawing. The tendency to retrogradation of gels prepared from N and AC wheat starches was calculated by determining the syneresis (percentage of water loss) during storage at 4°C (Figure [6](#fsn31115-fig-0006){ref-type="fig"}). During the freezing cycles, the crystal size was enlarged with the reduction in their number. In the thawing stage, the ice crystals were converted to water masses easily removed from the polymer network (syneresis) (Smith & Schwartzberg, [1985](#fsn31115-bib-0046){ref-type="ref"}). Furthermore, water leakage led to the formation of sponge texture in starch. N starch gels started to retrograde following 24 hr, indicated by the increase in the percentage of water from 42.64% to 63.47% during storage (8 cycles). Acetylated starch gels prevented the separation of water from starch noodle even after five freeze--thaw treatments, showing a better freeze--thaw stability due to high DS. The percentage of water separated from AC starch gels at frequencies 25 kHz, 40 kHz, and 25 + 40 kHz reached 1.10%, 3.36, and 0.39%, respectively, indicating the lower increment in syneresis compared with the N wheat starch gel, as illustrated in Figure [6](#fsn31115-fig-0006){ref-type="fig"}. The highest resistance to freeze--thaw cycle(s) was observed in AC wheat starch at frequency 25 + 40 kHz (lowest % separated water), while the lowest belonged to N wheat starch (highest % separated water). Syneresis in freeze‐thawed gels is due to the reassociation of amylose molecules together and with amylopectin chain in starch granules at reduced temperatures which accelerates the exclusion of water from the gel structure. Amylose association takes place at the initial step of storage, while amylopectin association occurs in the following stages. Retrogradation may occur when a gelatinized starch is cooled. The increase in the amount of separated water indicates the lack of freeze--thaw stability. After acetylation, water molecules were absorbed by functional groups and were not able to freeze, thereby reducing syneresis (Han et al., [2012](#fsn31115-bib-0020){ref-type="ref"}). On the other hand, the introduction of large groups of acetyl into the starch chains reduced the inner and outer bonds. Moreover, the acetyl group is smaller in size, which is effective on water absorption and stability against water exclusion. By inserting the acetyl group, the hydrophilic nature of these reagents resulted in the electrostatic repulsion between the chains, preventing the amylose--amylose and amylose--amylopectin interactions (Kaur et al., [2004](#fsn31115-bib-0028){ref-type="ref"}; Lawal, [2010](#fsn31115-bib-0030){ref-type="ref"}). When the amylose molecules carrying acetyl group are mingled with amylopectin, the acetyl group sterically prevents the aggregation of amylopectin, entailing lower retrogradation and reduction in the percentage of water separated during freeze--thaw cycles (Kaur et al., [2004](#fsn31115-bib-0028){ref-type="ref"}; Lawal, [2010](#fsn31115-bib-0030){ref-type="ref"}). {#fsn31115-fig-0006} 4. CONCLUSIONS {#fsn31115-sec-0026} ============== In order to increase the DS and consumption of minimum reagents, a reaction mixture was carried out under ultrasonic processes. The molar substitution optimizing AC wheat starch was 1.22%, 1.37%, and 1.02% for 25 kHz, 25 + 40 kHz, and 40 kHz, consuming 5% of acetic anhydride. High DS is due to the ultrasonic conditions at dual frequency (25 + 40 kHz) which form fractures, pores, and cracks on the surface of granules, facilitating the penetration of reagents into granule starch. Acetylation ameliorates the solubility, water absorption, resistance to freeze and thaw cycle, and viscosity peak, and reduces the gelatinization temperature. CONFLICT OF INTEREST {#fsn31115-sec-0027} ==================== The authors declare that they do not have any conflict of interest. ETHICS STATEMENT {#fsn31115-sec-0028} ================ Human and animal testing is unnecessary in this study. The authors would like to thank Dr. Mohsen Ajdari, University of Fasa, for his technical assistance.

Introduction {#s1} ============ Lysosomes and the ubiquitin proteasome system (UPS) are major degradation machineries for damaged and unwanted cellular contents in eukaryotic cells. The UPS degrades damaged and unwanted proteins that are tagged with ubiquitin, whereas large protein aggregates and damaged or dispensable organelles are degraded in the lysosome. Besides housekeeping functions, regulatory roles of these two machineries in development and differentiation are emerging in a number of organisms. Malaria parasites critically rely on these two systems, as inhibitors of the UPS and of proteases of the food vacuole, the lysosome-equivalent of *Plasmodium*, have been shown to block parasite development. Hence, these two proteolytic machineries are attractive targets for antimalarial drug development \[[@B1]--[@B6]\]. Erythrocytic malaria parasites critically rely on hemoglobin for amino acids, which is hydrolyzed in the food vacuole by concerted action of several different types of proteases. The major food vacuole-resident hemoglobin degrading proteases are papain-like cysteine proteases falcipains, aspartic proteases plasmepsins, the metalloprotease falcilysin, dipeptidyl aminopeptidase I, and a M1-family alanyl aminopeptidase \[[@B7]--[@B13]\]. A large body of literature indicates that the papain-like cysteine proteases falcipains, particularly falcipain-2 (FP2) and falcipain-3 (FP3), are the major hemoglobin degrading enzymes in *P. falciparum* \[[@B5]\]. FP2, FP3 and their homologs in other malaria parasites prefer leucine at the P2 position in substrates and inhibitors \[[@B14]--[@B24]\]. Several peptidyl inhibitors of papain-like cysteine proteases, including leupeptin (acetyl-Leu--Leu--Arg-aldehyde, with P2 leucine and aldehyde warhead), block development of malaria parasites by inhibiting FP2 and FP3 \[[@B25]\]. This selectivity for a P2 leucine residue has been exploited for optimizing inhibitors of falcipain-like proteases of malaria parasites \[[@B26]--[@B28]\]. Malaria parasites appear to have a typical 26S proteasome \[[@B29]\]. A 26S proteasome is composed of two multi-subunit assemblies: a core protease complex, called the core particle (CP) or 20S proteasome; and a regulatory element, known as the 19S regulatory particle (RP) \[[@B30]\]. The CP is a barrel-shaped complex of four stacked rings, each with 7 α or 7 β subunits, which are arranged as "αββα". Three catalytically distinct β subunits (β1, β2, and β5) in each inner ring, whose active sites line the central lumen of the CP, form a proteolytic chamber where substrates are degraded. Substrates gain access to this chamber through the pores formed by α rings on either side of the CP \[[@B31],[@B32]\], which requires the RP for opening the pores and unfolding of the substrate. Active site mutagenesis of the three protease subunits together with the use of defined peptide substrates revealed that a typical 26S proteasome possesses three types of activities \[[@B33]--[@B35]\]: 1) caspase-like or peptidyl-glutamyl peptide hydrolyzing (PGPH) activity, which cleaves after acidic residues; 2) trypsin-like activity that cleaves after basic residues; and 3) chymotrypsin-like activity, which cleaves after large hydrophobic residues. The majority of available proteasome inhibitors, including MG132, epoxomicin, and lactacystin, block the chymotrypsin activity \[[@B6],[@B36]--[@B39]\]. Dual inhibition of falcipains and the UPS with optimal selectivity could offer higher potency and less risk of development of drug resistance by the parasite than individual inhibitors of these two proteolytic systems. One such compound might be MG132 (Z-Leu-Leu-Leu-CHO), as it contains P2 leucine, a falcipain preferred residue, and an aldehyde group that reacts with catalytic cysteine residue of cysteine proteases and threonine residues of protease units of the proteasome. MG132 is a first choice inhibitor for studying UPS in a variety of organisms, including malaria parasites. It is commonly used at micromolar concentrations to study the UPS in a variety of human cell lines, with reported 50% cytotoxic concentrations in the range of 2.5-21 µM depending on the cell type and treatment duration \[[@B40]--[@B43]\]. MG132 has also been shown to inhibit the papain-like cysteine proteases cathepsin L and B and the calcium dependent cysteine proteases calpains \[[@B44],[@B45]\]. However, the cysteine protease inhibitory property of MG132 remains underappreciated compared to its extensive use as an UPS inhibitor. Hence, we hypothesized that this compound is a dual-target inhibitor of malaria parasites, blocking hemoglobin degradation by inhibiting falcipains and also inhibiting UPS. To test this hypothesis, we assessed the effects of specific inhibitors of cysteine proteases (E64) and UPS (epoxomicin, lactacystin, and MG132) on parasite development, hemoglobin degradation, proteasome and cysteine protease activities in *P. falciparum* extracts, and activity of recombinant falcipains. We demonstrated that MG132 blocks asexual erythrocytic development of *P. falciparum* by inhibiting both hemoglobin degradation and the UPS. Materials and Methods {#s2} ===================== Materials {#s2.1} --------- The *P. falciparum* 3D7 strain was obtained from the Malaria Research and Reference Reagent Resource Centre (MR4). MG132, epoxomicin, and lactacystin were from Santa Cruz Biotechnology; all other biochemical reagents were from Sigma or Serva. Plasmid isolation kits were from Qiagen or MACHEREY-NAGEL; cell culture reagents were from Lonza and Invitrogen; restriction and DNA modifying enzymes were from New England Biolabs; and SYBR Green 1 was from Invitrogen. Human blood was collected from healthy volunteers after written consent under medical supervision at the medical dispensary of the institute, and the protocol (IEC-2/2010) for blood collection for this study has been approved by the Institutional Ethical Committee of CCMB. Parasite culture {#s2.2} ---------------- In vitro parasite culture was done according to the protocols approved by the Institutional Biosafety Committees of CCMB and UCSF. *P. falciparum* was cultured in human erythrocytes at 2% haematocrit in the presence of a gas mixture (5% CO~2~, 5% O~2~, and 90% N~2~) in RPMI 1640 medium supplemented with 41.1 mg/litre hypoxanthine, 300 mg/litre glutamine, 2.5% human serum, and 0.5% Albumax II \[[@B46]\]. Synchrony was maintained by serial treatment with 5% D-sorbitol \[[@B47]\]. For parasite isolation, the culture was centrifuged at 894g for 5 min, the supernatant was aspirated off, the pellet was treated with ice-cold 0.05% saponin (in PBS) for 5 min to lyse erythrocytes, and the sample was centrifuged at 12,096g for 5 min at 4°C. The supernatant was discarded, the pellet was washed twice with ice-cold PBS to remove residual hemoglobin, and parasites were recovered by centrifugation at 12,096g for 5 min at 4°C. Genomic DNA (gDNA) was isolated from late trophozoite/schizont stage parasites using the Puregene Blood Core Kit B (Qiagen). Parasite growth inhibition assays {#s2.3} --------------------------------- Inhibitors of the proteasome (epoxomicin, lactacystin, and MG132), cysteine proteases (L-trans-epoxysuccinyl-leucyl-amido(4-guanidino)butane; E64), and aspartic proteases (pepstatin) were assessed alone and in combinations for inhibition of *P. falciparum* erythrocytic stage development. 200 x stocks of inhibitors were made in DMSO and serially diluted 2-fold in 100 µl culture medium across rows of a 96 well tissue culture plate. Control wells contained DMSO (0.05%) or 500 nM chloroquine. 100 µl of parasite suspension (1% ring-infected erythrocytes at 4% haematocrit) was added to each well, and plates were incubated in a modular incubator chamber (Billups-Rothenberg, Inc.) with the gas mixture at 37°C for 50 hours. At the end of incubation, 100 µl culture medium was carefully removed from each well, and 100 µl lysis buffer (20 mM Tris-Cl, 5 mM EDTA, 0.008% saponin, 0.08% Triton X-100, pH 7.5) with SYBR Green 1 (at the manufacturer's recommended dilution) was added to each well. The plate was incubated at 37°C for 30 min, and fluorescence was measured (Ex: 485 nm, Em: 530 nm, gain setting: 50) using an Infinite M200 multimode microplate reader (TECAN) as described previously \[[@B48]\]. The fluorescence of chloroquine-treated culture was subtracted from inhibitor-treated and DMSO-containing cultures to account for background fluorescence. Fluorescence values of inhibitor-treated cultures were normalized as percentage of DMSO-treated cultures, plotted against inhibitor concentrations, and analyzed using nonlinear regression analysis to determine IC~50~ concentrations for the inhibitors as described earlier \[[@B12]\]. For assessing the effect of inhibitor treatment on parasite morphology, ring stage parasites were cultured with DMSO (0.1%) or inhibitors (21.7 µM E64, 0.024 µM epoxomicin, 0.820 µM lactacystin, 0.133 µM MG132, and 220.6 µM pepstatin; concentrations are nearly 3 times the IC~50~) for 48 hours. Samples were collected at various time points, and thin smears were stained with Giemsa and observed under a 100 x lens using a bright field microscope. To investigate the effect of inhibitors on different developmental stages, highly synchronized parasites (3% rings) were cultured for 60 hours in the presence of DMSO (0.25%) or inhibitors (21.7 µM E64, 0.024 µM epoxomicin, and 0.133 µM MG132; concentrations are nearly 3 times the IC~50~), which were added at different time points of the development. DMSO was added in the beginning (T0), and inhibitors were added in the beginning and at 12, 24 and 36 hour time points. Thin smears were prepared after 48 hours and 60 hours, stained with Giemsa, and observed using a bright field microscope under a 100 x objective. At least 200 parasites were observed to evaluate parasite morphologies and the number of different parasite stages. To determine if MG132 irreversibly blocks parasite development, inhibitor washout experiments were performed for three development cycles (^≈^150 hours). Ring stage parasites (3% parasitemia) were cultured in the presence of DMSO (0.25%) or chloroquine (500 nM) or inhibitors (21.7 µM E64, 0.024 µM epoxomicin, 0.133 µM MG132; concentrations are nearly 3 times the IC~50~) for 50 hours in a 96-well tissue culture plate (250 µl culture volume, four wells for each treatment). The culture media was removed at the end of the incubation period; cells were washed twice with fresh culture medium, and then resuspended in 250 µl of fresh culture medium. 100 µl of this culture was used for measurement of parasite growth by the SYBR Green 1 assay, 50 µl was transferred to a well of another 96-well plate containing 200 µl culture medium with fresh erythrocytes (at 2% haemetocrit) for assessing parasite viability and growth, and the remaining culture was used for making smears for Giemsa-staining. For measuring parasite viability and growth, the plate was incubated in a modular incubator chamber (Billups-Rothenberg, Inc.) with the gas mixture at 37°C for 50 hours, and then each culture was processed for measurement of parasite growth by the SYBR Green 1 assay, assessing parasite viability and growth in the next development cycle, and making smears for Giemsa-staining as described above. The fluorescence values of chloroquine-treated, inhibitor-treated, and DMSO-containing cultures were compared to assess parasite growth. In parallel, Giemsa-stained smears were observed to assess parasite morphology and determine parasitemia. At least 1000 cells were counted to determine parasitemia, and 200 parasites were observed to assess morphology. Effect on hydrolysis of hemoglobin and ubiquitinated proteins {#s2.4} ------------------------------------------------------------- To test if MG132 targets hemoglobin hydrolysis and the UPS, a 100 ml culture of early trophozoite stage *P. falciparum* parasites (at 11% parasitemia) was divided into 4 equal parts; DMSO (0.1% final) was added to one part, and inhibitors (21.7 µM E64, 0.024 µM epoxomicin, and 0.133 µM MG132; all at concentrations nearly 3 times the IC~50~) were added to the remaining 3 parts. The cultures were grown for 10 hours, and parasites were isolated as described above. The parasite pellet from each culture was suspended in 150 µl reducing SDS-PAGE sample buffer, boiled for 10 min, and centrifuged at 27,216g for 15 min at room temperature, and the supernatants were then separated. For hemoglobin analysis, supernatants (corresponding to ^≈^2.5 x 10^7^ parasites) were run in 15% SDS-PAGE, and the gel was stained with coomassie blue. For analysis of ubiquitinated proteins, supernatants (corresponding to ^≈^5 x 10^7^ parasites) were run in 10% SDS-PAGE, and resolved proteins were transferred onto an Immobilon-P membrane (Millipore). The membrane was blocked (5% nonfat dry milk and 3% BSA in TBST (100 mM Tris-Cl, 0.5 M NaCl, 0.05% Tween-20, pH 7.5), incubated with mouse anti-ubiquitin antibodies (Santa Cruz Biotechnology, Cat No. sc-8017; 1/500 dilution in the blocking buffer), and then incubated with HRP-conjugated goat anti-mouse IgG (Invitrogen; diluted 1/10,000 dilution in blocking buffer). The signal was developed with the SuperSignal West Pico Chemiluminescent kit (Pierce) on X-ray film. MG132 cytotoxicity for human peripheral blood mononuclear cells (PBMCs) {#s2.5} ----------------------------------------------------------------------- 10 ml human blood was collected aseptically in heparinized tubes, mixed with 10 ml complete medium (RPMI1640 with 10% FBS), layered over an equal volume of Histopaque-1077, and centrifuged at 400g for 20 minutes at room temperature. The interphase containing PBMCs was transferred to a fresh tube, mixed with 10 ml complete medium, and centrifuged at 350g for 10 minutes. The cells were washed twice with 10 ml complete medium. The washed cells were suspended in complete medium (5.5 x 10^5^ cells/ml) and grown for 20 hours at 37°C in a CO~2~ incubator. The cells were suspended in fresh culture medium, 100 µl cell suspension was added to the wells of a 96 well tissue culture plate (7 x 10^4^ cells/well), and the plate was incubated for 4 hours at 37°C in a CO~2~ incubator. 100 µl complete medium with 0.025% DMSO or varying concentrations of MG132 was added to the wells (three wells for DMSO and each concentration of MG132), and the plate was incubated in a 37°C CO~2~ incubator for 48 hours. MG132 cytotoxicity was determined following the quick protocol for MTT assay (Invitrogen). Briefly, 20 µl MTT (3-(4, 5-dimethylthiazolyl-2)-2, 5-diphenyltetrazolium bromide; 5 mg/ml in Hanks' balanced salt solution (HBSS)) was added to each well, and the plate was incubated in a 37°C CO~2~ incubator for 3 hours to label cells. The plate was centrifuged at 894g for 5 minutes, the culture medium was removed, and the cells were washed with 150 µl of HBSS. The plate was centrifuged again at 894g for 5 minutes, and the supernatant was removed. The insoluble MTT product formazan in cells was solubilized with DMSO (150 µl/well), and absorbance was read at 540 nm for formazan and at 690 nm for plate background. The plate background was subtracted from the 540 nm reading, and the absorbance values of MG132-treated cultures were normalized as percentage of DMSO-treated cultures, plotted against MG132 concentrations, and analyzed using the Four Parameter Logistic Function dose response curve (Sigma Plot) to determine the 50% cytotoxic concentration. Effect of MG132 on falcipain-2 knockout parasites {#s2.6} ------------------------------------------------- To determine susceptibility of FP2 knockout (FP2KO) parasites to MG132, the knockout and parental wild type parasites were cultured in the absence (control) or presence of various concentrations of MG132 or chloroquine in RPMI 1640 containing 10% human serum at 37°C for 50 hours. Parasitemias in control and inhibitor-treated cultures were determined by flow cytometry and compared to determine IC~50~ values of inhibitors as described earlier \[[@B7],[@B12]\]. Effect of inhibitor combinations on parasite growth {#s2.7} --------------------------------------------------- To determine the effect of inhibitor combinations on parasite growth, 300 x stocks of combinations (E64-MG132, epoxomicin-lactacystin, E64-epoxomicin, and E64-lactacystin in the ratio of 1:5, 1:3, 1:1, 3:1, and 5:1 of their IC~50s~; E64-pepstatin, MG132-pepstatin, and epoxomicin-pepstatin in the ratio of 2:1 and 1:1 of their IC~50s~) were serially diluted 2-fold in 100 µl culture medium across rows of a 96 well tissue culture plate. 100 µl of parasite suspension (1% ring-infected erythrocytes at 4% haematocrit) was added to each well, cultures were grown for 50 hours, and the 50% inhibitory concentration of each inhibitor in the combination was determined as described above. The 50% inhibitory concentration of each inhibitor in the combination was used to determine the fractional inhibitory concentration (FIC) as described previously (FIC = concentration of a compound that caused 50% inhibition in the combination/concentration of the compound required for 50% inhibition when used alone) \[[@B49]\]. The FICs obtained at different combination ratios of compound 1 and compound 2 were plotted using linear regression analysis (GraphPad Prism) to construct an isobologram, and the two axes of the isobologram were connected at points corresponding to FIC 1.0. The sum of FICs (ΣFIC: FIC of compound 1 + FIC of compound 2 in the combination) at all combinations was used to determine mean ΣFIC, which was used to determine the nature of interactions (ΣFIC 0.5-1.0, \<0.5, and \>1.5 are indicative of additive, synergistic and antagonistic interactions, respectively). Inhibition assay with parasite extracts {#s2.8} --------------------------------------- A synchronized *P. falciparum* culture with trophozoite/schizont stage parasites (10% parasitemia) was processed for isolation of parasites as described above. The parasite pellet was suspended in 600 µl lysis buffer (10 mM Tris, pH 7.0, 1 mM EDTA, 1 mM PMSF, 1 mM DTT, 1 mM ATP, and 10 µM pepstatin), subjected to 2 cycle of freeze-thaw (in liquid N~2~) and sonication for 3 min (pulses of 9 seconds on/off and 20% amplitude at 4°C; Sonics). The lysate was centrifuged at 13,000g for 15 min at 4°C, the supernatant was transferred to a fresh tube, its protein content was estimated using Bio-Rad protein assay reagent (Bio-Rad), and it was used for inhibition assays. Identical aliquots of the extract (corresponding to 16 µg protein) were added to 100 µl cysteine protease assay buffer (100 mM sodium acetate, pH 5.5, 10 mM DTT, 1 mM EDTA, 1 mM PMSF, 10 µM pepstatin) or proteasome assay buffer (50 mM Tris, 1 mM ATP, 5 mM MgCl~2~, 40 mM KCl, 1 mM DTT, 1 mM PMSF, 10 µM pepstatin, pH 7.5), which contained DMSO (1.25% final) or individual inhibitors (10 µM E64, 1 µM epoxomicin, or 1 µM MG132) or combinations of inhibitors (10 µM E64 + 1 µM epoxomicin, 10 µM E64 + 1 µM MG132, or 1 µM epoxomicin + 1 µM MG132). The reactions were incubated for 10 min at 37°C, substrates (Z-LR-AMC for cysteine protease activity, Suc-LLVY-AMC for chymotrypsin-like proteasome activity, Ac-nLPnLD-AMC for caspase-like proteasome activity, and Ac-RLR-AMC for trypsin-like proteasome activity) were added to the reactions, and protease activities were determined by monitoring substrate hydrolysis for 30 min at 37°C using an Infinite M200 (TECAN) or SpectraMax M5 (Molecular Devices) microplate reader (excitation 355 nm; emission 460 nm). Enzyme activities with Z-LR-AMC and Ac-RLR-AMC were expressed as relative fluorescence units/min (RFU/min), and those with Suc-LLVY-AMC and Ac-nLPnLD-AMC were expressed as total increase in RFU over 30 min. Enzyme activities of inhibitor-containing reactions were compared with those of DMSO-containing reactions, expressed as percent inhibition, and plotted against the inhibitors. Assays with recombinant falcipains {#s2.9} ---------------------------------- For production of recombinant FP2 and FP3, portions of the FP2 and FP3 genes corresponding to the carboxy-terminus regions of the prodomains (FP2, 36 amino acid residues; FP3, 35 amino acid residues) and the entire mature protease domains were PCR amplified from 3D7 gDNA and expressed in bacteria using pRSETA (for FP2) and pQE-30 (for FP3) expression plasmids. Recombinant proteins were purified from isopropyl-1-thio-β-D-galactopyranoside (IPTG) induced bacteria by nickel-nitrilotriacetic acid (Ni-NTA) chromatography under denaturing conditions as described earlier \[[@B15],[@B50]\]. The purified FP2 and FP3 were refolded into active forms as described earlier \[[@B15],[@B50]\], with a modification that involved dialysis of denatured proteins against refolding buffers, and used for inhibition assays. Concentrations of active FP2 and FP3 were determined by active site titration using E64. Briefly, enzymes were incubated with DMSO (1% final) or varying concentrations of E64 in sodium acetate buffer (100 mM sodium acetate, 10 mM DTT, pH 5.5) for 30 min at room temperature, Z-LR-AMC was added (25 µM final) to the reactions, and enzyme activity was determined by monitoring the release of AMC upon substrate hydrolysis over 30 min at 37°C as described above. Enzyme activities as RFU/min for DMSO and inhibitor-containing reactions were plotted against inhibitor concentrations, and analyzed using the ORIGIN program (OriginLab) to determine enzyme concentrations. To determine if FP2 and FP3 are inhibited by proteasome inhibitors, FP2 (5 nM) or FP3 (10 nM) was incubated with DMSO (1% final) or with inhibitors (MG132, epoxomicin, lactacystin, E64, and pepstatin; each at 10 µM) in sodium acetate assay buffer for 30 min at room temperature. Z-LR-AMC was added to the reaction (25 µM final), and enzyme activity was measured by monitoring substrate hydrolysis for 30 min at 37°C as described above. Enzyme activities of inhibitor-containing reactions were compared with those of DMSO-containing reactions, expressed as percent inhibition, and plotted against the inhibitors. To determine association rate constants of inhibitors for FP2 and FP3, enzymes (1 nM FP2 or 2 nM FP3) were added to 200 µl reactions (100 mM sodium acetate, pH 5.5, 10 mM DTT, 25 µM Z-LR-AMC) with DMSO (1% final) or varying concentrations of E64 (at concentration \>10 times the enzyme concentration) or MG132 (2-10 nM), and enzyme activity was measured by monitoring the release of AMC upon substrate hydrolysis for 30 min at 37°C as described above. The activity progress curves (fluorescence vs time) without and with inhibitors were analyzed by non-linear regression analysis (GraphPad software) using the pseudo-first-order equation. *y* = *A*\[1 − exp(−*k* ~obs~ *t*)\] + *B*; where *y* is the fluorescence intensity at time *t*, *A* is the amplitude of the reaction, and *B* is the offset. *k* ~obs~ vs inhibitor concentration \[I\] was linear. The association rate constant *k* ~ass~ was determined by linear regression using *k* ~ass~ = (*k* ~obs~/\[I\])(1 + \[S\]/*K* ~m~); where \[S\] is the substrate concentration. Results and Discussion {#s3} ====================== MG132 inhibits *P. falciparum* development, hemoglobin degradation and the UPS {#s3.1} ------------------------------------------------------------------------------ To determine if MG132 exerts antimalarial activity by inhibition of hemoglobin degradation and/or by inhibiting the UPS, we assessed the effects of inhibitors of falcipain cysteine proteases (E64), the UPS (epoxomicin, lactacystin, and MG132), and plasmepsin aspartic proteases (pepstatin) on *P. falciparum* erythrocytic development. All these inhibitors blocked parasite development ([Table 1](#tab1){ref-type="table"}), but the inhibited parasites showed different morphologies. As has been previously described, E64 caused enlargement of food vacuoles ([Figure 1](#pone-0073530-g001){ref-type="fig"} \[[@B9]\]), which is indicative of accumulation of undegraded hemoglobin due to inhibition of falcipains. MG132, but not epoxomicin or lactacystin, also caused enlargement of the food vacuole, suggesting that MG132, like E64, inhibited hemoglobin degradation. However, MG132-treated parasites were much smaller than E64-treated parasites. Pepstatin, an inhibitor of the food vacuole-resident plasmepsins, resulted in pyknotic parasites without swollen food vacuoles. This result is consistent with previous studies demonstrating normal hemoglobin catabolism in plasmepsin knockout parasites \[[@B51],[@B52]\]. ###### Inhibition of intraerythrocytic parasite development. **Inhibitors** **IC~50~** ^a^ (µM) ---------------- --------------------- E64 7.6 ± 1.22 Epoxomicin 0.0077 ± 0.00027 Lactacystin 0.276 ± 0.038 MG132^b^ 0.0476 ± 0.0058 Pepstatin 74.3 ± 8.8 ^a^ Mean IC~50~ with standard deviation from three independent experiments, each in duplicate. ^b^ The IC~50~ for PBMCs was 10.8±0.04 µM. {#pone-0073530-g001} The 50% cytotoxic concentration of MG132 for PBMCs was 10.8 µM ([Table 1](#tab1){ref-type="table"}), which is remarkably higher than its IC~50~ for inhibition of parasite growth (0.0476 µM), and indicates its selective toxicity for malaria parasites. The previously reported 50% cytotoxic concentration of MG132 for PMBCs was somewhat higher (20 µM), but was based on a shorter (24 hour) incubation with the inhibitor \[[@B43]\]. To determine if MG132-treated parasites had accumulated undegraded hemoglobin, total lysates of control and inhibitor-treated parasites were analyzed by SDS-PAGE. Both E64- and MG132-treated parasites, but not controls and those treated with epoxomicin, showed significant accumulation of undegraded hemoglobin ([Figure 2A](#pone-0073530-g002){ref-type="fig"}), confirming inhibition of hemoglobin degradation by MG132. Immunoblots with anti-ubiquitin antibodies showed a high molecular weight signal, which was markedly more intense in lanes containing MG132- and epoxomicin-treated parasites than lanes with control and E64-treated parasites ([Figure 2B](#pone-0073530-g002){ref-type="fig"}), indicating accumulation of ubiquitinated proteins upon inhibition of the UPS. Thus, unlike the other inhibitors tested, MG132 appears to inhibit both UPS and hemoglobin degradation, and both of these activities likely contribute to its action against asexual *P. falciparum* erythrocytic development. {#pone-0073530-g002} MG132 is effective against all asexual erythrocytic stages and irreversibly blocks parasite development {#s3.2} ------------------------------------------------------------------------------------------------------- To determine if MG132 is effective against all erythrocytic stages, parasites at different stages of development (ring, T0 and T12; early trophozoite, T24; late trophozoite, T36) were treated with inhibitors (E64, epoxomicin, and MG132), and the number of newly formed rings was counted at the end of the development cycle (48 hours) and 12 hours after the end of the cycle (60 hour). The control parasites developed normally and formed new rings upon completion of the development cycle ([Table 2](#tab2){ref-type="table"}). Treatment with MG132 and epoxomicin at ring and early trophozoite stages completely blocked the development, and that with E64 resulted in a very small number of new rings ([Table 2](#tab2){ref-type="table"}). Treatment at the late trophozoite stage also had significant inhibitory effect, as only 3-11% of the total parasite population was of new rings ([Table 2](#tab2){ref-type="table"}). The number of rings in the control increased from 40% at 48 hour time point to 80% at 60 hour time point; the inhibitor-treated parasites had 0-11% rings at 48 hour time point, and this number almost remained the same (0-12%) at 60 hour time point regardless of the stage when inhibitors were added, which suggests that inhibitors blocked maturation of parasites. Both E64 and MG132, but not epoxomicin, caused food vacuole enlargement in the majority (88%) of parasites regardless of the treatment time point. Thus, MG132 inhibits development of all asexual erythrocytic stages. ###### Stage-specific effects of inhibitors. **Post synchronization stage at which treatment started** **% Rings (R) at 48 hours and 60 hours post-synchronization in cultures containing DMSO (control) or inhibitors** ----------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------- ----------------- ----------------- ----------------- **48/60 hours** **48/60 hours** **48/60 hours** **48/60 hours** **0 Hour** 41 (±15)/80 (±12) 5 (±3)/5 (±0) 0/0 0/0 **12 Hour** \- 6 (±6)/5 (±2) 0/0 0/0 **24 Hour** \- 6 (±6)/6 (±1) 0/0 0/0 **36 Hour** \- 11 (±3)/12 (±0) 3 (±3)/10 (±0) 6 (±6)/0 Ring stage parasites were treated with DMSO (control) or inhibitors (21.7 µM E64, 0.024 µM epoxomicin, and 0.133 µM MG132; all at concentrations nearly 3 times the IC~50~) at different stages of development (ring, T0 and T12; early trophozoite, T24; late trophozoite, T36), and the number of rings was counted at the end (48 hours) and 12 hours after the end (60 hours) of the development cycle. At least 200 parasite-infected cells were observed, and the table shows the percentage of rings for control and the inhibitor-treated cultures. The data are mean with standard deviation from three (for 48 hour) and two independent experiments (60 hour) done in duplicates. ^a^ DMSO was added at the start of the experiment (T0) only. To test if the inhibition of parasite development by MG132 is permanent we performed inhibitor washout experiments in which parasites were cultured for one cycle in the presence of DMSO (control) or chloroquine (an irreversible inhibitor) or inhibitors (MG132, E64, and epoxomicin), and then without any treatment for two cycles. In SYBR Green-based assay, which is a measure of overall parasite growth, control parasites showed almost 3.5-6 times higher fluorescence than chloroquine-treated parasites in all three cycles ([Figure 3](#pone-0073530-g003){ref-type="fig"}), which indicated robust growth in the control culture. The E64-treated parasites showed similar to 2.5 times higher fluorescence than chloroquine-treated parasites, particularly in cycle 1, which suggested partial growth during treatment and recovery of parasites after E64 was washed out. The fluorescence values of MG132- and epoxomicin-treated parasites were similar to that of chloroquine-treated parasites at the end of all three cycles, which suggested that both MG132 and epoxomicin irreversibly blocked parasite development in cycle 1. Giemsa-stained smears of the control culture showed 3.5-6 fold increase in parasitemia (10 to 18% parasitemia) over the starting parasitemia in all three cycles, which is consistent with the results of SYBR Green assay. Parasitemia did not increase in cultures containing the inhibitors and chloroquine at the end of cycle 1. Dark-stained dots (cycle 2) or no parasites (cycle 3) were observed in MG132- and epoxomicin-treated cultures after the inhibitors had been washed out, indicating that MG132 and epoxomicin, like chloroquine, irreversibly block parasite development. On the other hand, washing out E64 led to partial recovery of a fraction of parasites in subsequent cycles (^≈^1% parasitemia). Thus, both SYBR Green assay and the assessment of Giemsa-stained parasite smears indicate that MG132 irreversibly blocks parasite development. {#pone-0073530-g003} MG132 inhibits the UPS and cysteine protease activities in parasites {#s3.3} -------------------------------------------------------------------- To further test the predicted actions of MG132, we determined whether cysteine protease and proteasome activities of a total soluble parasite extract are inhibited by MG132. FP2 and FP3 contribute almost exclusively to the acidic cysteine protease activity of *P. falciparum* erythrocytic stages; this is routinely measured by monitoring hydrolysis of Z-LR-AMC and is blocked by the cysteine protease inhibitor E64. Malaria parasites appear to contain a typical 26S proteasome, but the proteasome-associated activities have not been characterized. A typical 26S proteasome possesses three types of activities \[[@B33]--[@B35]\]: peptidyl-glutamyl peptide hydrolyzing (PGPH) activity that cleaves after acidic residues; trypsin-like activity that cleaves after basic residues; and chymotrypsin-like activity that cleaves after large hydrophobic residues. These three activities are routinely measured using fluorogenic peptide substrates (Z-nLPnLD-AMC for PGPH, Ac-RLR-AMC for trypsin, and Suc-LLVY-AMC for chymotrypsin) \[[@B53]\]. The chymotrypsin activity is inhibited by the majority of proteasome inhibitors, including MG132 and epoxomicin \[[@B36]--[@B39]\]. Control inhibitors of other proteases (EDTA, metalloproteases; PMSF, serine proteases; and pepstatin, aspartic proteases) were included in all reactions. E64 and MG132 completely inhibited Z-LR-AMC hydrolyzing activity of the extract, suggesting that MG132, like E64, inhibits falcipains ([Figure 4A](#pone-0073530-g004){ref-type="fig"}). Epoxomicin alone did not inhibit Z-LR-AMC hydrolyzing activity of the extract, but it completely inhibited the same activity when combined with E64 or MG132, suggesting that it does not inhibit falcipains. {#pone-0073530-g004} The hydrolysis of the chymotrypsin substrate Suc-LLVY-AMC was inhibited by E64, MG132, and epoxomicin to varying degrees ([Figure 4A](#pone-0073530-g004){ref-type="fig"}). As E64 and epoxomicin specifically target falcipains and UPS, respectively, it seems that both falcipains and the UPS contribute to the total Suc-LLVY-AMC hydrolysing activity. MG132 alone was as effective (83.6%) as E64-epoxomicin combination (84.3%) in inhibiting the Z-LR-AMC hydrolyzing activity of the extract, suggesting that MG132 has both cysteine protease and proteasome inhibitory activities. This dual-inhibitory property of MG132 also explains comparable inhibition of Suc-LLVY-AMC hydrolysis by MG132 alone and together with E64 (83%) or epoxomicin (88%). The parasite extract showed negligible hydrolysis of the PGPH substrate Z-nLPnLD-AMC (1.75 RFU/min), which may be due to different substrate preference of the Plasmodium proteasome. All three inhibitors marginally inhibited Z-nLPnLD-AMC hydrolyzing activity, and the inhibition levels were not clearly distinguishable, hence the data is excluded from [Figure 4A](#pone-0073530-g004){ref-type="fig"}. Hydrolysis of the trypsin substrate Ac-RLR-AMC was almost completely inhibited by E64 and MG132, and only slightly inhibited by epoxomicin, suggesting that falcipains primarily contributed to the hydrolysis of Ac-RLR-AMC. As in the case of Z-LR-AMC hydrolysis, combinations of E64, MG132, and epoxomicin were as effective as E64 or MG132 alone. Complete inhibition of Ac-RLR-AMC hydrolysing activity by MG132 alone, as by the specific cysteine protease inhibitor, further supports the falcipain-inhibitory property of MG132. The hydrolysis of Ac-RLR-AMC by falcipains was not surprising, as these and other papain-family enzymes readily hydrolyze substrates with Arg or Lys at the P1 position \[[@B18]\]; falcipains prefer Leu at the P2 position and can also degrade peptide substrates containing P3 Arg \[[@B19]\]. Taken together, the demonstration of inhibition of both cysteine protease and UPS activities by MG132 and the accumulation of undegraded hemoglobin and ubiquitinated proteins in parasites after treatment with this compound are all consistent with the conclusion that MG132 blocks parasite development by inhibiting hemoglobin degradation and the UPS. MG132 is a potent falcipain inhibitor {#s3.4} ------------------------------------- As MG132 caused accumulation of undegraded hemoglobin in parasites and inhibited cysteine protease activity of the parasite extract, we assessed it for inhibition of recombinant FP2 and FP3. MG132 and E64 inhibited hydrolysis of Z-LR-AMC by both enzymes ([Figure 4B](#pone-0073530-g004){ref-type="fig"}), confirming that MG132 is a falcipain inhibitor. Identical concentrations of epoxomicin and pepstatin did not inhibit, and of lactacystin marginally inhibited the enzymes. A kinetic analysis revealed that, compared to E64, MG132 is 18 and 100 times more potent in the inhibition of FP2 and FP3, respectively ([Table 3](#tab3){ref-type="table"}). Furthermore, MG132 was twice as effective against FP3 as FP2, which is of importance for drug development efforts, because only FP3 seems to be essential for erythrocytic development of *P. falciparum* \[[@B7],[@B12]\]. ###### Association rate constants (M^-1^. S^-\ 1^) for recombinant falcipains^a^. **Inhibitors** **FP2** **FP3** ---------------- ------------------ ------------------ E64 10,260 ± 714 4,090 ± 277 MG132 188,570 ± 24,840 412,460 ± 35,080 ^a^ Means ± standard deviations for at least three independent experiments, each in duplicate. Previous studies have demonstrated that loss of FP2 in FP2 knockout parasites renders them almost 2.5 times more sensitive than wild type parasites to E64 \[[@B7],[@B12]\]. We evaluated FP2 knockout parasites for sensitivity to MG132. The FP2 knockout parasite was as susceptible as the wild type parasite to the control drug chloroquine (IC~50~: FP2KO, 5.3±1.5 nM; wild type, 4.5±0.3 nM), but almost twice as susceptible to MG132 (IC~50~: FP2KO, 15.5±1.4 nM; wild type, 26.6±4.1 nM), which further supports inhibition of falcipains by MG132. Joint inhibition of falcipains and UPS has an additive antiparasitic effect {#s3.5} --------------------------------------------------------------------------- A comparison of the IC~50s~ of MG132 and E64 for parasite inhibition ([Table 1](#tab1){ref-type="table"}) indicated that MG132 is 160 times more potent than E64, which could be explained by superior potency for fa1cipains, improved cell permeability/intracellular accumulation, and simultaneous inhibition of both falcipains and UPS by MG132. Malaria parasite genomes encode 115--137 putative proteases of all five classes (aspartic, cysteine, metallo, serine, and threonine), and MG132 may target other proteases in addition to falcipains and UPS \[[@B54],[@B55]\]. To further explore effects on the two targets, we evaluated antiparasitic effects of inhibitors of the proteasome (epoxomicin, lactacystin, and MG132), falcipains (E64), and plasmepsins (pepstatin). Combinations of proteasome-specific inhibitors (epoxomicin-lactacystin) showed additive effects, which was expected as both the inhibitors target the same enzyme ([Figure 5](#pone-0073530-g005){ref-type="fig"}). Combinations of E64 with epoxomicin, lactacystin or MG132 also had additive effects ([Figure 5](#pone-0073530-g005){ref-type="fig"}). Effects were more remarkable with pepstatin, as combination with MG132 showed synergy, with E64 showed a trend towards synergy, and with epoxomicin showed antagonism ([Figure 6](#pone-0073530-g006){ref-type="fig"}). Synergistic antiparasitic effects of MG132-pepstatin and E64-pepstatin agree with proposed joint roles of falcipains and plasmepsins in hemoglobin degradation, and were consistent with a previous result showing synergism of E64 and pepstatin \[[@B56]\]. The antagonistic effect of pepstatin-epoxomicin is unexplained. One explanation could be the accumulation of plasmepsins, which are major food vacuole aspartic proteases, in the presence of epoxomicin, as plasmepsin II has been shown to be a target for ubiquitin-proteasome-mediated degradation \[[@B57]\]. The accumulation of plasmepsins would decrease susceptibility of parasites to pepstatin, potentially mediating the antagonistic effect of the pepstatin-epoxomicin combination. The additive to slightly synergistic anti-parasitic effect of the UPS and falcipain inhibitors suggests therapeutic benefits of either combined chemotherapy based on the UPS and falcipain targets or a single agent, such as MG132, that inhibits both targets. {#pone-0073530-g005} {#pone-0073530-g006} MG132 has also been shown to inhibit calpains, with IC~50~ values of 1.2 µM for purified and \>100 µM for intracellular calpains \[[@B45]\]. However, as MG132 inhibits parasite growth at much lower concentrations (0.0476 µM), it is unlikely that inhibition of parasite and/or host calpains by MG132 contributes to its antimalarial effects. Inhibition of the UPS is being actively pursued as a strategy to treat cancer; bortezomib is a proteasome inhibitor approved for treating multiple myeloma. Bortezomib has also been shown to kill *P. falciparum* intraerythrocytic stages with an IC~50~ concentration 2-4 times lower than levels achieved at recommended doses in human plasma \[[@B1]\], supporting consideration of proteasome inhibitors as antimalarials. MG132 is an inexpensive and potent inhibitor of the UPS, and for these reasons it has been widely studied as a proteasome inhibitor in a number of cell types, including several protozoan parasites. The reported 50% cytotoxic concentrations of MG132 for mammalian cell lines are in the range of 2.5-21 µM depending on the cell type and treatment duration \[[@B40]--[@B43]\]. In our studies MG132 was over 200 times more potent for inhibition of development of malaria parasites (IC~50~ 0.0476 µM) than for human PBMCs (10.8 µM), indicating that it is selectively toxic to the parasite. This selective toxicity could be due to the simultaneous inhibition of the two vital proteolytic systems of the parasite, the UPS and the falcipains, establishing it as a dual-target inhibitor of malaria parasites. MG132 may thus be considered a lead compound to design potent antimalarials acting against two parasite proteolytic systems. The authors thank Dr. V. Vijayalakshmi, National Institute of Nutrition, India, for providing access to a fluorescence microplate reader. We thank Dr. Sunit Kumar Singh and Ms. Ruhi Verma for advice on isolation and culture of PBMCs. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RP A PJR PSS. Performed the experiments: RP A VKK JL NS RN. Analyzed the data: RP A PJR PSS. Wrote the manuscript: RP A PJR PSS.

Background ========== Endothelial cell adhesion molecules (\'ECAMs\') play essential roles in the development of chronic inflammation by recruiting leukocytes, especially lymphocytes to tissues. ECAMs support several forms of leukocyte adhesion including rolling, firm adhesion and extravasation \[[@B1]\]. Infiltration of tissues by leukocytes is a common hallmark of many chronic inflammatory states that include the inflammatory bowel diseases (IBD), ulcerative colitis (UC), and Crohn\'s disease (CD). In the setting of IBD, the expression of ECAMs like ICAM-1, VCAM-1, and MAdCAM-1 (mucosal addressin cell adhesion molecule-1) is observed in experimental models of colitis, \[[@B2]-[@B5]\] and also within the inflamed human colon in Crohn\'s disease and ulcerative colitis \[[@B6],[@B7]\]. Among the adhesion molecules up-regulated in IBD, MAdCAM-1, the mucosal cell adhesion molecule, is thought to be preeminent in the development of chronic gut inflammation. MAdCAM-1 is normally expressed in the gut, and its expression is dramatically amplified during inflammation \[[@B2],[@B3]\]. The functional significance of increased appearance of MAdCAM-1 in IBD is supported by several reports which show that immunoneutralization of either MAdCAM-1 or its ligand, the α4β7 integrin, attenuate inflammation and mucosal damage in animal models of colitis \[[@B8]-[@B10]\]. However, since monoclonal antibodies directed against other ECAMs, particularly VCAM-1, can as well reduce disease activity in colitis models \[[@B11]-[@B14]\], the literature suggests that MAdCAM-1 is probably necessary, but insufficient for the maximal penetrance of experimental and probably also clinical IBD. Based on these findings, it is apparent that a better understanding of the mechanisms regulating ECAM expression, especially that of MAdCAM-1, might help to devise improved therapies for colitis. It has been suggested that pathologic activation of the mucosal immune system in response to antigens is a key factor in the pathogenesis of IBD. Furthermore, changes in leukocyte migration and cytokine production appear to contribute the perpetuation of IBD \[[@B15]\]. Based on modern advances, recombinant anti-inflammatory cytokines (i.e. IL-10) treatment is now being developed for experimental colitis and human IBD. IL-10 produced by macrophages and monocytes appears to limit chronic inflammation \[[@B16]-[@B18]\], through a decreased release of inflammatory factors (IL-1, IL-6, IL-12, TNF-α, GM-CSF, GCSF), suppression of adhesive determinants (MHC class II molecule, β7), and by blocking ICAM-1 induction \[[@B19]-[@B24]\]. Conversely, IL-10 gene-knockout mice develop a chronic colitis that is extremely similar to IBD \[[@B25]\]. IL-10 treatment can reduce inflammation in several models of colitis \[[@B26]-[@B30]\] and human IBD \[[@B31]-[@B35]\]. However, the clinical efficacy of systemically administered IL-10 for patients mild to moderately active Crohn\'s disease has not been as effective as hoped \[[@B33]-[@B35]\]. Furthermore the efficacy of IL-10 administration in mouse colitis models is contentious \[[@B36]\]. We have described in vitro that exogenous IL-10 can block the expression of MAdCAM-1 in response to TNF-α, and attenuates lymphocyte adhesion to lymphatic node derived endothelium under cytokine stimulating conditions via NF-kB inhibition \[[@B5]\]. The purpose of the current study was to show that induction of endothelial expression of IL-10 through an IL-10 expression vector attenuates MAdCAM-1 expression in response to TNF-α and optimistically suggests the possibility of targeted Th2-cytokine gene therapy in IBD. Methods ======= Reagents -------- Recombinant mouse TNF-α was purchased from ENDOGEN (Stoughton, MA) amd plasmid containing human IL-10 (phIL-10) was generous gift from Dr. Meng X (Thomas Jefferson University, PA). Cell culture ------------ The SVEC4-10 line is an endothelial cell line derived by SV40 (strain 4A) transformation of murine small vessel endothelial cells, originally isolated from the axillary lymph node vessels of an adult male C3H/Hej mouse \[[@B37],[@B38]\]. These cell types were all maintained in Dulbecco\'s modified Eagle\'s medium (DMEM) with 10% fetal calf serum with 1% antibiotic/ antimycotic. Cells were seeded into 24-well tissue culture plates at approximately 20,000 cells/cm^2^, and cultures were used immediately upon reaching confluence (usually 3--4 days after seeding). Lymphocytes ----------- The mouse CD8^+^T cell lymphoma TK-1 cells (that constitutively expresses the α4β7 integrin \[[@B39]\]) were obtained as a generous gift from Dr. Eugene Butcher (Stanford University, CA). These cells were cultured in RPMI-1640 medium supplemented with 10% FCS and 0.05 mM 2-mercaptoethanol (minus antibiotic/ antimycotic). IL-10 gene transfer ------------------- The pCYIL-10 vector is an expression vector which was established by Dr. Xianmin Meng based on the pCY4B vector \[[@B40]\]. Cultured SVEC endothelial cells seeded in 24 wells were transfected at 70% confluency with SuperFect (Qiagen). Briefly, 1 μg plasmid DNA was diluted with 60 μl DMEM and then 5 μl Superfect (3 mg/ml) was added. After incubation at room temperature for 10 min, the complex was mixed with 350 μl medium and added to the each wells. After 2 h at 37°C, 5% CO2, the complex was aspirated and washed 2 times with medium following replace by 1 ml of medium. After incubation at 37°C for 24 h cells were stimulated with TNF-α. Western analysis of cell lysates -------------------------------- 24 h of IL-10 gene transfer, monolayers were treated with TNF-α (1 ng/ml, 24 h). All cell samples were harvested at 24 hours. Equal quantities of protein (75 μg) from each sample were electrophoretically separated on 7.5% SDS-PAGE gels. Gels were transferred to nitrocellulose membranes (Sigma) and blocked with 5% milk powder in PBS at 4°C (overnight). These membranes were washed twice for 10 min with wash buffer (0.1% milk powder in PBS). Primary rat anti-mouse MAdCAM-1 mAb was added at a concentration of 10 μg/ml and incubated at room temperature for 2 h. These membranes were washed twice with wash buffer. Secondary rabbit anti-rat horseradish peroxidase conjugated secondary antibody (Sigma) was added at a 1:2000 dilution for 2 h. Lastly, membranes were washed 3 times and developed using the enhanced chemiluminescence (ECL) detection system (Amersham, La Jolla, CA). The density of MAdCAM-1 staining was measured by scanning the 60 kD band, using a HP ScanJet™ flatbed scanner. Images were analyzed for density using Image Pro Plus™ image analysis software (Media Cybernetics). The data are expressed as a percentage of TNF-α-induced level of density. In each protocol, treatments were performed at least in triplicate. TK-1 lymphocyte adhesion assay ------------------------------ Briefly, TK-1 cells were suspended in culture medium and fluorescence labeled by incubating TK-1 cells at 2 × 10 ^6^cells/ml with 0.02 mg fluorescein diacetate (FDA) (Sigma) at 37°C for 30 min. The cells were then washed twice with ice-cold HBSS, spun at 250 g for 5 min to remove unincorporated fluorescence and suspended in HBSS. The TK-1 lymphocyte cell line used in this assay expresses high levels of the α4β7 integrin, \[[@B39],[@B41]\] which can interact with multiple ligands including mucosal addressin-1 (MAdCAM-1), as well as VCAM-1, L-selectin and fibronectin \[[@B42]\]. In this system, TNF-α stimulated TK-1 adhesion to SVEC4-10 endothelial cells is at least 50% MAdCAM-1 dependent \[[@B41]\]. SVEC monolayers were grown in 48-well plates as described, and to activate endothelium, the monolayers were incubated with TNF-α (1 ng/ml) for 24 h. Cytokine treated endothelial cells were washed three times with media. Labeled TK-1 cells were then added to the endothelium at a 5:1 lymphocyte to endothelial cell ratio \[[@B43]\] and allowed to bind for 30 min under static conditions. At the end of the incubation period, the supernatant was removed and the monolayers were washed twice with HBSS. Plates were read on a Fluoroskan Ascent (Labsystems, Helsinki, Finland) set for excitation at 485 nm, and emission at 515 nm. Blank wells (0% TK-1 adhesion) were run as controls that did not contain labeled TK-1 cells. The data are expressed as a percentage of TNF-α-induced level of fluorescence. In each protocol, treatments were performed at least in triplicate. Statistical analysis -------------------- All values are expressed as mean ± SD. Data were analyzed using multiple comparisons. Probability (*P*) values of \< 0.05 were considered significant. Results ======= Secretion of human IL-10 concentration by transfected endothelial cells ----------------------------------------------------------------------- To screen for the efficacy of IL-10 transfected SVEC, we initially measured the IL-10 concentration in the medium prior to gene transfection. There was no detectable human IL-10 signal in the non-transfected cell medium (control), nor any mouse IL-10 (*data not shown*). However, the medium from the IL-10 transfected SVEC medium showed a large and significant increase in the IL-10 concentration (1209 ± 2 pg/ml) at 48 h after IL-10 gene transfection (Figure [1](#F1){ref-type="fig"}). Importantly, SVEC cells transfected with this vector showed no evidence of toxicity, altered proliferation and morphologically were indistinguishable from mock-transfected and untreated controls. {#F1} MAdCAM-1 expression in IL-10 gene transfected endothelial cells --------------------------------------------------------------- The anti-inflammatory effect of IL-10 gene transfection in endothelial cells was measured by the reduction in the expression of the endothelial cell adhesion molecule MAdCAM-1 induced by TNF-α (1 ng/ml, 24 h). TNF-α strongly induced expression of MAdCAM-1. This effect was significantly reduced by IL-10 gene transfection (Fig. [2](#F2){ref-type="fig"}). Alone, the IL-10 gene transfection had no effect on MAdCAM-1 expression. {#F2} α4β7 integrin lymphocyte adhesion to IL-10 gene transfected endothelial cells ----------------------------------------------------------------------------- Having established that IL-10 gene transfection exerts a significant protective effect against TNF-α stimulated endothelial MAdCAM-1 induction, we next examined the effects of IL-10 gene transfection on the adhesion of α4β7 expressing lymphocytes (using the mouse cell line TK-1) to endothelial monolayers. TNF-α stimulation (1 ng/ml, 24 h) significantly increased the adhesion of TK-1 lymphocytes to SVEC monolayers. IL-10 gene transfection significantly reduced TK-1 adhesion in response to TNF-α stimulation at 24 h (Fig. [3](#F3){ref-type="fig"}). Alone, IL-10 gene transfection did not modify the basal level of lymphocyte adhesion to the endothelium without TNF-α treatment. {#F3} Discussion ========== MAdCAM-1 is a 60 kDa endothelial cell surface molecule that is strongly expressed by mucosal endothelial cells, particularly following exposure of these cells to pro-inflammatory cytokines such as TNF-α. Expression of MAdCAM-1 has also been reported in the brain, and in the heart \[[@B44],[@B45]\]; based on these findings, it is now been suggested that MAdCAM-1 might play roles in chronic inflammation of these organs as well. With respect to inflammatory bowel disease, MAdCAM-1 appears to be essential to the lymphocyte homing to the mucosa associated lymphoid tissue (MALT) \[[@B6],[@B37],[@B38],[@B46]\]. Since MAdCAM-1 is normally expressed mainly within the gut microvasculature, and is dramatically amplified during IBD, it has been suggested that increased MAdCAM-1 expression contributes to the etiology of IBD through its ability to direct homing of lymphocytes to the gut. This notion is well supported by several reports that show that antibodies directed against either MAdCAM-1, or its lymphocyte ligand, the α4β7 integrin will significantly attenuate several indices of gut damage in experimental models of colitis \[[@B9],[@B47]\]. Several studies have indicated that T helper (Th1) immune response has important roles in the development of IBD \[[@B48]-[@B50]\]. Moreover, dysregulation of cytokine networks is involved in Th1-dominant immune responses in IBD \[[@B48]-[@B50]\]. Among the Th1 cytokines, TNF-α is thought to be perhaps the most important cytokine responsible for driving the onset and evolution of IBD. Because of this prime role of TNF-α in IBD, anti-TNF-α antibody therapy has been very successfully used in IBD to reduce both colonic injury and expression of ECAMs in IBD \[[@B51]\]. IL-10, a cytokine produced by activated macrophages and Th2-type T cells, has crucial inhibitory effect against several Th-1 type immune responses, such as the antigen-presenting functions of monocytes and macrophages \[[@B52],[@B53]\]. IL-10 may play an important role in preventing the induction of IBD, since animals deficient in IL-10 will develop colitis spontaneously, and low levels of IL-10 are positively correlated with recurrences of Crohn\'s disease \[[@B25],[@B54]\]. However, unlike TNF-α based therapies, the administration of recombinant IL-10 (rIL-10) shows poor clinical efficacy. This may reflect the fact that TNF-α therapies for IBD are aimed at efficiently clearing TNF-α, while IL-10 therapies must *increase*IL-10 and recombinant IL-10 is likely too rapidly cleared from the circulation after *in vivo*administration to provide a uniform protection \[[@B55]\]. On the other hand, IL-10 *gene transfer*technology has been used with some success in models of colitis, however its effectiveness is also variable. One reason for this variability may be that the final serum IL-10 concentration of gene-transfected mice might be below the critical threshold needed for gut protection \[[@B55],[@B56]\]. Therefore targeting of the IL-10 gene to the inflamed colon should ideally exploit tissue (i.e. gut) specific promoters to control selective organ gene transfer technology, endothelial specific promoters and also organ specific intra-arterial injection of vector to activate some genes in specific locations \[[@B57]\]. Administration of IL-10 *in vitro*prevents TNF-α stimulated expression of MAdCAM-1, and also blocks lymphocyte adhesion to endothelial cells to the same extent as dexamethasone treatment, currently a key therapy in IBD \[[@B5]\]. Interestingly, the low basal levels of MAdCAM-1 expressed by control (cytokine unstimulated) cells were not affected by IL-10 transfection. This suggests that basal expression of MAdCAM-1, is IL-10 independent, while cytokine stimulated MAdCAM-1 *is*modulated by IL-10. While we previously showed that delivery of IL-10 to the endothelium *in vitro*is protective against TNF-α \[[@B5]\], *in vivo*IL-10 has not been as effective. There are at least two regulating pathways in ECAM expression, one is a cytokine-dependent pathway which is widely studied and the other is cytokine-independent \[[@B58]\]. IL-10 apparently prevents cytokine-induced NF-kB binding activity to the ECAM promoter \[[@B59]\] and blocks the excessive expression of ECAMs in cytokine-mediated inflammatory conditions. In this regard, IL-10 has been used as a treatment in inflammatory diseases, like IBD for precisely this reason: because it has few effects at sites remote from inflammation. Similarly, we found that IL-10 transfection did not affect cell morphology, total protein level, or the expression of actin and vimentin in these cells (measured by Ponceau red S staining density on immunoblots). Therefore methods like endothelial gene transfection/transfer *in vivo*may effectively maintain adequate IL-10 concentrations near the endothelial cells that finally achieve the protection against cytokines that is not uniformly produced by systemic IL-10 administration. The most important index for how well gene mediated recombinant IL-10 delivery might work in IBD appears to be measurement of the inhibition of lymphocyte-endothelium binding mediated by TNF-α induced MAdCAM-1 expression. In this experiment, IL-10 induction in the endothelium efficiently blocked TNF-α induced MAdCAM-1 expression and α4β7-dependent lymphocyte adhesion on SVEC endothelial cells. IL-10 deficiency in colitis increases expression of several ECAMs including ICAM-1, VCAM-1 and MAdCAM-1 \[[@B60]\]. We performed an experiment, where under non-stimulated conditions only 55.7 ± 5.4% as many of the TK-1 lymphocytes bound to SVEC compared to TNF-α stimulation (100%, or \'maximal\' binding). TK-1 were seen to interact with SVEC through ICAM-1 as well MAdCAM-1, but not VCAM-1, since an antibody against MAdCAM-1 reduced adhesion to 27 ± 4.6% of maximum and an anti-ICAM-1 reduced adhesion to 54.6 ± 8.3 % of maximum; anti-VCAM-1 reduced adhesion to 91% of maximum (blocks lymphocyte adhesion to endothelial cells (*data not shown*). Therefore we feel that adhesion in this system is mostly MAdCAM-1 dependent. IL-10 transfection of endothelial cells reduced MAdCAM-1 expression by 42% (following TNF-α stimulation) and reduced TK-1 adhesion by 50%. Therefore, these data suggest that the reduction of TK-1 adhesion is approximately 84% (42%/50%) MAdCAM-1 dependent, and that the remaining 16% of adhesion depends on other ECAMs, (such as ICAM-1). Although tissue specific promoters might further increase the organ specificity and transgene delivery of this approach, they have not been evaluated in this report. Our findings suggest that lymphatic or gut endothelial transfection with Th2 cytokines like IL-10 may be an effective method to reduce important clinical symptoms associated with IBD which are mediated by MAdCAM-1 dependent lymphocyte adhesion. Abbreviations ============= MAdCAM-1 (mucosal addressin cell adhesion molecule-1), IL-10 (Interleukin-10), HEV (high endothelial venular), ECAMs (endothelial cell adhesion molecules), IBD (inflammatory bowel diseases), UC (ulcerative colitis), CD (Crohn\'s disease), ICAM-1 (intracellular adhesion molecule 1), VCAM-1 (vascular adhesion molecule 1), NF-kB (nuclear transfection factor kB), TNF-α (tumor necrosis factor alpha), GM-CSF (granulocyte-macrophage colony-stimulating factor), GCSF (granulocute colony stimulating factor), ECL (enhanced chemiluminescence) Competing interests =================== None declared. Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-230X/3/3/prepub> Acknowledgements ================ This work supported by NIH grants HL47615, DK43785 and DK47663.

Introduction {#S1} ============ Skin with underlying subcutis is armed with neuroendocrine capabilities and represents the largest and one of the most complex organs in the human body ([@R58]). Strategically located at the interface with external environment, skin detects, integrates, and responds to stressors including ultraviolet radiation (UVR) ([@R8]; [@R26]; [@R47]; [@R58]; [@R59]). UVB (290--320 nm) radiation has powerful biological actions not only on cutaneous biology but it also impacts many regulatory pathways involved in immune homeostasis that are both vitamin D-dependent and independent ([@R1]; [@R2]; [@R8]; [@R12]; [@R19]; [@R23]; [@R27]; [@R31]). The mechanisms of UVB-induced immune suppression ([@R63]) are not completely understood. However, evidence is accumulating that DNA damage and other mechanisms such as the photoisomerization of urocanic acid, free-radical formation, and signal transduction-mediated activation of transcription factors and induction of neuroendocrine signaling may also contribute to the resulting pathological conditions as well ([@R8]; [@R23]; [@R39]; [@R58]; [@R63]). The main regulatory algorithm to maintain body homeostasis is the hypothalamic-pituitary-adrenal (HPA) axis, which requires activation of a complex range of responses involving the endocrine, nervous, and immune systems, collectively known as the stress responses ([@R4]; [@R34]; [@R60]; [@R62]). Neural signals encoded by the limbic system as stressors trigger neurons of the paraventricular nucleus (PVN) of hypothalamus to produce and release corticotropin releasing hormone (CRH) into the hypophyseal portal circulation ([@R60]; [@R62]). Next, CRH binds to CRH receptor type 1 (CRH-R1) on pituitary corticotropes, and induces the release of POMC-derived adrenocorticotropic hormone (ACTH) into the systemic circulation. The melanocortin receptor type 2 (MC2R), expressed in the adrenal cortex, stimulates glucocorticoid (GC) synthesis and secretion after binding of ACTH ([@R4]; [@R21]; [@R60]). GC (i.e., cortisol (COR) in humans and corticosterone (CORT) in rodents) maintain metabolic and stress-responses, suppress immune activity and are self-regulated, with negative feedback to the hypothalamus and pituitary to mute the HPA axis ([@R4]; [@R21]). The skin has neuroendocrine capabilities that also encompasses all elements of the "cutaneous HPA axis" that follow the organization of the central HPA axis \[reviewed in ([@R50])\]. This concept was based on the evidence that vertebrate skin expresses CRH with functional CRH-R1 \[reviewed in ([@R59])\] and related POMC macromolecule, which is further processed to ACTH ([@R14]; [@R29]; [@R39]; [@R42]; [@R48]), which, after interaction with MC2R, induces steroidogenesis with the final production of highly adaptable COR or CORT ([@R5]; [@R14]; [@R35]; [@R41]; [@R51]; [@R52]; [@R53]; [@R54]; [@R64]). Furthermore, Hiramoto et al ([@R11]) have demonstrated that exposure of the eye to the UVB increases plasma α-MSH levels with systemic stimulation of epidermal melanocytes. Based on the above we have decided to test the hypothesis that UVB acting on the skin can regulate body homeostasis through activation of central HPA. Using mouse model we evidence that UVB activates the HPA axis both on the local (skin) and systemic (brain, adrenal and plasma) levels, with a latter requiring an intact pituitary. The UVB induced increases in corticosterone production explain immunosuppressive effects of the UVB, while that of β-endorphin could explain a phenomenon of "UV addiction". Results {#S2} ======= General design of UVB exposure {#S3} ------------------------------ To prevent retinal or non-retinal eye signal transmission, the heads including eyes, were covered with aluminum foil, and the skin on the back was irradiated with UVB ([Figure 1](#F1){ref-type="fig"}). Dose and Time Dependent Effects of UVB in the Skin {#S4} -------------------------------------------------- Previously, we have documented that UVB (290--320 nm), but not less energetic UVA (320--400 nm), is effective in stimulating HPA axis elements in human and mouse skin organ culture *in vitro* ([@R35]; [@R36]; [@R37]). In current experiments we first measured the CORT in the skin *in vivo* using different doses and time after UVB exposure and found that the dose of 400 mJ/cm^2^ (2.1 minimal erythema doses (MED), [Table S1](#SD1){ref-type="supplementary-material"}) and 12 and 24 h after exposure were most optimal for enhancement of the CORT levels ([Figure 2a,b](#F2){ref-type="fig"}). Lower dose (100 mJ/cm^2^) and shorter times of observation (3 and 6 h) showed significantly lower stimulatory effect. Similarly, markedly stronger stimulation of plasma CORT levels was observed at 400 in comparison to 100 mJ/cm^2^ and at 12 in comparison to 3 h after UVB exposure ([Figure S1](#SD1){ref-type="supplementary-material"}). The histopathological analysis demonstrated that UVB at 400 mJ/cm^2^ did not produce noticeable epidermal necrosis nor trigger marked/moderate inflammatory infiltrate ([Figure 2c](#F2){ref-type="fig"}). However, small increase in infiltrating neutrophils and eosinophils was observed at 1--6 h after UVB exposure, which after 12 or 24 h returned to the control ([Figure 2d](#F2){ref-type="fig"}). Therefore, based on the previous ([@R36]) and current *in vivo* experiments we have chosen the dose of 400 mJ/cm^2^ (2.1 MED) and a time of 12 and 24 h after UVB exposure for further experiments. UVB effects on cutaneous expression of CRH and Ucn {#S5} -------------------------------------------------- Since the CRH gene is not expressed in C57BL/6 skin ([@R49]; [@R59]), and can be replaced by Ucn in activating HPA axis elements ([@R46]; [@R59]), we examined Ucn cutaneous expression as a potential triggering regulator. We also measured the concentration of CRH since this peptide can be released locally from cutaneous nerve fibers ([@R28]; [@R40]). UVB radiation stimulated the Ucn expression at the gene ([Figure 3a](#F3){ref-type="fig"}) and peptide levels ([Figure 3l](#F3){ref-type="fig"}) with the similar patterns at 12 and 24 h post-radiation. *In situ* localization studies showed increased expression of Ucn in the main skin compartments including the epidermis, adnexal structures and stratum paniculosum ([Figure 3g](#F3){ref-type="fig"}). There was also an increase in CRH peptide concentration in the skin after UVB exposure, as evaluated by ELISA ([Figure 3k](#F3){ref-type="fig"}). UVB effects on cutaneous expression of POMC, ACTH and β-END {#S6} ----------------------------------------------------------- Next, we checked cutaneous POMC expression, a "pituitary" element of the systemic HPA axis. UVB light stimulated a 2.5-fold increase in expression of POMC mRNA after 12 h, and which was still present after 24 h, although at lower levels ([Figure 3b](#F3){ref-type="fig"}). Immunoblotting with antibodies directed against ACTH, which recognize the 33 kDA POMC precursor, confirmed increased expression of this molecule ([Figure 3e](#F3){ref-type="fig"}). UVB-induced increased ACTH production was also confirmed with quantitative IHC ([Figure 3h](#F3){ref-type="fig"}) as was expression of POMC-derived β-END stimulated by UVB ([Figure 3i](#F3){ref-type="fig"}). UVB effects on cutaneous expression of MC2R, CYP11A1, StAR, 3β-HSD and CORT {#S7} --------------------------------------------------------------------------- The expression of the next crucial element of the HPA axis, the MC2R (responsible for initiation of steroidogenesis upon ACTH activation), was upregulated 1.5 times after 12 h and almost two times after 24 h ([Figure 3c](#F3){ref-type="fig"}). We also investigated StAR gene expression, which is required to transfer cholesterol from the outer to the inner mitochondrial membrane, and showed that its up-regulation occurred after 12 and 24 hrs post-UVB exposure ([Figure 3d](#F3){ref-type="fig"}). Moreover, we evaluated the expression of the rate-limiting enzyme of steroidogenesis, the cytochrome P450scc (CYP11A1), which cleaves the cholesterol side chain to produce pregnenolone, a precursor of all steroids. Western blot analysis revealed high expression of P450scc at 12 h and 24 h post-UVB exposure, compared to appropriate controls ([Figure 3f](#F3){ref-type="fig"}). Furthermore, immunohistochemistry for 3β-HSD, (the enzyme that transforms pregnenolone to progesterone), showed that this antigen is highly expressed in cutaneous adnexal structures and in the stratum paniculosum, but weakly expressed in the epidermis of untreated skin. UVB radiation enhanced 3β-HSD expression, especially in epidermal cells ([Figure 3j](#F3){ref-type="fig"}). The final product of HPA axis activation, CORT was produced at significantly high level at 12 h and increased further at 24 h after UVB irradiation ([Figure 3m](#F3){ref-type="fig"}). UVB effects on CRH expression in the Hypothalamus {#S8} ------------------------------------------------- The area corresponding to the hypothalamus (Bregma \~ −0.34 to −2.70 mm) was dissected out and processed either for CRH mRNA expression or peptide measurements. There was a significant increase of CRH gene expression ([Figure 4a](#F4){ref-type="fig"}) and CRH peptide production ([Figure 4b](#F4){ref-type="fig"}) both 12 and 24 h after UVB exposure. Furthermore, immunohistochemistry showed an increased number of CRH-immunopositive neurons and nerve fibers surrounding the PVN area. The highest immunopositive signal was observed at 12 h after UVB exposure, and its relative values are calculated and presented as an insert to [Figure 4c](#F4){ref-type="fig"}. Changes in Adrenals after UVB exposure {#S9} -------------------------------------- QPCR analyses showed that MC2R mRNA was up-regulated after 12 and 24 h of UVB radiation ([Figure 5a](#F5){ref-type="fig"}). Similarly, StAR mRNA was increased at the same time points ([Figure 5b](#F5){ref-type="fig"}). Expression of the CYP11B1 gene was also up-regulated but only after 24 h of UVB exposure ([Figure 5c](#F5){ref-type="fig"}). Changes in Plasma after UVB exposure {#S10} ------------------------------------ CRH peptide content was highly increased at 12 and 24 h after UVB irradiation, compared to that in sham-irradiated animals ([Figure 5d](#F5){ref-type="fig"}). A similar effect was observed for Ucn ([Figure 5e](#F5){ref-type="fig"}). ACTH concentrations were markedly stimulated at 12 and 24 h after UVB exposure [Figure 5f](#F5){ref-type="fig"}). Interestingly, UVB enhanced plasma concentrations of another peptide, β-END ([Figure 5g](#F5){ref-type="fig"}), that results from POMC cleavage. The final component of the HPA stress-axis in rodents, CORT, was highly elevated after 12 and 24 h post-UVB exposure ([Figure 5h](#F5){ref-type="fig"}). UVB effects in Hypox Animals {#S11} ---------------------------- First, we tested the skin, and showed that both basal and UVB-stimulated cutaneous HPA activity was pronounced in mice with intact pituitary (sham-hypox controls) than in hypophysectomized (lacking the pituitary, hypox) mice ([Figure 6a](#F6){ref-type="fig"}). Histological evaluation showed no significant change in the skin morphology between hypox and sham-hypox mice skin after UVB exposure ([Figure 6b](#F6){ref-type="fig"}). Second, changes evoked by UVB at the plasma level in mice with intact pituitary were markedly different from those of mice lacking the pituitary (hypox). Although, CRH and Ucn followed similar pattern of activation in both animal groups ([Figure 6c,d](#F6){ref-type="fig"}), hypox mice had much lower plasma levels of ACTH and CORT ([Figure 5e,f](#F5){ref-type="fig"}) and UVB failed to stimulate plasma levels of ACTH and CORT ([Figure 6e,f](#F6){ref-type="fig"}). Discussion {#S12} ========== This manuscript shows, that the exposure of the skin to UVB can activate the systemic HPA axis culminating in increased plasma levels of corticosterone, which requires a functional pituitary. These studies parallel an important discovery by Hiramoto's group that exposure of the eyes to UVB stimulates plasma levels of α-MSH and increases number of epidermal melanocytes ([@R11]). Although the mechanism of this stimulation remains to be established (evidence for activation of any the hypothalamic nucleus awaits experimental demonstration), the requirement for an intact pituitary was demonstrated ([@R11]). However, the final product of UVB induced in the eye axis is α-MSH, accompanied by an increased expression of pituitary prohormone convertase 2 ([@R9]; [@R10]), which contrasts the classical HPA axis. In the HPA axis, after pituitary stimulation by hypothalamic CRH, the main product of POMC process is ACTH that acts on the adrenal glands to stimulate glucocorticosteroidogenesis ([@R61]). While Hiramato et al show a novel mechanism of UVB activity at the central level that is different from the HPA axis, our studies demonstrate activation of the classical HPA axis by skin derived factors acting at different entry points [Figure S2](#SD1){ref-type="supplementary-material"}, for detailed discussion see below). We also show that UVB can induce in mouse skin *in vivo* expression of all the elements of the HPA axis including CRH, POMC and corticosteroidogenic pathway. Since UVB did not induce CRH gene expression in skin cells, CRH must be delivered from nerve endings supplying dermo-epidermal or follicular junctions. This is consistent with previous studies showing increases in skin CRH after trauma1-induced hair cycling in C57BL/6 mouse ([@R25]; [@R28]; [@R40]; [@R45]; [@R49]; [@R56]), and detection of CRH in skin nerve bundles ([@R28]). CRH, together with Ucn (of which both gene and protein expression are up-regulated by UVB), can interact with CRH-R1 and CRH-R2 to promote local POMC activity ([@R59]) followed by the stimulation of local steroidogenesis ([@R51]; [@R57]). Concomitantly, UVB enhanced ACTH and β-END production in the skin, with increased expression of crucial regulators of streoidogenesis such as MC2R (receptor for ACTH), StAR (transporter of cholesterol) and steroidogenic enzymes (CYP11A1/P450scc and 3β-HSD) culminating with the final production of CORT. Thus, UVB activates all elements of the HPA axis in mouse skin *in vivo*, which is in agreement with previous studies on UVB induction of cutaneous POMC ([@R3]; [@R30]) and of all or selected elements of the HPA axis in skin cells ([@R38]; [@R55]; [@R65]) or skin organ culture ([@R35]; [@R36]). The most striking and previously unreported observations were the stimulation of CRH in the brain localized to the PVN of the hypothalamus and plasma increases of CRH, Ucn, ACTH, β-END and CORT that were accompanied by up-regulation of adrenal MC2R, StAR and CYP11B1. Activation of all of these elements upon skin exposure to UVB clearly indicates that activation of the systemic stress response is centered in the HPA axis and is invoked by skin signals induced by UVB. The crucial role of skin factors in this activation is documented by our experimental design that shielded the head from UVB exposure, preventing retinal signal transmission. The possible mechanisms of this activation are outlined in [Figure S1](#SD1){ref-type="supplementary-material"}, which includes both the neural route and humoral signals sent from the skin to the central regulatory elements. UVB enhancement of the production of CRH mRNA and peptide in the hypothalamus with localization at the PVN supports the hypothesis that cutaneous signals are conveyed via DRG- spinal cord- dorsal column to the PVN, the center where the HPA axis begins ([@R59]; [@R60]). However, the detailed mapping of this routing is beyond the confines of this project and represents a future challenge. The nature of humoral signaling from the skin to increase systemic CORT is more complex because in addition to production of ACTH, Ucn and CRH (see above), UVB stimulates cytokine production (IL1, IL6 and TNFα) and release into circulation ([@R16]; [@R22]; [@R33]), all of which can activate the pituitary POMC ([@R4]; [@R20]; [@R48]; [@R61]). Although dissection of potential contributions of cytokines signals can be difficult or impossible to quantitate in this experimental setting, the most logical explanation would be the activation of CRH-R1 in anterior pituitary by the circulating natural ligands: CRH and Ucn, induced by UVB, to initiate the following cascade of HPA activity: CRH + Ucn→CRH-R1→POMC →ACTH→MC2R→CORT To substantiate this hypothesis, we employed hypophysectomized mice. After UVB exposure, there was significant enhancement of CRH and Ucn peptide levels in the skin and plasma of hypox and sham-hypox animals, with plasma increases of ACTH and CORT seen only in sham-hypox (control) mice. This documents that the pituitary is necessary for UVB activation of a final element of the HPA axis in rodents, CORT. Moreover, the elevated CRH, Ucn and ACTH in the skin after UVB exposure were insufficient for direct (omitting pituitary) humoral activation of adrenals. However, they were capable of stimulating cutaneous CORT but without the corresponding increases in the plasma. Thus, UVB is another trigger of the HPA axis that requires a functional pituitary to stimulate secretion of ACTH and plasma glucocorticoids, while activation of local (skin) HPA activity is restricted to this organ without systemic involvement, as previously predicted ([@R48]). The net phenotypic consequences of these UVB-induced processes will include homeostatic and immunosuppressive effects resulting from the action of glucocorticoids and POMC-derived peptides, consistent with the established role and function of the HPA axis in the regulation of body homeostasis ([@R4]; [@R34]; [@R59]; [@R60]; [@R61]). These activities would explain the powerful systemic UVB immunosuppressive effects ([@R18]; [@R31]) as well as systemic beneficial effects in attenuation of autoimmune processes that are independent of vitamin D3 ([@R1]; [@R15]; [@R32]). Thus, it may be possible to attenuate severity of autoimmune diseases such as rheumatoid arthritis, multiple sclerosis, inflammatory bowel disease or scleroderma, by stimulation of endogenous glucocorticoids in an organized fashion through UVB induction of the HPA axis. The stimulation of β-END levels in the skin and plasma offers a mechanistic explanation for the recently described phenomenon of "UV addiction" ([@R17]; [@R24]), as a secondary effect of UVB-induced production of POMC-derived s-END ([@R58]). The above data on β-END are supported by most recent report from Dr. Fisher group showing UVB mediated stimulation of β-END in the skin and serum ([@R7]). In summary, we provide the evidence that UVB can activate the central HPA axis and, based on its organization and function, we propose that this mode of action represents unique mechanism regulating body homeostasis in response to UVB spectrum of solar light. Materials and Methods {#S13} ===================== Animals {#S14} ------- All procedures involving mouse experiments and tissue handling were approved by the IACUC at the UTHSC, Memphis TN. Eighty seven week-old females C57BL/6 mice (*n*=6) were purchased from the Taconic Farms (Hudson, NY). After arrival, animals were kept for 4 days to avoid transportation stress with free access to standard laboratory chow and water, maintained on a 12:12 light/dark cycle and room temperature (RT) ranged from 20 to 24 C. Eight week-old animals with all hairs at the telogen stage as judged by a lack of skin pigmentation ([@R42]) were used for the experiments. This was further confirmed by histological examination of the hair cycle stage confirming that hairs were at the telogen (resting-quiescence) stage where the expression of neuroendocrine factors is expected to be the lowest ([@R42]; [@R48]) and, therefore, the predicted UVB stimulation would be the highest. To test the role of the pituitary in UVB-induced HPA-activation, hypophysectomized and control (sham hypophysectomized) ninety eight (n=6) B6 females, 7 weeks old, were purchased from Charles River Laboratories (Wilmington, MA) and were kept on 5% sucrose-supplemented water according to the vendor's guidelines. General Experimental Design {#S15} --------------------------- Details of experimental design are in [Figure 1](#F1){ref-type="fig"} and its legend. Shortly, after irradiation, animals (3 animals per cage) were kept for 12 or 24 h, according to the group assignment. Finally, under deep anesthesia (isoflurane 4 %) blood was collected by retro-orbital phlebotomy into EDTA tubes, and animals were euthanized by cervical dislocation. Plasma was obtained by centrifugation (1,600 g, 5 min, 4 C), skin, brain and adrenals harvested and frozen immediately at −80 C or fixed with paraformaldehyde. We repeated experiments 3 times during the spring, summer and winter to avoid confounding effects caused by the seasonal fluctuation of the HPA axis activity ([@R6]). To test the role of the pituitary, we employed the hypophysectomized (hypox) and sham operated controls (sham-hypox) mice. The animals were treated as above. Blood and tissues were collected in the morning (6 -- 7 am, 12 and 24 h after irradiation). Irradiation {#S16} ----------- UVB (290--320 nm) irradiation was performed with a Spectroline XX-15A lamp (Spectronics Corp., Westbury, NY) equipped with a UVB waveband bulb (USHIO G15T8E) from the distance of 2.5 inches (see [Table S1](#SD1){ref-type="supplementary-material"}. The UV dosimetry was described in ([@R37]). During irradiation, the bulb was covered with a cellulose triacetate sheet (Kodacel filter, Kodacel™, Eastman Kodak, Rochester, NY) which cuts-off wavelengths shorter than 290 nm, as described in ([@R37]). The time of UVB irradiation was calculated upon the formula Time (s) = Dose (J/cm^2^)/Intensity (W/cm^2^) and presented as a standard erythema dose (SED) and MED in [Table S1](#SD1){ref-type="supplementary-material"}. One SED is equivalent to an erythemal effective radiant exposures of 100 J/m^2^ (MKS) or 0.01 J/cm^2^ (CGS)([@R13]). Quantitative real time RT-PCR (QRT-PCR) {#S17} --------------------------------------- A detailed description of QRT-PCR is given in Supplemental materials and methods. The list of primers used for amplification with SYBR Green polymerase (Kapa Biosystems, Inc., Woburn, MA) is in [Table S2](#SD1){ref-type="supplementary-material"}. Results are presented as fold change based on the ΔΔ ct method ± SD. ELISA/EIA {#S18} --------- A detailed description is given in Supplemental materials and methods. Commercially available kits used in this study are listed in [Table S3](#SD1){ref-type="supplementary-material"}. Results are presented as mean ± SD in either pg/mL or ng/mL after recalculations by total protein concentration (brain, skin) or dilution (plasma). Immunohistochemistry {#S19} -------------------- Detailed protocols are in supplementary file. Briefly, the hypothalamus was isolated at the level of anterior Bregma +1 mm up to posterior Bregma −2.70 mm, by the use of the Brain Slicer Matrix (Zivic Instrument, Pittsburgh, PA). Region resembling the whole hypothalamus (Bregma \~ −0.34 to −2.70) were characterized under light microscope based on the The Allen Reference Atlas (<http://mouse.brain-map.org/static/atlas>) and 10 µm coronal sections were mounted onto silanized slides (Dako, Carpinteria, CA), and subjected to immunofluorescence protocols described in supplementary file. Western Blot {#S20} ------------ A detailed description is in ([@R35]). Briefly, equal amounts of protein from a combination of 3 skins for each sample was denatured with Laemli buffer, subjected to SDS/PAGE, and proteins were transferred to a PVDF membrane and incubated with antibodies listed in [Table S4](#SD1){ref-type="supplementary-material"}. Next, the membrane was incubated with secondary IgG-HRP, and detection of immunocomplexes was performed with chemiluminescence. Statistics {#S21} ---------- Data are presented as means ± SD and are analyzed using Prism 4.00 (GraphPad Software, San Diego, CA). Statistically significant differences are denoted by \* (Student's *t*-test for two groups) or \# (one-way ANOVA Tuckey test for more than two groups), where *p*\<0.05 is considered as statistically significant. Supplementary Material {#SM} ====================== The paper is dedicated to the memory of Dr Wylie Vale. We gratefully acknowledge careful reading of the manuscript and recommendations by Dr Jean Rivier (Salk Institute), Dr John Pawelek (Yale University) and Drs Arnold Postlethwaite and Robert Williams (UTHSC). We also thank Ms Deborah Doan (Salk Institute) for the editing of the manuscript. This study was supported by grants from the National Science Foundation (\# IOS-0918934) and the National Institutes of Health (\#1R01AR056666-01A2 and R01AR052190) to AS. **Conflict of interest:** The authors state no conflict of interest. ACTH : adrenocorticotropic hormone α-MSH : α-melanocyte stimulating hormone β-END : β-endorphin B6 : C57BL/6 mice CIE : Commission Internationale de L\'Éclairage (International Commission on Illumination) CORT : corticosterone COR : cortisol CRH : corticotropin releasing hormone CRH-R1 : CRH receptor type 1 CYP11B1 : gene coding of Steroid 11β-hydroxylase P450scc : cytochrome P450 side-chain cleavage enzyme DRG : dorsal root ganglia MC2R : Melanocortin receptor 2 POMC : proopiomelanocortin PVN : paraventricular nucleus SED : standard erythemal dose StAR : steroidogenic acute regulatory protein Ucn : Urocortin 3β-HSD : 3-β-hydroxysteroid dehydrogenase {#F1} {#F2} {#F3} {#F4} {#F5} {#F6}

INTRODUCTION {#s1} ============ One of the most frequently deregulated pathways in human cancers is the phosphoinositide 3-kinase (PI 3-K) and Akt signaling cascade \[[@R1]\]. This is particularly evident in breast cancer where mutations exist in virtually all of the proteins that lead to activation of PI 3-K and its downstream effectors. Activation of cell surface receptors, particularly receptor tyrosine kinases (RTKs), leads to activation of class I PI 3-K, which catalyzes the synthesis of PtdIns-3,4,5-P3 (PIP3) from PtdIns-4,5-P2 (PIP2). PIP3 functions as a true second messenger by recruiting multiple effector molecules, one of which is the Akt/PKB protein kinase that directly binds to PIP3 through its pleckstrin homology (PH) domain. This binding facilitates the phosphorylation of Akt at Thr308 and Ser473 mediated by phosphoinositide-dependent kinase-1 (PDK-1) and mammalian target of rapamycin complex 2 (mTORC2), respectively \[[@R2]\]. Finally, activated Akt translocates to multiple cellular compartments where it phosphorylates a large number of substrates that transduce the signal to a variety of cellular responses that are intimately associated with malignancy, including cell proliferation, growth, motility, survival and metabolic reprogramming \[[@R3]\]. The Akt family comprises three isoforms - AKT1 (PKBα), AKT2 (PKBβ), and AKT3 (PKBγ) - which have non-redundant functions in various physiological as well as pathophysiological conditions \[[@R4]\]. The most frequent genetic lesions in this pathway comprise oncogenic mutations in *PIK3CA*, the gene that encodes the p110α isoform of class I PI 3-K \[[@R5]\]. Other breast cancer mutations prevalent in this pathway include mutational or epigenetic inactivation of Phosphatase and Tensin Homolog (*PTEN*), a lipid phosphatase that terminates PI 3-K signaling by dephosphorylating PIP3. Both oncogenic *PIK3CA* mutations and inactivation of *PTEN* lead to excessive and constitutive activation of Akt and downstream signaling, resulting in uncontrolled proliferation and increased cellular survival \[[@R6]\]. Mutations and amplifications in the Akt genes themselves have also been identified in various human solid tumors. In the context of breast cancer, the *AKT1(E17K)* somatic mutation was first identified in breast cancer but is also found in lung, bladder, endometrial, urothelial and prostate cancers \[[@R7]-[@R13]\]. The frequency of the *AKT1(E17K)* mutation in breast cancers ranges from 4-8%. This oncogenic mutation renders Akt constitutively active by broadening the lipid specificity of the Akt PH domain \[[@R14]\], thus enabling its transforming capacity in fibroblasts *in vitro* and leukemias *in vivo \[[@R7]\]*. Interestingly, in breast cancer *AKT1(E17K)* is mutually exclusive with *PIK3CA* and *PTEN* mutations \[[@R15]\], although in other cancers such as endometrial carcinoma, these mutations frequently co-exist in the same tumor \[[@R12]\]. Furthermore, *AKT1(E17K)* has been found predominantly in estrogen receptor (ER)-positive breast tumors \[[@R16]\]. However, *in vitro* studies have provided inconclusive information regarding the functional advantages this oncogenic mutation confers \[[@R17]\]. Expression of AKT1(E17K) has been shown to enhance cell migration and resistance to chemotherapeutic agents in luminal breast cancer cells \[[@R17], [@R18]\]. Similarly, knock-in of the *AKT1(E17K)* mutation into MCF-7 ER-positive cells in which oncogenic *PIK3CA(E545K)* has been restored to the wild-type allele restores proliferation and tumor growth *in vivo*, arguing that at least in ER-positive luminal breast cancer cells, *AKT1(E17K)* can function as a *bona fide* oncogene \[[@R19]\]. It is also worth noting that an analogous E17K mutation has been identified in *AKT2* in one breast cancer patient \[[@R20]\] and in *AKT3* in melanomas \[[@R21]\]. Moreover, a recurrent MAGI3-Akt3 fusion protein that results in a truncated form of the *MAGI3* gene fused in frame to *AKT3* at the E17 residue of Akt3 has been identified in breast cancers \[[@R22]\]. The mechanisms by which any of these somatic mutations contribute to malignancy have yet to be reported. To date, no studies have examined the capacity of *AKT1(E17K)* to drive mammary cancer in a genetically engineered mouse model. Previous studies have addressed the contribution of AKT1 activity to mammary tumorigenesis using constitutively active AKT1 transgenes driven by the mouse mammary tumor virus (MMTV) promoter. MMTV-MyrAKT1 mice treated with DMBA to induce chemical carcinogenesis develop ER-positive mammary tumors \[[@R23]\]. In addition, transgenic mice harboring a phospho-mimetic *AKT1(T308D/S473D)* mutant in combination with *HER2* display a decrease in tumor latency and accelerated tumor growth, but decreased incidence of metastases, consistent with AKT1 functioning as a metastasis suppressor \[[@R24], [@R25]\]. Studies using AKT1 and AKT2 knockout mice have arrived at similar conclusions \[[@R26]\]. Since any association between AKT1 and ER has not been explored *in vivo,* and there are no models to evaluate the contribution of *AKT1(E17K)* to mammary tumorigenesis, we generated a mammary-specific inducible *AKT1(E17K)* transgenic mouse. We present evidence indicating that *AKT1(E17K)* is not sufficient for transformation *in vivo*, but induces mammary gland hyperplasia in both virgin and multiparous females. In addition, combination of MMTV-driven *AKT1(E17K)* with MMTV-*HER2* overexpression prevents *HER2-*driven mammary tumor formation, in part through negative feedback inhibition of RTK signaling mediated by AKT1(E17K). RESULTS {#s2} ======= AKT1(E17K) escapes negative feedback inhibition but does not enhance proliferation of mammary epithelial cells *in vitro* {#s2_1} ------------------------------------------------------------------------------------------------------------------------- To study the functional significance of the *AKT1(E17K)* mutation in breast cancer, we developed a system to stably express either wild-type *AKT1* or *AKT1(E17K)* in a doxycycline-inducible manner in the non-tumorigenic immortalized MCF10A breast epithelial cell line. Cells were serum-starved overnight and stimulated with 5% serum. Consistent with previous studies \[[@R17]\], basal phosphorylation of AKT1(E17K) at Ser473 and Thr308 is moderately elevated compared to wild-type AKT1 (Figure [1A](#F1){ref-type="fig"}). However, this does not translate into significantly enhanced phosphorylation of downstream Akt substrates as measured with a substrate-directed Akt motif antibody, as well as antibodies against known Akt substrates (Figure [1A](#F1){ref-type="fig"}). This is despite the fact that in a cell-free system, isolated AKT1(E17K) has significantly elevated protein kinase activity toward the model substrate GSK-3β, again when compared to wild-type AKT1. Apparently, this enhanced intrinsic kinase activity is not sufficient to propagate signals to constitutive downstream substrate phosphorylation in the absence of stimuli. Consistently, AKT1(E17K) cannot promote the proliferation of cells in the absence of serum and growth factors (Figure [1C](#F1){ref-type="fig"}), nor does it provide a proliferative advantage in full growth media (data not shown). {#F1} Since multiple feedback loops exist in the PI 3-K and Akt pathway to maintain homeostatic control, and oncogenic mutations in genes that modulate this pathway can often escape this feedback regulation, we next evaluated the kinetics of AKT1(E17K) activation in MCF10A cells. Cells expressing wild-type AKT1 show a robust induction of Akt phosphorylation in response to IGF-1 as early as 2 min, translating into Akt substrate phosphorylation (pPRAS40). This activation attenuates by 1 h post-stimulation and returns to basal levels by 24 h (Figure [2A](#F2){ref-type="fig"}), as would be expected by feedback inhibition. By contrast, cells expressing AKT1(E17K) show sustained Akt activation (pSer473, pThr308, pPRAS40) out to 24 h post-stimulation (Figure [2B](#F2){ref-type="fig"}). The kinetics of endogenous Akt2 phosphorylation are still subject to feedback inhibition as judged by phosphorylation of pSer474 in both wild type and AKT1(E17K) expressing cells. This indicates that AKT1(E17K) specifically escapes feedback inhibition, allowing for sustained signal propagation. {#F2} AKT1(E17K) transgenic mice develop mammary hyperplasia associated with increased estrogen receptor expression {#s2_2} ------------------------------------------------------------------------------------------------------------- The above data demonstrate that the *AKT1(E17K)* mutation is not sufficient to drive proliferation in non-tumorigenic breast epithelial cells *in vitro*. However, MCF10A cells do not express estrogen receptor, and the *AKT1(E17K)* mutation is strongly associated with ER-positive tumors in patients \[[@R16]\]. Furthermore, the majority of ER-positive breast cancer cell lines also harbor *PIK3CA* or *PTEN* mutations, which are mutually exclusive with *AKT1(E17K)* \[[@R15]\]. Therefore, to evaluate the contribution of *AKT1(E17K)* to mammary tumorigenesis in the context of ER expression, we generated a tissue-specific, *AKT1(E17K)* transgenic mouse line by cloning human HA-*AKT1(E17K)* (*hAKT1(E17K)*) into the tetracycline-responsive (tet-off) pTET splice vector. Two founder lines (Tg \#4 and Tg \#9) were selected based on protein expression by crossing candidate founders determined by PCR with a VE-Cadherin-tTA driver mouse for constitutive expression of the transgene (Figure [3A](#F3){ref-type="fig"}). This allowed us to perform expression analysis on livers harvested from newborn pups to expedite the initial screening of protein levels among founder lines. To validate the responsiveness to tetracycline, we added tetracycline to the drinking water of VE-Cadherin-tTA;*hAKT1(E17K)* mice to turn off protein expression (Figure [3B](#F3){ref-type="fig"}). Upon selection of two *hAKT1(E17K)* expressing lines, we validated activity of the transgene in the mammary epithelium using MMTV-tTA driver mice, revealing expression of the transgene using anti-HA IHC in control MMTV-tTA or MMTV-tTA;*hAKT1(E17K)* lines (data not shown). {#F3} To address whether *AKT1(E17K)* is sufficient to drive oncogenic signaling in the mammary epithelium, we initially assessed transformation in virgin female transgenic mice. MMTV-tTA;*hAKT1(E17K)* mice were monitored for the presence of palpable mammary tumors over the course of 1 year of constitutive transgene expression. No palpable tumors were detected in any of the virgin females. Whole mount and histological analyses of mammary glands confirmed the absence of tumors as a result of prolonged MMTV-driven *hAKT1(E17K)* expression, compared to control mice harboring MMTV alone (Figure [4A and 4B](#F4){ref-type="fig"}). However, all mammary glands of transgenic mice demonstrated marked acinar hyperplasia beginning at 6 months of age (Figure [4B](#F4){ref-type="fig"}). Immunohistochemical analysis revealed that the hyperplasia results from an expansion of the myoepithelium, since the hyperplastic mammary acini are negative for CK8, a luminal marker, but maintain expression of CK14, a myoepithelial marker ([Supplementary Figure S1](#SD1){ref-type="supplementary-material"}). Furthermore, expression of the *hAKT1(E17K)* transgene results in increased expression of ER (Figure [4D](#F4){ref-type="fig"}, bottom panel), concomitant with expression of the transgene as revealed by staining for anti-HA and anti-pSer473 (Figure [4D](#F4){ref-type="fig"}, top and middle panels), again compared to control mice (Figure [4C](#F4){ref-type="fig"}). This is consistent with published data showing that the *AKT1(E17K)* somatic mutation is associated with ER-positive breast tumors \[[@R16]\]. {#F4} We next determined whether signaling events contributing to the proliferation and morphogenesis of the mammary gland as a result of pregnancy provide the necessary factors to promote transformation. MMTV-tTA;*hAKT1(E17K)* mice were paired multiple times and monitored for 1 year for evidence of palpable tumors. Similar to virgin females, whole mount and histological analyses confirmed the absence of tumors. However, multiparous females display a more dramatic mammary gland hyperplasia at 1 year compared to virgin females and control MMTV mice (Figure [5A and 5B](#F5){ref-type="fig"}). In addition, expression of *hAKT1(E17K)* also leads to increased ER expression in hyperplastic mammary acini compared to control (Figure [5C](#F5){ref-type="fig"} and Figure [5D](#F5){ref-type="fig"}, bottom panel). Taken together, these data demonstrate that constitutive expression of *hAKT1(E17K)* causes mammary hyperplasia in both virgin and multiparous females with an associated increase in ER expression. We also determined if mammary hyperplasia occurs as a result of *hAKT1(E17K)* signaling during early events in mammary gland development. Tetracycline was added to the drinking water to turn off transgene expression just after the onset of puberty in virgin females at approximately 6-7 weeks of age. Mice were administered tetracycline for a total of 6 months. Mammary glands analyzed by whole mount and histological analysis appear normal, and hyperplasia is not observed (data not shown). These data indicate that the events driving mammary hyperplasia in MMTV-tTA;*hAKT1(E17K)* transgenic animals do not occur as a result of alterations in early mammary developmental programs. {#F5} Estrogen does not enable *AKT1(E17K)* to promote mammary tumorigenesis {#s2_3} ---------------------------------------------------------------------- Numerous studies have indicated multiple interactions between PI 3-K/Akt signaling and ER function in breast cancer. Growth factor signaling promotes ER phosphorylation, which alters receptor conformation, affinity and transcriptional activity \[[@R27]\]. ERα can also directly bind the p85 subunit of PI 3-K, allowing estrogen stimulation to potentiate PI 3-K activity leading to Akt activation \[[@R28]\]. Moreover, overexpression of HER2 simultaneously with ERα in breast cancer modulates endocrine resistance \[[@R29]\]. Stimulation of mammary epithelial cells with estradiol results in increased expression and activity of the estrogen receptor \[[@R30]\], and prolonged exposure to estrogen accelerates mammary transformation in mice \[[@R31]\]. Since the *AKT1(E17K)* somatic mutation is only detected in patients with ER-positive breast tumors \[[@R16]\], we evaluated whether exposure to estradiol accelerates the MMTV-tTA;*hAKT1(E17K)* mammary gland phenotype. Multiparous MMTV-tTA;*hAKT1(E17K)* transgenic mice were exposed to a slow-release pellet of 17β-estradiol for 63 days. Whole mount and histological analyses revealed that prolonged exposure to 17β-estradiol results in an arrested state of lactation in both MMTV-tTA;*hAKT1(E17K)* and MMTV-tTA control mice ([Supplementary Figure S2A and S2B](#SD1){ref-type="supplementary-material"}). We were unable to detect mammary tumors in either MMTV-tTA;*hAKT1(E17K)* or MMTV-tTA control mice; however, hyperplasia was observed in MMTV-tTA;*hAKT1(E17K)* mice ([Supplementary Figure S2C and S2D](#SD1){ref-type="supplementary-material"}). Interestingly, increased pAkt Ser473 staining is observed as a result of estradiol stimulation in the acini of MMTV-tTA control mammary glands, with a further increase in MMTV-tTA;*hAKT1(E17K)* mice (Figure [4C and 4D](#F4){ref-type="fig"}). As expected, elevated expression of ER is also observed in both transgenic and control mammary glands as a result of prolonged estradiol stimulation. Taken together, these results further support a role for *AKT1(E17K)* in mediating upstream and downstream ER signaling. *AKT1(E17K)* suppresses *HER2*-mediated mammary tumorigenesis {#s2_4} ------------------------------------------------------------- Integrative genomic studies have indicated that the *AKT1(E17K)* mutation in breast cancer is mutually exclusive with *HER2* \[[@R16]\]. However, transgenic expression of activated phospho-mimetic or myristoylated AKT1 mutants in the mammary gland in combination with MMTV-*HER2* significantly decreases tumor latency, suggesting that signaling through Akt can enhance the effects of *HER2* \[[@R25], [@R32]\]. By contrast, we find that tumor formation in double MMTV-tTA;*hAKT1(E17K)*;MMTV-*HER2* transgenic mice is completely abolished compared to MMTV-*HER2* alone (Figure [6A-6C](#F6){ref-type="fig"}). However, the MMTV-tTA;*hAKT1(E17K)*;MMTV-*HER2* mammary glands display mammary gland hyperplasia similar to that observed in MMTV-tTA;*hAKT1(E17K)* transgenic mice (Figure [6C](#F6){ref-type="fig"}, compared to Figure [4B](#F4){ref-type="fig"}). MMTV-tTA;*hAKT1(E17K)*;MMTV-*HER2* mice exposed to tetracycline (to turn off transgene expression) in the drinking water develop mammary tumors as expected (data not shown). A previous study has shown that expression of an activated Akt allele in a *HER2*-driven background in the mouse mammary gland leads to decreased expression and phosphorylation of a number of RTKs, including HER3 and EGFR \[[@R32]\]. Consistent with this, we find that MMTV-tTA;*hAKT1(E17K)*;MMTV-*HER2* mice express significantly lower total and tyrosine-phosphorylated EGFR, HER2 and HER3, to the same extent that is observed in MMTV-tTA;*hAKT1(E17K)* mice (Figure [6D](#F6){ref-type="fig"}). Moreover, MMTV-tTA;*hAKT1(E17K)*;MMTV-*HER2* mice show significantly improved survival to the same extent that is seen in control or MMTV-tTA;*hAKT1(E17K)* mice, when compared to MMTV-*HER2* mice (Figure [6E](#F6){ref-type="fig"}). These data are consistent with AKT1(E17K) attenuating RTK signaling through feedback inhibition. {#F6} To determine if AKT1(E17K) controls negative feedback, we examined dose-dependent activation of endogenous Akt2 in MCF10A cells expressing AKT1(E17K). As predicted, activation of endogenous Akt2 is suppressed by AKT1(E17K) expression in cells stimulated with IGF-1, compared to cells expressing wild-type AKT1 (Figure [7A](#F7){ref-type="fig"}). No changes in the total levels of AKT or IGF-1-1R are observed. Finally, we analyzed mRNA expression and reverse-phase protein array (RPPA) data from the breast cancer TCGA data set for tumors harboring *AKT1(E17K)* \[[@R33]\]. Consistent with the findings from the MMTV-tTA;*hAKT1(E17K)*;MMTV-*HER2* mice, analysis of RTKs and RTK-related proteins in the microarray and RPPA data sets reveals that a significant proportion of these show a downward trend in mRNA expression, protein levels, and phosphorylation, although this does not reach statistical significance due to the relatively small number of *AKT1(E17K)* cases (Figure [7B](#F7){ref-type="fig"}). Moreover, total ER (*ESR1*) is elevated in *AKT1(E17K)* tumors (Figure [7C](#F7){ref-type="fig"}), consistent with elevated ER in the MMTV-tTA;*hAKT1(E17K)* mice (Figure [4D](#F4){ref-type="fig"}). Intriguingly, we also note a significant up-regulation of several tumor suppressors in *AKT1(E17K)* tumors, including neurofibromin 2 (*NF2*), *PTEN* and tuberous sclerosis 2 (*TSC2*) (Figure [7C](#F7){ref-type="fig"}). The significance of this up-regulation remains to be determined, but is consistent with *AKT1(E17K)* suppression of *HER2*-driven tumorigenesis. Taken together, these data support a model in which *AKT1(E17K)* is not sufficient to promote signals that promote tumorigenesis because it attenuates RTK signaling through negative feedback control. {#F7} DISCUSSION {#s3} ========== In this study we generated the first genetically engineered mouse model for the oncogenic *AKT1(E17K)* somatic mutation in the mammary gland. This mutation has been detected in approximately 4-8% of breast tumors, with a strong association with ER expression. Although *AKT1(E17K)* can drive transformation of fibroblasts *in vitro* \[[@R7]\] and enhances the survival and migration of luminal breast epithelial cells \[[@R18]\], we find that expression of *AKT1(E17K)* in MCF10A cells does not strongly activate downstream signaling, nor does it enhance proliferation, consistent with previous findings \[[@R17]\]. Consistently, transgenic mice expressing *AKT1(E17K)* driven by the MMTV promoter develop hyperplastic lesions that do not progress to carcinoma. This hyperplasia is concomitant with an increase in ER expression, suggesting that ER activity may enable *AKT1(E17K)* to promote proliferation *in vivo*. Previous studies that have evaluated the functional significance of the *AKT1(E17K)* mutation *in vitro*, including our own, have been primarily performed in ER-negative cell lines. More recently, a study demonstrated that replacing the oncogenic *PIK3CA(E545K)* mutation with wild-type *PIK3CA* in MCF7 luminal breast cancer cells that are ER-positive in the context of *AKT1(E17K)* knock-in leads to increased transformation and xenograft tumor volume, demonstrating that *AKT1(E17K)* can function as an oncogene in the context of an ER-positive, luminal breast cancer \[[@R19]\]. Together with the observation that *AKT1(E17K)* expression leads to different phenotypes in mammary myoepithelial *versus* luminal cells \[[@R18]\], it is likely that this mutation is selected for in human breast cancer under specific genetic and cellular contexts that enable it to function effectively as an oncogene. The finding that *AKT1(E17K)* suppresses *HER2*-mediated mammary tumorigenesis is in stark contrast with previous findings that have examined activated AKT1 transgenes in the same context. Specifically, both phospho-mimetic *AKT1(T308D/S473D)* and Myr-AKT1 transgenes accelerate *HER2*-driven mammary tumor formation \[[@R25], [@R32]\]. There are likely a number of reasons why *AKT1(E17K)* does not phenocopy artificially-activated AKT1 with respect to *HER2*-mediated tumor progression. First, both *AKT1(T308D/S473D)* and Myr-AKT1 have a quantitatively higher constitutive protein kinase activity that is not recapitulated by *AKT1(E17K)*. *AKT1(T308D/S473D)* is constitutively basally hyperactive and not subject to inactivation by dephosphorylation, and similarly Myr-AKT1 is constitutively membrane-localized and phosphorylated. Moreover, molecular dynamics simulations have revealed that the increased membrane association of the AKT1(E17K) mutant is due to rapid conformational changes in the PH domain that likely explain the pathological consequences attributed to this mutation \[[@R34]\]. Another likely explanation that accounts for the suppressed *HER2*-driven tumors in the context of *AKT1(E17K)* is feedback inhibition of RTK signaling. Several lines of evidence support this conclusion. For example, studies have shown that inhibition of Akt leads to up-regulation of several RTKs, including insulin receptor, IGF-1R and HER3 \[[@R35]\]. By extension, one would predict that in the converse situation, *AKT1(E17K)* would suppress RTK expression, and indeed we observe this correlation in both MMTV-tTA;*hAKT1(E17K)*;MMTV-*HER2* mice (Figure [6D](#F6){ref-type="fig"}) and human tumors harboring *AKT1(E17K)* (Figure [7B](#F7){ref-type="fig"}). Although Myr-AKT1 transgenic mice show decreased RTK expression \[[@R32]\], this hyperactive allele is sufficient to propagate downstream signaling to enhance phenotypes that control tumor progression, even in the presence of the negative feedback. By contrast, the relatively lower constitutive activity of AKT1(E17K) is not sufficient to promote downstream signaling (Figure [1A](#F1){ref-type="fig"}). Therefore, *AKT1(E17K)* suppresses HER2 signaling in a manner that is not overcome by its own protein kinase activity, ultimately leading to suppression of tumorigenesis. Since *AKT1(E17K)* expression in the mouse mammary gland does not lead to tumor formation, this begs the question as to why this mutation is detected in 4-8% of breast cancer patients \[[@R7], [@R16]\]. One study revealed that the *AKT1(E17K)* somatic mutation occurs early in breast tumor progression, and therefore one can speculate that subsequent and additional genetic lesions are required for *AKT1(E17K)* to function as an oncogene in breast carcinoma \[[@R36]\]. Since *AKT1(E17K)* escapes feedback inhibition and displays sustained activation kinetics (Figure [2](#F2){ref-type="fig"}), it is likely that additional genetic lesions may potentiate its protein kinase activity. The nature of these genetic lesions remains to be determined, but are not likely to include oncogenic *PIK3CA* mutations or *PTEN* inactivation, which have been shown to be mutually exclusive with *AKT1(E17K)*, at least in breast tumors \[[@R16]\]. Based on the correlation of RTK suppression in the MMTV-tTA;*hAKT1(E17K)*;MMTV-*HER2* mice and in human *AKT1(E17K)* tumors, we also predict that cooperating mutations likely do not occur in signaling pathways that function through RTKs, particularly those in the EGFR family. The findings presented in this study have implications for therapeutic intervention and development of targeted therapies for both RTKs as well as PI 3-K and Akt. There are presently numerous phase I and II clinical trials with a variety of allosteric and catalytic Akt small molecule inhibitors for therapeutic benefit in breast and other solid tumors \[[@R1], [@R37]\]. Since inhibition of Akt using catalytic inhibitors leads to the up-regulation of RTK signaling \[[@R35]\], and *AKT1(E17K)* tumors show suppressed levels and phosphorylation of RTKs, one implication is that breast cancer patients with *HER2* amplification would not benefit from treatment with Akt inhibitors. Instead, combination therapy with Akt inhibitors followed by RTK inhibition may be a more effective therapeutic strategy \[[@R38]\]. It is also important to note that the E17K somatic mutation has also been identified in *AKT2* in breast cancer, albeit at lower frequency, and also in *AKT3* in human melanoma. Whether the same mechanism of RTK suppression and escape of feedback inhibition applies to these mutants remains to determined. Regardless, these findings underscore the importance of understanding the mechanisms by which oncogenes escape feedback inhibition to lead to tumor initiation and progression. MATERIALS AND METHODS {#s4} ===================== Cell lines {#s4_1} ---------- MCF10A cells were obtained from the American Type Culture Collection (ATCC) and authenticated using short tandem repeat (STR) profiling. Cells were maintained in DMEM/Ham\'s F12 supplemented with 5% equine serum (Gibco), 10 mg/mL insulin, 500 ng/mL hydrocortisone (Sigma-Aldrich), 20 ng/mL EGF (R&D Systems), and 100 ng/mL cholera toxin (Sigma-Aldrich). Cells were passaged for no more than 6 months and routinely assayed for mycoplasma contamination. Plasmids {#s4_2} -------- The *AKT1(E17K)* mutation was generated by site-directed mutagenesis (Qiagen) from HA-*AKT1*/pcDNA3 (Addgene). For doxycycline-inducible overexpression of *AKT1* and *AKT1(E17K)*, HA-*AKT1*/pTRIPZ and HA-*AKT1(E17K)*/pTRIPZ were constructed. HA-*AKT1* and HA-*AKT1(E17K)* cDNA was amplified by PCR from HA-*AKT1*/pcDNA3 and HA- *AKT1(E17K)*/pcDNA3. The resulting PCR product was digested with restriction enzymes AgeI and ClaI, followed by insertion into the pTRIPZ lentiviral vector (Thermo Scientific). Immunoblotting {#s4_3} -------------- Cells were washed with PBS at 4°C and lysed in radioimmunoprecipitation assay buffer (1% NP-40, 0.5% sodium deoxycholate, 0.1% SDS, 150 mmol/L NaCl, 50 mmol/L Tris-HCl (pH 7.5), proteinase inhibitor cocktail, 50 nmol/L calyculin, 1 mmol/L sodium pyrophosphate, and 20 mmol/L sodium fluoride) for 15 minutes at 4°C. Cell extracts were pre-cleared by centrifugation at 13,000 rpm for 10 minutes at 4°C, and protein concentration was measured with the Bio-Rad DC protein assay. Lysates were then resolved on 10% acrylamide gels by SDS-PAGE and transferred electrophoretically to nitrocellulose membrane (Bio-Rad) at 100 V for 90 minutes. The blots were blocked in Tris-buffered saline (TBST) buffer (10 mmol/L Tris-HCl, pH 8, 150 mmol/L NaCl, and 0.2% Tween 20) containing 5% (w/v) nonfat dry milk for 1 hour, and then incubated with the specific primary antibody diluted in blocking buffer at 4°C overnight. Membranes were washed three times in TBST and incubated with HRP-conjugated secondary antibody for 1 hour at room temperature. Membranes were washed three times and developed using enhanced chemiluminescence substrate (EMD Millipore). *In vitro* protein kinase assays {#s4_4} -------------------------------- MCF10A cells expressing HA-*AKT1* or HA-*AKT1(E17K)* were serum starved for 16 hours. HA-*AKT1* or HA-*AKT1(E17K)* were immunoprecipitated from cell extracts with an anti-HA antibody and incubated with 300 ng GSK-3 fusion protein peptide (Cell Signaling Technology) in the presence of 150 μmol/L cold ATP in a kinase buffer for 40 min at 30°C. The kinase reaction was terminated by the addition of SDS-PAGE sample buffer. Antibodies {#s4_5} ---------- All antibodies except the anti-HA and anti-p85 antibodies were purchased from Cell Signaling Technology. The anti-HA monoclonal antibody was purified from the 12CA5 hybridoma. The anti-p85 antibody has been described \[[@R39]\]. Proliferation assays {#s4_6} -------------------- MCF10A cells were seeded into 96-well plates at a density of 1500 cells per well in 100 μL medium. The medium was replaced with serum-free medium after 16 h. Cell viability was measured 0, 1, 2, and 3 days after the media change using the water soluble tetrazolium salt WST-1 assay (Clontech) according to the manufacturer\'s protocol. Generation of hAKT1(E17K) transgenic mice {#s4_7} ----------------------------------------- HA-tagged *hAKT1(E17K)* (Addgene) was cloned into the Tetracycline-regulated (Tet-off) vector pTET Splice \[[@R40]\]. DNA was prepared for microinjection, performed at the Transgenic Animal Core Facility at Beth Israel Deaconess Medical Center under IACUC approved protocols. Resulting founder animals generated on an FVB background were genotyped using genomic DNA purified from a tail biopsy and performing PCR utilizing the following primer sequences: FW: 5′-CTGGAATTCATGTACCCATACGATGTTCCAG-3′ RV: 5′-CCT CTTCTTGAGGCCGTCAGCCACAGTCTGG-3′ Validation of protein expression {#s4_8} -------------------------------- HA-*hAKT1(E17K)* protein expression was confirmed by crossing animals positive for the transgene with VE-Cadherin-tTA driver mice \[[@R40]\]. Livers were collected from pups at P10 and screened for protein expression using immunoblot for HA (12CA5 Hybridoma) as previously described \[[@R41]\]. Mammary specific transgene expression {#s4_9} ------------------------------------- *hAKT1(E17K)* transgenic mice that screened PCR positive were crossed with MMTV-tTA mice generated by Henninghausen et. al. as previously described \[[@R42]\], and the offspring were re-genotyped for the presence of the *hAKT1(E17K)* transgene as well as for MMTV-tTA using the primer sequences: FW: 5′-GACGCCTTAGCCATT AGAT-3′ RV: 5′-CAGTAGTAGGTGTTTCCCTTTCTT-3′. Double positive transgenic females were either maintained as virgins or were multi-paired. Double positive males were bred with MMTV-*HER2* homozygous females (Jackson Laboratories, \#002376) and offspring were re-genotyped to confirm the presence of the transgenes. Mammary gland wholemount staining {#s4_10} --------------------------------- Mammary fat pads were removed and placed on glass microslides and allowed to dry for 5 min, and then fixed overnight in Carnoy\'s fixative (60% ethanol, 30% chloroform, 10% acetic acid) at room temperature. After fixation, mammary glands were stained with Carmine alum. Immunohistochemistry {#s4_11} -------------------- Mammary glands were removed and fixed in 10% NBF and processed for histology and evaluated by H&E staining at the Harvard Medical School Rodent Histopathology Core and the Beth Israel Deaconess Medical Center Histology Core. Immunohistochemistry on paraffin sections was performed for α-HA (12CA5 Hybridoma) utilizing a M.O.M kit (Vector labs, BMK-2202). Immunohistochemistry for α-ER (Millipore, \#04-227), α-pAkt Ser473 (Cell Signaling Technology \#4060), α-HER2 (Cell Signaling Technology \#4290), α-pHER2 pTyr877 (Abcam \#ab47262), α-HER3 (Cell Signaling Technology \#12708), α-pHER3 pTyr1289 (Cell Signaling Technology \#4791), α-EGFR (Cell Signaling Technology \#4267), pEGFR pTyr1068 (Cell Signaling Technology \#3777), α-CK8 (Developmental Studies Hybridoma Bank clone Troma-I), and α-CK14 (Biolegend PRB-155) was performed as follows: Briefly, sections were deparaffinized and rehydrated. Antigen unmasking was performed using target retrieval solution (DAKO \#S2367). Endogenous peroxidases were blocked using 3% H~2~0~2~ and immunostaining performed using the Rabbit Vectastain ABC kit (Vector labs \#PK-4001) and developed using a DAB peroxidase substrate kit (Vector labs \#SK-4100). SUPPLEMENTARY FIGURES {#s5} ===================== Grant Support: NIH grants CA177910 (A.T.), F32 CA134056 (M.L.M.) and National Science Foundation Graduate Research Fellowship Program Grant No. DGE1144152 (E.C.L.). We thank Joel Lawitts in the BIDMC Transgenic Core Facility, Roderick Bronson in the HMS Rodent Histopathology Core facility, Andrew Beck for the TCGA analysis and members of the Wenyi Wei and Toker laboratories for advice. **CONFLICTS OF INTEREST** The authors disclose no potential conflicts of interest

1. Introduction {#sec1-ijms-17-01969} =============== Glycosylation is a posttranslational modification that is found on about 50% of all proteins, in particular on secreted and transmembrane proteins of eukaryotes, archaea and to a lesser extent in prokaryotes \[[@B1-ijms-17-01969],[@B2-ijms-17-01969],[@B3-ijms-17-01969]\]. Eukaryotic proteins require glycosylation for proper folding, oligomerization and solubility, while glycans significantly prolong the stability and half-life time in many cases by protection against proteolysis \[[@B4-ijms-17-01969],[@B5-ijms-17-01969]\]. Although *N*-glycosylation is more frequent, *O*-glycosylation can similarly protect against general and specific proteolysis \[[@B6-ijms-17-01969],[@B7-ijms-17-01969],[@B8-ijms-17-01969]\]. Protein trafficking, i.e., the sending of proteins to cellular compartments or to the extracellular matrix, depends on specific, covalently linked glycans \[[@B9-ijms-17-01969]\]. In addition, glycans play an important role in the interaction and recognition of proteins, such as in the context of immunity and cell adhesion \[[@B10-ijms-17-01969],[@B11-ijms-17-01969],[@B12-ijms-17-01969]\]. Glycosylation may even protect against molecular damage by free radicals \[[@B13-ijms-17-01969]\]. In recent years, increasing evidence was found that glycans have distinct effects on the activity of many enzymes, in particular as regulatory modules for substrate binding and turnover. This study gives an overview on the most relevant types of glycosylation of proteases, regarding the structural knowledge and the functions of glycans. The importance of this little investigated field lies in the enormous diversity of possible glycosylation variants and the altered functionality of proteins under healthy or disease conditions. *N*-glycosylation at sequons of the Asn-Xaa-Ser/Thr type is widespread in proteins of archaea and eukaryotes, whereby proline is largely excluded as Xaa and disfavored as residue following Ser/Thr \[[@B14-ijms-17-01969],[@B15-ijms-17-01969]\]. Some rare sequons are Asn-Xaa-Cys (1%), Asn-Gly (0.5%) and Asn-Xaa-Val (\<0.5%) \[[@B16-ijms-17-01969]\]. The process of *N*-glycosylation is extensively described in the literature on glycobiology \[[@B17-ijms-17-01969]\]. Essentially, a newly synthesized polypeptide emerging from a ribosome binds with a signal peptide to a signal recognition particle, which docks to a receptor in the endoplasmic reticulum (ER) membrane and forms a complex with the Sec machinery, which transfers the polypeptide through a transmembrane channel into the lumen of the ER \[[@B18-ijms-17-01969],[@B19-ijms-17-01969]\]. A signal peptidase cleaves the N-terminal signal peptide and the oligosaccharyltransferase complex attaches a GlcNAc~2~Man~9~Glc~3~ precursor at a suitable sequon of the Asn-Xaa-Ser/Thr type \[[@B20-ijms-17-01969]\]. Subsequently, the *N*-glycosylated polypeptide folds in the oxidizing environment of the ER, supported by protein disulfide isomerase for disulfide formation and by various chaperones \[[@B21-ijms-17-01969],[@B22-ijms-17-01969]\]. Afterwards, glucosidases and mannosidases trim the *N*-glycan precursor to Man~5~GlcNAc~2~ or GlcNAcMan~3~GlcNAc~2~ core glycans, which are extended by glucosyltransferases, accompanied by protein quality control and followed by sorting and further processing on their way through the ER Golgi intermediate compartment into the Golgi \[[@B23-ijms-17-01969],[@B24-ijms-17-01969]\]. Final modifications of the *N*-glycans in the Golgi comprise extensions by transferases that attach *N*-acetyl-glucosamine (GlcNAc), fucose, galactose, mannose, and sialic acid sugars, before sorting to secretory vesicles \[[@B25-ijms-17-01969]\]. Variations of branching generate a large diversity of *N*-glycans with distinct composition under physiological and pathological conditions ([Figure 1](#ijms-17-01969-f001){ref-type="fig"}A) \[[@B26-ijms-17-01969]\]. Earlier structural database analyses reported a relatively low percentage of only 27% *N*-glycosylation of all sequons in human proteins, of which 96% belonged to secreted and membrane proteins and 4% to cytoplasmic and nuclear proteins \[[@B28-ijms-17-01969],[@B29-ijms-17-01969]\]. More recent data suggested around 85% occupancy of sequons, with 50% glycosylation of all proteins in the SWISS-PROT sequence data bank \[[@B2-ijms-17-01969]\]. Thorough mass spectrometric analyses demonstrated that in the mouse *N*-glycoproteome the majority of identified sequons is occupied, e.g., 99% of predicted membrane protein *N*-glycosylation sites \[[@B16-ijms-17-01969]\]. Glycosylated Asn residues in human proteins are preferentially located in turns (78%) compared to β-sheets (12%) and α-helices (10%), which resembles the situation in murine proteins \[[@B16-ijms-17-01969],[@B28-ijms-17-01969]\]. The strict structural constraints for *N*-glycosylation are reflected by a similar localization in proteins of fish, insects, plants and lower eukaryotes \[[@B28-ijms-17-01969],[@B30-ijms-17-01969]\]. About 12% of all glycosylated proteins are exclusively *O*-glycosylated, while about 10% of them are both *N*- and *O*-glycosylated \[[@B2-ijms-17-01969]\]. Seven types of *O*-glycosylation have been found in humans ([Figure 1](#ijms-17-01969-f001){ref-type="fig"}B). The mucin-type with *N*-acetylgalactosamine (GalNAc) linked to Ser or Thr in membrane or secreted proteins is more common than *O*-glycosylation with xylose, galactose, glucose, fucose, and mannose, whereas mostly proteins with *O*-linked GlcNAc localize to the cytoplasm and the nucleus \[[@B27-ijms-17-01969]\]. Although no distinct consensus sequence is known, Pro-rich sequences are favored, such as Pro-Ser/Thr-Pro-Xaa-Pro \[[@B31-ijms-17-01969],[@B32-ijms-17-01969]\]. Usually, the first step of *O*-glycosylation of the mucin-type is performed in the Golgi by a GalNAc transferase, followed by extensions and branching, which results in eight core glycan subtypes \[[@B27-ijms-17-01969],[@B33-ijms-17-01969]\]. *O*-glycosylation of the mucin-type is important for tissue development and immune reactions \[[@B34-ijms-17-01969]\]. Among the rarer types of glycosylation is *O*-mannosylation for quality control and protection against proteolysis \[[@B35-ijms-17-01969]\]. Similarly, *O*-glycosylation of hydroxylysine can occur in mammals, in contrast to *O*-glycosylation of hydroxyproline, which is found in other eukaryotes \[[@B36-ijms-17-01969],[@B37-ijms-17-01969]\]. The uncommon *C*-mannosylation of Trp residues plays a role in protein folding, secretion and signaling \[[@B38-ijms-17-01969],[@B39-ijms-17-01969]\]. Glycosylphosphatidyl-inositol-anchored proteins (GPI-APs) are linked to glycolipids as membrane anchors, which is sometimes termed glypation \[[@B9-ijms-17-01969],[@B40-ijms-17-01969]\]. By contrast, glycation is the uncatalyzed covalent linkage of glucose or fructose with amino groups, which is involved in diabetes and ageing related diseases \[[@B41-ijms-17-01969],[@B42-ijms-17-01969]\]. Apart from this unregulated process, most glycans play very well defined roles in physiology, such as in the regulation of enzymatic activity. 2. Glycosylated Proteases {#sec2-ijms-17-01969} ========================= Currently, the UniProtKB database lists about 250 glycosylated human proteases out of more than 700, with changing numbers due to putative or ambiguous classification \[[@B43-ijms-17-01969]\]. They exhibit either experimentally identified or predicted *N*-, *O*-, and *C*-glycosylation sites, using the NetNGlyc and NetOGlyc tools, which are currently applicable to mammalian entries and the mucin-type of *O*-glycans \[[@B29-ijms-17-01969],[@B44-ijms-17-01969]\]. The MEROPS database of proteolytic enzymes includes about 1200 known and putative human proteases, including inactive protease homologs \[[@B45-ijms-17-01969]\]. 2.1. Aspartic Proteases {#sec2dot1-ijms-17-01969} ----------------------- An overview of the glycosylated human aspartic proteases shows the 11 confirmed or potentially glycosylated members out of altogether 28, which agrees with their prevalent extracellular localization ([Table 1](#ijms-17-01969-t001){ref-type="table"}). The prototypic aspartic protease pepsin is not glycosylated, despite its secretion into the stomach, where it cleaves proteins between hydrophobic residues, such as P1-Leu and P1'-Leu according to the Schechter--Berger nomenclature \[[@B46-ijms-17-01969]\]. The related, brain expressed β-secretase 1 (BACE1) is a key player in neuronal regulation and a drug target in Alzheimer´s disease \[[@B47-ijms-17-01969]\]. Although the UniProtKB database annotates BACE1 only as *N*-glycosylated according to manual assertion, a recombinant variant carries *N*-glycans according to mass spectrometry and PNGase F treatment \[[@B48-ijms-17-01969]\]. 2.2. Cysteine Proteases {#sec2dot2-ijms-17-01969} ----------------------- Another important enzyme class is formed by the cysteine proteases, of which only 12 out of 158 human members exhibit glycosylation sites, among them several cathepsins ([Table 1](#ijms-17-01969-t001){ref-type="table"}). This phenomenon can be easily explained by the predominant intracellular location with reducing conditions that are required for the activity of many cysteine proteases. Cysteine cathepsins occur in endolysosomes and under pathological conditions even more often outside cells, where they can be regulated by glycosaminoglycans \[[@B49-ijms-17-01969]\]. For example, the important protease families of the apoptosis- and inflammation-related caspases and of the Ca^2+^-signal mediating calpains do not possess any ascertained glycosylation site \[[@B50-ijms-17-01969],[@B51-ijms-17-01969]\]. However, legumain, a monomeric caspase-like cysteine protease, is found in the acidic lysosomes and outside cells, carrying at least three *N*-glycans in its recombinant form, which was generated by *Leishmania* expression \[[@B52-ijms-17-01969],[@B53-ijms-17-01969]\]. 2.3. Metalloproteases {#sec2dot3-ijms-17-01969} --------------------- The mostly Zn^2+^-dependent metalloproteases are secreted from cells to a large extent, with 108 of 163 members carrying at least one glycan ([Table 2](#ijms-17-01969-t002){ref-type="table"}). Among them are the extracellularly located, soluble or membrane-anchored matrix metalloproteinases (MMPs), which play important roles in development, wound healing, brain processes, and cancer \[[@B54-ijms-17-01969]\]. Their major task is the degradation of matrix proteins by cleaving at larger hydrophobic P1′ residues, which can be supported by collagen-binding hemopexin domains \[[@B55-ijms-17-01969]\]. The majority of MMPs is *N*-glycosylated, although six of them appear to be non-glycosylated, while MMP-25 is potentially GPI-anchored \[[@B56-ijms-17-01969]\]. Several MMPs possess *N*-glycosylation sites, which are required for protein interaction and secretion as observed for MMP-9 \[[@B57-ijms-17-01969]\]. MMP-9 is the remarkable example of a protease with a heavily *O*-glycosylated, Pro-rich linker domain, containing more than a dozen branched mucin-type core 2 glycans \[[@B58-ijms-17-01969]\]. Intriguingly, tumor cells produce MMP-9 mainly with *O*-glycans of mucin core 1, which reduces the binding affinity to its ligand galectin-3 \[[@B59-ijms-17-01969]\]. The two families of transmembrane and GPI-anchored ADAM and ADAMTS proteases, which cleave extracellular portions of other membrane proteins and matrix proteins, are involved in cell adhesion, growth factor shedding, cell migration, morpho-genesis and cancer \[[@B60-ijms-17-01969],[@B61-ijms-17-01969]\]. An outstanding example is ADAM17 or tumor necrosis factor α converting enzyme (TACE), which sheds ectodomains of numerous substrates depending on *O*-glycosylation of cleavage sites \[[@B62-ijms-17-01969]\]. Similarly, the heavily *N*-glycosylated meprin-α and -β act as sheddases and cleave pro-inflammatory cytokines \[[@B63-ijms-17-01969],[@B64-ijms-17-01969]\]. The membrane-bound meprin-β subunits tend to form dimers, whereas α and β subunits can form heterodimers and α subunits may form even higher oligomers \[[@B65-ijms-17-01969]\]. Metallocarboxypeptidases are another relevant subgroup of secreted neurohormone and cytokine processing enzymes, which have been in implicated in diseases of the pancreas, diabetes and cancer and are considered nowadays as promising drug targets \[[@B66-ijms-17-01969],[@B67-ijms-17-01969]\]. To date, the developmentally and pathophysiologically relevant Wnt-signal regulating Tiki 1 and 2 proteases belong to a superfamily of hardly characterized, unique Co^2+^/Mn^2+^-dependent enzymes \[[@B68-ijms-17-01969]\]. 2.4. Serine Proteases {#sec2dot4-ijms-17-01969} --------------------- Serine proteases constitute another large group with about 106 glycosylated members out of 144 according to UniProtKB, with family S1 being the largest one ([Table 3](#ijms-17-01969-t003){ref-type="table"}). The prototypic digestive proteases trypsin, chymotrypsin and elastase are produced in the pancreas and define the corresponding major substrate specificities, referring to the S1 subsites. Among the digestive proteases, only the membrane-anchored enteropeptidase, also known as enterokinase, is heavily *N*-glycosylated and activates trypsinogen, which in turn activates other digestive enzymes \[[@B69-ijms-17-01969]\]. The blood coagulation cascade involves several trypsin-like proteases, which leads to the activation of prothrombin by factor Xa, resulting eventually in thrombin cleavage of fibrinogen and formation of fibrin clots, which stop bleeding \[[@B70-ijms-17-01969]\]. *N*-glycosylated variants of factor VII propeptides prolong its half-life significantly, while corresponding variants of factor IX can also play a role in the treatment of hemophilia \[[@B71-ijms-17-01969],[@B72-ijms-17-01969]\]. Regulation of the fibrinolytic plasmin depends on differential *N*- and *O*-glycosylation, which alters the structure significantly \[[@B73-ijms-17-01969],[@B74-ijms-17-01969]\]. In addition, both the tissue-type plasminogen activator (tPA) and urokinase-type plasminogen activator (uPA) possess *N*- and *O*-glycans, whose roles have been partially elucidated \[[@B75-ijms-17-01969],[@B76-ijms-17-01969]\]. Kallikrein 1 (KLK1) and the other fourteen kallikrein-related peptidases (KLKs) fulfill numerous crucial tasks in tissue development, reproduction, wound healing or neuronal processes \[[@B77-ijms-17-01969]\]. Since several KLKs exhibit altered expression in various cancers, they are used as biomarkers, e.g., KLK3/PSA (prostate specific antigen), and investigated as drug targets \[[@B78-ijms-17-01969],[@B79-ijms-17-01969]\]. Moreover, most human KLKs are *N*-glycosylated proteins (except KLK14), for which potential functional roles have been described \[[@B80-ijms-17-01969],[@B81-ijms-17-01969]\]. Secretory proprotein convertases (PPCs), among them the *N*-glycosylated furin and PPCs 4, 5, 6, 7, and 9, process many substrates upon secretion ([Table 4](#ijms-17-01969-t004){ref-type="table"}) \[[@B82-ijms-17-01969]\]. Up to now, no distinct functions of their *N*-glycans were reported, whereas *O*-glycosylation near the cleavage sites of their substrates contributes to the regulation of turnover \[[@B83-ijms-17-01969]\]. As observed for many glycoproteins, *N*-glycosylation is critical for folding, secretion and stability of tripeptidyl peptidase I \[[@B84-ijms-17-01969]\]. Interestingly, the hereditary disease late infantile neuronal ceroid lipofuscinosis is caused by a mutation resulting in the loss of an *N*-glycan and all enzymatic activity \[[@B85-ijms-17-01969]\]. The α/β-hydrolases dipeptidyl peptidase-4 (DPP4) and prolyl endopeptidase (FAP) are multifunctional transmembrane enzymes, which cleave substrates after P1-Pro residues and serve as cell surface receptors \[[@B86-ijms-17-01969],[@B87-ijms-17-01969]\]. Apparently, the heavy *N*-glycosylation of DPP4 does not influence dimerization, activity, and T-cell protection \[[@B88-ijms-17-01969]\]. 2.5. Threonine Proteases {#sec2dot5-ijms-17-01969} ------------------------ Threonine proteases are both rare and widespread, due to the central role of the ubiquitin-proteasome system for eukaryotic protein degradation, which is highly conserved from yeast to mammals ([Table 4](#ijms-17-01969-t004){ref-type="table"}) \[[@B89-ijms-17-01969]\]. Besides 14 inactive α-subunits, the proteasome comprises 14 β-subunits in the two inner rings, of which only six are active and cleave after hydrophobic (β1), acidic (β2), and basic (β5) S1 side chains, which is slightly modified in immunoproteasomes \[[@B90-ijms-17-01969]\]. Proteasomes generate peptides that are presented by most cells as epitopes on MHC class I molecules to cytolytic T-cells \[[@B91-ijms-17-01969]\]. Seemingly, *O*-glycosylation is present in mammalian, but not in yeast 26 proteasomes, while the role of the mucin-type *O*-glycans on four α-subunits and the β1 and probably the β6 subunit remains unclear \[[@B92-ijms-17-01969],[@B93-ijms-17-01969]\]. 3. Effects of Glycosylation on Proteases {#sec3-ijms-17-01969} ======================================== 3.1. Effects of Glycosylation on Folding, Sub-Cellular Distribution and Secretion of Proteases {#sec3dot1-ijms-17-01969} ---------------------------------------------------------------------------------------------- Usually, *N*-glycosylation is required for sufficient expression, efficient secretion, and protein trafficking, which holds true for numerous proteases. Glycans facilitate the folding of polypeptides, enhance the solubility of proteins and prevent aggregation \[[@B94-ijms-17-01969]\]. Chymotrypsin C, a pancreatic regulator of trypsin, exhibits two *N*-glycans in the propeptide and the catalytic domain, which serve folding and secretion, but have no impact on activation and activity \[[@B95-ijms-17-01969]\]. A comparison of a fully *N*-glycosylated *Pseudomonas* elastase with its triple mutant Asn43Gln/Asn212Gln/Asn280Gln in the catalytic domain showed virtually no differences in enzyme activity, in contrast to significantly reduced secretion levels \[[@B96-ijms-17-01969]\]. Inspection of this elastase structure (1EZM) revealed that the glycans are remote from the active site, in line with the assumption that the glycans are responsible for proper folding alone as basis for secretion \[[@B96-ijms-17-01969],[@B97-ijms-17-01969]\]. In addition, *N*-glycosylation can be crucial in recombinant expression, as demonstrated for the trypsin-like human KLK5, which could be only expressed in insect cells with a single core *N*-glycan, while expression in *E. coli* cells and subsequent in vitro refolding was not feasible \[[@B98-ijms-17-01969]\]. Similarly, human legumain expression for structure--function studies was largely improved in *Leishmania* cells, which generate short core *N*-glycans \[[@B52-ijms-17-01969]\]. In the case of ADAMTS-9, three *N*-linked glycosylation sites in the propeptide are critical for proper secretion and subcellular localization \[[@B99-ijms-17-01969]\]. Similarly, a mutagenesis study of the human aspartic peptidase renin, which regulates blood pressure by cleavage of angiotensinogen, revealed that *N*-glycosylation at Asn5 and Asn75 plays only a role for secretion \[[@B100-ijms-17-01969]\]. In addition, natural *N*-glycosylation regulates uptake and metabolic clearance, as shown for variants of rat renin \[[@B101-ijms-17-01969]\]. 3.2. Effects of Glycosylation on Activation and Stability of Proteases {#sec3dot2-ijms-17-01969} ---------------------------------------------------------------------- *N*-Glycosylation influences the activation of proteases, both with glycans that are located in the propeptides and in the catalytic domain. Several human proteases possess *N*-glycosylated propeptides, such as KLKs 10 and 13 ([Table 3](#ijms-17-01969-t003){ref-type="table"}). These glycans may regulate protease activation, similar as in the trypsin-like Der p 3, a dust mite allergen, whose *N*-glycan at position P3 with respect to the cleavage site slows down the activation process \[[@B102-ijms-17-01969]\]. In addition, glycans in the catalytic domain of a proform can be crucial as demonstrated by tunicamycin induced loss of *N*-glycosylation at Asn968 and Asn1087 of rat corin, whose activation failed completely \[[@B103-ijms-17-01969]\]. However, for a recombinant form of the human cysteine protease cathepsin B, no effect of *N*-glycosylation on activation, stability and enzymatic activity was found \[[@B104-ijms-17-01969]\]. Many glycosylation sites protect against proteolysis in sensitive regions, which represents often autoproteolysis of the investigated proteases. For example, the aspartic protease cathepsin E lost rapidly activity and stability when its *N*-glycans were enzymatically cleaved \[[@B105-ijms-17-01969]\]. Enzymatic removal of two *N*-glycans in the 75- and 99-loops from human mast cell chymase did not alter kinetic parameters, only the enzymatic activity decreased faster, suggesting that the glycans protect the exposed surface loops from autoproteolysis \[[@B106-ijms-17-01969]\]. Even the branching of *N*-glycans can contribute to protein stability, as demonstrated for matriptase, which was more resistant to proteolysis by trypsin with fucose β1-6 linked to GlcNAc than with the unbranched *N*-glycan in the protease domain \[[@B107-ijms-17-01969]\]. 3.3. Effects of Glycosylation on Substrate Binding and Turnover {#sec3dot3-ijms-17-01969} --------------------------------------------------------------- Regardless of the protease type, it is more likely that glycans in the vicinity of the active site influence substrate binding in the specificity subsites, which depend on numerous biophysical interactions ([Figure 2](#ijms-17-01969-f002){ref-type="fig"}A). Several examples of aspartic, metallo- and other proteases, which exhibit enzyme substrate (Michaelis) complex and tetrahedral intermediates during catalysis, confirm that the presence or absence of *N*-glycans can considerably alter substrate binding and turnover ([Figure 2](#ijms-17-01969-f002){ref-type="fig"}B--D). Nevertheless, the overall presence or absence of *N*-glycosylation plays no role for the function of some proteases, such as human complement factor I \[[@B108-ijms-17-01969]\]. Numerous examples of *N*- and *O*-glycosylated and non-glycosylated variants of enzymes demonstrate that the presence of glycans can have very diverse effects. Single mutations of the glycosylated Asn632 and Asn651 of the metalloprotease endothelin-converting enzyme had no effect, while the double mutant was completely inactive \[[@B109-ijms-17-01969]\]. A series of sequon mutants of mouse meprin-α showed that six out of ten *N*-glycans are critical for catalysis \[[@B110-ijms-17-01969]\]. The completely deglycosylated snake venom RVV-X, a metalloprotease, which can activate coagulation factor X, exhibited no change in *K*~M~ and a 130-fold reduced *k*~cat~ \[[@B111-ijms-17-01969]\]. Since all *N*-glycans of RVV-X are distant from the active site (PDB code 2E3X) and substrate binding was not hampered, one has to conclude that the overall conformation was significantly affected by their absence \[[@B112-ijms-17-01969]\]. Complete deglycosylation of MT1-MMP (MMP-14) in the *O*-glycan rich linker resulted in a stable and active protease, which, however, failed to bind the tissue inhibitor of metalloproteinases-2 (TIMP-2) and could not form the MT1-MMP/TIMP-2/pro-MMP-2 activation complex \[[@B113-ijms-17-01969]\]. By contrast, removal of glycans enhances the activity of several proteases, most likely because the accessibility of the active site increases. Insect cell expressed ADAM17 with short *N*-glycans has an up to 30-fold increased catalytic efficiency compared to a mammalian cell expressed variant with more complex *N*-glycans, which could interfere with substrate binding \[[@B121-ijms-17-01969]\]. Nevertheless, *O*-glycosylation near the scissile bond of ADAM17 substrates enhances the turnover significantly \[[@B122-ijms-17-01969]\]. *N*-glycosylated plasma kallikrein (KLKB1) has a lower catalytic efficiency than deglycosylated KLKB1, which exhibits an altered cleavage pattern for the substrate insulin \[[@B123-ijms-17-01969]\]. The deletion of a single *N*-glycan in the kringle 2 domain of tPA form II enhanced fibrin binding and the resulting fibrinolytic activity by plasmin \[[@B124-ijms-17-01969]\]. Fully deglycosylated tPA is nearly 4-times more active against small chromogenic substrates, whereby a higher mannose content or removal of sialic acids stimulates the activity as well \[[@B125-ijms-17-01969],[@B126-ijms-17-01969]\]. Since the *O*-glycan in the EGF-like domain and the *N*-glycans in both kringle domains can hardly interfere with small substrates, the glycosylated Asn173 in the 176-loop of the protease domain is most likely critical for the observed effect. 3.4. Effects of Glycosylation on Protease Structures {#sec3dot4-ijms-17-01969} ---------------------------------------------------- Structural studies explain to some extent how glycosylation influences the protein conformation. Small angle X-ray scattering data of an *N*-glycosylated fungal enzyme and its deglycosylated counterpart suggested that the glycans limit the torsion angle range of the polypeptide in general \[[@B127-ijms-17-01969]\]. Currently, only a few studies provide sufficient structural and functional information on glycosylated enzymes and their non-glycosylated variants. Several studies of mammalian lipases, which are closely related to proteases, emphasize the stabilizing role of *N*-glycans, in particular of those close to the active site, for the conformation with the highest activity \[[@B128-ijms-17-01969],[@B129-ijms-17-01969],[@B130-ijms-17-01969],[@B131-ijms-17-01969]\]. Thus, glycans that are located distant from the active site, as seen for the aspartic protease cathepsin D, cannot directly influence substrate recognition and turnover, while these *N*-glycans are important for protein interactions in lysosomal targeting ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}A) \[[@B132-ijms-17-01969]\]. A major consequence of the conformational stabilization conferred by *N*-glycosylation might be an efficient interaction with substrates and regulation of substrate access, which can enhance the specificity. For both cathepsin C and meprin-β a single *N*-glycan at the substrate binding cleft near the catalytic residues appears to be crucial for structure stabilization and substrate binding ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}B,C) \[[@B110-ijms-17-01969],[@B133-ijms-17-01969]\]. Although in some cases the exact role of individual glycans is unclear, it can be deduced that they support oligomerization or interaction with other proteins. The *O*-glycans of recombinant Carboxypeptidase *N* are compatible with substrate binding and tetramerization, however, their function has not been defined yet ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}D) \[[@B134-ijms-17-01969]\]. Double Ala mutants of the *O*-glycosylated Ser52 and Ser60 in the EGF-like domain of coagulation factor VIIa exhibited about 14% of the coagulant activity of wild-type FVIIa, while the amidolytic activity was unchanged, indicating that both *O*-glycans are relevant for the association with tissue factor ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}E) \[[@B135-ijms-17-01969]\]. A comparison of recombinant FVII and plasma FVII confirmed that *N*-glycans with terminal GalNAc instead of sialic acids enhance activity and association with tissue factor \[[@B136-ijms-17-01969],[@B137-ijms-17-01969]\]. Subtle variations of the *O*-glycans at Thr346 (form 1) or Ser248 (form 2) and of the *N*-glycan at Asn288 in plasmin influence binding of the inhibitor α~2~-antiplasmin and the substrate fibrin ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}E) \[[@B138-ijms-17-01969],[@B139-ijms-17-01969]\]. As in several other kallikrein-related peptidases, KLK2 exhibits a single *N*-glycan in the 99-loop close to the substrate binding cleft, which regulates substrate turnover and could increase the specificity for larger protein substrates ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}G) \[[@B140-ijms-17-01969]\]. *N*-Glycosylation in the nearby 62-loop is known for KLK3, thrombin and human neutrophil cathepsin G, where it may modulate substrate binding in the prime side region ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}H) \[[@B141-ijms-17-01969]\]. Eventually, DPP4 is a dimeric transmembrane α/β-hydrolase with various receptor functions beyond its protease activity \[[@B142-ijms-17-01969]\]. As for other integral membrane proteases its *N*-glycosylation may prevent unspecific protein interaction, such as aggregation, but individual glycans may have specific functions ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}I). 3.5. Effects of Glycosylation on Protease Mechanisms {#sec3dot5-ijms-17-01969} ---------------------------------------------------- Kinetic and structural data can provide valuable information on enzymatic mechanisms. The basic proteolytic mechanisms are well known and kinetic parameters are available for many substrates, e.g., in the BRENDA database \[[@B143-ijms-17-01969]\]. However, both functional and structural data of glycosylated and non-glycosylated enzymes are scarce. Thus, a model of the possible effects of glycosylation on catalysis by a trypsin-like serine protease shall be described. In general, the Michaelis complex corresponds to the enzyme substrate complex (ES) and the first tetrahedral intermediate is close to the transition state (TS), while the acyl intermediate is the first intermediate product ([Figure 2](#ijms-17-01969-f002){ref-type="fig"}B). The distinct conformational states of these reaction steps were calculated for trypsin with a combined approach of quantum mechanics/molecular mechanics with molecular dynamics/free energy perturbation calculations, resulting in a free energy profile, which was extended by analyzing the role of Asp102 \[[@B118-ijms-17-01969],[@B144-ijms-17-01969]\]. In addition, free energy profiles for the single steps of this mechanism have been determined, including enzyme product complexes, using rate constants from Michaelis−Menten kinetics, viscosity and isotope kinetic parameters ([Figure 4](#ijms-17-01969-f004){ref-type="fig"}) \[[@B145-ijms-17-01969]\]. Although *K*~M~, *k*~cat~, and *k*~cat~/*K*~M~ do not directly correspond to free energies in the reaction profile, they correlate, such as a low *K*~M~ with a high ∆*G*~bind~ of the enzyme substrate complex formation, a high *k*~cat~ with a low ∆*G*^‡^, and *k*~cat~/*K*~M~ with the overall change of the free energy ∆*G* = ∆*G*^‡^ − ∆*G*~ES~ \[[@B146-ijms-17-01969],[@B147-ijms-17-01969]\]. Under certain conditions the free activation energy can be calculated from the catalytic efficiency \[[@B152-ijms-17-01969]\]. Thus, free energy differences between a glycosylated protease and a glycan-free counterpart can be treated as the ones derived from the catalytic efficiency of enzyme mutants \[[@B151-ijms-17-01969]\]. Based on the known catalytic mechanism, such experimental data could be employed to refine the corresponding mechanistic calculations in more detail as shown here. *N*-glycosylated and glycan-free KLK2 were analyzed with respect to turnover of several small synthetic substrates, resulting always in a 5-fold increased *K*~M~ compared with the *K*~M~ of glycan-free KLK2, while the *k*~cat~ was less affected \[[@B140-ijms-17-01969]\]. This effect could be explained with an influence of the core glycan (GlcNAc~2~Man~3~) at Asn95 in the 99-loop close to the active site, favoring the formation of a type I β-turn over the Asx turn of glycan-free KLK2 \[[@B153-ijms-17-01969],[@B154-ijms-17-01969]\]. Apparently, the flexible 99-loop is wide open in the glycan-free KLK2 crystal structure, whereas it may adopt a closed conformation due to the presence of the *N*-glycan, as in the related KLK1 structure ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}G) \[[@B155-ijms-17-01969],[@B156-ijms-17-01969]\]. The open loop allows for rapid binding of the substrate in the non-prime region at the specificity pockets S4 to S2, which can be explained by a higher *k*~on~ rate, resulting in a lower *K*~M~ = (*k*~off~ + *k*~cat~)/*k*~on~. In addition, substrate binding requires a higher free activation energy ∆*G*~ES~\*, since the energy barrier of the *N*-glycosylated, closed loop has to be overcome. Interestingly, the lid-like 99-loop may also serve in fixing the substrate, as observed in a KLK3 acyl intermediate complex structure ([Figure 5](#ijms-17-01969-f005){ref-type="fig"}) \[[@B157-ijms-17-01969]\]. Remarkably, *N*-glycosylated KLK2 was more efficient than the glycan-free variant in autoactivation and to some extent in degrading large protein substrates \[[@B140-ijms-17-01969]\]. This finding can be partially explained by a stronger influence of the 99-loop on small synthetic substrate binding in the non-prime side, but the catalytic efficiency of *N*-glycosylated KLK2 might be enhanced by an optimally shaped active site for stabilizing the transition states. Thus, the decreased ∆*G*^‡^ could depend on an increased *k*~cat~ of the glycosylated enzyme as well. By contrast, *N*-glycans in close to the active site can have a different effect, as shown for porcine pepsin, in which artificial sequons were *N*-glycosylated, resulting in an overall stabilizing effect and a *k*~cat~ decrease to about 30%, which was explained with a more rigid "flap" loop in favorable contact with the S1 and S2 specificity pockets \[[@B158-ijms-17-01969]\]. However, the snake venom thrombin-like enzyme regulates its enzymatic activity, in particular the substrate access to the active site, by a distortion of the *N*-glycosylated 99-loop \[[@B159-ijms-17-01969]\]. Similarly, the presence of the *N*-glycan in the 99-loop of KLK2 appears to regulate the substrate turnover by favoring the closed state (E) over the open E\* state, as proposed by the conformational selection model, which is opposed to the induced fit model \[[@B155-ijms-17-01969],[@B160-ijms-17-01969]\]. Thorough analyses of these two mechanistic principles conclude with the combined view of induced fit and conformational selection as extremes of one flux model \[[@B161-ijms-17-01969],[@B162-ijms-17-01969]\]. This model may require further adaptations, as suggested by a molecular dynamics calculation of thrombin carrying a single *N*-glycan at Asn60G, resembling the one of KLK3 ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}H). Seemingly, the *N*-glycan rigidifies distant surface regions that become more flexible upon heparin binding, such as the 62-, 148-, and 176-loops \[[@B163-ijms-17-01969]\]. This phenomenon might be related to the long range interaction between protein sectors, i.e., residues in different segments of hydrolytic enzymes with remarkable functional consequences \[[@B164-ijms-17-01969]\]. 4. Conclusions and Outlook {#sec4-ijms-17-01969} ========================== Notably, not only the presence of a single glycan can influence kinetic parameters, but its composition as well. In 1988, a groundbreaking NMR study on the glycosylation of porcine, bovine and human plasmin, revealed the positions and composition of the *O*- and *N*-glycans \[[@B165-ijms-17-01969]\]. The now well defined form 1 of human plasmin, in which the *N*-glycan at Asn288 with terminal sialic acids was altered to a high mannose glycan (GlcNAc~2~Man~9~), exhibited a *k*~cat~/*K*~M~ of about 6%, due to interference with substrate binding in the kringle 3 domain ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}F) \[[@B139-ijms-17-01969],[@B166-ijms-17-01969]\]. Glycosylation variants of plasminogen and its activator tPA play a significant role in the activation and activity of the fibrinolytic system \[[@B167-ijms-17-01969],[@B168-ijms-17-01969]\]. Glycans are important for the fine tuning of substrate recognition and binding, as demonstrated by two hybrid plasminogen activator variants, in which the epidermal growth factor-like domain of uPA preceded the kringle 2 and catalytic domain of tPA \[[@B169-ijms-17-01969]\]. A summarizing overview of the most relevant effects of glycosylation on the physiology of proteases is depicted in [Figure 6](#ijms-17-01969-f006){ref-type="fig"}. Further evidence for the medical relevance of research on glycosylated proteases can be found in the kallikrein field. Natural, inhomogeneously glycosylated KLK3 and glycan-free KLK3 were equally active, recombinant glycosylated KLK3 was three times more active \[[@B170-ijms-17-01969],[@B171-ijms-17-01969]\]. Intriguingly, glycosylation patterns of KLK3 and KLK6 seem to correlate with prostate and ovarian cancer stages \[[@B172-ijms-17-01969],[@B173-ijms-17-01969],[@B174-ijms-17-01969]\]. Mass spectrometry determined forty glycan variants of KLK3 and eleven of KLK6 \[[@B174-ijms-17-01969],[@B175-ijms-17-01969],[@B176-ijms-17-01969]\]. Recently, a glycan-specific immunoassay for cancer-related KLK3 variants has been established \[[@B177-ijms-17-01969]\]. Already a decade ago, cancer-related glycans had been proposed as therapeutic targets \[[@B178-ijms-17-01969]\]. Recently, it was demonstrated that variations of *N*-glycans reflect gene methylation and expression in cancer cells \[[@B179-ijms-17-01969]\]. Altered glycosylation patterns were discovered in other diseases, such as diabetes, and may serve as markers or as targets for future therapies \[[@B180-ijms-17-01969]\]. Biotechnological efforts aim at stabilizing proteases against autodegradation and "humanizing" glycosylation patterns of expression systems for pharmaceutical proteins, in order to abolish unfavorable immunogenic glycan-epitopes \[[@B5-ijms-17-01969],[@B181-ijms-17-01969]\]. The study of effects of glycosylation on protease mechanisms could capitalize on the variation of Cys-linked glycans in the active site, as performed with neoglycoprotein variants of a bacterial subtilisin \[[@B182-ijms-17-01969]\]. Despite the wealth of structural information from X-ray crystallography, this method struggles with glycosylated proteins, since they hamper crystallization and are often too flexible for building models of more than two sugar molecules. Novel methods or old techniques with new applications may facilitate the analysis of glycans on natural proteins, in particular NMR \[[@B165-ijms-17-01969],[@B184-ijms-17-01969]\]. Thus, we are looking forward to a new era of combined efforts from different structural biological methods, which can elucidate the architecture and behavior of naturally glycosylated proteins. Eventually, this knowledge will be highly valuable for the development of better compounds with strongly reduced side-effects and biologically most compatible pharmaceuticals. This study was supported by the Austrian Science Fund (FWF) with the grant P25003-B21. I would like to thank Hans Brandstetter for his encouragement in writing this review. I am also grateful to Brigitta Elsaesser for the preparation of [Figure 2](#ijms-17-01969-f002){ref-type="fig"}A. The author declares no conflict of interest. The founding sponsors had no role in the design of the study; in the analyses and interpretation of the literature; in the writing of the manuscript, and in the decision to publish the results. ![Examples of the most relevant types of glycosylation according to the literature \[[@B3-ijms-17-01969],[@B27-ijms-17-01969]\]. (**A**) *N*-glycosylation of asparagine in sequons with the consensus sequence Asn-Xaa-Ser/Thr. *N*-glycans are generated by trimming and extending the common precursor GlcNAc~2~Man~9~Glu~3~. Small core glycans are mostly intermediates in mammalian glycan synthesis, but often occur in more primitive eukaryotes and insects, as used for recombinant expression. Mammalian *N*-glycans exhibit an enormous diversity, due to many possible combinations of branching sugars; (**B**) *O*-glycosylation at Ser and Thr is found in all kingdoms of life. There is no distinct consensus sequence, but proline-rich regions are favored, e.g., a typical *O*-glycan site would be Pro-Ser/Thr-Xaa-Yaa-Pro. A very common mammalian *O*-glycan is the mucin-type that starts with GalNAc and is extended by galactose and sialic acids or GlcNAc, with eight different cores known. In addition, the *O*-xylose linked, non-branched glucosamine glycans (GAG) or proteoglycans are a large and diverse glycan family. The displayed chondroitin can be phosphorylated and heavily sulfated, comprising up to fifty disaccharide units. *O*-GlcNAc glycans occur inside cells, even in the nucleus, while *O*-galactosylation is found at hydroxylysine residues (Hyl) of collagens.](ijms-17-01969-g001){#ijms-17-01969-f001} ![Substrate recognition and catalytic steps in proteases. (**A**) Two-dimensional interaction map of the model peptide AWTRVR-SILMHY with the specificity subsites S6-S6′ of a KLK protease, calculated with the MOE software \[[@B114-ijms-17-01969]\]. The tryptic specificity is based on the electrostatic interaction of P1-Arg and Asp189 in the S1 pocket. Chymotryptic proteases prefer hydrophobic P1 side chains, such as Tyr or Phe; (**B**) The aspartic protease mechanism requires a pair of Asp residues, with one being the general base that activates a water molecule as in BACE-1 \[[@B115-ijms-17-01969]\]. After substrate binding, the nucleophilic water attacks the scissile bond between P1 and P1′ at the carbonyl C atom; (**C**) Three major metalloprotease mechanisms are known, such as the favored one for MMP-3 \[[@B116-ijms-17-01969]\]; and (**D**) Catalysis of a serine protease with chymotrypsinogen numbering \[[@B117-ijms-17-01969]\]. Other serine protease clans exhibit different arrangements of the triad with similar mechanisms. The catalytic triad activates the Ser Oγ as nucleophile via an acid (Asp102) and a general base (His57), which activates a water molecule for hydrolysis of the acyl intermediate \[[@B118-ijms-17-01969]\]. In addition, cysteine protease mechanisms are related, but often require only Cys-His dyads, since the Sγ is more nucleophilic than the Ser Oγ \[[@B119-ijms-17-01969]\]. Similarly, threonine proteases have a nucleophilic Thr Oγ, while the N-terminus and/or a Lys side chain serve as bases \[[@B120-ijms-17-01969]\]. MEROPS lists variations of catalytic residues and rare protease types \[[@B45-ijms-17-01969]\].](ijms-17-01969-g002){#ijms-17-01969-f002} {ref-type="fig"}). Active site residues are colored dark red, while inhibitors or substrates are shown as ball-and-stick models. (**A**) The aspartic protease cathepsin D with two *N*-glycans distant from the active sit cleft (1LYA/1LYB); (**B**) The papain-like cysteine protease cathepsin C has four *N*-glycans, one is located near the substrate binding region (1K3B/2DJF); (**C**) The metalloprotease meprin β carries seven *N*-glycans, with the one at Asn254 located at the substrate binding cleft near the catalytic Zn^2+^ (4GWM); (**D**) Carboxypeptidase N is another Zn^2+^-dependent protease, with three *O*-glycans distant to the active site (2NSM); (**E**) The trypsin-like factor VII is an important activator in the blood coagulation cascade, exhibiting an *N*-glycan in the catalytic domain and two *O*-glycans in the EGF-like domain 1 (1QFK); (**F**) Plasmin, another trypsin-like protease degrades fibrin clots and is *O*-glycosylated between kringle domains 3 and 4 (4DUR); (**G**) The trypsin-like KLK1 exhibits one *N*-linked GlcNAc of a core glycan (1SPJ), which is sufficient to rigidify the flexible 99-loop (yellow) at the active sit cleft, filled with the PPACK inhibitor of the closely related KLK2 (4NFF), in order to explain the effect of glycosylation on substrate binding. In glycan-free KLK2, the 99-loop is more flexible and open; (**H**) The chymotrypsin-like KLK3/PSA carries a triantennary *N*-glycan, which may enhance binding of natural substrates and inhibitors (3QUM/2ZCK). The flexible 99-loop resembles the one of KLK1 ([Figure 3](#ijms-17-01969-f003){ref-type="fig"}G) and is depicted in yellow; (**I**) DPP4 is a membrane anchored, dimeric α/β-hydrolase, with the shielded active site located inside a cavity. The glycans may prevent aggregation and could play a role in the receptor function of DPP4 (1N1M).](ijms-17-01969-g003){#ijms-17-01969-f003} ![Free energy profile for substrate turnover by the serine protease trypsin \[[@B118-ijms-17-01969],[@B144-ijms-17-01969],[@B148-ijms-17-01969]\]. Various approaches find shifted energy levels and additional energy minima of intermediates and the enzyme product complex \[[@B149-ijms-17-01969],[@B150-ijms-17-01969]\]. The y-axis represents the Gibb´s free energy of the process (∆*G*), which is temperature dependent and related to the reaction enthalpy and entropy: ∆*G* = ∆*H* − *T*∆*S*. The reaction coordinate represents the progress of the reaction, not a real time process. E, S, and P denote enzyme, substrate and products, while TS1 to TS4 are transition states. Direct conversion of *k~cat~* and *K*~M~ into free energy values is not feasible, while they show an inverse correlation, e.g., a high *k*~cat~ with a lower free activation energy ∆*G*^‡^. Currently, the influence of glycosylation on the single mechanistic steps can only be estimated, but differences in the binding energy of the transition states can be calculated as for mutant enzymes \[[@B151-ijms-17-01969]\]. For the substrate Bz-Pro-Phe-Arg-pNA with the *k*~cat~/*K*~M~ values of *N*-glycosylated KLK2 (76,930 M^−1^·s^−1^) and of glycan-free KLK2 (5780 M^−1^·s^−1^) with ∆*G* = −*RT*·ln(\[*k*~cat~/*K*~M~\]~glyc~/\[*k*~cat~/*K*~M~\]) the result is−6.7 KJ·mol^−1^ \[[@B140-ijms-17-01969]\].](ijms-17-01969-g004){#ijms-17-01969-f004} ![The effect of *N*-glycosylation at Asn95 on the active site conformation of KLK2 according to crystal structure derived models and kinetic data \[[@B140-ijms-17-01969]\]. (**A**) Glycan-free KLK2 expressed in *E. coli* exhibits a wide open 99-loop and access of substrates, depicted as green ball-and-stick model bound to the specificity subsites (S4 to S2′ specificity subsites are labeled); (**B**) *N*-glycosylation at Asn95 favors a closed 99-loop, which covers the non-prime side region, left to Ser195 (dark red) in the standard orientation, which prevents substrate binding; (**C**) The *N*-glycosylated 99-loop of KLK2 opens to a lesser extent than in the glycan-free variant. Thus, substrate binding to glycosylated KLK2 requires more free energy, resulting in a lower *k*~on~ rate and higher *K*~M~.](ijms-17-01969-g005){#ijms-17-01969-f005} ![Effects of glycosylation on human proteases. After protein synthesis (**A**) unfolded proteases enter the endoplasmic reticulum (ER), *N*-glycans are linked (**B**); which supports folding (**C**); The *N*-glycans are trimmed and extended, accompanied by quality control (**D**); *N*-glycans are further modified and *O*-glycans are attached in the Golgi (**E**); Sorting leads to membrane anchoring or storage in secretory vesicles (**F**); After secretion, glycosylation prevents aggregation and unspecific binding (**G**); as well as proteolysis, which increases the stability and lifetime of proteases (**H**); Glycosylation regulates binding of: activators (**I**); cofactors (**J**) oligomer partners (**K**); inhibitors (**L**); and substrates (**M**); Eventually, glycosylation fine tunes turnover and kinetic parameters in enzymatic reactions (**N**) \[[@B183-ijms-17-01969]\].](ijms-17-01969-g006){#ijms-17-01969-f006} ijms-17-01969-t001_Table 1 ###### Human aspartic and cysteine proteases with experimentally confirmed or predicted glycosylation sites (*O*-glycosylation indicated in parentheses as "*o*"). MEROPS families are A1, A28 (clan AA), A22 (clan AD), C1, C12 (clan CA), C13 (clan CD), and C26 (PC). UniProtKB entries refer to the first example, following ones continue numerical or alphabetical.Selected structures are shown with Protein Data Bank (PDB) code in bold and numbers of *N*- and *O*-glycans per monomer. Family UniProtKB Aspartic Proteases PDB (*N-/O-*Glycans) -------- ------------- ------------------------------------------------ --------------------------------- A1 BACE1_HUMAN β-Secretase 1, 2 \- RENI_HUMAN Renin **1HRN** (1N) CATD_HUMAN Cathepsin D(*o*), E D: **1LYA** (2N), **1LYB** (2N) NAPSA_HUMAN Napsin-A \- A22 SPP2A_HUMAN Signal peptide peptidase-like 2A, 2B, 2C \- HM13_HUMAN Minor histocompatibility antigen H13 \- A28 APRV1_HUMAN Retroviral-like aspartic protease 1 \- **Cysteine Proteases** C1 CATB_HUMAN Cathepsin B, C, F, H, K, L2, S, W, Z C: **1K3B** (4N), **2DJF** (4N) C12 UCHL1_HUMAN Ubiquitin C-terminal hydrolase isozyme L1(*o*) \- C13 LGMN_HUMAN Legumain **4AW9** (3N) C26 GGH_HUMAN γ-Glutamyl hydrolase \- ijms-17-01969-t002_Table 2 ###### Human metalloproteases with glycosylation sites (*o*: *O*-glycans) according to UniProtKB. MEROPS families are M1, M2, M10, M12, M13, M43 (clan MA), M14 (MC), M16 (ME), M19 (MJ), M20, M28 (MH), M24 (MG) and M50 (MM). M87 and M96 have not been assigned to a clan. UniProtKB entries refer to the first example, following ones continue numerical or alphabetical. Selected structures are shown with PDB code in bold and numbers of *N*- and *O*-sites per monomer. MB: membrane-bound. Family UniProtKB Metallo Proteases PDB (*N-/O-*Glycans) -------- ------------- ------------------------------------------------------------------ --------------------------------------- M1 AMPE_HUMAN Aminopeptidase E, N(*o*), Q E: **4KX7** (8N), N: **4FYT** (8N) ERAP1_HUMAN Endoplasmic reticulum aminopeptidase 1, 2 1: **2YD0** (3N), 2: **5AB0** (8N) LCAP_HUMAN Leucyl-cystinyl aminopeptidase **4P8Q** (8N) TRHDE_HUMAN TRH-degrading ectoenzyme \- M2 ACE_HUMAN Angiotensin-converting enzyme 1, ACE 2 **1O8A** (6N), ACE2: **2AJF** (3N) M10 MMP1_HUMAN MMP 1--3, 8, 9(*o*), 12, 13, 14(*o*), 15--17, 19, 21, 23, 26--28 \- M12 ADAM2_HUMAN ADAM 2, 7--10, 12, 15, 17--23, 28--30, 32, 33 33: **1R55** (1N) ADEC1_HUMAN ADAM DEC1 \- ATS1_HUMAN ADAM-TS 1--4, 5(*o*), 6-10, 12, 13(*o*), 14--20 5: **2RJQ** (1N) 13: **3GHN** (3N/1O) BMP1_HUMAN Bone morphogenetic protein 1 \- TLL1_HUMAN Tolloid-like protein 1, 2 \- MEP1A_HUMAN Meprin A subunit α, β(*o*) **4GWM** (7N) M13 ECE1_HUMAN Endothelin-converting enzyme 1, 2, -like 1 \- NEP_HUMAN Neprilysin **5JMY** (4N) KELL_HUMAN Kell protein \- MMEL1_HUMAN Membrane metallo-endopeptidase-like 1 \- PHEX_HUMAN Phosphate-regulating neutral endopeptidase \- M14 CBPA4_HUMAN Carboxypeptidase A4, A6, B2, D, E A4: **2BOA** (1N),B2: **3D68** (4N) Carboxypeptidase M, N(*o*), O, X1, Z M: **1UWY** (1N) , N: **2NSM** (3O) M16 IDE_HUMAN Insulin-degrading enzyme \- M19 DPEP1_HUMAN Dipeptidase 1, 2, 3 1: **1ITQ** (2N) M20 P20D1_HUMAN Carboxypeptidase PM20D1 \- CNDP1_HUMAN β-Ala-His dipeptidase \- M24 MAP2_HUMAN Methionine aminopeptidase 2(*o*) \- XPP2_HUMAN Xaa-Pro aminopeptidase 2 \- M28 CBPQ_HUMAN Carboxypeptidase Q \- ERMP1_HUMAN Endoplasmic reticulum metallopeptidase 1 \- FOLH1_HUMAN Glutamate carboxypeptidase 2 **2C6C** (6N) NALD2_HUMAN NAALADase 2, NAALADaseL 2: **3FED** (4N), L: **4TWE** (6N) M43 PAPP1_HUMAN Pappalysin 1, 2 \- M50 MBTP1_HUMAN MB transcription factor site protease 1, 2 \- M87 CLCA1_HUMAN Ca-activated chloride channel regulator 1, 2, 4 \- M96 TIKI1_HUMAN Metalloprotease TIKI1, TIKI2 \- ijms-17-01969-t003_Table 3 ###### Glycosylated human serine protease family S1 (clan PA) members (*o*: *O*-glycans). UniProtKB entries refer to the first example, following ones continue numerical or alphabetical. Selected structures are shown with PDB code in bold and numbers of *N*- or *O*-glycosylation sites per monomer. SP: serine protease. Family UniProtKB Serine Proteases PDB (*N-/O-*Glycans) -------- ------------- ------------------------------------------------------- ------------------------------------------------------- S1 C1R_HUMAN Complement C1r, C1s, C2 C1r subcomponent-like protein 1r: **1GPZ** (2N), 1s: **1ELV** (1N) 2: **2I6S** (5N) CFAB_HUMAN Complement factor B, I B: **2OK5** (4N), I: **2XRC** (6N) CELA1_HUMAN Chymotrypsin-like elastase family A1, 3A, 3B \- CMA1_HUMAN Chymase **1NN6** (2N) CORIN_HUMAN Atrial natriuretic peptide-converting enzyme \- CTRC_HUMAN Chymotrypsin-C, CTR-like protease 1 \- ELNE_HUMAN Neutrophil elastase **1PPG** (2N) ENTK_HUMAN Enteropeptidase (Enterokinase) FA7_HUMAN Coagulation factor V, VII(*o*), IX(*o*), X(*o*) 7: **1QFK** (1N/2O) 9: **3KCG** (3N) Coagulation factor XI, XII(*o*) 11: **5EOK** (4N) 12: **4XE4** (1N) GRAA_HUMAN Granzyme A, B, H, M B: **1IAU** (2N) HABP2_HUMAN Hyaluronan-binding protein 2 \- HEPS_HUMAN Serine protease hepsin \- HGFA_HUMAN Hepatocyte growth factor activator **2R0L** (1N) KLK1_HUMAN Kallikrein-related peptidase 1(*o*), 2, 3(*o*), 4, 5 1: **1SPJ** (1N) 3: **3QUM** (1N/1O) 6--13, 15 5: **2PSX** (1N) KLKB1_HUMAN Plasma kallikrein \- MASP1_HUMAN Mannan-binding lectin serine protease 1 **3DEM** (1N) NETR_HUMAN Neurotrypsin \- OVCH1_HUMAN Ovochymase-1, 2 \- PLMN_HUMAN Plasminogen(*o*) **4DUR** (1O) POLS2_HUMAN Polyserase-2 \- PROC_HUMAN Vitamin K-dependent protein C(*o*) \- PRS23_HUMAN SP 23, 27, 29, 35, 38, 41, 42, 44, 45, 47, 48, 55--58 57: **4Q7Y** (2N) PRSS8_HUMAN Prostasin \- PRTN3_HUMAN Myeloblastin **1FUJ** (4N) ST14_HUMAN Matriptase \- TEST_HUMAN Testisin \- THRB_HUMAN Prothrombin **5E8E** (1N) TMPS2_HUMAN Transmembrane (TM) protease serine 2--7, 9 \- TMPSC_HUMAN TM protease serine 12, 13 \- TM11A_HUMAN TM protease serine 11A, 11B, 11D, 11E \- TPA_HUMAN Tissue-type plasminogen activator(*o*) \- TRYB1_HUMAN Tryptase α/β-1, β-2, γ, δ B1: **2F9N** (2N) UROK_HUMAN Urokinase-type plasminogen activator(*o*) **2FD6** (3N) ijms-17-01969-t004_Table 4 ###### Glycosylated human serine and threonine proteases (*o*: *O*-glycans). Included are families S8, S53 (clan SB), S9, S10, S28 (SC), S60 (SR), T1 and T3 (clan PB) according MEROPS. UniProtKB entries refer to the first example, following ones continue numerical or alphabetical. Selected PDBs with glycans are indicated with numbers of *N*- or *O*-glycosylation sites per monomer. PPC: Proprotein convertase. Family UniProtKB Serine Proteases PDB (*N-/O-*Glycans) -------- ------------- --------------------------------------------- ---------------------- S8 FURIN_HUMAN Furin \- NEC1_HUMAN Neuroendocrine convertase 1(*o*), 2 \- PCSK4_HUMAN PPC subtilisin/kexin type 4--7, 9 9: **4NE9** (1N) S9 DPP4_HUMAN Dipeptidyl peptidase 4 **1N1M** (6) SEPR_HUMAN Prolyl endopeptidase FAP **1Z68** (5N) S10 PPGB_HUMAN Cathepsin A **4MWS** (3N) RISC_HUMAN Retinoid-inducible serine carboxypeptidase \- S28 DPP2_HUMAN Dipeptidyl peptidase **3JYH** (4N) PCP_HUMAN Lysosomal Pro-X carboxypeptidase **3N2Z** (5N) TSSP_HUMAN Thymus-specific serine protease \- S53 TPP1_HUMAN Tripeptidyl-peptidase 1 **3EE6** (4N) S60 TRFL_HUMAN Lactotransferrin **1LGB** (1N) **Threonine Proteases** T1 PSA1_HUMAN Proteasome subunit α type-1(*o*), 5--7(*o*) \- PSB1_HUMAN Proteasome subunit β type-1(*o*), 6(*o*) \- T3 GGT1_HUMAN γ-Glutamyltransferase 1, 3, 5--7 1: **4GDX** (6N)

Visual working memory (VWM) is a cornerstone of human visual cognition. It temporarily retains visual information and makes it accessible for cognitive operation, report, and action control (Eriksson et al. [@CR10]; Oberauer [@CR32]; Poth and Schneider [@CR38]; Schneider [@CR48]). VWM has only limited capacity (Luck and Vogel [@CR23]; Shibuya and Bundesen [@CR50]; Sperling [@CR53]). Efficient use of this capacity dictates selectivity: currently relevant information must enter VWM with priority over less relevant information. This prioritization is performed by mechanisms of visual attention (Bundesen [@CR2]; Bundesen et al. [@CR3]; Duncan and Humphreys [@CR9]; Schneider [@CR47]). The bulk of attention research focused on prioritization up to the time of encoding into VWM (Bundesen et al. [@CR4]; Duncan [@CR8]; Poth and Schneider [@CR37]). However, flexible visual cognition requires that changes of priority can be accommodated also when they happen after information has entered VWM. Indeed, more recent research demonstrated that prioritization continues after encoding into VWM. This research made use of the retro-cuing paradigm (Griffin and Nobre [@CR13]; Landman et al. [@CR18]). Participants memorized a set of visual stimuli, the *memory display*, over a retention interval which was followed by a probe stimulus. The task was to report whether the probe matched an item from the memory display. A so-called *retro*-*cue* (i.e., a "retrodictive" cue) was shown after the memory display but before the probe appeared. In the experiments of current interest, retro-cues could be valid or neutral (Astle et al. [@CR1]; Kuo et al. [@CR17]). A valid retro-cue predicted which of the items from the memory display was going to be relevant for the upcoming comparison with the probe. A neutral retro-cue did not contain any predictive information regarding this comparison. The central result is that valid retro-cues improved comparison performance relative to neutral retro-cues. Over a decade of research accumulated evidence for beneficial effects of valid retro-cues in different versions of the basic paradigm (Astle et al. [@CR1]; Griffin and Nobre [@CR13]; Landman et al. [@CR18]; Makovski and Jiang [@CR25]; Makovski et al. [@CR26]; Souza et al. [@CR52]). Thus, it seems safe to conclude that valid retro-cues prioritize an item from a preceding memory display, while the memory display is retained in VWM. Still controversial, however, is the question which mechanisms underlie the prioritization within VWM (Souza and Oberauer [@CR51]). Current accounts assume that valid retro-cues improve comparison performance by manipulating the representations of the memory display in VWM. Specifically, some authors propose that they strengthen the VWM representation of the cued item, increasing the utility of this item for the comparison (Kuo et al. [@CR16]; Lepsien et al. [@CR21]; Nobre et al. [@CR31]). Others propose that they free VWM capacity and reduce interference within VWM by having uncued items removed from VWM (Souza et al. [@CR52]; Williams et al. [@CR60]). Again others suggest that valid retro-cues protect the cued item against decay (Matsukura et al. [@CR27]; Murray et al. [@CR30]; Pertzov et al. [@CR34]) or new interfering information (such as from the probe; Makovski et al. [@CR26]; Makovski and Jiang [@CR25]). Finally, some suggest that valid retro-cues grant the cued items priority in the process of being compared to the probe (Astle et al. [@CR1]; Makovski et al. [@CR26]; Nobre et al. [@CR31]). Fundamental to all these accounts is that retro-cues are, as the term implies, retroactive. That is, all accounts assume that valid retro-cues engage mechanisms that, in one way or the other, prioritize information from the past which is now retained in VWM. Here, we ask whether retro-cues facilitate memory retention in VWM, or whether they enhance the future perceptual processing in service of the comparison task, or both. To this end, we introduce a novel paradigm which allows to disentangle such retrospective and prospective effects of retro-cues (Fig. [1](#Fig1){ref-type="fig"}). Participants briefly viewed a memory display of two colored squares and memorized them over a retention interval. This interval outlasted iconic memory traces and thus called for retention in VWM (for a review, see Irwin and Thomas [@CR14]). Afterward, a probe stimulus was presented that either matched or did not match an item from the memory display, with each alternative occurring in half the trials. The probe was presented at a location different from the stimuli of the memory display. This ensured that comparisons of probe and memorized stimuli needed to rely on VWM, as opposed to more fragile location-specific forms of short-term memory (see Pinto et al. [@CR36]). Participants' task was to indicate whether or not the probe matched an item from the memory display. A retro-cue appeared after the retention interval but before the probe. The retro-cue was either valid or neutral. A valid retro-cue pointed at the location of one of the items from the preceding memory display, the one that was going to be relevant for the upcoming comparison with the probe. Specifically, if the probe matched an item from the memory display, it was always the one indicated by the retro-cue. A neutral retro-cue did not contain any specific location information. We varied the presentation duration of the probe and terminated it with a pattern mask. This enabled us to assess performance as a function of the presentation duration. To disentangle the retrospective and the prospective effects of retro-cues, we fit these data with an exponential model (cf. Bundesen [@CR2]; Wickelgren [@CR59]) and compared the estimated parameters between valid and neutral retro-cues. The exponential model comprised three parameters. First, the level of asymptotic performance which is reached when the probe is shown long enough (see the upper asymptotes of the smooth curves in Fig. [2](#Fig2){ref-type="fig"}). The perceptual processing of the probe should improve with increasing presentation duration (e.g., Petersen and Andersen [@CR35]; Shibuya and Bundesen [@CR50]). At the asymptote, however, performance stops to increase with increasing presentation duration of the probe. Therefore, when the asymptote was reached, perceptual processing (encoding) of the probe should be over and variations of the asymptote should be due to post-perceptual factors. In the present case, variations in the asymptote should reflect the performance level for retaining the items of the memory display in VWM. Second, the rate at which performance increases with increasing presentation duration of the probe toward asymptotic performance (see how steeply the smooth curves increase with increasing presentation duration in Fig. [2](#Fig2){ref-type="fig"}). This is a well-established measure of processing speed (Bundesen [@CR2]; Bundesen et al. [@CR4]; Shibuya and Bundesen [@CR50]; Wickelgren [@CR59]). In contrast to reaction times, this measure has the important advantage that it reflects the specific perceptual (or cognitive) process without contributions of processes such as response selection or motor execution (that are always included in reaction times, see Finke et al. [@CR12], for a related discussion). Applied to the present case, the measure of processing speed should refer to the speed with which the probe is perceptually processed in order to accomplish the comparison with the items of the memory display. Third, a temporal threshold, reflecting the presentation duration of the probe that must be exceeded to increase performance above chance level (cf. Bundesen [@CR2]; Wickelgren [@CR59]).Fig. 1Paradigm of Experiment A. The paradigm of Experiment B was similar but differed in the display durations (and other aspects, see the Methods). Participants fixated a fixation square, after which a memory display with two colored squares appeared (the squares appeared at two out of four possible positions, whereby each pairing occurred equally often). After an interstimulus interval (ISI), a retro-cue was shown. If the retro-cue was valid, it indicated the location of the previous item from the memory display that was going to be relevant for the current trial. If it was neutral, it did not contain any predictive information. After another ISI, a probe was presented whose presentation duration was parametrically varied across trials. The probe was terminated by a pattern mask. In the end of a trial, a question mark prompted participants to indicate whether or not the probe matched an item from the memory displayFig. 2Performance of an individual participant of Experiment A (left) and average performance for the participants of Experiment A (middle) and B (right). Data points represent (average) performance (*d*′) in indicating whether or not probes matched an item from the memory display at each of the presentation durations of the probes. Error bars indicate ± one standard error for within-subjects designs (Loftus and Masson [@CR22]). The two retro-cue conditions are depicted separately. Smooth curves represent least-squares fits of the exponential model to the data of the two retro-cue conditions. As model fitting was performed for each individual in each condition, the curves for the average performance result from averaging the parameters fitted to the data of individual participants If retro-cues facilitate memory retention in VWM, then valid retro-cues should result in a higher asymptotic performance than neutral ones. In contrast, if retro-cues enhance processing for the upcoming comparison task, then valid retro-cues should lead to a higher processing speed of the probe than neutral ones. We first tested these hypotheses in Experiment A. Next, we performed Experiment B as a replication experiment further substantiating the results of Experiment A. Methods {#Sec1} ======= Participants {#Sec2} ------------ Eleven participants were paid to perform Experiment A. One additional participant was excluded because of problems with the eye tracker calibration (see below) and corrupted eye-tracking data files. The participants were between 21 and 34 years old (*MD* = 24 years), eight were female, three were male, and all reported normal or corrected-to-normal (contact lenses) visual acuity and normal color vision. Thirteen participants received course credit for performing Experiment B. They were between 18 and 27 years old (*MD* = 20 years), eleven were female, two male, and all reported normal or corrected-to-normal visual acuity and normal color vision. All participants gave written informed consent before participation; the experiments were approved by Bielefeld University's ethics committee and complied with the ethical guidelines of the German Psychological Association (DGPs). Apparatus and stimuli {#Sec3} --------------------- Both experiments took place in a dimly lit room. A chin-and-forehead rest (Experiment A) and a chin rest (Experiment B) ensured that participants viewed the CRT monitors from a distance of 71 cm. In Experiment A, the CRT monitor (G90fB, ViewSonic, Brea, CA, USA) ran at a refresh rate of 100 Hz and a resolution of 1024 × 768 pixels (corresponding to physical dimensions of 36 × 27 cm). For control purposes, a video-based desktop-mounted eye tracker sampled participants' right eyes at 1000 Hz (Eyelink 1000, SR Research, Mississauga, Ontario, Canada; 9-point grid calibration). Experiment B did not use eye tracking, because it was not available in the employed laboratory setup. In Experiment B, the CRT monitor (Trinitron MultiScan G420, Sony, Park Ridge, NJ, USA) ran at a refresh rate of 85 Hz and a resolution of 1024 × 768 pixels (corresponding to physical dimensions of 36 × 27 cm). Responses were collected using standard computer keyboards (QWERTZ layout). The experiments were controlled by the Psychophysics toolbox (3.0.12; Kleiner et al. [@CR15]; Pelli [@CR33], and in Experiment A, the Eyelink toolbox, 3.0.12; Cornelissen et al. [@CR7]) extensions for Matlab (R2014b in Experiment A, R2013b in Experiment B, The MathWorks, Natick, MA, USA). Stimulus luminance was measured using an LS-110 luminance meter (MINOLTA, Osaka, Japan). Stimulus luminance is reported for Experiments A and B side by side (i.e., luminance A/luminance B). An empty light-gray square was used as fixation stimulus (0.67 × 0.67° of visual angle, with a linewidth of 4 pixels; 59/45 cd/m^2^), henceforth called fixation square. Valid retro-cues consisted in a brightening of two lines of this square, neutral ones in a brightening of the whole square (102/114 cd/m^2^). The stimuli of the memory display were squares (0.67 × 0.67°) of the following eight colors: red (RGB: 255, 0, 0; 34/23 cd/m^2^), magenta (RGB: 255, 0, 255; 43/34 cd/m^2^), yellow (RGB: 255, 255, 0; 112/101 cd/m^2^), orange (RGB: 255, 145, 0; 59/44 cd/m^2^), blue (RGB: 0, 0, 255; 17/12 cd/m^2^), cyan (RGB: 0, 255, 255; 103/92 cd/m^2^), green (RGB: 0, 255, 0; 100/80 cd/m^2^), and black (RGB: 0, 0, 0; 1/\< 1 cd/m^2^). For each individual participant, 99 pattern masks were algorithmically created in the beginning of the experiment. Masks consisted of a square composed of a 4 × 4 matrix of smaller squares (0.30° × 0.30° each) whose colors were randomly chosen from the set of colors with the constraint that each color occurred twice in each mask (see Fig. [1](#Fig1){ref-type="fig"} for an example mask). The gray background had a luminance of 34/22 cd/m^2^. The white question mark of the response screen was written in Arial (0.50° × 1.00°; 102/114 cd/m^2^). Procedure and design {#Sec4} -------------------- Figure [1](#Fig1){ref-type="fig"} illustrates the paradigm of Experiment A (the paradigm of Experiment B was similar in most respects, see below). Participants started each trial by pressing the space bar. In the beginning of a trial, the fixation square was shown at screen center for 400 ms. In Experiment A, the eye tracker monitored if eyes were open (i.e., pupils visible) until the end of this fixation period, and if they were not, the period was prolonged until the next screen refresh after the eyes were open again. In Experiment B, eye behavior was not recorded. The fixation square stayed on for the most of a trial. After the fixation period, the memory display containing two differently colored squares was shown for 100 ms in Experiment A and for 94 ms in Experiment B. The colors of the squares were randomly chosen from the set of used colors. The squares appeared at two out of four possible positions (2° from screen center horizontally to the left or right × vertically to the left or right, see Fig. [1](#Fig1){ref-type="fig"}), and this was randomized across trials with each pairing of positions occurring equally often. The memory display was followed by an interstimulus interval (ISI) of 1000 ms, after which the retro-cue was presented for 100 ms in Experiment A and 94 ms in Experiment B. Valid retro-cues consisted in a selective brightening of two sides of the fixation square, forming an arrow pointing at one of the two locations of the squares of the previous memory display. Across trials, each location was cued equally often. Neutral retro-cues consisted in a brightening of the whole fixation square. The retro-cue was followed by another ISI of 1000 ms. The probe then replaced the fixation square at screen center. In Experiment A, the probe was shown for eight different durations (10, 20, 30, 40, 60, 90, 120, and 180 ms). In Experiment B, it was shown for six different durations (12, 24, 35, 59, 106, and 141 ms). The presentation of the probe was terminated by a central pattern mask lasting for 300 ms in Experiment A and 306 ms in Experiment B. Afterward, a central question mark was presented until participants responded. Participants were instructed to respond with the F9-key if the probe matched an item from the memory display and the F1-key if it did not match any of them. There was no time limit for the response. On half of the trials, the color of the probe matched the color of one of the items from the memory display. On the other half, it had one of the colors that did not appear on this trial. Participants were informed that if a valid retro-cue was shown and the probe matched the color of an item from the memory display, then this would be the color of the item indicated by the retro-cue. Participants performed Experiment A in a single session of 768 trials, 48 trials per retro-cue condition (valid vs. neutral) and per presentation duration of the probe. Participants performed Experiment B in two sessions (on separate days) of 576 trials each, yielding 1152 trials in total, 96 trials per retro-cue condition and per presentation duration of the probe. Trials were administered in randomized order in Experiment A and in both sessions of Experiment B. Participants performed 30 training trials (randomly chosen with replacement from all trial types) before each participation. Results {#Sec5} ======= To control for response biases, performance in indicating whether the probe matched an item from the memory display was assessed as *d*′ (the *z*-transformed rate of "yes"-responses to probes matching an item from the memory display minus the *z*-transformed rate of "yes"-responses to probes not matching an item from the memory display; 0.5 was added to the data cells on which rates were computed to prevent infinite values of *d*′, see Macmillan and Creelman [@CR24]). Performance was assessed as a function of presentation duration of the probe. For each participant and each retro-cue condition, these data were fit with an exponential model of the type$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d^{\prime } = \omega \left( {1 - \exp \left( { - v*\left( {t - t_{0} } \right)} \right)} \right)$$\end{document}$$where *ω* is the upper asymptote of performance (in $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$d^{\prime }$$\end{document}$), *v* is the rate parameter of the exponential distribution which measures processing speed of the probe within the comparison with the items of the memory display (in items/s; cf. Bundesen [@CR2]; Bundesen et al. [@CR4]; Wickelgren [@CR59]). The third parameter, *t*~0~, is a temporal threshold consisting in the presentation duration of the probe that has to be exceeded to increase performance above chance level (in ms; cf. Bundesen [@CR2]). Fitting was accomplished using the nonlinear least-squares (nls) method implemented in R (3.6.3; R Core Team [@CR44]). Figure [2](#Fig2){ref-type="fig"} depicts the performance in each retro-cue condition and corresponding model fits, for one participant in Experiment A and for the average data of both experiments. Table [1](#Tab1){ref-type="table"} provides descriptive statistics of estimated parameters for the two retro-cue conditions of both experiments. Goodness-of-fit was quantified as Pearson's correlation *r* between the predicted values based on the fitted model and participants' observed values (see Table [1](#Tab1){ref-type="table"}).Table 1Descriptive statistics of Experiments A and BExperiment AExperiment BValid retro-cueNeutral retro-cueValid retro-cueNeutral retro-cueω3.51 (0.50)3.10 (0.64)3.49 (0.81)3.07 (0.89)*v*34.70 (11.08)27.80 (10.53)47.42 (20.22)36.72 (14.72)*t*~0~8 (3)9 (5)11 (3)10 (3)*r*.945 (.030).917 (.036).962 (.027).961 (.029)Descriptive statistics of estimated parameters for the two retro-cue conditions of Experiments A and B. Provided are means and standard deviations (in parentheses) across participants for the asymptotic performance level *ω* (in *d*′), the processing speed *v* (in items/s), the temporal threshold *t*~0~ (in ms), and for Pearson's correlation between the values predicted by the fitted model and the observed values The parameter estimates were compared between the retro-cue conditions using paired-samples *t* tests (two sided, with a significance criterion of *p* \< .05, and *d*~*z*~ as effect size Cohen [@CR6], and for which the assumption of normally distributed differences was assessed beforehand using Shapiro--Wilk tests). *t* tests were supplemented with corresponding Bayes Factors (BF~10~, Rouder et al. [@CR46], whereby values greater than one favor the alternative and values smaller than one favor the null hypothesis). Figure [3](#Fig3){ref-type="fig"} illustrates the parameter comparisons between the retro-cue conditions of the two experiments.Fig. 3Estimated model parameters in the two retro-cue conditions of Experiment A (upper panel) and Experiment B (lower panel). Depicted are the means of the retention performance ω (in *d*′), of the processing speed *v* (in items/s), and of the temporal threshold *t*~0~ (in ms), across participants. Error bars indicate ± one standard error for within-subjects designs (Loftus and Masson [@CR22]). n.s. nonsignificant. \**p *\<.05. \*\**p* \< .01 In Experiment A, the retention performance ω was significantly higher when retro-cues were valid than when they were neutral, *t*(10) = 3.212, *p* = .009, *d*~z~ = 0.97, BF~10~ = 6.42. Experiment B replicated this result, *t*(12) = 2.808, *p* = .016, *d*~*z*~ = 0.78, BF~10~ = 3.99. These findings demonstrate a retrospective effect of retro-cues. They indicate that valid retro-cues facilitated the retention of the relevant item from the memory display as compared with neutral retro-cues. The processing speed *v* was significantly higher when retro-cues were valid compared with neutral, and this was likewise the case in Experiment A, *t*(10) = 2.400, *p* = .037, *d*~z~ = 0.72, BF~10~ = 2.14, and Experiment B, *t*(12) = 2.589, *p* = .024, *d*~*z*~ = 0.72, BF~10~ = 2.89. These findings reveal a prospective effect of retro-cues on the future visual processing. Specifically, compared with neutral retro-cues, valid retro-cues increased the speed with which an upcoming probe was processed, in order to assess whether it had been contained in the preceding memory display. There were no differences between the valid and the neutral retro-cue condition regarding the temporal threshold, *t*~0~, neither in Experiment A, *t*(10) = − 0.629, *p* = .543, *d*~*z*~ = − 0.19, BF~10~ = 0.35, nor in Experiment B, *t*(12) = 0.787, *p* = .447, *d*~*z*~ = 0.22, BF~10~ = 0.36. Taken together, the findings of Experiment A and the replication Experiment B converged closely. This was further substantiated by mixed (within between) analyses of variances that included the experiment as a between-factor. For none of the three parameters, the effect of retro-cue condition (valid vs. neutral) and the effect of experiment (A vs. B) interacted, all *Fs*(1, 22) \< 1, all *p*s \> .33. For the retention performance, the absolute values did not differ between the two experiments, *F*(1, 22) = .007, *p* = .936. For the temporal threshold, the absolute values differed between the experiments, *F*(1, 22) = 5.533, *p *= .028 (see Fig. [3](#Fig3){ref-type="fig"}). For the processing speed, the difference between the absolute values in the experiments approached significance, *F*(1, 22) = 3.807, *p* = .064. It is important to note, however, that these between-subjects differences do not conflict with the findings of interest, namely the within-subjects effects of valid versus neutral retro-cues. Indeed, the between-subjects differences in temporal threshold and processing speed are not surprising, because the two parameters have previously been shown to be subject of considerable interindividual variation (see, for example, the individual model curves in Poth and Schneider [@CR40] and Poth et al. [@CR43]). Discussion {#Sec6} ========== The present experiments reveal that retro-cues impact on two distinct components of VWM performance. First, valid retro-cues improve memory retention in VWM. This was evident from the higher retention performance when retro-cues were valid compared with neutral. We assessed retention performance as the asymptote of performance as a function of the presentation duration of the probe. Asymptotic performance reflects a component of VWM performance that is independent from perceptual processing of the probe, because performance does not improve further when the probe is presented longer. Second, valid retro-cues enhanced the speed with which probe stimuli were perceptually processed in order to be compared to the items retained in VWM. We assessed the speed of processing the probe for this memory-based comparison as the rate of performance increase with increasing presentation duration of the probe. Valid retro-cues improved memory retention. This indicates that prioritization modulates representations in VWM independently of the time available to process the probe for the upcoming comparison task. This is in line with several not mutually exclusive accounts assuming that retro-cues impact on VWM-based performance by directly modulating VWM representations. Valid retro-cues may strengthen the representations of cued items in VWM (Kuo et al. [@CR16]; Lepsien et al. [@CR21]; Nobre et al. [@CR31]). They may remove uncued items from VWM, thereby freeing capacity and reducing interference within VWM (Souza et al. [@CR52]; Williams et al. [@CR60]). They may protect the representations of cued items in VWM against decay (Matsukura et al. [@CR27]; Pertzov et al. [@CR34]) or interfering new information (Makovski et al. [@CR26]; Makovski and Jiang [@CR25]). Crucially, valid retro-cues also accelerated the perceptual processing of probe stimuli for performing the upcoming comparison task. This indicates that prioritization in VWM has effects beyond pure memory retention. In this way, this present finding calls for an extension of current accounts of the beneficial effects of valid retro-cues on VWM-based performance. The finding can be interpreted in two ways. The first explanation is that the presentation duration of the probe stimulus determined the quality of its representation in VWM. The rate of performance increase with increasing probe presentation duration was higher when retro-cues were valid than neutral. Thus, valid retro-cues may have compensated for the low representational quality of the probe at short presentation durations, for example, by enhancing the VWM representations of the cued items, which would be in line with previous accounts (e.g., Kuo et al. [@CR16]; Lepsien et al. [@CR21]; Nobre et al. [@CR31]). The present findings would then show that valid retro-cues improve a component of VWM that can be traded-off for the representational quality of the probe stimulus. However, with the effects on memory retention, the findings would also indicate that valid retro-cues improve a VWM component beyond this, one that does not interact with the representational quality of the probe. The second explanation assumes that the acceleration of the processing of the probe resulted from a prospective monitoring process. A valid retro-cue indicates which of the items in VWM will be relevant for the comparison to an upcoming probe. Consequently, the environment may be monitored for the features of the cued item already before the probe appears. It has previously been shown that monitoring processes rely on visual attention (Poth et al. [@CR41]). In the present case, the features of the cued item may be monitored for by engaging the pigeonholing ("attention to features") mechanism from Bundesen's theory of visual attention (TVA; Bundesen [@CR2]; Bundesen et al. [@CR4]). Pigeonholing influences the speed with which visual features of objects are processed for being encoded into VWM. This should happen by up- or down-regulating a perceptual bias for categorizing any given object as having a certain feature. The perceptual bias is internal, meaning that it is independent of what objects are actually viewed. Specifically, the perceptual bias for a certain feature may be implemented by increasing or decreasing the firing rates of visual neurons preferentially coding for this feature (Bundesen et al. [@CR3]). Increasing the perceptual bias for the features of the cued item would increase the speed of encoding the probe into VWM, if the probe has these features. This would imply that the present processing acceleration stemmed from the trials on which the probe matched the cued item. Indeed, effects of retro-cues have been found more pronounced for such matches (Lepsien et al. [@CR20]; Nobre et al. [@CR31]). However, a processing acceleration on trials on which the probe did not match an item from the memory display could still be explained in terms of pigeonholing by assuming an additional decision deadline (cf. Bundesen's [@CR2] explanation of target-absent response times in visual search). In this scenario, valid retro-cues would lower the deadline for processing the probe in order for deciding that it did not match an item from the memory display. In sum, valid retro-cues offered benefits for VWM performance both retrospectively (for retention) and also prospectively (for perceptual processing). To further investigate the mechanisms by which retro-cues achieve this, future studies could include invalid retro-cues that indicate an item that would not be probed afterward. In this way, the studies could contrast beneficial effects of valid (vs. neutral) retro-cues with detrimental effects of invalid (vs. neutral) retro-cues. Based on recent research, one might speculate whether active or passive working memory processes underlay the effects of valid retro-cues. Both types of processes rely on the sensory recruitment hypothesis, stating that the retention of information in VWM is enacted by the same visual brain areas that encode this information at first (Ester et al. [@CR11]; Miller et al. [@CR28]; Serences et al. [@CR49]; Supèr et al. [@CR57]). The hypothesis of active working memory processes relies on the assumption that retention in VWM is performed by maintaining the spiking activity of the neurons coding for the retained items and their features (Chelazzi et al. [@CR5]). The retrospective effects of valid retro-cues on memory retention in VWM could consist in an increase in activity in neurons coding for the features of the cued item, or a decrease in those coding for features of other items, or both (Lepsien et al. [@CR21]; Trapp and Lepsien [@CR58]). An increase in spiking activity of these neurons could at the same time provoke the prospective effects of valid retro-cues. The increased firing would support the future encoding of stimuli by these neurons, explaining why processing of the probe was accelerated in the present experiments. The hypothesis of passive working memory processes is grounded on recent evidence questioning maintained spiking activity as the sole neuronal implementation of retention in VWM (for an overview, see Stokes [@CR54]). Findings from single-cell neurophysiology (Stokes et al. [@CR55]) and brain imaging and stimulation (e.g., Rose et al. [@CR45]; for a review, see Larocque et al. [@CR19]) suggest a passive (V)WM (Schneider [@CR48]), which presumably relies on increased synaptic connectivity rather than maintained spiking activity (Mongillo et al. [@CR29]; though synaptic connectivity may stem from an initial increase in neuronal firing, e.g., Stokes [@CR54]). Valid retro-cues could also exert their effects here, by modulating the synaptic connectivity in visual brain areas at short time scales. Valid retro-cues could increase the synaptic connectivity of neurons coding for the features of the cued item. Enhanced synaptic connectivity could make VWM representations more robust, offering another explanation why valid retro-cues improved retention performance. Interestingly, if valid retro-cues increased the short-term synaptic connectivity of these neurons, this might also increase their efficiency of encoding new stimuli (Stokes et al. [@CR55]; Sugase-Miyamoto et al. [@CR56]). Hence, this provides another explanation of why valid retro-cues accelerated processing of probes. It is important to note, however, that both active and passive working memory processes are feature based. Therefore, both processes would only be able to operate if the features of probes matched the features of the cued items in VWM. This would call for an additional process implementing a decision deadline, in the same way as the prospective monitoring process that relied on Bundesen's ([@CR2]) pigeonholing mechanism. To conclude, the present study shows that priority within VWM not only affects the retention of past information but also future processing in a comparison task. Visual attention seems not only to set processing priorities for encoding into VWM (Bundesen [@CR2]) and for selection within VWM (Griffin and Nobre [@CR13]; Landman et al. [@CR18]), but at the same time also for processing upcoming information. In this vein, visual attention may provide a bridge between episodes of visual processing that are composed of encoding into and retention in VWM (as proposed by Schneider [@CR48]; see also Poth et al. [@CR42]; Poth and Schneider [@CR39]). Handling editor: Luca Simione (ISTC-CNR, Rome). Reviewers: Salvatore Gaetano Chiarella (Sapienza University of Rome) and Enrico Di Pace (Sapienza University of Rome). **Publisher\'s Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Open Access funding provided by Projekt DEAL. This research was supported by the Cluster of Excellence Cognitive Interaction Technology "CITEC" (EXC 277) at Bielefeld University, which is funded by the German Research Foundation (DFG). A previous (non-peer reviewed) version of this paper has been part of the author's Ph.D. thesis, which is available at <https://pub.uni-bielefeld.de/record/2911816>. I thank Werner X. Schneider, who supervised the project and co-authored a previous draft of the paper and generously left the paper for me to publish as a single author. I also thank Rebecca Helling for help with the data collection of Experiment B.

{#sp1 .971} {#sp2 .972}

Introduction ============ CTLs are crucial for the defense against many invading organisms and certain tumors. Presentation of antigenic peptides bound to MHC class I molecules is a prerequisite for stimulation of a CTL response, and therefore plays a pivotal role in providing CTLs with the capacity to respond to foreign antigens [1](#R1){ref-type="bib"}. Peptides that meet the restrictive binding characteristics of MHC class I molecules for presentation to CTLs are generated after intracellular protein degradation by cytosolic proteases. The central enzyme responsible for protein degradation is the proteasome [2](#R2){ref-type="bib"}. Because of their intimate involvement in antigen processing and presentation [3](#R3){ref-type="bib"}, detailed knowledge on the cleavage preferences of proteasomes will be crucial for understanding CTL epitope generation and thus, for the regulation of specific immune responses. The 20S proteasome represents the proteolytic core of the larger 26S proteasome complex that encompasses either one or two regulatory particles of at least 18 subunits [4](#R4){ref-type="bib"}. The eukaryotic 20S particle is composed of 14 different but related subunits organized in a barrel-shaped complex with the stoichiometry α~7~β~7~β~7~α~7~. Three subunits of the two inner β-rings (β1, β2, and β5) participate directly in peptide bond cleavage. They represent three distinct proteolytic activities, designated as the chymotrypsin (ChT)-like, trypsin-like, and peptidyl-glutamylpeptide--hydrolyzing (PGPH) activities [5](#R5){ref-type="bib"} [6](#R6){ref-type="bib"}. As the NH~2~-terminal threonine residues responsible for peptide bond cleavage do not prefer certain peptide bonds over others, the basis for the three distinct proteolytic activities most likely resides in the characteristics of the amino acids in the vicinity (pockets) of each active NH~2~-terminal threonine [7](#R7){ref-type="bib"}. Upon IFN-γ exposure of cells, the three active β-subunits that are constitutively expressed in 20S proteasomes can be replaced by three IFN-γ--inducible homologues, low molecular weight protein (LMP)-2 (=β1i) (for Y \[β1\]), multicatalytic endopeptidase complex--like (MECL)-1 (β2i) (for Z \[β2\]), and LMP-7 (β5i) (for X \[β5\]). Although there is extensive sequence homology, these replacements alter the nature of peptides that are generated by proteasomes [8](#R8){ref-type="bib"} [9](#R9){ref-type="bib"} [10](#R10){ref-type="bib"} [11](#R11){ref-type="bib"} [12](#R12){ref-type="bib"}. Proteasomes harboring these IFN-γ--inducible subunits are also called immunoproteasomes, as opposed to the constitutively expressed "constitutive" proteasomes, because immunoproteasomes were found to process a number of viral epitopes with greater efficacy in vitro [13](#R13){ref-type="bib"} [14](#R14){ref-type="bib"} [15](#R15){ref-type="bib"} [16](#R16){ref-type="bib"}. Using several artificial fluorogenic substrates in vitro, it was found that immunoproteasomes display a better capacity to cleave after hydrophobic and basic residues but are less well equipped for cleavage after acidic amino acids [17](#R17){ref-type="bib"}. The finding that proteasomes are responsible for the generation of the correct COOH terminus of several CTL epitopes [18](#R18){ref-type="bib"} [19](#R19){ref-type="bib"} and the notion that hydrophobic or positively charged amino acids serve in most cases as COOH-terminal anchor residues of MHC class I ligands led to the concept that immunoproteasomes contribute to more efficient MHC class I antigen processing. While this holds true for most viral antigens, recent studies have shown that some antigenic peptides are efficiently produced by constitutive proteasomes but cannot be generated by immunoproteasomes [20](#R20){ref-type="bib"}. This demonstrates that immunoproteasomes are not necessarily better suited for the processing of all MHC class I ligands. To better understand the reasons why certain MHC class I ligands are generated with greater efficiency and others with lower efficiency in cells expressing different sets of proteasomes, we have performed an in-depth analysis of peptide fragments generated after proteasomal cleavage. We have employed a strictly quantitative method to analyze a large collection of peptide fragments produced by either set of proteasome. Our observations allowed the identification of certain amino acids (or their characteristics) in positions distant, or directly flanking the cleavage sites selected by either set of proteasomes. The (quantified) mapping of cleavage sites using a large protein substrate provides the basis for a better understanding of proteasomal cleavage specificity, allowing a refined proteasomal cleavage prediction, which will be helpful for the identification of new CTL epitopes, the design of new (recombinant) vaccines, and for better insight into immunity against infection. Materials and Methods ===================== Purification of 20S Proteasomes. -------------------------------- 20S proteasomes were isolated as described previously [9](#R9){ref-type="bib"}. Frozen pellets of LCL-721 cells or LCL-721.174 cells were lysed in a buffer containing 0.1% Triton X-100 on ice and homogenized in a Dounce homogenizer. The 40,000× *g* supernatant of the lysate was bound to DEAE Sephacel. After elution, the protein fraction was concentrated and loaded onto a 10--40% sucrose gradient. After centrifugation, gradient fractions were tested for protease activity using the fluorogenic substrates succinyl-leucyl-leucyl-valyl-tyrosyl-7-amino-4-methylcoumarin (Suc-LLVY-AMC) and succinyl-tyrosyl-valyl-alanyl-aspartyl-7-amino-4-methylcoumarin (Suc-YVAD-AMC). Active fractions were pooled and further purified by anion exchange chromatography on a MonoQ HR5/5 FPLC column (Amersham Pharmacia Biotech). The purity of the proteasome preparates, checked by SDS-PAGE, was \>95%. Quantification of native proteasome protein was determined by a variation of the Lowry Method (protein assay; Bio-Rad Laboratories) and BSA as a standard. Immunoblotting. --------------- 5 μg of purified proteasome polypeptides were separated by 12% SDS-PAGE and transferred to polyvinyldifluoride (DuPont) with a semidry transfer system. Human LMP-7 was detected using a rabbit polyclonal antiserum in conjunction with chemiluminescence (PW8200; Affiniti Research Products, Ltd.). In Vitro Degradation of Enolase-1. ---------------------------------- 150 μg of yeast enolase-1 were incubated in digestion buffer (20 mM Hepes/KOH, pH 7.6, 2 mM MgAc2, and 0.01% SDS) with proteasomes at a molar ratio of 150:1. Digestions were stopped by freezing the samples at −80°C when ∼50% of the substrate was digested (usually after ∼48 h). Separation and Analysis of Cleavage Products. --------------------------------------------- For the separation of degradation products, unfractionated enolase digests were subjected to μRP SC 2.1/10 columns (Amersham Pharmacia Biotech) on a Microbore HPLC system (SMART; Amersham Pharmacia Biotech). Buffer A contained 0.1% trifluoroacetic acid; buffer B contained 0.081% trifluoroacetic acid and 80% acetonitrile. Gradients were 0% for 5 min, in 40 min to 40% in buffer B, in 8 min to 75% in buffer B, and up to 85% in another 7 min at a flow rate of 150 μl/min. Fractions were collected by peak fractionation with a maximal volume of 500 μl/peak. Peak fractions were dried and dissolved in 25 μl of 40% methanol, 1% formic acid, and subsequently analyzed by matrix-associated laser desorption ionization (MALDI) time of flight mass spectrometry (MS) (G2025A; Hewlett Packard) and NH~2~-terminal sequencing (Edman degradation) (pulsed liquid protein sequencer procise 494A; Applied Biosystems). Alternatively, peptides were analyzed on a hybrid quadruple orthogonal acceleration tandem mass spectrometer (Micromass). All these techniques were applied as described previously [21](#R21){ref-type="bib"}. Pmol amounts for each peptide detected in the HPLC fraction were determined by Edman sequencing and used for the quantitative analysis of the data. Statistical Analysis - Frequencies of Amino Acids. -------------------------------------------------- To detect statistically significant features in the amino acid distribution flanking the cleavage sites, we compared percent values using a classic chi-squared test for four tables (variance assumed due to counting). This method was used to compare constitutive and immunoproteasomes fragments with each other and with enolase. For a more thorough comparison of the absolute pmol amounts of constitutive and immunoproteasomes, we accounted for the experimental variability and, according to the quasilikelihood approach of Wedderburn et al. [22](#R22){ref-type="bib"}, assumed a mean variance structure. We assumed the variance to be proportional to the mean and fitted the proportionality constant α from all the data. Then, the usual chi-squared test variable was scaled by 1/α, which led asymptotically to a test variable that is chi-squared distributed with one degree of freedom. The results of the latter was a more thorough approach correlated to the approach neglecting experimental variability and using percent values. Only chi-squared values \>3.841 are considered to be significant. Statistical Analysis - Comparison of Amino Acid Characteristics. ---------------------------------------------------------------- To compare the characteristics of amino acids, the observed frequencies of amino acids at P6 to P6′ around cleavage sites in both proteasomes were compared with each other using the chi-squared test. Hydrophobicity, bulkiness, and flexibility characteristics [21](#R21){ref-type="bib"} of both proteasomes and enolase were compared by translating the percentage of amino acids found to the corresponding hydrophobicity, bulkiness, and flexibility scales. This resulted in spectra per cleavage site, and these were compared by means of regression analysis. Mice, CTL Assays, and Skin Grafting. ------------------------------------ LMP-7^−/−^ mice were generated as described previously [23](#R23){ref-type="bib"} and maintained at the animal facilities of the Basel Institute for Immunology or the Institute for Cell Biology. C57BL/6 mice were obtained from Charles River Laboratories. RMA and RMA-S cells were cultured in RPMI 1640 containing 10% FCS, 2-Me, [l]{.smallcaps}-glutamine, and antibiotics. CTLs were generated by intraperitoneal immunization of mice with 10^7^ irradiated spleen cells (33Gy) in 300 μl PBS. 10 d later spleens were removed and the splenocytes were stimulated with irradiated spleen cells (33Gy). CTL lines were generated by weekly restimulations with irradiated spleen cells as described previously [24](#R24){ref-type="bib"}. CTL assays were performed as described previously [24](#R24){ref-type="bib"} using RMA, RMA-S, and concanavalin A blasts of spleen cells as target cells. In addition, CTL lines were tested on the LMP-7^−/−^ cell line BII-103 (unpublished data) and BII-103 transfected with LMP-7 wild-type and LMP-7 T1A DNA. The LMP-7 gene was cloned from a 129/Ola cDNA using oligonucleotides 5′ AGGATCCACCATGGCGTTACTGGATCTGTGCGG (with BamH1 site at the 5′ end) and 3′ TGAATTCTCACAGAGCGGCCTCTC (with EcoR1 site at the 3′ end). The DNA fragment was then digested with EcoR1 and BamH1 endonucleases and cloned into similarly digested LZRSpBMN-linker-IRES-green fluorescent protein plasmid (unpublished data). Mutagenesis of LMP-7 to replace T by A at position 1 (in the processed subunit) was done using oligonucleotides 5′ GGCCCACGGCGCAACCACACTCGCC and 3′ GGCGAGTGTGGTTGCGCCGTGGGCC. LMP-7 wild-type and LMP-7 T1A expressing retroviral vectors were transfected into packaging cell lines and supernatants were used to infect the LMP-7^−/−^ cell line BII-103. After appropriate selection, comparable LMP-7 wild-type and LMP-7 T1A transcription was detected by Northern blot analysis and comparable green fluorescent protein expression by FACS^®^ analysis (FACSCalibur™; Becton Dickinson). Skin grafts were performed on the back of the mice. Pieces of skin (∼7 mm of diameter) were obtained with a punch from the back of shaved donors and grafted onto anesthesized recipients where holes had also been made with a punch. Each animal had two separated grafts, foreign and autologous, as an experimental control. The skin was discreetly glued (histoacryl; B. Braun Surgical AG) and bandaged. After a week, the bandage was removed to allow the observation of the putative graft rejection. Results ======= Isolation of Proteasomes. ------------------------- Proteasomes were isolated from EBV-transformed B cells. Constitutive proteasomes were purified from cells (LCL-721.174) lacking LMP-2 and LMP-7 due to a chromosomal deletion in the MHC locus [20](#R20){ref-type="bib"} [25](#R25){ref-type="bib"}. The lack of LMP-2 excludes the incorporation of MECL-1 in 20S proteasomes. Therefore, this cell line contains only proteasomes carrying active constitutive subunits [26](#R26){ref-type="bib"}. The immunoproteasome preparation was isolated from the parental line (LCL-721) that served for the generation of LCL-721.174. As expected, only proteasomes isolated from LCL-721 cells expressed immuno subunits, as exemplified by the presence of LMP-7 ([Fig. 1](#F1){ref-type="fig"}, a and b). Moreover, these proteasomes showed a reduced ability to release fluorogenic groups linked to acidic amino acids compared with proteasomes isolated from LCL-721.174 cells (data not shown). Together, these data indicate that proteasomes purified from LCL-721, although containing some constitutive proteasomes, behave like immunoproteasomes with very little PGPH activity, whereas LCL-721.174--derived proteasomes can be classified as constitutive proteasomes with high PGPH activity. Digestion of Enolase. --------------------- Although the ability of constitutive and immunoproteasomes to cleave a set of standard fluorogenic substrates or some CTL epitope containing peptides has been well documented, little is known about the selection of cleavage sites during the degradation of proteins, especially on a quantitative basis. To obtain further insight into these cleavage preferences, we used the complete protein enolase-1 from yeast as substrate which can be digested by proteasomes in vitro without prior modifications. Enolase is a 436 amino acid--long protein in which the frequency of amino acids resembles the average amino acid frequency in proteins [27](#R27){ref-type="bib"}. Digestion of enolase was performed by incubation of enolase with constitutive or immunoproteasomes at a molar ratio of 150:1. The reaction was stopped when ∼50% of the substrate was degraded to ensure comparable substrate turnover. This degree of degradation was obtained after 48 h, indicating similar rates of digestion for both constitutive and immunoproteasomes. Subsequently, enolase fragments were separated by reverse-phase HPLC. Comparison of two independent digests obtained after incubation with two independent constitutive proteasome batches revealed that highly comparable degradation profiles were obtained. The generation of similar degradation products was confirmed by MALDI-MS analyses of several fractions that eluted at the same time (data not shown). After having established the reproducibility of the digestion profiles, the peptide fragments in all fractions were analyzed by MALDI-MS, in combination with Edman degradation, and compiled in a digestion map ([Fig. 2](#F2){ref-type="fig"}). Approximately 50% of cleavages generated by constitutive proteasomes were not produced by immunoproteasomes. Indeed, when the identity of fragments produced by either set of proteasome was compared on a qualitative basis only ∼25% of peptides produced by immunoproteasomes were also found in constitutive proteasome digests. Therefore, the pool of peptides generated by cells expressing immunoproteasomes is very different from the peptide pool generated by cells harboring constitutive proteasomes only. Quantification of Digestion Profiles. ------------------------------------- For careful examination of proteasomal cleavage preferences, it is important to know the quantity of each fragment to calculate how often particular cleavage sites are selected. In contrast to MS data, data acquired by Edman sequencing are quantitative and can thus be used to determine the amount of peptide liberated. The combination of MS analysis and the quantified Edman sequencing data identified the absolute amount of each peptide detected in HPLC fractions. In the constitutive proteasome digests, a total of 136 fragments was detected, representing 6,135 pmol of peptide ([Table](#T1){ref-type="table"}). By adding the pmol of all fragments starting or ending at a particular cleavage site (and then choosing the higher one of the two sums), pmol amounts of peptide generated from a given cleavage site, and thus the frequency of cleavage site utilization, was determined. The most frequently used cleavage sites were found at amino acid positions 278 and 404, resulting in the liberation of 265 pmol peptide each ([Table](#T1){ref-type="table"}, top). As usage of many other cleavage sites resulted in only 5 pmol of peptide, these data indicate that the relative usage of cleavage sites within one protein can differ substantially. Similar data were obtained when the immunoproteasome digest was analyzed (123 peptides representing 6,370 pmol of peptide; [Table](#T2){ref-type="table"}). The most prominent cleavage site was found at amino acid position 419, resulting in the generation of 250 pmol peptide ([Table](#T2){ref-type="table"}, top). As for constitutive proteasomes, many other cleavage sites were used less often, indicating that both proteasome species prefer certain peptide bonds over others as cleavage sites. The almost identical amounts of peptides generated by both proteasomes (6,135 pmol versus 6,370 pmol) demonstrate a comparable substrate turnover, allowing a direct comparison of pmol amounts of amino acids at different positions around cleavage sites as well as a comparative statistical analysis of both digests. Comparison on a qualitative and quantitative basis of the fragment length distributions revealed that both proteasome species generated peptides with an average length of 7--9 amino acids (c20S, 7.4 aa; i20S, 8.6 aa) with 30% of the peptides produced by constitutive proteasomes being identical to those generated by immunoproteasomes ([Fig. 2](#F2){ref-type="fig"}). Together, these findings strongly indicate that in cells harboring constitutive or immunoproteasomes, respectively, the pool of peptides generated from proteasomal protein turnover will differ substantially. Analysis of Cleavage Site Usage. -------------------------------- As outlined above, some cleavage sites are used often, whereas others are used less frequently or not at all by one or both proteasome species. Until now, most studies addressing the specificity of proteasomal peptide/protein degradation did not take the frequency of cleavages into account. To study more accurately the influence of all 20 amino acids flanking proteasomal cleavage sites, we determined the frequency of amino acids around cleavage sites (P6 to P1 NH~2~-terminal of cleavage site; P1′ to P6′ COOH-terminal of cleavage site) using the quantified data set described above ([Table](#T1){ref-type="table"} and [Table](#T2){ref-type="table"}, and [Fig. 3](#F3){ref-type="fig"}). No statistically significant differences in cleavage site selection were observed between two independent digests performed by two independent immunoproteasome batches, indicating that the fragment pattern obtained after proteasomal degradation was highly reproducible (data not shown). Examination of cleavages performed by immunoproteasomes revealed several deviations from randomness at the amino acid level (i.e., the background in enolase) as determined by chi-squared analyses (data not shown). L as P1 residue has the highest preference (chi-squared = 21) and reflects a pronounced ChT-like activity within immunoproteasomes. Weaker preferences for D at P2′ (chi-squared = 6.2) are also evident. On the other side, several residues are found less frequently than expected with the hydrophobic amino acids I, L, and V (chi-squared = \>4.3) at P1′ being the most prominent. This finding reflects the preference of proteasomes to cleave behind rather than in front of L residues, for example. Apart from these amino acids at the P1′ position, L at P2 was found less abundantly than expected. When amino acids are grouped according to their characteristics, it was found that there was a positive correlation between cleavage and the presence of hydrophobic, and bulky amino acids in P1 (*P* \< 0.0002). An analogous examination for the cleavages selected by constitutive proteasomes revealed that the frequency of S as P3 and the positively charged amino acid K as P1 and P2′ residue is reduced (all chi-squared = 3.9), indicating that these amino acids are disfavored by constitutive proteasomes at these positions. Constitutive proteasomes show a weak enrichment of P at P4 (chi-squared = 3.8) and, like immunoproteasomes, prefer A at P1′ and L as P1 residue (albeit to a lesser extent; chi-squared = 4.2). Similarly, an enrichment for D at P2′ was noted (chi-squared = 8.5), indicating that both proteasome species have a preference for negatively charged amino acids at the P2′ position. When comparing characteristics of amino acids at these positions, a preference for hydrophobic, bulky amino acids in P1 and P2′ (*P* \< 0.0002) was observed. There is a negative correlation between cleavage site selection and the presence of hydrophobic, bulky amino acids in P1′ (*P* \< 0.0002). Effect of Immunosubunit Incorporation on Cleavage Site Selection. ----------------------------------------------------------------- The data described above reveal that both proteasome species exhibit a different, but partially overlapping (L in P1, A in P1′, and D in P2′) cleavage preference. However, they do not address directly the influence of immunosubunit incorporation on cleavage site selection. Therefore, we compared relative amino acid frequencies around cleavage sites used by immunoproteasomes to those used by constitutive proteasomes. As a control, we evaluated cleavage site selection of two independent immunoproteasome batches yielding no statistically significant differences (data not shown). For comparison between constitutive and immunoproteasomes for positions P6-P6′, significant differences were found at position P1 (*P* \< 0.004) and, to a lesser extent, also at P1′. When testing for amino acid characteristics, we found that immunoproteasomes prefer bulky, hydrophobic amino acids at P1 (*P* \< 0.003) but dislike flexible amino acids at this position (*P* \< 0.001). At position P1′ we found that hydrophobic, bulky amino acids were favored more by constitutive proteasomes (*P* \< 0.04). Thus, comparison of peptide bonds used as cleavage sites by immunoproteasomes to the ones used by constitutive proteasomes revealed a pronounced ChT-like activity of immunoproteasomes, as is reflected by a strong enrichment of hydrophobic amino acids in P1. When comparing the two proteasome species for individual amino acid usage, a strong enrichment of L (chi-squared = 7.1 and, to a lesser extent F, chi-squared = 3.8) was detected at P1 in peptides generated by immunoproteasomes. The strong enrichment of L became most prominent when we analyzed in detail those cleavages exclusively generated by immunoproteasomes, as 43% of these cleavages were performed after L (data not shown). Acidic amino acids D and E in P1 exert a negative influence on cleavage site selection by immunoproteasomes (chi-squared = 4.5 and 8.1, respectively). G in P1′ is enriched in peptides generated by immunoproteasomes (chi-squared = 7.8), pointing to a preference of immunoproteasomes for this small amino acid in P1′. Analysis of amino acid frequencies at positions P2 to P6 and P2′ to P6′ showed that L at P2′ is disfavored by immunoproteasomes (chi-squared = 4.3). Effect of Immunosubunit Incorporation on CTL Induction. ------------------------------------------------------- The observed major differences in the peptide repertoire produced by constitutive and immunoproteasomes could have important biological implications that go beyond differential proteasomal processing of certain CTL epitopes [8](#R8){ref-type="bib"} [13](#R13){ref-type="bib"} [14](#R14){ref-type="bib"} [15](#R15){ref-type="bib"} [16](#R16){ref-type="bib"} [20](#R20){ref-type="bib"}. To study the biological consequences of immunosubunit incorporation in vivo, we tested the immune response of LMP-7^−/^− mice lacking functional immunoproteasomes [28](#R28){ref-type="bib"} against wild-type cells. If immunoproteasome incorporation affects the nature of the CTL epitopes presented, it is expected that wild-type cells expressing both constitutive and immunoproteasomes will induce a CTL response in mice expressing constitutive proteasomes only. The opposite direction should not lead to a CTL response, as no unique ("immuno") peptides are presented in this setting. Indeed, skin grafting experiments showed that wild-type mice were not able to reject skin from LMP-7^−/^− animals [23](#R23){ref-type="bib"}, which are not able to express functional immunoproteasomes because in the absence of the LMP-7 subunit, LMP-2 and MECL-1 subunits are only incorporated in 20S proteasomes as inactive precursors [28](#R28){ref-type="bib"}. However, LMP-7^−/^− mice rejected skin transplants from wild-type mice with even greater efficiency than observed for the rejection of male skin by female recipients ([Fig. 4](#F4){ref-type="fig"} a). The fact that the latter rejection is mediated over several antigenic differences [29](#R29){ref-type="bib"} supports the argument that LMP-7^−/^− mice react against multiple antigenic differences present on LMP-7 wild-type skin. The identical scenario was observed for the induction of CTL responses. Vaccination of wild-type mice with cells derived from LMP-7^−/^− mice did not generate a CTL response. In contrast, LMP-7^−/^− mice readily mounted a strong, transporter-associated with antigen processing (TAP)-dependent CTL response upon vaccination with wild-type cells harboring both proteasome species ([Fig. 4](#F4){ref-type="fig"} b). Cell lines expressing an inactive LMP-7 T1A subunit were recognized with a 30-fold lower efficiency compared to cells expressing the LMP-7 wild-type subunit despite equal expression levels of both gene products and H-2 class I molecules (data not shown). This finding argues against the possibility that CTL responses against a peptide from the LMP-7 protein itself have been induced. The fact that LMP-7 T1A transfectants are still recognized to some extent by CTLs from LMP-7^−/^− mice can be explained by the observation that some CTL epitopes, not produced in LMP-7^−/^− mice, can be generated by proteasomes carrying both catalytically active or inactive LMP-7 subunits, as reported for immunoproteasome-mediated CTL epitope generation from viral antigens [14](#R14){ref-type="bib"} [30](#R30){ref-type="bib"}. The recognition of these epitopes on LMP-7 T1A transfectants would result in a CTL activity above the one induced by LMP-7^−/^− cells but below the one induced by LMP-7 wild-type cells. Together, our findings indicate that the incorporation of immunosubunits strongly influences the repertoire of peptides presented to CTLs and thereby the overall specificity of the CTL response. Discussion ========== The proteasome is the key enzyme responsible for cytosolic protein degradation and generation of peptides that are presented by MHC class I molecules by liberation of their proper COOH termini. However, proteasomes also can prevent MHC class I presentation by destruction of potential MHC ligands or their inability to select the proper COOH-terminal cleavage site. Cleavage site selection might change substantially after incorporation of IFN-γ--inducible subunits, as indicated by experiments using small fluorogenic substrates or short synthetic peptides. Therefore, detailed knowledge on the specificity of protein degradation by either constitutive or immunoproteasomes is crucial to the understanding and prediction of CTL epitope generation. To analyze alterations of cleavage preference upon immunosubunit incorporation we used an entire protein as a substrate without modification of amino acids or bias to known CTL epitopes. The quantitative identification of \>120 peptide fragments generated by either constitutive or immunoproteasomes allowed a detailed analysis of the cleavage specificities. Our data reveal that especially the amino acids in P3 to P3′ have a strong influence on cleavage site selection as the observed frequencies of amino acids in these locations showed the largest discrepancy to their background frequencies in enolase. Nonetheless, amino acids further away most likely influence cleavage site selection as well, as was reported earlier for the effect of P at P4 on peptide bond cleavage by yeast and constitutive human proteasomes [21](#R21){ref-type="bib"} [31](#R31){ref-type="bib"} [32](#R32){ref-type="bib"} [33](#R33){ref-type="bib"}. The average length of fragments produced by either set of proteasome is the same, showing that spacing at which specific cleavage sites of the substrate are recognized is not changed by immunosubunit incorporation. By comparing flanking residues around cleavage sites, we found that both proteasomes display a partially overlapping but different cleavage specificity ([Fig. 3](#F3){ref-type="fig"} and [Fig. 5](#F5){ref-type="fig"}). They both prefer L as P1 residue and K in P3; A in P1′ and D in P2′ were also favored. Nonetheless, immunoproteasomes have a much stronger preference for L at P1, as well as other hydrophobic amino acids in this position. These findings point to a pronounced ChT-like activity after immunosubunit incorporation. In contrast, the acidic amino acids D and E were clearly disfavored by immunoproteasomes, whereas these amino acids were enriched at P1 in peptides generated by constitutive proteasomes. Some of the different characteristics regarding amino acid preferences at the P1 position were also observed previously in a small number of fragments generated from covalently modified lysozyme using proteasomes from bovine spleen and pituitaries [12](#R12){ref-type="bib"}. Our observations most likely reflect the characteristics of amino acids at the inner surface of the proteasome. Analysis of the contribution of individual active β subunits to cleavage site selection in yeast 20S proteasomes [5](#R5){ref-type="bib"} [21](#R21){ref-type="bib"}, revealed that the active site of β1 prefers to cleave after acidic residues, β2 after basic residues, and β5 after hydrophobic residues. Structural analysis of proteasomes has shown that the pockets around the active site threonine of β5 and β2 do not change after immunosubunit incorporation [7](#R7){ref-type="bib"}. Therefore, it can be expected that the cleavage site--flanking amino acids that are preferred by both proteasomes (K in P3, L in P1, A in P1′, and D in P2′) correlate with the characteristics of amino acids in the vicinity of the S-1 pockets of β5/ β5i and/or β2/β2i. The enhanced preference for hydrophobic amino acids at P1 of immunoproteasomes does not probably result from exchange of β5, which mediates the ChT-like activity, for β5i (LMP-7), as replacement of β5 for β5i does not alter the pocket surrounding the active site threonine. Moreover, functional data indicate that β5i influences the structural features of 20S proteasomes, thereby enhancing the activity of β1i (LMP-2; references [14](#R14){ref-type="bib"} and [30](#R30){ref-type="bib"}). The stronger preference of immunoproteasomes for hydrophobic amino acids at P1, however, correlates with the characteristics at the inner surface of the proteasome when β1 (Y) is replaced for β1i. Replacement of β1 for β1i is predicted to cause the formation of a more apolar pocket around the active site threonine. This most likely results in the accommodation of hydrophobic amino acids at P1, leading to an enrichment of peptides with hydrophobic COOH termini and a reduction of fragments with charged COOH termini [7](#R7){ref-type="bib"}. Thus, it is conceivable that the exchange of β1 for β1i causes a reduction of peptides with charged COOH termini (amino acids like D and E) and enhanced liberation of peptides with hydrophobic COOH termini. Structural analyses of proteasomes also predict that the active S-1 pocket of β5i is more constricted than the pocket of β5 [7](#R7){ref-type="bib"}. As most hydrophobic amino acids are rather bulky, it is possible that small and flexible amino acids, like G, in close vicinity to hydrophobic amino acids, support accommodation of these amino acids in the S1 pocket of β5i most efficiently. Therefore, we anticipate that the strong enrichment of G and the decrease of hydrophobic amino acids in P1′ of cleavage sites selected by immunoproteasomes are a direct consequence of these altered characteristics of the β5 immunosubunit. The observation that the P1 preference of immunoproteasomes closely resembles the preferences of the F pocket of most MHC class I molecules is thought to explain why immunoproteasomes are associated with more efficient processing of MHC class I ligands, either by generating CTL epitopes directly or by producing more CTL epitope precursors that are converted into CTL epitopes by the action of aminopeptidases. As some CTL epitopes are preferentially generated by immunoproteasomes [34](#R34){ref-type="bib"}, it is postulated that immunoproteasome expression is generally associated with more efficient CTL epitope generation. However, the correlation between epitope production and expression of immunoproteasomes is more subtle. Some MHC class I alleles harbor F pockets that allow the binding of amino acids with charged polar side chains (e.g., K and R). These MHC alleles are unlikely to profit from immunoproteasome expression, as the trypsin-like activity is not enhanced by immunosubunit incorporation. In view of our results, increased CTL epitope generation can now be explained and also predicted by several different mechanisms. Examples of quantified fragment generation from synthetic peptides by constitutive and immunoproteasomes are listed in [Table](#T3){ref-type="table"}. They include (a) enhanced liberation of the proper NH~2~ and/or COOH terminus by i20S proteasomes resulting from the combination hydrophobic amino acid, small (and polar) amino acid at P1 and P1′, respectively (hepatitis B core antigen 141--151, lymphocytic choriomeningitis virus \[LCMV\] pp89, and adeno virus E1B), (b) enhanced generation of TAP-compatible CTL epitope precursors due to increased i20S proteasome activity for the combination hydrophobic amino acid, small (and polar) amino acid at P1 and P1′, respectively (LCMV pp89, LCMV nucleoprotein \[NP\]), or (c) reduced PGPH activity of i20S proteasomes not destroying the CTL epitope (influenza A NP). However the combination hydrophobic amino acid, small (and polar) amino acid at P1 and P1′ within a CTL epitope can also favor its destruction by i20S proteasomes as shown for a CTL epitope derived from the ubiquitous self-protein RU1. The finding that the peptide pool generated by immunoproteasomes differs from the one produced by constitutive proteasomes leads to important biological consequences as also shown by the effect of immunosubunit incorporation on skin graft rejection and CTL induction ([Fig. 4](#F4){ref-type="fig"}). Recently, it was found that dendritic cells and other professional APCs express immunoproteasomes [20](#R20){ref-type="bib"}. In contrast, most cells outside the lymphoid system harbor mainly constitutive proteasomes, unless they are exposed to inflammatory stimuli. These findings indicate that the peptide pool available to MHC molecules, and thus for presentation to the immune system, differs substantially between professional APCs and nonlymphoid cells. Professional APCs are not only involved in T cell priming, they also play a pivotal role in central and peripheral T cell tolerization [35](#R35){ref-type="bib"} [36](#R36){ref-type="bib"} [37](#R37){ref-type="bib"}. They can tolerize against endogenously expressed antigens, but also against antigens acquired from exogenous sources (cross-tolerization; reference [38](#R38){ref-type="bib"}). Although these antigens are derived from nonlymphoid cells that only express constitutive proteasomes, tolerance will be induced against self-peptides that are generated by cells expressing immunoproteasomes. During viral infection, virus-infected cells but also neighboring cells will be exposed to inflammatory cytokines, leading to immunoproteasome expression. Because tolerance has been adjusted to the peptide pool expressed by cells harboring immunoproteasomes, self-reactive immune attack is thus reduced to a minimum. These considerations might also be relevant for induction of CTL immunity against ubiquitously expressed self-proteins. If tolerance to self-proteins was restricted to peptides generated by immunoproteasomes, nonlymphoid cells expressing constitutive proteasomes would display peptides for which no CTL tolerance has been induced. In this respect, it is noteworthy that all CTL epitopes derived from ubiquitous self-proteins (like RU1, Mage-1, gp100, and Melan-A; reference [20](#R20){ref-type="bib"}) identified thus far are generated less efficiently by immunoproteasomes, whereas presentation of viral epitopes is, in general, enhanced after immunoproteasome expression [8](#R8){ref-type="bib"} [13](#R13){ref-type="bib"} [14](#R14){ref-type="bib"} [15](#R15){ref-type="bib"} [16](#R16){ref-type="bib"}. The combination of MHC class I ligand motifs and constitutive/immunoproteasomal cleavage motifs for T cell epitope prediction will enhance progress in precise manipulation of specific immune responses. For example, it should now be possible to identify new CTL epitopes preferentially generated by constitutive proteasomes (like the RUI epitope). Such peptides are attractive candidates to be used as cancer vaccines, especially if they are derived from antigens that are, for example, overexpressed in tumors. This work was supported by grants of the Deutsche Forschungsgemeinschaft to H. Schild (Schi301/2-2 and SFB 510 C1) and the European Union to H.G. Rammensee (Biomed 95-0263), Merck KgaA (Darmstadt, Germany), and CTL Immunotherapies Corporation (Chatsworth, CA). The research of Dr. R.E.M. Toes has been made possible by a fellowship of the Royal Netherlands Academy of Arts and Sciences. The Basel Institute for Immunology was founded and is supported by Hoffmann-La Roche, Ltd. *Abbreviations used in this paper:* ChT, chymotrypsin; LCMV, lymphocytic choriomeningitis virus; LMP, low molecular weight protein; MALDI, matrix-assisted laser desorption ionization; MECL, multicatalytic endopeptidase complex--like; MS, mass spectrometry; NP, nucleoprotein; PGPH, peptidyl-glutamylpeptide-hydrolyzing; TAP, transporter-associated with antigen processing. R.E.M. Toes and A.K. Nussbaum contributed equally to this work. ###### Absolute Amounts of Amino Acids Found in Positions P6 to P1 and P1′ to P6′ of Peptides Generated by Constitutive Proteasomes Pos Pmol P6-P1 P1′-P6′ ------------ ------- -------- --------- ------- ------- ------- ------- ------- ------- ------- ------- ------- 278 265 LTGPQL ADLYHS 404 265 APARSE RLAKLN 383 185 TEDTFI ADLVVG 31 175 EKGVFR SIVPSG 142 165 SKSKTS PYVLPV 183 165 TFAEAL RIGSEV 230 165 LDLIVD AIKAAG 330 155 TNPKRI ATAIEK 146 150 TSPYVL PVPFLN 79 145 PAFVKA NIDVKD P6 P5 P4 P3 P2 P1 P1′ P2′ P3′ P4′ P5′ P6′ A 865 595 1,020 485 620 825 1,545 330 710 590 930 365 C 0 0 0 0 130 0 0 0 0 0 0 0 D 580 460 255 215 185 805 300 1,185 315 235 170 400 E 530 465 385 455 250 605 300 275 460 150 465 440 F 50 275 430 165 430 145 190 250 165 575 220 50 G 300 575 750 230 545 175 185 415 760 185 320 780 H 175 125 10 150 280 65 165 110 65 35 385 90 I 505 125 385 335 180 510 60 845 220 450 405 170 K 220 580 240 945 455 80 365 115 370 675 600 655 L 540 550 535 525 230 1,185 345 900 1,120 435 670 565 M 10 0 95 10 15 35 70 40 75 65 10 60 N 120 285 235 140 240 95 390 215 195 160 270 730 P 365 475 550 365 95 40 480 20 255 240 330 315 Q 45 260 25 20 340 60 75 60 155 305 35 75 R 125 45 45 425 290 275 595 95 155 145 80 60 S 575 365 380 90 535 395 500 215 170 530 490 555 T 735 460 140 365 365 140 150 430 140 360 250 105 V 270 380 365 920 840 450 275 410 695 705 355 695 W 40 15 35 30 20 30 0 50 0 0 50 10 Y 85 100 255 265 90 220 145 175 110 295 100 15 sum (pmol) 6,135 6,135 6,135 6,135 6,135 6,135 6,135 6,135 6,135 6,135 6,135 6,135 (Top) 10 most abundant cleavage sites from P6-P1 and P1′-P6′, the position of the cleavage site (pos), and the calculated amount of enolase that has been cleaved at this position (pmol; see also Materials and Methods). (Bottom) Based upon quantified data (as exemplified on top), the absolute amounts of amino acids (in pmol) around all identified cleavage sites are shown. ###### Absolute Amounts of Amino Acids Found in Positions P6 to P1 and P1′ to P6′ of Peptides Generated by Immunoproteasomes Pos Pmol P6-P1 P1′-P6′ ------------ ------ -------- --------- ------- ------- ------- ------- ------- ------- ------- ------- ------- 419 250 RIEEEL GDNAVF 390 200 DLVVGL RTGQIK 183 155 TFAEAL RIGSEV 313 155 HFFKTA GIQIVA 133 150 PLYKHL ADLSKS 142 150 SKSKTS PYVLPV 253 150 ASSEFF KDGKYD 285 150 DLYHSL MKRYPI 289 150 SLMKRY PIVSIE 383 150 TEDTFI ADLVVG P6 P5 P4 P3 P2 P1 P1′ P2′ P3′ P4′ P5′ P6′ A 680 615 940 420 590 745 1,440 665 560 775 480 810 C 0 0 0 0 0 0 0 0 0 0 0 0 D 725 360 190 225 160 150 335 1,210 255 325 220 375 E 300 370 515 870 580 155 255 320 410 270 400 515 F 0 495 280 105 665 550 50 100 120 260 190 250 G 500 550 485 180 360 170 880 475 970 265 465 675 H 300 35 80 285 205 130 140 40 50 40 340 285 I 440 345 290 470 145 585 20 800 295 420 730 395 K 90 475 400 1,065 620 175 450 275 290 405 590 830 L 560 975 450 235 160 2,305 65 370 1,025 410 105 510 M 20 0 215 10 80 50 295 0 0 100 0 50 N 245 285 290 165 205 95 520 290 560 575 190 290 P 325 370 255 380 35 0 375 175 305 180 440 80 Q 10 260 210 105 170 10 175 75 225 250 100 20 R 365 100 65 215 395 200 480 0 190 290 320 100 S 545 250 610 225 735 185 345 320 265 695 380 520 T 890 200 85 235 570 195 310 560 275 195 275 55 V 255 485 585 940 695 320 85 535 495 555 925 610 W 25 60 0 25 0 70 0 0 0 75 0 0 Y 95 140 425 215 0 280 150 160 80 285 220 0 sum (pmol) 6370 6,370 6,370 6,370 6,370 6,370 6,370 6,370 6,370 6,370 6,370 6,370 (Top) 10 most abundant cleavage sites from P6-P1 and P1′-P6′, the position of the cleavage site (pos), and the calculated amount of enolase that has been cleaved at this position (pmol; see also Materials and Methods). (Bottom) Based upon quantified data (as exemplified on top), the absolute amounts of amino acids (in pmol) around all identified cleavage sites are shown. ###### Correlation between Observed and Predicted Cleavages Preferentially Performed by c20S and i20S Proteasomes^a^ Source MHC restriction Influence of IFN-γ Sequence Reference ---------------- ----------------- -------------------- --------------------------------- ----------- ▾ NAPIL **STLPETTVVRR**RGRSPR HbcAg 141--151 HLA-Aw68 \+ 14 ▿ RGVQI **ASNENMD AM**DSRTLELR Influenza A NP H2-D^b^ \+ 35 ▾▾ LMY DM **YPHFMPTNL** GPSEKR LCMV pp89 H2-L^d^ \+ 8 ▾▿ KIM RTE **RPQASGVYM** GNLTAQ LCMV NP H2-L^d^ \+ 16 ▾ YKISKL **VNIRNCCYI** SGNGAE Adeno E1B H2-K^b^ \+ 13 ▾ TGSTA**VPYGSF KHV**DTRLQ RUI HLA-B51 − 20 ^a^Cleavages performed at comparable amounts by c20S and i20S proteasomes are not indicated. HBcAg, hepatitis B core antigen; ▾, i20S preference; ▿, c20S preference. ###### Immunosubunit incorporation into 20S proteasomes purified from LCL-721 cells (left) but not into 20S proteasomes derived from LCL-721.174 cells (right). 20S proteasomes were isolated from LCL-721 and LCL-721.174 cells as described in Materials and Methods. Proteasome subunits were separated by SDS-PAGE (a) and probed with an LMP-7--specific antiserum (b). LMP-7 was only detected in 721 proteasomes. i or c indicate subunits exclusively present or overexpressed in i20S or c20S proteasomes, respectively.   ###### Digestion map generated from degradation of enolase by constitutive proteasomes (a) and immunoproteasomes (b). Vertical lines, cleavage sites determined by Edman degradation and/or MS; black bars, degradation products identified by Edman degradation in combination with MS; white bars, degradation products identified by Edman degradation only (COOH terminus of peptide not identified).   ###### Relative frequencies of amino acids in position P1 (a) P1′ (b). The absolute amount of amino acids found in a defined positions (P6 to P6′) around cleavage sites used by constitutive proteasomes and immuoproteasomes was divided by the total amount of peptides detected in the digests resulting in the relative frequency of amino acid usage in that position. Big white bars, relative frequency of amino acids found in enolase; black bars, relative frequency of amino acids at P1 or P1′ positions in peptide fragments generated by constitutive proteasomes; grey bars, relative frequency of amino acids at P1 or P1′ in peptide fragments generated by immunoproteasomes.   {#F5} ###### \(a\) Skin of the indicated donors were grafted on the back of the indicated recipient mice according to the protocol described in Materials and Methods. Graft rejection was controlled daily. Each dot represents a single recipient. This graph shows a representative result of four independent experiments. (b) LMP-7^−/^− mice and wild-type littermates (129/Ola background) were immunized with wild-type or LMP-7^−/^− splenocytes. CTL activity was tested after in vitro stimulation on LMP-7^−/^− and wild-type concanavalin A blasts, RMA, and RMA-S cells as described in Materials and Methods. This graph shows a representative result of three independent experiments.  

Background ========== Soft tissue defects around the foot and ankle region often present an awkward problem to the orthopaedic surgeons. Tendons and bones are frequently exposed after trauma because of the thinness of the subcutaneous tissue. Traditionally, local flaps were used for resurfacing but their use is limited by their size and arc of rotation. Reverse flow flaps such as anterior tibial artery flap and posterior tibial artery flap require the sacrifice of major arteries. Free flap is currently the treatment of choice for large soft tissue defects of the distal extremity and it solves the problem of donor site morbidity in the immediate vicinity of the flap. It is however a technically demanding procedure for surgeons with less microsurgical experience. Furthermore in a few cases of trauma with damaged or occluded major vessels, where a free flap may be potentially hazardous, the reverse sural artery flap can prove to be one of the few safe options for soft tissue coverage \[[@B1]\]. The accompanying arteries of the lesser saphenous vein and sural nerve have been utilized with success for harvest of reverse flow sural flap \[[@B2]\]. The sural nerve remains the anatomic landmark for the inclusion of vessels in pedicle of the flap. The sural artery reverse flow flap is nourished by the lowermost perforating branch of the peroneal artery. Since the introduction of this flap by Masquelet \[[@B3]\] in 1992, it has become one of the favourite among reconstructive surgeons. It is the flap of first choice in our centre. In the present report, the results of reverse flow sural neurocutaneous flaps done at our centre is presented along with some thought provoking ideas for the future. Methods ======= This descriptive case series was conducted between 1997 and 2003, at Queen Mary Hospital, The University of Hong Kong, Division of Hand Surgery, Department of Orthopaedics and Traumatology. It is a tertiary care centre for cases requiring microsurgery of the extremities. From 1997 to 2003, we performed ten reverse sural neurocutaneous flaps on ten patients (See Table [1](#T1){ref-type="table"}). The age ranged from 19 to 85 years, with a mean age of 59.8 years. There were six females (60%) and four male patients (40%). The soft tissue defects were located on non weight bearing heel in four patients (40%), tendo Achilles in two patients (20%), distal tibia in two patients (20%), lateral malleolus in one patient (10%) and medial aspect of the midfoot in one patient (10%). No patient underwent resurfacing of the weight bearing heel. The minimum and maximum flap size harvested (length × breadth in cm.) was 5 × 4 and 14 × 6 respectively. A midline cuff of gastrocnemius was included in one of the flap harvested from the proximal calf area. All benefits and disadvantages of the operation were discussed with the patient before the operation, including the sensory loss in the distribution of the sural nerve. A detailed questionnaire was developed using variables of interest. ###### Patients\' data. Demographic features, etiology, defect site and size, comorbids, flap type and outcome **Sex** **Age** **Etiology** **Site of defect** **Flap size (cm)** **Comorbids** **Flap type** **Complications** --------- --------- --------------- ------------------------------- -------------------- --------------- --------------- ------------------------ F 72 Ulcer Medial aspect of mid foot 9 × 7 DM Sural Mild venous congestion M 74 Pressure sore Lateral malleolus 5 × 4 Paraplegia Sural F 68 Ulcer NWB heel 8 × 6 DM Sural F 62 Trauma NWB heel and medial malleolus 11 × 7 Sural M 48 Trauma Tendoachilles 8 × 6 Sural F 53 Trauma Shin 8 × 6 Sural F 19 Trauma NWB heel 14 × 6 Sural M 52 Trauma Distal tibia 9 × 7 Sural F 85 Ulcer Achilles tendon 7 × 4 PVD, DM Sural M 65 Ulcer Posterior heel 8 × 5 DM Sural Preoperative evaluation included identification of the site of peroneal perforators above the lateral malleolus using Doppler flow meter. Two or three perforators were identified above the lateral malleolus. The pivot point of the pedicle was chosen according to the need of distal coverage, but it was limited by the lower most perforator, about 05 cm. from the tip of the lateral malleolus. The skin was incised close to the midline, so as to be sure to include the short saphenous vein in the pedicle in all patients. The skin flap was then elevated with the deep fascia. The pedicle was kept wide, around 03 to 04 cm. in diameter. The flap was rotated 180 degrees through an open tunnel in all but one patient. Split thickness skin graft was used to cover the exposed pedicle in all but one patient. All patients were kept in the \'modified plaster of paris boot\' in the post operative period (Figure [1](#F1){ref-type="fig"}). {#F1} Results (Also see Table [1](#T1){ref-type="table"}) =================================================== Survival -------- All flaps survived. Venous congestion ----------------- Only one flap developed mild venous congestion in the early post operative period, which resolved with foot elevation. Weight bearing problems ----------------------- None of our patients had resurfacing to the weight bearing part of the heel. There was no interference with walking. Donor and recipient site appearance ----------------------------------- The donor site had no complications except for the relatively unsightly scar. It was acceptable to all patients. All patients were satisfied with the cosmesis of the recipient site. Numbness/Neuroma ---------------- Transient numbness along the sural nerve distribution was experienced by the patients. It resolved in all patients by three months. No painful neuroma developed as we routinely buried the nerve stump in the deep muscular plane. Need for debulking of the flap ------------------------------ No patients required debulking for cosmetic reasons or shoe wear problems. Case reports ============ Case report 1 (Figures [2A](#F2){ref-type="fig"} and [2B](#F2){ref-type="fig"}) ------------------------------------------------------------------------------- {#F2} 72 years diabetic lady suffering from ulcer on the dorsomedial aspect of the right midfoot was successfully covered using the reverse sural flap from proximal calf area with the modification of inclusion of the midline cuff of gastrocnemius, as described by Al Qattan MM \[[@B4]\]. Case report 2 (Figures [3A](#F3){ref-type="fig"}, [3B](#F3){ref-type="fig"}, [4A](#F4){ref-type="fig"}, [4B](#F4){ref-type="fig"} and [4C](#F4){ref-type="fig"}) ---------------------------------------------------------------------------------------------------------------------------------------------------------------- {#F3} {#F4} 19 years lady sustained road traffic accident resulting in skin necrosis of the posterior aspect of the left heel. Reverse sural flap size 14 × 6 cm. was used to successfully cover the defect. Discussion ========== The results of this study show that the reverse sural neurocutaneous flap is an effective method to resurface soft tissue defects around the foot and ankle region. We would recommend some measures that we adopted in our cases and which have been advised in the literature to ensure safe flap elevation and improve the venous congestion of the flap, hence decrease the risk of failure. 1\) Preoperative identification of the peroneal artery perforators by Doppler should be performed and the peroperative search for a perforator should be avoided which can be potentially dangerous \[[@B1]\]. 2\) A \'modified plaster of paris boot\' (Figure [1](#F1){ref-type="fig"}), that we describe in our cases can be used to avoid pressure on the pedicle in the postoperative period. This helps the patients to remain supine, allows assessment of the flap circulation and assists in the elevation of the foot to relieve venous congestion. 3\) The pedicle should be kept wide (3 to 4 cm.) \[[@B1],[@B3]\] and skin grafting should be done on the pedicle if the subcutaneous tunnel becomes tight and impairs circulation. In our series, in nine out of ten patients the pedicle was passed through an open tunnel and skin grafted. Although skin grafting added slightly to the donor site morbidity, it prevented pressure on the pedicle in the post operative period. 4\) The lesser saphenous vein should be included in the pedicle in all cases \[[@B2]\]. 5\) There are a few studies in which this flap is used to cover defects in the distal part of the foot by slight modification of the flap design, by inclusion of the midline cuff of gastrocnemius muscle, in flaps harvested from proximal calf area \[[@B4]\]. We performed one flap in this manner to cover the medial aspect of the midfoot in an old lady with diabetic ulcer (Figure [2A](#F2){ref-type="fig"} and [2B](#F2){ref-type="fig"}). We postulate that the whole skin of the posterior calf can be harvested without any major problem. Sural neurofasciocutaneous flaps as large as 17 × 16 cm. have been reported \[[@B5]\]. 6\) Leech therapy can also be used to decrease venous congestion \[[@B6]\]. We did not require the use of leech therapy in cases of reverse sural neurocutaneous flaps but did found it of use in a reverse flow saphenous neurocutaneous flap. Anastomosis with one of the donor veins from the foot may also be an alternative, but we do not see the need for this intervention at the moment. The vascular anatomy and clinical application of the reverse sural neurocutaneous flap has been well studied \[[@B2],[@B7]-[@B10]\]. The artery existed as an axial pattern vessel in only 50 percent of our patients. Self limiting numbness in the distribution of sural nerve is not a major concern. The patients should be counselled preoperatively and the problem is usually resolved on an average in three months in all patients. To avoid a painful neuroma, the nerve stump needs to be buried in the deep muscular plane. Donor site morbidity was minimal in our patients. Unaesthetic donor site, perhaps, may be a major concern to the young female patients. It can potentially be avoided by using an adipofascial flap, rather than harvesting skin island in larger flaps. Future Concern -------------- Based on the sensate character and same quality skin, medial plantar flap has been proposed to be superior for heel coverage over sural artery reverse flow flaps \[[@B11],[@B12]\]. However no substantial clinical or laboratory data is available for reinnervating the reverse sural artery flaps. Reverse flow homodigital island flaps for finger tip injuries have been rendered sensate by including a segment of palmar digital nerve in the flap design and anastomosing it to the opposite side digital nerve \[[@B13]-[@B15]\]. Based on the above observations we postulate end to side anastomosis of the sural nerve to a plantar digital nerve, in order to provide sensory reinnervation to the flap especially in cases of weight bearing heel reconstruction. Conclusion ========== We highly recommend this flap to resurface soft tissue defects in the foot and ankle region because it is easy to harvest, reliable and can be used in patients with peripheral vessel disease and trauma patients with damaged major vessels \[[@B1]\]. List of abbreviations ===================== POP: plaster of paris, CR: case report, pre op: pre operative, post op: post operative, debrima: debridement, NWB: non weight bearing, F: female, M: male, PVD: peripheral vascular disease, DM: diabetes mellitus, cm: centimetres Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= SKA carried out patient follow up, concept design, data collection, literature search, manuscript writing and critical revision. BKKF carried out surgery, patient follow up, concept design, manuscript writing and literature search. WYI carried out surgery, patient follow up, literature search. MF carried out patient follow up, data collection. SPC carried out surgery, review and final approval of the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ We acknowledge the help of Mr. Henry Leung in the process of formatting figures and table according to the Journal\'s guidelines. As no individual patient\'s identity is revealed by the photographs, verbal consent for publication was obtained from the patient or their relative, either telephonic or on a follow up visit.

{#sp1 .277} {#sp2 .278} {#sp3 .279} {#sp4 .280} {#sp5 .281} {#sp6 .282} {#sp7 .283} {#sp8 .284} {#sp9 .285} {#sp10 .286} {#sp11 .287} {#sp12 .288} {#sp13 .289} {#sp14 .290} {#sp15 .291} {#sp16 .292} {#sp17 .293} {#sp18 .294} {#sp19 .295} {#sp20 .296} {#sp21 .297} {#sp22 .298} {#sp23 .299}

Background ========== Increased population density and movement of people around the globe have generated a rise in the number of outbreaks of infectious diseases and led to the emergence of new infectious diseases \[[@B1]\]. Worldwide, 3.575 million people die each year from water-related diseases \[[@B2]\]. The water and sanitation crises claim more lives through disease than any warfare \[[@B2]\]. A key step in the prevention of outbreaks of communicable diseases is the early detection of virulent particles \[[@B3]\]. Rapid detection of active viral pathogens is of central importance for public health risk assessment and environmental protection. Waterborne viruses are particularly important for public safety monitoring due to their environmental stability and low infectious dose; a single virion is sufficient to initiate illness in previously unexposed, healthy adults \[[@B4]\]. Enteroviruses (family Picornaviridae) are a genus of waterborne viruses that infect humans and other mammals. They are a health problem worldwide, leading to 10 to 15 million cases of symptomatic infection in humans annually in the United States alone \[[@B5]\]. Enteroviruses are single, positive-strand RNA viruses that include polioviruses, Coxsackieviruses and echoviruses, among others. Some enteric virus groups have emerged as waterborne pathogens because of their high levels of resistance to current water treatment processes, which include ultraviolet light inactivation and heat inactivation \[[@B6],[@B7]\]. Poliovirus was used here as a model virus because a large body of research data exists on the physical, chemical and biological properties of the virus, vaccination is available, and its ease of cell culturing \[[@B8]-[@B10]\]. In addition, poliovirus remains endemic in four countries. During 2002 the rejection of polio immunization led to a worrying resurgence of polio in some areas of Nigeria, followed by re-infection in 21 other countries; resurgence of the disease was also observed in India. Auxiliary vaccination actions were restarted and by 2007 most re-infected countries had become polio-free again. The goal of global polio eradication was re-set to 2010, but concerns continue to be expressed about the progress of this eradication program \[[@B11]\]. Current methods for enterovirus detection use mammalian cell culture and require complex analyses (visible monolayer cytopathic effects) that require several days of laboratory time \[[@B12]\]. Polymerase chain reaction (PCR) methods for the detection of viruses have been developed, offering specificity, speed and cost advantages over cell culture methods \[[@B13],[@B14]\]. PCR methods alone do not, however, differentiate between the presence of physical (inactive) virus particles and viable (active) virus particles \[[@B6],[@B15]\]. The major disadvantage of most current methods of virus detection is the inability to provide information about whether a viral particle can start an infection or not. Faster methods with increased sensitivity and specificity are needed to quantify active viral pathogens from medical and environmental samples. Virus-infected individuals can excrete over 1 billion (10^9^) viruses/g of feces. Some enteric viruses can also be excreted in urine from infected individuals. The presence of these viruses in a human population is variable and reflects current epidemic and endemic conditions \[[@B16]\]. In general, the level of infectious enteric virions in sewage ranges from 100 to 10,000 infectious units/L \[[@B17]-[@B23]\]. In contaminated surface water, levels of 1-100 infectious enteric virions/L are common. In less polluted surface water, their numbers are closer to 1-10/100 L. Groundwater sources have been shown to have between 0 and 200 infectious enteric virions/100 L, depending on the level of contamination; however, most contaminated groundwater systems are thought to have very low levels (\< 2/100 L) \[[@B24]\]. These concentrations were generally obtained through targeted studies, since water and wastewater sources are not routinely monitored for enteric viruses. These measurements are typically performed after an extensive concentrating step which reduces sample volume by 100 to 1000-fold before virus detection is performed. FTIR spectroscopy is a noninvasive measurement method that has previously been applied for identifying various biological components of cells by detecting vibrations of molecules leading to spectral patterns \[[@B12],[@B25]\]. It can be used as part of a sensitive method for the detection of specific cellular molecular changes \[[@B12],[@B26]-[@B30]\]. Quantitative infrared absorption methods such as FTIR spectroscopy differ from ultraviolet/visible molecular spectroscopic methods because of the greater information content of the spectra. FTIR spectroscopy has been applied in medicine, particularly to study the process of herpes virus infection \[[@B31]\] and in the diagnosis of cancers and other disorders \[[@B32]-[@B35]\]. In addition, FTIR spectroscopy has been used for the quantification of blood serum components such as glucose, protein, cholesterol and urea \[[@B36]\]. Cell based sensors detect changes in the physiological state of cells following exposure to an environmental stimulus. Changes in cell state can provide information about the stimulus; for example, cells have been used to sense toxins in water samples \[[@B37],[@B38]\]. Cell based sensors have recently been combine with spectroscopy and applied to viral detection \[[@B39],[@B40]\]. Cantera et al. \[[@B41]\] used an optical system that involved the use of molecular beacons as a way to detect infective virus particles, resulting in a detection limit of 1 PFU. Using live cells to assist in identifying and quantifying viruses in samples helps to bring the detection closer to an *in vivo*setting, allowing the natural and complex interactions between cell and virus to be part of the experimental setup. Here we present the development of a novel strategy for virus detection using a combination of FTIR spectroscopy and live BGMK cells (Figure [1](#F1){ref-type="fig"}). Specific absorbance patterns are monitored following changes in cell components (such as lipids, proteins, nucleic acids and sugars) subsequent to the virus infection, effectively using the cells as biosensors \[[@B42],[@B43]\]. {#F1} Results ======= Microscopy of cell structure on crystal --------------------------------------- The way in which cells attach in monolayer culture depends on the cell type and the characteristics of the surface. BGMK cells were confirmed to be biocompatible with the ZnSe crystals for transmission measurements. The organization of the actin cytoskeleton exhibited a good distribution throughout the cell volume with many focal adhesion points that resulted in a spread phenotype on the stiff surface, indicating actin assembly by healthy cells. The cell architecture of the BGMK cells attached to the ZnSe crystal can be seen in Figure [2a](#F2){ref-type="fig"}. Actin, vinculin and cell nuclei were stained to study cell adhesion of BGMK cells on the ZnSe crystal. Figure [2b](#F2){ref-type="fig"} shows the cells under bright field microscopy, where they can be seen elongated and spread out over the crystal surface. {#F2} Optimal time for virus detection -------------------------------- BGMK cells were infected with poliovirus PV1 at different multiplicities of infection (m.o.i.) of 10 PFU (0 - 10^6^PFU/ml) and studied at 1, 1.5, 2, 4, 5, 6, 8 and 12 h.p.i. Virus infection regression models were developed to correlate changes in spectral features with time of infection. Changes in the spectra varied depending on the progress of the viral infection, with biochemical alterations appearing in poliovirus infected cells within 2 h.p.i. Example regression models for 1.5, 4, 6 and 8 h.p.i. are shown in Figure [3](#F3){ref-type="fig"} and a summary of the regression model parameters are given in Table [1](#T1){ref-type="table"}. This table shows the comparison between the error of calibration (RMSEC) and the root mean square error of crossvalidation, RMSECV, which is a mesure of a model\'s ability to predict samples that were not used to build the model (leave-one-out-crossvalidation). RMSECV was analyzed to determine the optimum number of latent variables (LVs) to include in the PLS model. The number of LVs of a PLS model is usually optimized by performing a cross-validation and minimizing the corresponding RMSECV. The RMSECV decreases with the inclusion of each additional initial factor, reaching a minimum value with a certain number of latent variables. The best choice of number of latent variables is also supported by the variance capture in Y. The goal is to identify a subset of the measured variables that gives the lowest RMSECV, which is the most useful and accurate regression model. General calibration procedure consists in the selection of the pretreatment (preprocessing of the spectra), wavelength intervals, and number of latent variables to be driven by minimizing the RMSECV \[[@B44]\]. An infection time of 8 h was selected for subsequent experiments based on these results and on the reported viral replication time \[[@B44]\]. For more information on the spectra of 8 h and 12 h uninfected control cells see Additional File [1](#S1){ref-type="supplementary-material"} and Additional File [2](#S2){ref-type="supplementary-material"} respectively. {#F3} ###### Comparison of poliovirus prediction models using different infrared regions and virus titers Method IR region used h.p.i Virus titers (PFU) Latent variables R^2^ RMSEC RMSECV ------------- ------------------- ------- ------------------------ ------------------ ------- -------- -------- **Initial** 650 - 1650 cm^-1^ 1.5 0, 10^2^- 10^4^ 4 0.910 0.4225 0.7093 650 - 1650 cm^-1^ 4 0, 10^1^, 10^3^, 10^4^ 3 0.891 0.5225 0.8900 650 - 1650 cm^-1^ 6 0, 10^1^, 10^2^, 10^4^ 6 0.868 0.4914 0.8800 650 - 1650 cm^-1^ 8 0 - 10^4^ 7 0.911 0.4290 0.8471 **Revised** 650 - 3600 cm^-1^ 8 0 - 10^4^ 7 0.819 0.6480 0.9187 650 - 1650 cm^-1^ 8 0 - 10^3^ 7 0.917 0.3298 0.5726 iPLS 9 regions 8 0 - 10^3^ 7 0.903 0.3577 0.5581 iPLS 9 regions 8 0, 10^2^- 10^4^ 7 0.944 0.4019 0.6628 iPLS 9 regions 8 0 - 10^2^ 7 0.964 0.1640 0.3163 650-1650 cm^-1^ 12 0 - 10^3^ 7 0.716 0.5963 1.2866 Table shows root mean square error of calibration (RMSEC), root mean square error of cross validation (RMSECV) and correlation coefficient for each model. Viral infection at 8 h.p.i -------------------------- Spectra of cells at 8 h.p.i infected with different m.o.i. of PV1 are shown in Figure [4](#F4){ref-type="fig"}. Interval Partial Least Square (iPLS) was performed to determine the optimal regions of the spectra used in the virus detection model, an interval size of 10 cm^-1^and a maximum of 8 latent variables were chosen. These 9 areas of the spectra correspond to the following wavenumbers: 660.90 - 667.26, 806.11 - 823.46, 844.68 - 862.03, 979.67 - 997.03, 1095.38 - 1112.74, 1191.80 - 1209.16, 1230.37 - 1267.02 and 1326.80 - 1344.16 cm^-1^. {#F4} Changes in absorbance can be correlated with the development of poliovirus infection. The region between 600 - 900 cm^-1^corresponds to C2\' endo/anti (B-form helix) conformation, DNA and RNA molecules. 1000 to 1300 cm^-1^relates to symmetric stretching mode of dianionic phosphate monoester in phosphorylated proteins and left handed helix DNA (Z form) \[[@B45]\]. Figure [5](#F5){ref-type="fig"} shows a regression model of poliovirus at 8 h.p.i. that correlates changes in absorbance spectra with virus infection titer. This model has a root mean square error of cross validation (RMSECV) of 0.57 log. Similar results were achieved using the whole spectral region or the nine regions selected by iPLS (data not shown). This method can detect a viral titer of 10^1^- 10^2^PFU/ml with a RMSECV of 17 PFU/ml; at higher titers, 10^2^- 10^4^PFU/ml, a RMSECV of 2009 PFU/ml for 8 h.p.i. was achieved. {#F5} Effect of virus titer on characteristic spectra peak height at 8 h.p.i ---------------------------------------------------------------------- The characteristic peaks of the BGMK cell line showed changes in relative absorbance following infection with different viral titers. These peaks correspond to different biomolecules. A graphical representation of the absorbance values of the eighteen characteristic peaks of each data set for the different infection titers (10^6^, 10^5^, 10^4^, 10^3^, 10^2^and 10^1^PFU/ml) for the 8 h.p.i. time point is given in Figure [6](#F6){ref-type="fig"}. A summary of the change in peak height upon virus infection and the corresponding biomolecules represented at the peak wavenumbers are given in Table [2](#T2){ref-type="table"}. {#F6} ###### Observed changes in relative peak absorbance values with corresponding biomolecules ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Peak wavenumber (cm^-1^) Corresponding biomolecules Observed changes in cell absorbance following PV1 infection relative to uninfected control cells ------------------------------ ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **700** (Broad) cis-C-H out-of-plane bend General increase **835 - 840** C2\' endo/anti (B-form helix) conformation and left handed helix DNA (Z form) Gradual increase corresponding to increased PV1 titer **900 - 1350** Phosphodiester stretching bands (from absorbance due to collagen and glycogen) Decrease; trend is less clear at higher PV1 titers **1040 - 100** Symmetric stretching vibration PO~2~due to RNA and DNA General increase **1079/80** Symmetric phosphate stretching modes or ν(PO2 −) symmetric. Phosphate stretching modes originate from the phosphodiester groups in nucleic acids. Sharp increase in BGMK cells infected with 10^1^PFU/ml titer followed by a decrease in the 10^2^PFU/ml infected cells then a gradual height incremental with the higher titers. **1153** Stretching vibrations of hydrogen bonding\ Virus titer 10^2^PFU/ml shows a decrease C-OH groups **1236/7** Amide III, stretching PO~2~−asymmetric (phosphate I) Decrease in the peak height for the cells infected with 10^2^and 10^3^PFU/ml; higher titers shown an increase **1312-1317** Collagen related, amide III band components of proteins General increase **1455/6** Asymmetric CH~3~bending modes of the methyl groups of proteins General decrease **2852, 2873, 2924/5, 2956** Lipids region (CH~2~symmetric, symmetric stretching vibration of CH~3~of acyl chains, stretching C-H and asymmetric stretching vibration of CH~3~of acyl chains respectively) Decrease for infection titers 10^2^- 10^3^PFU/ml **3084** Stretching N-H symmetric General increase **3293** OH stretching (associated) General increase; increase gradually less from 10^1^- 10^6^PFU/ml titers ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ Some trends are evident for the change in peak heights at different titers. For example, lower virus concentrations (10^2^and 10^3^PFU/ml) showed negative values relative to controls at 1399 - 2956 cm^-1^, indicating a decrease in the amount of a specific biochemical component. In contrast, the mean absorbance of the cells with higher virus titers (10^4^, 10^5^and 10^6^PFU/ml) only showed average increases in absorbance across all peaks relative to uninfected controls. The most significant differences were found for the peak at 3293 cm^-1^, assigned to OH stretching (One-way ANOVA, d.f. 6, F = 7.63, p \< 0.0001), for which the 10^6^PFU/ml sample had significantly lower absorbance than the 10^1^PFU/ml sample (Tukey-Kramer HSD, α = 0.05). The 10^6^PFU/ml sample and uninfected control had significantly lower absorbance than the 10^1^- 10^5^PFU/ml samples at this wavenumber (Tukey-Kramer HSD, α = 0.05). A significant difference was found for the peak at 1043 cm^-1^assigned to glycogen (One-way ANOVA, d.f. 6, F = 2.47, p = 0.032) \[[@B46],[@B47]\] for which the absorbance of the 10^1^PFU/ml sample was significantly higher than the absorbance of the uninfected control (Tukey-Kramer HSD, α = 0.05). Alterations in glycogen have been described in neurons of monkeys infected with poliovirus \[[@B48]\]. The mean absorbance values for the different titers at the 700 cm^-1^(ANOVA, d.f. 6, F = 2.37, p = 0.0383) and 835 cm^-1^(ANOVA, d.f. 6, F = 2.36, p = 0.0387) peaks also formed significantly separate groups. The absorbance for the 1079 cm^-1^(ANOVA, d.f. 6, F = 2.08, p = 0.0659), 1313 cm^-1^(ANOVA, d.f. 6, F = 1.99, p = 0.0781), 2852 cm^-1^(ANOVA, d.f. 6, F = 1.89, p = 0.0942), 2924 cm^-1^(ANOVA, d.f. 6, F = 2.16, p = 0.0575) titers showed consistent trends. CPE analysis ------------ The error associated with a standard CPE analysis was compared to the error of the new FTIR spectroscopy with cell culture method. The results of this assay are shown in Table [3](#T3){ref-type="table"}. Standard error was calculated from duplicate experiments except where there were too many PFUs to count accurately. The average standard error was 12.0% of the mean estimated number of PFU/ml. For comparison, the error of cross validation was \~17% of the mean across titers using the FTIR with cell culture method. Note that these calculations assume a perfect serial dilution to obtain the starting known viral titers. ###### CPE Analysis for Poliovirus at a virus titer of 10^7^PFU/ml. Sample of Known Titer (PFU/ml) Dilution Factor Estimated PFU/ml **S.E**. -------------------------------- ----------------- ------------------ ---------- **1000000** -6 7000000 2000000 **100000** -5 550000 50000 **100000** -4 380000 40000 **10000** -4 35000 5000 **10000** -3 39500 2500 **1000** -3 4000 0 **1000** -2 3300 100 **10** -1 5 5 Discussion ========== Cells infected with poliovirus PV1 showed consistent alterations in their infrared spectral features. The broad, undulating features observed relatively weakly in the spectra of entire cells has been previously attributed to Mie scattering of the cellular nuclei \[[@B49],[@B50]\]. The overall variations in intensity in the initial method are most likely due to variations in the thickness of the cell, as well as the nucleus/cytoplasm (N/C) ratio. Multiple chemometric models were developed using a variety of pre-processing approaches, different spectral regions and different latent variables. This method could detect a viral titer of 10^1^- 10^2^PFU/ml with a RMSECV of 17 PFU/ml in cells at 8 h.p.i. At higher titers the accuracy decreased slightly, for example at 10^2^- 10^4^PFU/ml, a RMSECV of 2009 PFU/ml for 8 h.p.i. was achieved. Changes in cell spectral features become apparent initially at 2 h.p.i, but detection accuracy is best at 8 h.p.i. Changes in cell absorbance spectra were correlated with specific cell components to better understand the progress of the poliovirus infection. The most significant changes were associated with an increase in the peak assigned to OH stretching, particularly at low virus titers. Another significant change associated with infection was an increase in the peak assigned to symmetric stretching vibration of PO~2~due to RNA and DNA, which was most evident in the 10^1^PFU/ml virus titer. Broad cis-C-H out of plane bend and center handed helix of DNA also showed significant changes associated with PV1 infection. To obtain the high level of sensitivity required substantial method development steps. Initial experiments resulted in models with a high level of error. In the initial approach cells were grown in a culture flask, trypsin was used to detach the cells, they were centrifuged and transferred to the ZnSe crystals for transmission measurements. Growing cells directly on the thin ZnSe crystals used a greatly simplified protocol, removing many steps where error and sample handling variation could occur. The direct cultivation method produced cleaner, more stable spectral patterns and had a higher reproducibility compared to the initial method. This virus detection method was consistently more accurate for lower virus titers. This could be because the lower titers cause less variation in the viral infection process of adsorption and penetration inside of each individual cell. For example, at low titers the virus particles may be invading the cells in synchrony, causing more equal changes in the cell components among cells. With many virus particles present there might be more than one virus infecting each cell, meaning that the stage of infection is not synchronized among cells. Cantera et al. \[[@B41],[@B51]\] have reported greater success for detection of lower virus titers of poliovirus after 12 h.p.i. It may be that for lower virus infection it is necessary to wait longer until the cells reach some steady infection state so that the changes in cell components are saturated, causing the overall signal pattern to stabilize. An infection period of eight hours had the lowest RMSECV of the times tested here (Table [1](#T1){ref-type="table"}, Figure [7](#F7){ref-type="fig"}). Poliovirus replicates and lyses the cell in approximately 8 h \[[@B44]\] and a previous study using molecular beacons also found this time period to be the best for viral quantification \[[@B41]\]. Eight hours of infection was found to be the best for detection of poliovirus with a method based on engineered BGMK cells expressing fluorescent proteins undergoing fluorescence resonance energy transfer (FRET) \[[@B51]\] and \[[@B52]\]. Perhaps this is the time point when the most dramatic changes in cell components are occurring \[[@B51]\], meaning that the eight hour time point in our protocol may be just prior to virus release from infected cells, producing a better prediction model of the infection (Figure [7](#F7){ref-type="fig"}). Note that our method does not require any modification to the sacrificial cells and as such can have the recognition element readily changed. {#F7} A CPE assay is the standard method for enterovirus detection and requires three to fourteen days to perform depending on the virus type. The poliovirus assay takes three to ten days during which the presence of a virus is marked by the death of animal cells in culture \[[@B3]\]. The cells are grown in a monolayer in a semi-solid media such that new virus particles infect surrounding cells. Cells killed by the virus form a plaque and the remaining living cells are stained with a dye. Another method, integrated cell culture-PCR (ICC-PCR), is a faster molecular detection method that can identify low levels of viable virus \[[@B3]\]. A sample is applied to cell culture and then molecular methods are used to detect replicating virus in the cell culture. This method takes 1 to 3 days from infecting the cells, depending on the replication time of the virus \[[@B13]\]. The cell culture-FTIR process has a distinct advantage over standard PCR methods since it provides information on viral infectivity and is inherently a multiplexed method; it does not however provide information on the specific type of viral infection. We envision this approach as being on the front end of a comprehensive screening process upstream of a more time consuming but more specific method such as cell culture-PCR to later determine the identity of the virus only for samples that indicate the presence of such an infective component. Conclusions =========== Here we present a fast new method for the detection of low titers of viable poliovirus using mammalian cells as a biosensor combine with FTIR spectroscopy to detect changes in cell components. Virus titers from 10^1^- 10^6^PFU/ml were assayed over a range of infection times from 1 h.p.i. - 12 h.p.i., with detection most accurate at low titers from 10^1^-10^4^PFU/ml. Prediction models of infected cells were best detected as early as 6 h.p.i., and had the most accurate prediction of virus titer at 8 h.p.i. The model gave a RMSECV of 17 viral particles. The major changes in cell components following virus infection were an increase in the peak 3293 cm^-1^and negative trend in the absorbance for lipids region (3200 - 2800 cm^-1^) in the poliovirus infected cells with 10^2^- 10^3^PFU/ml. The cell culture method is still considered the standard for viral diagnosis as it has the advantages of detecting infectious viral particles and the ability to achieve low detection limits \[[@B6]\]. Combining the cell culture method with FTIR spectroscopy, we can enhance the cell culture method with the increased speed of FTIR spectroscopy. This new method has the potential to be extended for the detection of other viruses and adapted into a portable, automated system for detection of viruses from environmental samples. Methods ======= Protocol development -------------------- An overview of the method is presented in Figure [1](#F1){ref-type="fig"}. Experiments were performed in a biosafety cabinet under laminar flow to maintain aseptic conditions. Our first experiments involved growing cells on the plastic surface of 24-well plates, removing the cells with an enzymatic treatment, pelleting the cells by centrifugation and then transferring them to ZnSe crystals for transmission measurements, allowing them to dry on the crystal to scan using FTIR spectroscopy. Spectra were collected in a transmission mode with light passing through both the cells and the underlying ZnSe crystal. Several changes were made to the initial method to optimize both the laboratory steps and the spectra results by reducing the room for variation between experimental replicates. We present both methods for comparison. Cell culture ------------ BGMK cells were selected due to their high sensitivity to enteroviruses and previous use \[[@B53]\]. Cells were grown in T25 flasks to a confluent monolayer in Dulbecco\'s Modified Eagle Medium (DMEM) with 584 mg/L L-glutamine (Cellgro) containing 0.1% NaHCO~3~, 10,000 units/ml penicillin, 10,000 μg/ml streptomycin (HyClone) and 10% (vol/vol) fetal bovine serum (FBS; HyClone) and buffered with 12 mM HEPES. Reagents were from Sigma Chemical Co. unless otherwise noted. Cells were cultured at 37°C in a humidifying incubator with 5% carbon dioxide and passaged when confluent, approximately every two days. Cells were detached from the growth flask with 2.5% EDTA-trypsin, centrifuged at 18 g for 8 minutes, resuspended in media and transferred into a flask of fresh media. Initial method -------------- T25 flasks with 95% confluent cell monolayers were trypsinized, resuspended in media and seeded in the eight central wells of a 24-well plate. The wells were individually seeded with 125,000 cells in 1 ml of media. Cells were grown to 100% confluence by incubating for 24 h at 37°C and 5% CO~2~. 0.5 ml of media was removed from the well and replaced with 0.5 ml of virus. A minimum of two control samples per experiment were inoculated with virus-free media using the same incubation time for infected and uninfected cells. Following viral exposure, cells were detached with 70 µL of trypsin per well and pelleted by centrifugation at 18 × g for 8 minutes. The cells were washed with 1 ml of 0.9% NaCl to remove any residual cell media then resuspended in 20 μL of 0.9% NaCl saline solution. Revised method -------------- Several changes to the initial method were made to reduce the room for variation between experimental replicates, including growing the cells directly on the ZnSe crystals for transmission measurements. Individual ZnSe crystals were located in the eight central wells of a 24-well plate and seeded with 150,000 cells with a final volume of 1 ml DMEM media per well in a 24-well plate and incubated for 24 h to allow cell attachment to the crystal surface. After the 24 h incubation time, 0.5 ml of media was removed from the well and replaced with 0.5 ml of the virus titer. A minimum of two control samples per experiment were inoculated with virus-free media and incubated for the same period of time. Following the virus infection, ZnSe crystals were removed from the aqueous medium allowing them to dry at room temperature for a minimum of 8 h. Spectra were collected in a transmission mode with light passing through both the cells and the underlying ZnSe crystal. An estimated 100,000 cells were scanned per ZnSe crystal. ZnSe crystals were washed with soap, pure acetone and sterilized in 70% EtOH between uses. Experiments with different infection times were performed with the initial method. The 8 h.p.i. and 12 h.p.i. experiments were subsequently performed with the revised method. Virus titers and infection times -------------------------------- Multiple serial dilutions were made from the initial stock concentration of 10^7^PFU/ml of purified vaccine strain PV1 (LSc-2ab) using different time of infection experiments. Previous studies observed molecular and morphological changes in poliovirus infected cells within 2 h.p.i. \[[@B44]\]. This infection time was therefore used as a starting point for identification of the optimum incubation time. BGMK cells were infected with PV1 at m.o.i. of 10 PFU (0 - 10^6^PFU/ml) and studied at 1, 1.5, 2, 4, 5, 6, 8 and 12 h.p.i. Different viral titers were made using DMEM with 10% NCS. FTIR spectroscopy ----------------- Infrared spectra of healthy cells and cells post-infection were collected in transmission mode on a ThermoNicolet Magna 560 FTIR spectrometer equipped with a liquid nitrogen-cooled MCT-A detector, KBr beamsplitter and infrared light source. The spectral collection parameters used were 128 co-added scans and a spectral resolution of 4 cm^-1^with a 20% aperture opening, performed using OMNIC 7.3 (Thermo Electron Corporation). Preprocessing of spectra ------------------------ Preprocessing was performed to reduce noise and other interferences in the data. This included computation of the first derivative of spectral intensity with respect to wavenumber, vector normalization using amide I as the maximum absorbance equal to unity and mean center. Spectral expansion emphasized the most diagnostic \"mid-IR\" region between 650 and 1700 cm^−1^and eliminated a large spectral range (1800 - 2800 cm^−1^), which contains no vibrational spectroscopic information. Data analysis ------------- Collection of the spectra was performed using OMNIC 7.3 (Thermo Electron Corporation). Analysis of spectra was carried out using the PLS toolbox. PLS is a multivariate method to analyze noisy, strongly collinear (correlated) data with numerous X-variables \[[@B54],[@B55]\]. iPLS assisted in the selection of the most informative spectral regions. OMNIC 7.3 was used to find the absorbance values of the eighteen characteristic peaks for the different infection titers for 8 h.p.i. Differences among peak heights were analyzed using JMP version 8.0 with a one-way ANOVA. Pair-wise comparisons of peak height were performed using a Tukey-Kramer HSD test that corrects for multiple comparisons. The analysis utilizes a \"leave-one-out\". In brief, the entire data set serves as the calibration, save for one sample that is employed as the prediction or validation sample. The ability to predict the concentration in this one sample is evaluated and tabulated then the one sample is returned to the calibration set, another sample is selected, and the process is repeated. In this way each sample serves as a blind assessment of prediction capability. The method the review suggests would only evaluate a small number of samples and so would not be as robust of an assessment. Microscopy of cell structure on crystal --------------------------------------- An Actin Cytoskeleton and Focal Adhesion Staining Kit (Millipore) was used to visualize the architecture of cells attached to titanium oxide (data not shown) and ZnSe crystals and to perform immunohistochemistry to observe the focal adhesion points and actin filaments in the cells. Actin was stained using TRITC conjugated Phalloidin and nuclei with DAPI. The average cell height on titanium oxide was determined using a z-axis stack of cell images. Cytopathic effect assay ----------------------- A CPE assay was conducted as described previously \[[@B3]\]. Briefly, a poliovirus sample of 10^7^PFU/ml was diluted in series. 1 ml of 1% (wt/vol) carboxymethylcellulose in maintenance medium (with 2% FBS) was overlaid onto an infected-cell monolayer. After 48 h at 37°C, plaques were stained with 0.8% (wt/vol) crystal violet in 3.7% (vol/vol) formaldehyde. List of abbreviations ===================== (CPE): cytopathic effect; (FTIR): Fourier transform infrared; (BGMK): buffalo green monkey kidney; (PFU): plaque forming units; (DMEM) Dulbecco\'s Modified Eagle Medium; (NCS): new calf serum; (PCR): polymerase chain reaction; (RNA): ribonucleic acid; (DNA): deoxyribonucleic acid; (RMSECV): root mean square error of cross validation; (RMSEC): root mean square error of calibration; (PV1): poliovirus 1; (FBS): fetal bovine serum; (iPLS): Interval Partial Least Square; multiplicity of infection: (m.o.i.); hours post-infection: (h.p.i.). Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= FLM conducted the experiments, performed the data analysis, participated in the design and coordination of experiments and drafted the manuscript. MR and KR conceived of the study and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1 **Mean Absorbance Spectra of Uninfected BGMK Cells Incubated for 8 h**. Mean absorbance of uninfected cells for the spectral region between 3600 - 650 cm**^-1^**after 8 h of incubation on ZnSe crystals. Blue dashed lines show standard deviation. ###### Click here for file ###### Additional file 2 **Mean Absorbance Spectra of Uninfected BGMK Cells Incubated for 12 h**. Mean absorbance of uninfected cells for the spectral region between 3600 - 650 cm**^-1^**after 12 h of incubation on ZnSe crystals. Blue dashed lines show standard deviation. ###### Click here for file Acknowledgements ================ FLM was funded by a CONACYT fellowship. This work was supported in part by the NIEHS sponsored Southwest Environmental Health Sciences Center \#P30 ES06694. Dr. Charles Gerba and Dr. Kelly Reynolds generously provided purified vaccine strain poliovirus type 1 (PV1, LSc-2ab), at University of Arizona. Doug Cromey and Dr. Brooke Beam helped with the confocal images and Jonathan Sexton assisted with CPE analysis of the virus samples used in the development of this method.

Introduction {#Sec1} ============ Body size, a key trait for all organisms, affects various aspects of survival and reproduction of individuals in their interactions with their physical and biological environments. In animals, divergence of body size among related populations is often related to adaptation to differing environmental conditions^[@CR1]^ and/or mitigation of interactions with other species through resource competition and/or reproductive interference^[@CR2]--[@CR4]^. Revealing the genetic basis for body size differences between closely related populations is important in understanding the adaptive divergence of populations, which may ultimately result in speciation. Genes involved in the control of animal body size have been examined in various mammals and insects^[@CR5],\ [@CR6]^. Several pathways have been investigated in *Drosophila melanogaster*, including the insulin signalling pathway, which contributes to the regulation of growth rates and body size in response to environmental cues, such as the presence of nutrients^[@CR6]^. For example, overexpression of an insulin-like peptide during development can increase body size^[@CR7]^. Mutations in genes involved in insulin signalling and other developmental pathways regulating larval growth^[@CR8]^ may also be involved in the evolution of overall body size in insects. However, the genetic basis for body size differences among populations or closely related species remains largely unknown. Although several genetic loci or markers in *Drosophila* show strong associations with geographic variation in body size^[@CR9]--[@CR14]^, the relationships between genes or alleles and the regulation of body size that affect body size evolution remain unclear^[@CR6]^. Here, we explored the genetic basis for inter-population body size differences in the carabid ground beetle *Carabus* (*Ohomopterus*) *japonicus* (Fig. [1a](#Fig1){ref-type="fig"}). The subgenus *Ohomopterus*, which is endemic to the Japanese islands, shows marked variation in body size (body length) within and among species^[@CR15],\ [@CR16]^. Notably, two or three species with different body sizes co-occur in a large region of the subgenus' range; the size differences are the main contributors to the avoidance of hybridisation^[@CR4]^. *C*. *japonicus* is sympatric with one or two larger species existing within the majority of its range^[@CR16]^; however, it exhibits larger bodies in several solitary islands, consistent with the pattern of character release^[@CR17]^. Such inter-population divergence might be an initial step in speciation via recurrent secondary contacts between populations of differently sized individuals.Figure 1The body lengths of adult P, F~1~, and F~2~ *Carabus japonicus* beetles. (**a**) Males and females from Mt. Aburayama and Kabeshima Island were used in the present study. Measurements of body dimensions are shown. BL, body length; HW, head width; HL, head length; TW, thoracic width; TL, thoracic length; EW, elytral width; EL, elytral length. (**b**) BLs of parents and F~1~ and F~2~ individuals in the crossing experiment. In this study, we used a traditional approach of accessing the loci responsible for quantitative traits (i.e., QTL mapping) combined with whole genome sequencing and gene prediction to determine the genomic region and gene loci responsible for inter-population body size differences. First, we assembled the genomic sequence of *C*. *japonicus* and predicted protein-coding genes on the scaffolds. Second, we obtained F~2~ individuals by crossing females and males from populations with large and small bodies and measured their body size dimensions. Then, we obtained a large number of restriction site-associated DNA (RAD) sequences for F~2~ individuals to construct linkage maps and to perform QTL mapping. The locations of QTLs for body size on the linkage maps were estimated from the correlations between the phenotypic values of a body dimension and genotypes among F~2~ individuals. Finally, we searched for protein-coding genes potentially related to the body size differences by exploring the scaffolds of the *C*. *japonicus* genomic sequence containing QTLs for body size with reference to the known functions of annotated genes. Results {#Sec2} ======= By experimental crossings, we obtained 40 male and 40 female F~2~ adults (Fig. [1b](#Fig1){ref-type="fig"}). Although the initial first instar weight (BW1) and total development time from larval hatching to adult eclosion (TDT) showed no difference between the sexes (BW1, *F* ~1,75~ = 0.073, *P* = 0.7879; TDT, *F* ~1,77~ = 2.327, *P* = 0.1312), the growth rate in terms of increasing body weight during the immature stages was larger in females than in males (*F* ~1,74~ = 7.875, *P* = 0.0064), leading to larger body sizes in females. Excluding the effect of sex, the geometric mean of all adult body dimensions (GM) was positively related to the initial larval body weight (proxy of egg size), developmental time from oviposition to adult emergence, and growth rate (Fig. [2](#Fig2){ref-type="fig"}; Table [1](#Tab1){ref-type="table"}). Of these three variables, growth rate had the largest effect, while the remaining two variables had comparable effects. Overall body length (BL; Fig. [1a](#Fig1){ref-type="fig"}) was closely correlated with GM (*r* = 0.962, *P* \< 0.0001) and showed almost identical responses to the four variables (results not shown). Because of the sexual dimorphism in body dimensions, we used sex-adjusted dimensions in the QTL analyses. Hereafter, symbols for individual body dimensions (HL, HW, TL, TW, EL and EW; Fig. [1a](#Fig1){ref-type="fig"}) and overall body size dimensions (GM and BL) are used for log~10~ transformed, sex-adjusted values. All of these variables were significantly correlated with each other except for two cases (Fig. [3](#Fig3){ref-type="fig"}).Figure 2Correlations between developmental trait values and body size dimensions. Open and closed circles indicate male and female individuals, respectively. BW1, BWA and BL are log~10~ transformed values, and GM (geometric mean of all body dimensions) was calculated as the arithmetic mean of all log~10~ transformed body dimensions except BL. Table 1Effects of sex, developmental time (DT), larval weight at hatching, and growth rate on the geometric mean of all adult body dimensions (GM; log10 transformed).VariableEffectSEdf*tP*Sex \[female-male\]0.4190.0631,70.116.67\<0.0001Initial larval weight (BW1)0.6130.0721,70.258.49\<0.0001Development time (TDT)0.1870.0691,70.112.730.0080Growth rate0.8240.0881,70.179.37\<0.0001All variables (other than sex) were standardised. Family was incorporated as a random variable. Figure 3Correlations between body size dimensions. The values in millimetres were log~10~ transformed and adjusted for sex differences. Open and closed circles indicate male and female individuals, respectively. Pearson correlation coefficients (*r*) and *P* values are given in the panels. The body dimensions are described in the legend of Fig. [1](#Fig1){ref-type="fig"}. Sequence assembly resulted in the construction of 76,540 scaffolds with an N50 of 734,069 bp (max. 6,537,065) and total length of 191,032,029 bp (N, 8%; GC, 35.0%). A total of 14,929 protein-coding genes were predicted; of these, 10,624 (71%) were assigned to known genes by a blast search. We constructed 23 autosomal linkage groups with a total length of 1,029 cM from 319 RAD loci and an X-chromosome linkage group of 91 cM from 56 RAD loci (a total of 1,120 cM with 375 markers; Fig. [4](#Fig4){ref-type="fig"}). Based on shared scaffolds, the number of autosomal linkage groups could be reduced to 17, which was still larger than the actual haploid autosome number of 13.Figure 4Linkage map of *Carabus japonicus* showing the locations (peak and range) of the body dimension quantitative trait loci (QTLs). Two-way arrows connect linkage groups probably located on the same chromosomes. Among the autosomal linkage groups, we found two QTLs of large effect for body length (BL) and geometric mean of all body dimensions (GM) on LG5 and LG6 (*R* ^2^ = 0.26--0.35; Figs [4](#Fig4){ref-type="fig"} and [5](#Fig5){ref-type="fig"}; Table [2](#Tab2){ref-type="table"}). Of the components of BL, elytral length (EL), which comprises 63.1% of BL on average, had two QTLs corresponding to those of BL on LG5 and LG6 and an additional QTL on LG6; these QTLs had large effects (*R* ^2^ = 0.36−0.44). In addition, a QTL for GM occurred on LG17 (*R* ^2^ = 0.43) and was related to peaks of other dimensions, which were not significant. We found one QTL for head width (HW) on LG2, one QTL for head length (HL) on LG22, and one QTL for thoracic width (TW) on LG5, but no QTL for thoracic length (TL). All QTLs, except that for HW, exhibited large dominance effects (Table [2](#Tab2){ref-type="table"}). The X chromosome contained a QTL for elytral width (EW), but it had a relatively low logarithm of odds (LOD) score (Table [2](#Tab2){ref-type="table"}; Fig. [5](#Fig5){ref-type="fig"}).Figure 5LOD scores for QTLs of body dimensions in the composite interval mapping analysis. Horizontal lines indicate threshold LOD values at α = 0.05. Arrows indicate peaks of LOD score. Table 2Significant QTLs for body size dimensions.TraitLGPosition (cM)LODEffect*R* ^2^PeakRangepeak value*ad*BLLG57.30.0--37.14.2830.01110.02160.2862BLLG60.00.0--10.04.2990.01030.01910.3514GMLG516.60.0--18.34.3900.01870.02120.2733GMLG541.341.3--51.14.2160.02160.01990.2620GMLG60.00.0--11.94.3770.01870.02080.3546GMLG17.23.20.0--23.24.466-0.02260.01760.4304HLLG220.00.0--3.03.3500.04370.02460.3355HWLG245.845.5--46.83.567-0.02910.00490.4790TWLG545.33.0--68.76.0560.02970.03450.3078ELLG56.00.0--25.15.2950.00980.02610.4397ELLG60.00.0--5.04.1190.01250.02560.4101ELLG633.519.8--41.14.2740.00790.03410.3606EWLGX55.655.5--56.62.655-0.0118---0.2373Trait: body size dimensions, which were log10-transformed and sex-adjusted.LG: linkage group.Effect: *a*, additive; *d*, dominance. *R* ^2^, proportion of variance explained. From 2,304 gene loci involved in 24 scaffolds with body size QTLs, we obtained 1,566 RefSeq protein IDs and determined 110 candidate genes by referring to the gene ontology terms (Table [3](#Tab3){ref-type="table"}; see also Table [S3](#MOESM1){ref-type="media"}, Supplementary Information, for details of all candidate genes found in the scaffolds with QTL regions). In the two linkage groups (LG5, LG6) with major QTLs, LG5 contained *Wnt*/*wingless* genes and several other genes related to the regulation of cell proliferation, cell size, and wing size; namely, the *vein* locus existed near the peak position of the QTL for BL/EL, and the *ribosomal protein L8*, *E2F transcription factor 1* and *topoisomerase 1* loci occurred around the peak position of the QTL for GM (Fig. [6a](#Fig6){ref-type="fig"}; Table [S3](#MOESM1){ref-type="media"}, Supplementary Information). The beetle elytra correspond to the forewings of flies; therefore, the *wingless* gene may be related to the regulation of elytral size. The peak position of the QTL for EL is located in the *terribly reduced optic lobes* (*trol*) gene region; however, whether this gene is related to the regulation of body size is unknown. The QTL region for HW/GM on LG5 (not shown in Fig. [6a](#Fig6){ref-type="fig"}) contained the *CG11940* (*pico*), *humpty dumpty*, *CG4207* (*bonsai*) and *Stat92E* loci, which are involved in regulation of cell proliferation/size and growth (Table [S3](#MOESM1){ref-type="media"}, Supplementary Information). For LG6 (Fig. [6b](#Fig6){ref-type="fig"}), the composition and order of scaffolds are unclear. Near the QTL for BL/EL/GM on LG6, we found the *menin 1* and *syntaxin 5* genes, which are involved in regulation of cell proliferation. The QTL region for EL in LG6 contains the *pudgy* loci; this gene negatively regulates the insulin receptor signalling pathway. Additionally, the *decapentaplegic* gene, a key morphogen affecting cell proliferation, exists near the EL QTL region. For the QTL for HW on LG2, we found the *dumpy* and *Keren* genes, which are involved in regulation of wing size and cell proliferation, respectively (Table [S3](#MOESM1){ref-type="media"}, Supplementary Information). For the QTL for HL on LG22, we found the *Smurf* gene, involved in cell size regulation, and the *frizzled 2* gene of the Wnt signalling pathway. Also, the *CG7467* (*osa*) gene, which affects the Wnt signalling pathway, was found in the QTL for EW (Table [S3](#MOESM1){ref-type="media"}, Supplementary Information).Table 3Scaffolds of *C*. *japonicus* with body size QTLs.Linkage groupScaffold no.QTL traitsLength, bpNo. gene lociNo. candidate genesLG2226HW\*5110100LG2350HW424748232512LG550BL, TW151137212210LG551BL\*, GM, TW, EL\*1040234684LG552BL, GM\*, TW, EL252459929217LG5119GM\*, TW653579660319LG51761GM, TW\*146260LG569886TW348051120LG6216BL\*, GM\*, EL\*167963151LG6588GM226143302LG619857EL413910LG669888EL15868380LG669783EL\*236463230LG669899EL599710510LG6224EL15994841587LG64437EL1158400LG1715GM341150270LG1754GM13098411013LG172833GM\*75834110LG173109GM4137450LG173143GM3420510LG2259HL\*323716426521LGX341EW799159765LGX644EW8749661049Total25779592304110Asterisks indicate that the LOD peaks of the traits exist on the corresponding scaffolds.For candidate gene numbers, the numbers accounting for multi-locus genes on the same scaffold are given. Figure 6Genes potentially related to body size QTLs on scaffolds of two linkage groups, LG5 (**a**) and LG6 (**b**). Red upward asterisks indicate restriction site-associated DNA (RAD) loci with significant QTL LOD scores. Genes shown here are those related to the regulation of cell proliferation, cell size/cell growth, and wing size, and growth; and those involved in or regulating the insulin receptor/insulin-like growth factor signalling pathway, the Wnt signalling pathway, and the epidermal growth factor/epidermal growth factor receptor signalling pathway. Open circles indicate functions known for the gene located at the position. See Table [S3](#MOESM1){ref-type="media"} (Supplementary Information) for precise gene names and gene ontology terms. Discussion {#Sec3} ========== The inferred genome size of *C*. *japonicus* is approximately 191 Mb, and the number of predicted genes is \~14,900. This moderate genome size and gene number may render this species suitable for the study of the genomic basis of adaptive traits. However, a disadvantage of this species is its low fecundity, which makes construction of a large F~2~ population difficult. In this study, the use of multiple parents was necessary to obtain F~1~ beetles and a total of 80 F~2~ beetles with a sex ratio of 1:1. The limited number of F~2~ individuals was a factor reducing the accuracy of the QTL analysis. Through the hybridisation of differently sized males and females from two populations, we obtained F~2~ individuals showing large variation in body size (Fig. [1b](#Fig1){ref-type="fig"}). Thus, the body size dimensions likely had a polygenic basis. In F~2~ individuals, larger body sizes were related positively to larger live body weight of first instars and larger growth rate in body weight from the first instar to the emerging adult stage (Fig. [2](#Fig2){ref-type="fig"}). Therefore, the larger initial size (egg size) and the larger immature growth rate contributed to the larger adult body size^[@CR17]^. We found that two regions on LG5 and LG6 regulated the overall body size (BL and GM), with relatively large effects. These regions contained QTLs for EL, which occupies a large proportion of BL, and a QTL for TW. There was another QTL for GM on LG17, where we found no QTL for body part dimensions, suggesting that this QTL regulates sizes of body parts in a different way from that of the QTLs on LG5 and LG6. If the body size of *C*. *japonicus* is actually controlled by a few gene loci with large phenotypic effects, body size divergence may occur readily in response to different habitat conditions, such as temperature and food resources^[@CR17]^. The pattern of body size QTL distribution is similar to that found in previous QTL studies for *D*. *melanogaster* and its allied species, in which a major QTL region for wing and other body sizes was located on chromosome 3 and an additional QTL region was located on chromosome 2^[@CR9],\ [@CR11],\ [@CR14]^. The *Drosophila* studies focused on latitudinal variations in body size, and adaptive body size divergence between beetles and flies may have a similar genetic basis. In *Drosophila* species, several genes with allelic differences are associated with latitudinal body size variations. However, the functions of those genes in controlling body size remain undetermined. In this *C*. *japonicus* study, determination of responsible genes was difficult due to the low marker density and limited linkage map. The QTL regions may involve some degree of sequence differences in several regulatory genes that affect cell proliferation, cell size, or growth rate. Future studies are required to detect the sequence differences in the QTL regions that correspond to body size differences by re-sequencing multiple samples from populations with different body sizes. Methods {#Sec4} ======= Experimental crossing and body size analysis {#Sec5} -------------------------------------------- We constructed an F~2~ adult population from four pairs of large females from Kabeshima Island, Saga (33°32′55″N, 129°52′56″E, 30 m a.s.l.), and small males from Mt. Aburayama, Fukuoka (33°30′43″N, 130°21′52″E, 580 m a.s.l.; Fig. [1a](#Fig1){ref-type="fig"}). By crossing F~1~ males and females, we obtained 40 adult males and 40 adult females in the F~2~ generation for linkage map construction and QTL analysis (Fig. [1b](#Fig1){ref-type="fig"}). The rearing methods were identical to those described by Okuzaki *et al*.^[@CR17]^. We measured the body weights of the first, second, and third instar and adult individuals just after hatching/eclosion. We also measured the duration of the various developmental stages to analyse the variations in growth and development patterns in F~2~ individuals. Using log~10~ transformed values for the initial first instar weight (BW1) and adult body weight at eclosion (BWA), and the total development time from hatching to adult eclosion (TDT; not log~10~ transformed), we calculated the larval growth rate as (BWA--BW1)/TDT. Approximately 1 month after adult emergence, F~2~ beetles were fixed in ethanol. We measured six body dimensions (Fig. [1a](#Fig1){ref-type="fig"}; Table [S1](#MOESM1){ref-type="media"}, Supplementary Information) using a digital calliper (Mitsutoyo, Japan): the BL, from the apex of the labrum to the apex of the elytra; the HW, represented by the distance between the outer margins of the eyes; the TW, the maximum width of the thorax; the TL, the length of the thorax; the EW, the maximum width of the elytra; and the EL, from the base to the tip of the right elytra. HL was determined as BL--TL--EL. All body size dimensions were log~10~ transformed. Although BL is the most straightforward measurement for overall body size and has been used in previous studies of *Carabus* beetles^[@CR4],\ [@CR15],\ [@CR16]^, we also used log~10~ transformed value of the geometric mean of all body dimensions except BL^[@CR18]--[@CR20]^. The logged geometric mean is effectively the arithmetic mean of log~10~ transformed dimensions and is denoted simply by GM in this paper. Consistent with sexual dimorphism in body size, the females were generally larger than males. Therefore, male body dimensions were adjusted so that their means were equal to the means of females. RAD sequences {#Sec6} ------------- Using the Genomic DNA Purification Kit (Promega) or DNeasy Kit (Qiagen), total DNA was extracted from the gonads and muscle tissues of the parents and F~2~ adults preserved in ethanol. The library for RAD sequencing was constructed as described by Etter *et al*.^[@CR21]^. The extracted DNA was adjusted to 25 ng/μl, and a 12-μl aliquot was digested with the endonuclease *Pst*I; the digested fragments were ligated with P1 adaptors containing a 5-bp barcode sequence. Subsequent library construction procedures and 100-bp single-end sequencing with Illumina HiSeq. 2000 were conducted by Hokkaido System Science Co. Ltd. (Sapporo, Japan). The RAD sequences obtained and their accession numbers in the DNA Data Bank of Japan are listed in Table [S2](#MOESM1){ref-type="media"} (Supplementary Information). Genome sequencing and gene prediction {#Sec7} ------------------------------------- Total DNA was extracted from the testes of two males from Kabeshima Island. A paired-end library with 170-bp insertion (j-1--170) and two mate-pair libraries with 2,000-bp and 5,000-bp insertions (j-1-2K, j-2-5K) were constructed and sequenced using one lane of a HiSeq. 4000 sequencer (Illumina). We obtained 297 M clean reads for j-1-170 (Q20, 98.6%), 117 M reads for j-1-2K (Q20, 98.3%) and 110 M reads for j-2-5K (Q20, 98.0%). The raw reads were deposited at the DNA Data Bank of Japan (DDBJ DRA; BioProject, PRJDB5403; DRR089090-089092). The sequences were assembled using Platanus version 1.2.1^[@CR22]^. We annotated the assembled scaffolds by combining evidence from multiple gene prediction procedures. Ab-initio gene prediction was conducted with Augustus version 3.1^[@CR23]^ and GeneMark-ES version 4.32^[@CR24]^. Augustus was run with parameters trained using *Tribolium castaneum*, which is the closest available genome. A self-training procedure implemented in GeneMark-ES was used to optimise the parameters for prediction. In addition to the ab-initio predictions, the coding sequences (CDS) of 15,607 genes predicted in the *Carabus uenoi* draft genome (T. Fujisawa *et al*., unpublished) were aligned to the scaffolds using Exonerate^[@CR25]^ (Slater & Birney 2005) with a percentage similarity of 80%. The evidence from ab-initio prediction and CDS alignment was combined by EVidenceModeler^[@CR26]^, with evidence weights of 5 for CDS alignment and 1 for ab-initio prediction. The translated sequences of the predicted gene set were searched against the RefSeq protein database and its *D*. *melanogaster* subset using BLASTp (BLAST + ver. 2.2.30)^[@CR27]^ to obtain orthological information. Linkage map construction {#Sec8} ------------------------ We used the genome sequence of *C*. *japonicus* to determine the RAD loci segregating between parents and genotyping of F~2~ individuals for the orthologous loci using the Stacks programme (ver. 1.35)^[@CR28]^. The karyotype of the genus *Carabus* is 2n = 28, with an XY sex chromosome system^[@CR29]^. After genotyping, we discriminated loci on the X chromosomes from those on autosomes in males, as the male X chromosome markers would have contained no heterozygote. We constructed autosomal linkage maps and an X chromosomal linkage map separately using the JoinMap^®^ 4.1 programme (Kyazma V.B., Wageningen, The Netherlands). For autosomal chromosomes, we used 1,463 loci and obtained 23 linkage groups with a total length of 1,029 cM from 319 loci at an LOD score of 6. For the X chromosome, we used 171 putative X-chromosome loci from males (haploid) and obtained a linkage map with 56 loci and 90.83 cM in length at an LOD score of 2. QTL mapping {#Sec9} ----------- We conducted composite interval mapping using Windows QTL Cartographer version 2.5_011^[@CR30]^ for autosomal loci with male and female data combined (F~2~ cross) and for the X chromosome with male data only because the linkage map was constructed exclusively from male data. We conducted composite interval mapping^[@CR31]^ with a walk speed of 1 cM (using a standard background control method with forward and backward regression, control marker number of 5, and probability of into/out of a 0.1 window size of 10 cM). The LOD threshold values at α = 0.05 were obtained by 1,000 permutations. Candidate genes {#Sec10} --------------- For the predicted genes located in the QTL regions, functional annotations were made using DAVID Bioinformatics Resources 6.8^[@CR32],\ [@CR33]^. The RefSeq protein ID for each locus based on *Drosophila melanogaster* genes was used for input, and three gene ontology terms (biological process, cell component, and molecular function), protein domains (InterPro), and pathway (KEGG) annotations were obtained. *D*. *melanogaster* was used as the background genome in the DAVID annotation. We used genes existing in scaffolds with the QTLs of body dimensions. We focused on genes related functionally to the regulation of cell proliferation, cell size, cell growth, wing size, and general growth. In addition, we considered genes involved in pathways related to the regulation of body size and growth, the Wnt signalling pathway, the insulin receptor signalling pathway, the insulin-like growth factor signalling pathway, and the epidermal growth factor/epidermal growth factor receptor signalling pathway^[@CR6],\ [@CR8]^. Electronic supplementary material ================================= {#Sec11} Supplementary Information **Electronic supplementary material** **Supplementary information** accompanies this paper at doi:10.1038/s41598-017-08362-7 **Publisher\'s note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supported by JSPS KAKENHI (nos 26251044, 25128707, 23128507). Conception of the study: T.S. and R.K. Experiment and data analysis: R.K., T.S and Y.O. Genome assembly and analyses: T.F. and T.S. Wrote the paper: T.S. Competing Interests {#FPar1} =================== The authors declare that they have no competing interests.

(J Am Heart Assoc. 2020;9:e015118 DOI: 10.1161/JAHA.119.015118.)31992159 Clinical PerspectiveWhat Is New?The present study is the first population‐based, untargeted metabolomic analysis of left ventricular diastolic function in a biracial (35% black, 65% white) community‐based sample of middle‐aged adults**.**We found that lipid‐ and amino acid--derived metabolites independently associate with patterns of mitral inflow and left ventricular relaxation in people without heart failure.What Are the Clinical Implications?This hypothesis‐generating study serves as an original groundwork to use metabolite biomarkers for the detection of subclinical heart failure in the general population. Introduction {#jah34809-sec-0008} ============ The prevalence of heart failure in the United States is projected to increase by over 40% to affect \>8 million individuals by 2030.[1](#jah34809-bib-0001){ref-type="ref"}, [2](#jah34809-bib-0002){ref-type="ref"} Together with global epidemiologic trends, this estimated rise in prevalence classifies heart failure as the most rapidly increasing form of cardiovascular disease (CVD) in the United States and globally.[3](#jah34809-bib-0003){ref-type="ref"} Among all heart failure cases, heart failure with preserved ejection fraction (HFpEF) is becoming the major disease subtype, presently responsible for at least 50% of all heart failure cases.[4](#jah34809-bib-0004){ref-type="ref"}, [5](#jah34809-bib-0005){ref-type="ref"} Such data underline a crucial need to characterize and study subclinical HFpEF phenotypes, which may not only allow for early HFpEF risk prediction but also lead to the development of low‐risk interventions to delay or prevent heart failure onset, improving quality of life in an aging population. Left ventricular diastolic dysfunction (LVDD), defined by impaired relaxation, decreased compliance, and/or increased left ventricular filling pressures, is one key subclinical precursor for HFpEF.[6](#jah34809-bib-0006){ref-type="ref"} This upstream phenotype represents a potential inflection point on the HFpEF causal disease pathway, as evidence suggests that LVDD is reversible.[7](#jah34809-bib-0007){ref-type="ref"}, [8](#jah34809-bib-0008){ref-type="ref"} Molecular phenotyping may represent a cost‐effective approach for identifying individuals with asymptomatic LVDD, while also implicating novel biological pathways in heart failure pathogenesis. Because metabolites reflect the collective output of both genetic and environmental factors[9](#jah34809-bib-0009){ref-type="ref"} and share a proximal relationship with clinical phenotypes, metabolomics profiling represents a particularly powerful molecular tool for such precision medicine initiatives. Unfortunately, few studies have applied this methodology to the study of LVDD.[10](#jah34809-bib-0010){ref-type="ref"}, [11](#jah34809-bib-0011){ref-type="ref"} Given the fact that both impaired systemic and myocardial metabolism have been implicated in HFpEF pathogenesis,[12](#jah34809-bib-0012){ref-type="ref"} there is a need to determine the role of systemic metabolites in left ventricular diastolic function (LVDF) to improve heart failure prevention. The current study used metabolomics profiling and echocardiography to conduct a metabolome‐wide association study of LVDF in a cross‐sectional analysis of BHS (Bogalusa Heart Study) participants. We hypothesized that serum metabolites, particularly those involved in biological pathways involving fatty acid metabolism, would be associated with diastolic function. Methods {#jah34809-sec-0009} ======= The authors declare that all supporting data are available within the article and its online supplementary files. Study Population {#jah34809-sec-0010} ---------------- The BHS, seated in Bogalusa, Louisiana, is a population‐based study examining the natural course of CVD across the life span. Between 1973 and 2016, 7 surveys of children aged 4 to 17 as well as 10 surveys of adults, who had been previously observed as children, were completed. The current BHS cohort includes 1298 participants, born between January 1959 and June 1979, who were examined twice in both childhood and adulthood for CVD risk factors and outcomes. Among these 1298 BHS participants, we selected the 1050 individuals who had echocardiographic, metabolomic, and respective covariable data measured contemporaneously at the most recent BHS study examination taking place from 2013 to 2016 (Figure [S1](#jah34809-sup-0001){ref-type="supplementary-material"}). The current study sample (n=1050) was highly similar to the overall BHS sample (n=1298) with respect to sociodemographic characteristics and traditional CVD risk factors (Table [S1](#jah34809-sup-0001){ref-type="supplementary-material"}). All study participants provided written informed consent at each examination, and study protocols were approved the Institutional Review Board of the Tulane University Health Sciences Center. Echocardiography Protocol {#jah34809-sec-0011} ------------------------- Two‐dimensional and tissue Doppler echocardiography were performed by trained cardiac sonographers at the BHS field office. Participants were placed in a partial left lateral decubitus position for echocardiographic assessment. Ten cycles of each 2‐dimensional and Doppler signal were recorded. Echocardiographic recordings were accomplished using an Aplio 300 ultrasound instrument (Toshiba America Medical Systems, Tustin, CA) with a linear array transducer of 7.5 mHz using a standard protocol.[13](#jah34809-bib-0013){ref-type="ref"}, [14](#jah34809-bib-0014){ref-type="ref"} Diastolic function parameters assessed included early left ventricular filling peak velocity (E), late left ventricular filling peak velocity (A), medial mitral annular velocity (e'), deceleration time of the E wave (DT), isovolumic relaxation time (IVRT), and left atrial maximum volume index.[15](#jah34809-bib-0015){ref-type="ref"} The apical 4‐chamber view was used to assess left atrial volume as well as patterns of mitral inflow including, DT, E wave, A wave, and medial e' wave velocities. The E and A wave velocities were assessed using pulsed‐wave Doppler echocardiography of transmitral flow at the mitral valve leaflet tips, while medial e' velocity was measured using pulsed‐wave tissue Doppler echocardiography of the mitral annulus. The apical 5‐chamber view was used to measure IVRT, specifically by placing the sample volume in the left ventricular outflow tract to concurrently evaluate aortic ejection and the onset of mitral inflow. Doppler sample volumes were placed between the mitral leaflet tips to measure DT. Left atrial volume was quantified by observing windows in which the left atrial length and transverse diameter were maximized. Left atrial volume was then divided by body surface area to estimate left atrial maximum volume index. Heart rate was assessed during the echocardiography protocol. Metabolomic Profiling {#jah34809-sec-0012} --------------------- Fasting serum samples were obtained between 2013 and 2016 from BHS participants and stored at −80°C prior to metabolite quantification. Metabolomic analysis of serum samples was consequently performed at Metabolon Inc (Durham, NC). Metabolites were quantified using ultra‐high‐performance liquid chromatography--tandem mass spectroscopy,[16](#jah34809-bib-0016){ref-type="ref"} providing a comprehensive analysis of serum metabolites. Metabolites were identified by automated comparison of ion characteristics to a reference library of chemical standard records that include retention time, molecular weight (m/z), preferred adducts, and in‐source fragments, as well as associated mass spectroscopy spectra. The majority of missing data in mass spectroscopy--based metabolomics experiments is due to detection limitations.[17](#jah34809-bib-0017){ref-type="ref"} Thus, if a metabolite value was missing, we assumed that it was below the detection limit and assigned it the lowest detectable value in the data set for the corresponding metabolite. Recent metabolomics studies have employed a similar technique.[18](#jah34809-bib-0018){ref-type="ref"}, [19](#jah34809-bib-0019){ref-type="ref"} Untargeted metabolomic analysis yielded 1466 metabolites, including 956 known biochemical compounds. These metabolites belonged to pathways related to amino acids (n=184), lipids (n=408), nucleotides (n=41), peptides (n=35), carbohydrates (n=25), cofactors and vitamins (n=34), xenobiotics (n=220), and energy (n=9). Furthermore, an additional 510 unnamed compounds presently lacking chemical standards were also identified. We excluded metabolites (n=213) detected in \<20% of participants or with limited intra‐assay reliability, defined by a reliability coefficient \<0.30 (n=51) in duplicate samples. After completion of quality control procedures, 1202 metabolites remained for analysis. Among these 1202 metabolites, 167 were missing or under the detection limit in 50% to 80% of the samples. These latter metabolites were categorized and analyzed according to 3 mutually exclusive groups: (1) missing or below the detection limit; (2) below the median of measured values; and (3) greater than or equal to the median.[18](#jah34809-bib-0018){ref-type="ref"}, [19](#jah34809-bib-0019){ref-type="ref"} For the remaining 1035 metabolites, the minimum observed value of each analyte was imputed for below‐the‐detection‐limit or missing values, with these metabolites examined as continuous variables. Covariable Assessment {#jah34809-sec-0013} --------------------- Rigorous protocols were employed to collect clinical data on BHS participants.[20](#jah34809-bib-0020){ref-type="ref"} Validated questionnaires were used to obtain demographic and lifestyle variables, specifically, age, race, sex, cigarette smoking, alcohol consumption, and antihypertensive medications. Fasting measures of low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, triglycerides, and hemoglobin A~1c~ were collected using standardized methods.[20](#jah34809-bib-0020){ref-type="ref"} Blood pressure was measured in triplicate, while height and weight were measured in duplicate at the time of the physical exam. Weight in kilograms was divided by height in meters squared to calculate body mass index. Serum creatinine was measured using the Kinetic Jaffe method. Serum creatinine was used to calculate estimated glomerular filtration rate via the Chronic Kidney Disease Epidemiology Collaboration equation.[21](#jah34809-bib-0021){ref-type="ref"} Statistical Analysis {#jah34809-sec-0014} -------------------- BHS study sample characteristics were presented as means and SDs for continuous variables, and categorical variables were expressed as percentages. The Student *t* test and chi‐square test were used to examine differences in normally distributed continuous variables and categorical variables, respectively, between blacks and whites. Normality of distribution was assessed via the Kolmogorov--Smirnov test. Both the E/A and E/e' ratios underwent natural logarithmic transformation to normalize the distribution of these outcome variables before the analysis. Before association analyses, we standardized the concentrations of metabolites by dividing each metabolite by its respective SD. Each metabolite was individually tested for association with the diastolic function parameters using multivariable‐adjusted linear regression models. In overall and race‐stratified multivariable models, 16 covariables were adjusted for, including age, sex, race (when appropriate), education, cigarette smoking, alcohol drinking, average systolic blood pressure, body mass index, low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, serum triglycerides, hemoglobin A~1c~, estimated glomerular filtration rate, heart rate, and antihypertensive medications. The analyses were corrected for multiple testing via the Bonferroni approach, with an alpha threshold of 4.16×10^−5^ (0.05/1202) used for determining statistical significance. To further reduce false‐positive associations, only findings significant in the overall analysis and in at least 1 race group, while displaying consistent effect direction and nominal significance (*P*\<0.05) in the other race group, were reported in the current analysis. We conducted Pearson pairwise correlation tests to assess the relationships between metabolites that achieved significance after Bonferroni correction (*P*\<4.16×10^−5^). Weighted correlation network analysis (WGCNA)[22](#jah34809-bib-0022){ref-type="ref"} was employed to discover networks of highly correlated serum metabolites among BHS participants. WCGNA is an unsupervised data reduction technique, but unlike principal components analysis, it allows correlations between components, and thus may more effectively capture biological pathways underlying identified metabolites.[22](#jah34809-bib-0022){ref-type="ref"}, [23](#jah34809-bib-0023){ref-type="ref"} Previous metabolomic studies have used this data reduction technique.[22](#jah34809-bib-0022){ref-type="ref"}, [24](#jah34809-bib-0024){ref-type="ref"} In short, an adjacency matrix was generated to represent the metabolite network based on weighted pairwise‐correlations of all metabolites.[25](#jah34809-bib-0025){ref-type="ref"} Densely interconnected metabolites, also referred to as modules, were identified using an unsupervised hierarchical clustering approach.[26](#jah34809-bib-0026){ref-type="ref"} An eigenmetabolite was created for each metabolite module, representing the module\'s first principal component and can be interpreted as a weighted average value of correlated metabolites. Because there was similar metabolite clustering across ethnic groups, metabolite modules were generated among all study participants using data derived from the 1202 metabolites that passed quality control. To identify biological pathways underlying each module, metabolites most highly correlated with each module\'s eigenmetabolite were identified (*r*\>0.70). We then used the subpathways associated with such metabolites to describe each module. For weighted correlation network analysis analyses, adjusted LVDF measures were created using the residual values generated by regressing each raw LVDF parameter on age, sex, race (when appropriate), education, cigarette smoking, alcohol drinking, average systolic blood pressure, body mass index, low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, serum triglycerides, hemoglobin A~1c~, estimated glomerular filtration rate, heart rate, and antihypertensive medications. We then examined the Pearson correlations among each of the 9 metabolite modules and the residual LVDF values, which were independent of the covariable effects. We used a Bonferroni‐corrected alpha threshold of 5.56×10^−3^ (0.05/9) to identify significant correlations among the 9 eigenmetabolite modules with the residual of each LVDF phenotype. Analogous to the single‐metabolite analysis, modules were considered significant upon achieving significance in the overall population and in at least 1 race group, with nominal significance (*P*\<0.05) and consistent effect direction in the other race group. Statistical analyses were performed in SAS (version 9.4; SAS Institute, Cary, NC) and R (version 3.4.3) software. Sensitivity Analysis {#jah34809-sec-0015} -------------------- Four post hoc sensitivity analyses were performed, one of which assessed the relationship of metabolites with diastolic function additionally adjusting for left ventricular mass, a variable hypothesized to be on the causal pathway of diastolic dysfunction. A second sensitivity analysis similarly evaluated the relationship of metabolites with diastolic function, after removing BHS participants (n=12) who self‐reported previous myocardial infarction and/or a diagnosis of congestive heart failure. We also performed a sensitivity analysis additionally adjusting for gait speed, one surrogate marker of skeletal muscle breakdown, to help elucidate if significant endogenous metabolites related to myocyte turnover were of skeletal or cardiac origin. Finally, we conducted a sensitivity analysis adjusting for lifetime mean systolic blood pressure, as opposed to the systolic blood pressure measurement at the most recent examination, in a subsample of study participants. Study participants (n=804) who had at least 4 measures of systolic blood pressure from previous BHS exams were included, and these 4 measures were used to calculate a mean lifetime systolic blood pressure, which was included as a covariate in linear regression models. Results {#jah34809-sec-0016} ======= Table [1](#jah34809-tbl-0001){ref-type="table"} presents the characteristics of the 1050 participants, including information on sociodemographic, lifestyle, medication, metabolic, and hemodynamic factors. The average age of the study cohort was 48 years, composed of 58% women and over one third black individuals. The study sample, on average, was obese (body mass index=31.38±7.80 kg/m^2^) and prediabetic (hemoglobin A~1c~=5.87±1.19), with evidence of dyslipidemia (low‐density lipoprotein cholesterol=115.42±35.89 mg/dL), elevated systolic blood pressure (systolic blood pressure=123.32±16.69 mm Hg), and preserved kidney function (estimated glomerular filtration rate=93.34±17.18 mL/min per 1.73 m^2^). There were significant differences between blacks and whites with respect to all covariables except for sex and resting heart rate. ###### Characteristics of 1050 Bogalusa Metabolome Study Participants Variable All (n=1050) Whites (n=684) Blacks (n=366) *P* Value[a](#jah34809-note-0003){ref-type="fn"} -------------------------------------------------------- ---------------- ---------------- ---------------- -------------------------------------------------- Sociodemographic Female, % 57.81 56.14 60.93 0.1668 Post--high school education, % 49.90 56.58 37.43 \<0.0001 Age, y, mean (SD) 48.13 (5.32) 48.46 (5.09) 47.52 (5.70) 0.0084 Lifestyle Smoking, n (%) 0.0352 Never 529 (50.38) 353 (51.61) 176 (48.09) Former 301 (28.67) 204 (29.82) 97 (26.50) Current 220 (20.95) 127 (18.57) 93 (25.41) Drinking, n (%) 0.0138 Never 116 (11.05) 60 (8.77) 56 (15.30) Former 332 (31.62) 219 (32.02) 113 (30.87) Current 602 (57.33) 405 (59.21) 197 (53.83) Metabolic BMI, kg/m^2^, mean (SD) 31.38 (7.80) 30.4 (6.94) 33.34 (8.86) \<0.0001 LDL cholesterol, mg/dL, mean (SD) 115.42 (35.89) 117.20 (33.93) 112.08 (39.12) 0.0350 HDL cholesterol, mg/dL, mean (SD) 51.88 (16.30) 50.97 (16.36) 53.57 (16.06) 0.0133 Hemoglobin A~1c~, %, mean (SD) 5.87 (1.19) 5.72 (0.91) 6.15 (1.54) \<0.0001 Serum triglycerides, mg/dL, mean (SD) 126.76 (68.15) 134.83 (71.48) 111.69 (58.61) \<0.0001 Hemodynamic Systolic BP, mm Hg, mean (SD) 123.32 (16.69) 121.13 (14.78) 127.42 (19.14) \<0.0001 Antihypertensive medication, n (%) 359 (34.19) 190 (27.78) 169 (46.17) \<0.001 eGFR, mL/min per 1.73 m^2^, mean (SD) 93.34 (17.18) 90.71 (15.02) 98.26 (19.71) \<0.0001 Cardiac Resting heart rate, beats/min, mean (SD) 71.67 (11.42) 71.39 (11.09) 72.19 (12.01) 0.2782 E/A ratio (SD)[b](#jah34809-note-0004){ref-type="fn"} 1.12 (1.34) 1.14 (1.32) 1.08 (1.35) 0.0227 E/e' ratio (SD)[b](#jah34809-note-0004){ref-type="fn"} 7.54 (1.32) 7.24 (1.30) 8.00 (1.36) \<0.0001 Isovolumic relaxation time, ms (SD) 93.49 (22.36) 89.86 (21.26) 100.26 (22.79) \<0.0001 Deceleration time, ms (SD) 186.61 (45.15) 188.81 (44.28) 184.76 (42.72) 0.0418 Left atrial maximum volume index (SD) 28.85 (8.25) 28.10 (7.90) 30.26 (8.70) \<0.0001 BMI indicates body mass index; BP, blood pressure; E/A, peak early filling velocity to peak late filling velocity; E/e', peak early filling velocity to mitral annular velocity eGFR, estimated glomerular filtration rate; HDL, high‐density lipoprotein; LDL, low‐density lipoprotein. *P* value for comparison between white vs. black participants. Presented as geometric mean and SD. LVDF of the study sample is also shown in Table [1](#jah34809-tbl-0001){ref-type="table"}. Compared with blacks, whites had statistically significant preserved diastolic function for patterns of mitral inflow, E/A (*P*=0.0227) and E/e' ratio (*P*\<0.0001), left atrial maximum volume index (*P*\<0.0001), and left ventricular relaxation, via IVRT (*P*\<0.0001). On the other hand, whites had significantly higher values for DT (*P*=0.0418), compared with blacks. There were 8 serum metabolites, 7 amino acid derived and 1 lipid derived, that were associated with diastolic function (Figure [1](#jah34809-fig-0001){ref-type="fig"}; Table [2](#jah34809-tbl-0002){ref-type="table"}). The 7 amino acid--derived metabolites were associated with the log‐transformed E/e' ratio phenotype and represented several different biological pathways, including histidine, lysine, methionine, polyamine metabolism, and the urea cycle. Among E/e' ratio--related metabolites, formiminoglutamate (B=0.07; *P*=5.60×10^−7^) had the largest effect size, followed by N‐formylmethionine (B=0.05; *P*=1.50×10^−7^); 1‐methylhistidine (B=0.05; *P*=1.60×10^−7^); N2, N5‐diacetylornithine (B=0.05; *P*=1.30×10^−7^); methionine sulfoxide (B=0.04; *P*=3.80×10^−6^); 5‐methylthioadenosine (B=0.04; *P*=1.40×10^−7^); and N‐trimethyl 5‐amino‐valerate (B=0.04; *P*=5.10×10^−6^). Butyrylcarnitine, a lipid‐derived metabolite, was significantly associated with IVRT (B=3.18; *P*=2.10×10^−6^). Of note and with the exception of methionine sulfoxide, all significant associations of metabolites with the log‐transformed E/e' ratio were driven by significance in blacks, with nominal significance in whites (Table [S2](#jah34809-sup-0001){ref-type="supplementary-material"}). The reverse of this latter pattern was observed for the diastolic function parameter of IVRT. {#jah34809-fig-0001} ###### Novel Metabolites Significantly Associated With Echocardiographic Measures of Diastolic Function Metabolite Sub‐Pathway Ln E/e' Ratio Ln E/A Ratio IVRT (ms) DT (ms) LAVI (mL/m^2^) ------------------------------ --------------------------------------------------- --------------- ------------------------------------------------- -------------- ------------------------------------------------- ---------------- ------------------------------------------------- -------------- ---------- -------------- ------------------------------------------------- Amino acid 1‐methylhistidine Histidine metabolism 0.05 (0.01) 1.60E‐07[a](#jah34809-note-0006){ref-type="fn"} 0.03 (0.01) 1.10E‐03[b](#jah34809-note-0007){ref-type="fn"} 1.41 (0.75) 6.10E‐02 1.35 (1.63) 4.10E‐01 0.40 (0.32) 2.20E‐01 Formiminoglutamate Histidine metabolism 0.07 (0.01) 5.60E‐07[a](#jah34809-note-0006){ref-type="fn"} 0.03 (0.01) 4.50E‐02[b](#jah34809-note-0007){ref-type="fn"} 2.68 (1.13) 1.90E‐02[b](#jah34809-note-0007){ref-type="fn"} 0.94 (2.46) 7.00E‐01 −0.09 (0.47) 8.40E‐01 5‐methylthioadenosine Poylamine metabolism 0.04 (0.01) 1.40E‐05[a](#jah34809-note-0006){ref-type="fn"} 0.03 (0.01) 1.10E‐03[b](#jah34809-note-0007){ref-type="fn"} 0.10 (0.74) 8.90E‐01 2.18 (1.60) 1.70E‐01 0.26 (0.28) 3.60E‐01 N‐trimethyl 5‐amino‐valerate Lysine metabolism 0.04 (0.01) 5.10E‐06[a](#jah34809-note-0006){ref-type="fn"} 0.02 (0.01) 2.00E‐02[b](#jah34809-note-0007){ref-type="fn"} 1.91 (0.75) 1.10E‐02[b](#jah34809-note-0007){ref-type="fn"} 0.15 (1.62) 9.30E‐01 0.19 (0.28) 5.00E‐01 N2, N5‐diacetylornithine Urea cycle; arginine and proline metabolism 0.05 (0.01) 1.30E‐07[a](#jah34809-note-0006){ref-type="fn"} 0.03 (0.01) 8.50E‐04[b](#jah34809-note-0007){ref-type="fn"} 1.62 (0.70) 2.10E‐02[b](#jah34809-note-0007){ref-type="fn"} 1.67 (1.51) 2.70E‐01 0.16 (0.31) 6.00E‐01 N‐formylmethionine Methionine, cysteine, SAM, and taurine metabolism 0.05 (0.01) 1.50E‐07[a](#jah34809-note-0006){ref-type="fn"} 0.04 (0.01) 1.30E‐05[a](#jah34809-note-0006){ref-type="fn"} 1.00 (0.78) 2.00E‐01 2.10 (1.69) 2.10E‐01 0.58 (0.29) 4.70E‐02[b](#jah34809-note-0007){ref-type="fn"} Methionine sulfoxide Methionine, cysteine, SAM, and taurine metabolism 0.04 (0.01) 3.80E‐06[a](#jah34809-note-0006){ref-type="fn"} 0.03 (0.01) 1.30E‐03[b](#jah34809-note-0007){ref-type="fn"} 0.55 (0.67) 4.10E‐01 0.90 (1.44) 5.30E‐01 0.55 (0.24) 2.50E‐02[b](#jah34809-note-0007){ref-type="fn"} Lipid Butyrylcarnitine Fatty acid metabolism, BCAA metabolism 0.03 (0.01) 1.40E‐04[b](#jah34809-note-0007){ref-type="fn"} −0.01 (0.01) 2.60E‐01 3.18 (0.67) 2.10E‐06[a](#jah34809-note-0006){ref-type="fn"} −1.69 (1.46) 2.50E‐01 −0.33 (0.25) 1.80E‐01 Adjusted for age, race, sex, education, smoking, alcohol drinking, antihypertensive medications, heart rate, estimated glomerular filtration rate, body mass index, systolic blood pressure, low‐density lipoprotein cholesterol, high‐density lipoprotein cholesterol, serum triglycerides, and hemoglobin A~1c~. E indicates peak velocity blood flow in early diastole due to gravity; e', medial mitral annular velocity; IVRT, isovolumic relaxation time; Ln, natural logarithm; SAM, s‐adenosylmethionine. Statistically significant *P* value (*P*\<4.16E‐05). Nominally significant *P* value (*P*\<0.05). Associations of each metabolite with the diastolic function phenotypes, according to super pathway, are displayed in Figures [S2](#jah34809-sup-0001){ref-type="supplementary-material"} through [S6](#jah34809-sup-0001){ref-type="supplementary-material"}. Such plots are depicted for the overall sample, blacks, and whites for each respective diastolic function outcome. These plots also provide the names of metabolites that were significant for diastolic function parameters in blacks or whites, but not in the overall sample, and thus shed light on potential race‐specific associations of metabolites with LVDF. Several metabolites that met stringent Bonferroni significance‐level criteria for their association with E/e' ratio had nominally significant associations with other diastolic function phenotypes (Table [2](#jah34809-tbl-0002){ref-type="table"}). N‐formylmethionine shared the most robust associations with LVDF, as this amino acid--derived metabolite was significantly associated with the log‐transformed E/A ratio (B=0.04; *P*=1.30×10^−5^), and had a nominally significant association with left atrial maximum volume index (B=0.58; *P*=4.70×10^−2^). Methionine sulfoxide also had nominally significant associations with the log‐transformed E/A ratio (B=0.03; *P*=1.30×10^−3^) and left atrial maximum volume index (B=0.55; 2.50×10^−2^), although with lower parameter estimates when compared to N‐formylmethionine. Finally, 3 metabolites, formiminoglutamate, N‐trimethyl 5‐amino‐valerate, and N‐trimethyl 5‐amino‐valerate, demonstrated nominally significant associations with both IVRT and the log‐transformed E/A ratio. Significant metabolites demonstrated modest to moderate intercorrelation (Figure [2](#jah34809-fig-0002){ref-type="fig"}). The highest pairwise correlation coefficients were observed between N‐formylmethionine and 1‐methylhistidine (*r*=0.64) and between N‐formylmethionine and 5‐methylthioadenosine (*r*=0.53). With respect to biological pathways, the 9 metabolite modules identified among study participants are presented in Tables [S3](#jah34809-sup-0001){ref-type="supplementary-material"} through [S7](#jah34809-sup-0001){ref-type="supplementary-material"}. A module containing metabolites involved in primary and secondary bile acid metabolism demonstrated a significant positive correlation with E/e' in the overall population (*r*=0.16; *P*\<0.0001) and in blacks (*r*=0.27; *P*\<0.0001), with a consistent effect direction in whites (*r*=0.06; *P*=0.09) (Table [S3](#jah34809-sup-0001){ref-type="supplementary-material"}). Figure [S7](#jah34809-sup-0001){ref-type="supplementary-material"} presents the Pearson correlations of the 8 metabolites that were most highly correlated with the primary and secondary bile acid metabolism eigenmetabolite value (*r*\>0.70). {#jah34809-fig-0002} Results from sensitivity analyses are presented in Table [S8](#jah34809-sup-0001){ref-type="supplementary-material"}, excluding individuals with self‐report myocardial infarction or congestive heart failure; Table [S9](#jah34809-sup-0001){ref-type="supplementary-material"}, adding left ventricular mass to the multivariable regression model; Table [S10](#jah34809-sup-0001){ref-type="supplementary-material"}, adding gait speed to the multivariable regression model; and Table [S11](#jah34809-sup-0001){ref-type="supplementary-material"}, adjusting for lifetime mean systolic blood pressure. Beta estimates and *P* values in all 3 sensitivity analyses were consistent with primary study results depicted in Table [2](#jah34809-tbl-0002){ref-type="table"}. N‐formylmethionine maintained its effect size with the log‐transformed E/e' phenotype when excluding individuals with self‐report myocardial infarction or congestive heart failure (B=0.05; *P*=2.60×10^−7^), adjusting for mean lifetime systolic blood pressure (B=0.04; *P*=4.10×10^−4^), and when adding left ventricular mass (B=0.05 *P*=4.80×10^−6^) or gait speed (B=0.04; *P*=3.90×10^−5^) to the multivariable regression model. For each standardized unit increase in butyrylcarnitine, we observed a significant 3‐ms higher IVRT across all sensitivity analyses, except for the sensitivity analysis that controlled for mean lifetime systolic blood pressure. Discussion {#jah34809-sec-0017} ========== While the increasing availability of noninvasive imaging methods has allowed for the assessment of LVDF in broader populations, the biochemical profile associated with this upstream heart failure (HF) phenotype remains largely unexplored. To address this critical gap in CVD prevention knowledge, we conducted one of the first epidemiologic studies that assesses the role of serum metabolites in LVDF, via an untargeted approach. Formiminoglutamate; 1‐methylhistidine; N2, N5‐diacetylornithine; N‐trimethyl 5‐aminovalerate; N‐formylmethionine; 5‐methylthioadenosine; and methionine sulfoxide were positively associated with the E/e' ratio, while butyrylcarnitine had a significant positive association with IVRT. In addition to single‐metabolite analyses, a network of correlated metabolites involved in primary and secondary bile metabolism was associated with the E/e' ratio. These results build on previous smaller, targeted metabolomic studies of diastolic function[10](#jah34809-bib-0010){ref-type="ref"}, [11](#jah34809-bib-0011){ref-type="ref"} and suggest that a higher concentration of both lipid and amino acid--derived metabolites, particularly those associated with mitochondrial impairment, may serve as early biomarkers for preclinical LVDD. Given the diagnostic and therapeutic challenges associated with HFpEF,[27](#jah34809-bib-0027){ref-type="ref"}, [28](#jah34809-bib-0028){ref-type="ref"}, [29](#jah34809-bib-0029){ref-type="ref"} findings from the current study may serve as a platform to develop more precise biological assays and risk stratification protocols for HFpEF. Findings from our analysis suggest that amino acid--derived metabolites hold a close relationship with the E/e' ratio, one indicator of left ventricular filling.[30](#jah34809-bib-0030){ref-type="ref"} The E/e' ratio has emerged as one promising echocardiographic parameter to help facilitate the diagnosis of HFpEF,[31](#jah34809-bib-0031){ref-type="ref"} as evident by its presence in the H~2~FpEF (F=filling pressure, Doppler echocardiographic E/e' \>9 equates to 1 point) score clinically recommended to differentiate cardiac versus noncardiac causes of heart failure.[32](#jah34809-bib-0032){ref-type="ref"} While IVRT has not been incorporated into diagnostic scoring criteria for HFpEF, increases in this cardiac time interval have been independently associated with risk of future ischemic heart disease, peripheral artery disease, and stroke.[33](#jah34809-bib-0033){ref-type="ref"} A similar relationship has also been described for increases in E/e' and the risk of incident primary cardiac events.[34](#jah34809-bib-0034){ref-type="ref"} Below, we discuss the relevance of each of the 8 robustly identified metabolites in LVDD and outline potential pathophysiological mechanisms underlying such associations by referencing previous work in this research space. Butyrylcarnitine and N‐Formylmethionine: Mitochondrial Impairment {#jah34809-sec-0018} ----------------------------------------------------------------- We speculate that mitochondrial uncoupling is driving metabolic alterations in subclinical HF, such that asymptomatic LVDD is associated with mitochondrial impairment and leads to elevations in serum butyrylcarnitine and N‐formylmethionine. In the present study, butyrylcarnitine and N‐formylmethionine were positively associated with IVRT and E/e', respectively. Butyrylcarnitine is a member of the acylcarnitine family, a group of metabolites functioning in fatty acid metabolism, particularly in mitochondrial fatty acid beta oxidation.[35](#jah34809-bib-0035){ref-type="ref"} Likewise, N‐formylmethionine is an initiator of protein synthesis in the mitochondria of all eukaryotic cells.[36](#jah34809-bib-0036){ref-type="ref"} Unlike other tissues and organs, fatty acids produce between 70% and 90% of ATP needed for heart function and metabolism.[37](#jah34809-bib-0037){ref-type="ref"}, [38](#jah34809-bib-0038){ref-type="ref"} Heart failure pathophysiology attributable to mitochondrial dysfunction is associated with both impaired fatty acid and glucose oxidation.[37](#jah34809-bib-0037){ref-type="ref"}, [38](#jah34809-bib-0038){ref-type="ref"}, [39](#jah34809-bib-0039){ref-type="ref"} Our results support previous work in this research space, which has demonstrated that individuals with HF have a higher venous concentration of mitochondrial‐dependent metabolites, succinate, 3‐hydroxybutyrate, and acetone, compared with controls.[40](#jah34809-bib-0040){ref-type="ref"} Notably, succinate, 3‐hydroxybutyrate, and acetone share a positive association of myocardial energy metabolism, further underlining the central role of altered mitochondrial metabolism in HF.[40](#jah34809-bib-0040){ref-type="ref"} Mitochondrial impairment may also trigger apoptosis, causing cardiac myocyte N‐formylmethionine to leak into circulation. These hypotheses are in part supported by research demonstrating that individuals with HFpEF have higher systemic concentrations of acylcarnitines compared with age‐matched non--heart failure controls.[41](#jah34809-bib-0041){ref-type="ref"}, [42](#jah34809-bib-0042){ref-type="ref"} Similarly, a 4‐component metabolite panel, including butyrylcarnitine, was found to better predict heart failure prognosis in individuals with stage B and stage C, when compared with brain natriuretic peptide.[43](#jah34809-bib-0043){ref-type="ref"} Systemic N‐formylmethionine release may occur after myocardial infarction,[44](#jah34809-bib-0044){ref-type="ref"} suggesting that ischemia may be one contributing factor in cardiomyocyte mitochondrial impairment in the setting of LVDD. These latter results, together with the findings of the present study, highlight the potential diagnostic value of serum butyrylcarnitine and N‐formylmethionine in LVDD and HFpEF. 1‐Methylhistidine and N‐Trimethyl 5‐Aminovalerate: Diet {#jah34809-sec-0019} ------------------------------------------------------- Exogenous environmental factors, specifically diet, have significant effects on the human metabolome.[45](#jah34809-bib-0045){ref-type="ref"} Our findings demonstrated that 2 diet‐associated metabolites, 1‐methylhistidine and N‐trimethyl 5‐aminovalerate, were positively correlated with the E/e' ratio. Derived from dietary anserine sources, 1‐methylhistidine has been reported as a marker of meat consumption, specifically red meat and chicken. Several previous studies have found robust linear relationships between meat intake and urinary 1‐methylhistidine.[46](#jah34809-bib-0046){ref-type="ref"}, [47](#jah34809-bib-0047){ref-type="ref"} On the other hand, external evidence suggests that N‐trimethyl 5‐aminovalerate may be a metabolite indicator of dairy milk intake.[48](#jah34809-bib-0048){ref-type="ref"}, [49](#jah34809-bib-0049){ref-type="ref"} Importantly, restriction of both dairy and red meat consumption, via the Dietary Approaches to Stop Hypertension and Mediterranean diets, are associated with a reduced risk of developing heart failure.[50](#jah34809-bib-0050){ref-type="ref"}, [51](#jah34809-bib-0051){ref-type="ref"}, [52](#jah34809-bib-0052){ref-type="ref"} While systolic blood pressure reduction may be the primary mechanism of the protective HF effects of these 2 diets, it is also important to consider other CVD risk factors in the setting of our metabolite findings. For example, N‐trimethyl 5‐aminovalerate positively associates with abdominal adiposity and intramuscular adiposity,[53](#jah34809-bib-0053){ref-type="ref"} risk factors for insulin resistance, type 2 diabetes mellitus, and coronary microvascular dysfunction. Notably, type 2 diabetes mellitus is a common comorbid disease that is present in up to one half of all patients with HFpEF.[54](#jah34809-bib-0054){ref-type="ref"} Thus, our work serves as preliminary biological evidence that may help explain the potential benefits of animal protein restriction in atherosclerosis and HF risk reduction. However, future longitudinal studies that combine nutritional and metabolome data with diastolic function are undoubtedly required to understand and test such hypotheses and biological phenomena. N2, N5‐Diacetylornithine and 5‐Methylthioadenosine: Endothelial Dysfunction {#jah34809-sec-0020} --------------------------------------------------------------------------- N2, N5‐diacetylornithine is a metabolite by‐product of the urea cycle, an important biological pathway to consider in the setting of HFpEF and diastolic function. Compared with individuals with HF with reduced ejection fraction, those with HFpEF have a higher chronic kidney disease--associated mortality rate.[54](#jah34809-bib-0054){ref-type="ref"} Urea‐to‐creatinine ratio, and urinary albumin‐to‐creatinine ratio, biochemical indices of renal function, associate with parameters of LVDD.[55](#jah34809-bib-0055){ref-type="ref"} Although BHS participants in the current study had preserved kidney function, it is possible that N2, N5‐diacetylornithine levels begin to accumulate due to urea cycle perturbations in subclinical renal disease, which may be accompanied by systemic microvascular and endothelial dysfunction.[56](#jah34809-bib-0056){ref-type="ref"} Findings from our study lead to the postulation that subclinical LVDD attributable to underlying endothelial dysfunction is associated with impaired nitric oxide synthesis and may lead to an elevation of systemic polyamines particularly, 5‐methylthioadenosine. Present in all mammalian tissues, 5‐methylthioadenosine is a polyamine and sulfur‐based nucleoside.[57](#jah34809-bib-0057){ref-type="ref"} The 5‐methylthioadenosine/methionine and polyamine pathways have previously been implicated in heart failure among a sample of 515 individuals.[43](#jah34809-bib-0043){ref-type="ref"} Additionally, animal models of ventricular hypertrophy and aortic stenosis demonstrate increases in systemic polyamines after tissue remodeling.[58](#jah34809-bib-0058){ref-type="ref"}, [59](#jah34809-bib-0059){ref-type="ref"} Nitric oxide, a vasodilatory and protective metabolite in CVD, inhibits polyamine production.[60](#jah34809-bib-0060){ref-type="ref"} Methionine Sulfoxide: Oxidative Stress {#jah34809-sec-0021} -------------------------------------- Methionine sulfoxide reductase is a powerful antioxidant that reduces methionine sulfoxide[61](#jah34809-bib-0061){ref-type="ref"} and has been implicated in vascular disease and cardiac ischemia.[62](#jah34809-bib-0062){ref-type="ref"} We observed a significant positive association between methionine sulfoxide and left ventricular filling. Methionine oxide results from the oxidation of methionine and is produced from activated neutrophils. Present in both the cytoplasm and mitochondria, methionine sulfoxide may be released from cardiac myocyte turnover in the presence of LVDD or systemic endothelial dysfunction. Formiminoglutamate: Impaired Liver Metabolism {#jah34809-sec-0022} --------------------------------------------- Formiminoglutamate is an intermediate of histidine metabolism, and high levels of this metabolite have been traditionally used to identify vitamin B~12~ and/or folate deficiency, as well as underlying liver disease.[63](#jah34809-bib-0063){ref-type="ref"}, [64](#jah34809-bib-0064){ref-type="ref"} Such nutritional deficiencies have been hypothesized to contribute to anemia observed in patients with congestive HF, yet scarce data exist. Megaloblastic anemia caused by vitamin B~12~ and/or folate deficiency are relatively rare in heart failure[65](#jah34809-bib-0065){ref-type="ref"}, [66](#jah34809-bib-0066){ref-type="ref"}; therefore, the positive relationship identified between formiminoglutamate and E/e' may point more toward co‐occurrence of subclinical liver disease, as opposed to anemia, in diastolic dysfunction and HF. Folate and vitamin B~12~ deficiency do not currently predict prognosis in heart failure[66](#jah34809-bib-0066){ref-type="ref"}; therefore, our results should be considered only as hypothesis‐generating observations and exploratory in nature. Furthermore, though our analyses adjusted for alcohol intake in BHS participants, regular alcohol consumption associates with inferior diastolic function[67](#jah34809-bib-0067){ref-type="ref"} and leads to higher urinary excretion of formiminoglutamate.[64](#jah34809-bib-0064){ref-type="ref"} In addition to single‐metabolite analyses, aggregation approaches also provided novel insights. Among metabolites with a significant relationship with LVDF, N‐formylmethionine and 1‐methylhistidine, as well as N‐formylmethionine and 5‐methylthioadenosine demonstrated moderate pairwise correlation. These results suggest that there may be an interaction between exogenous and endogenous pathways that influences the role of metabolites in diastolic dysfunction. Likewise, a correlated network of metabolites involved in primary and secondary bile acid metabolism significantly was associated with E/e' in the overall population and in blacks and had a consistent effect direction in whites. In addition to implicating the role of bile acid metabolites, such as glycohyocholate and taurocholate, with E/e', our results may further underline the importance of E/e' and left ventricular filling in the natural course of diastolic dysfunction and HFpEF. Our findings are consistent with and build on current knowledge regarding the role of bile acids in HF. The ratio of secondary to primary bile acids is increased in individuals with HF and this pattern has been associated with a reduced HF survival time.[68](#jah34809-bib-0068){ref-type="ref"} Likewise, an increased level of serum bile acids, together with diastolic dysfunction has been well described in cirrhotic cardiomyopathy.[69](#jah34809-bib-0069){ref-type="ref"} Previous Mechanistic Studies: Branched Chain Amino Acid Metabolism {#jah34809-sec-0023} ------------------------------------------------------------------ Much of the existing literature regarding the metabolomics of HF has assumed a biomarker‐based approach, assessing the ability of metabolites to diagnose and predict disease prognosis. The identification of biomarkers associated with HF phenotypes, however, has also provided valuable mechanistic information on the pathophysiology of HFpEF. On the whole, impaired fatty acid and glucose oxidation, urea cycle dysfunction, and anabolic‐catabolic imbalance are main pathophysiological processes underlying HF.[70](#jah34809-bib-0070){ref-type="ref"} Ventricular filling is an energy‐intensive process,[71](#jah34809-bib-0071){ref-type="ref"} and recent mechanistic studies have implicated branched chain amino acids (BCAAs), leucine, isoleucine, and valine, in the modulation of left ventricular function.[72](#jah34809-bib-0072){ref-type="ref"}, [73](#jah34809-bib-0073){ref-type="ref"}, [74](#jah34809-bib-0074){ref-type="ref"}, [75](#jah34809-bib-0075){ref-type="ref"} Branched‐chain keto acid accumulation in cardiomyocytes has been documented in mitochondrial targeted 2C‐type serine/threonine protein phosphatase knockout mice[75](#jah34809-bib-0075){ref-type="ref"} as well as in humans with dilated cardiomyopathy.[74](#jah34809-bib-0074){ref-type="ref"} On the other hand, animal models also suggest that supplementation with BCAA may improve cardiomyocyte survival and preserve ventricular function at middle age.[73](#jah34809-bib-0073){ref-type="ref"} Thus, while a broad physiological range of circulating BCAA concentrations may exist, these amino acids may become detrimental when BCAA catabolism is impaired in subclinical or overt HF. Hypertrophic hearts in particular exhibit a reduction in mitochondrial targeted 2C‐type serine/threonine protein phosphatase mRNA and protein expression,[75](#jah34809-bib-0075){ref-type="ref"} and future longitudinal omics studies of mitochondrial targeted 2C‐type serine/threonine protein phosphatase and LVDD may be valuable to further elucidate the biological relationships of BCAA metabolism, mitochondrial function, and diastolic function. We found that butyrylcarnitine, a lipid metabolite involved in BCAA metabolism, had a positive association with IVRT; an association that warrants further investigation. Conclusions {#jah34809-sec-0024} =========== This study had several important strengths. To our knowledge, this is one of the first metabolome‐wide association studies of LVDF. Furthermore, identified metabolites demonstrated consistent associations with diastolic function across race groups and in sensitivity analyses adjusting for left ventricular mass and gait speed and excluding individuals with self‐reported myocardial infarction and/or congestive HF. On the other hand, our study was not without limitations. Its cross‐sectional design prohibits temporal inferences. Nevertheless, metabolites reported here may serve as biomarkers for LVDD, making them particularly relevant for future metabolomics studies involving HFpEF. Furthermore, to minimize false‐positive findings, only metabolites that demonstrated consistent associations with diastolic function across race groups were reported here. Therefore, our study may miss race‐specific metabolite--diastolic function associations. To address this limitation, we reported all significant race‐specific metabolites in Figures [S2](#jah34809-sup-0001){ref-type="supplementary-material"} through [S4](#jah34809-sup-0001){ref-type="supplementary-material"}, to inform future studies and research regarding the potential effect modifying role of race on the association of metabolites and LVDF. In conclusion, we have observed that lipid‐ and amino acid--derived metabolites associate with LVDF in a racially diverse cohort of middle‐aged adults. Butyrylcarnitine was positively correlated with left ventricular relaxation, while N‐formylmethionine; formiminoglutamate; N2, N5‐diacetylornithine; N‐trimethyl 5‐aminovalerate; 5‐methylthioadenosine; and methionine sulfoxide had positive associations with left ventricular filling pressure. Such metabolites may serve as early biomarkers for subclinical HFpEF in the general population. Future replication and longitudinal studies are required to further understand the role of serum metabolites in LVDF and progression to HFpEF. Sources of Funding {#jah34809-sec-0026} ================== This research was supported by the National Institute on Aging as well as the National Heart, Lung, and Blood Institute of the National Institutes of Health under grant numbers R21AG051914 (Principal Investigator, Kelly), 5R01AG041200‐05 (Principal Investigator, Bazzano), and P20GM109036 (Principal Investigator, He). Dr Razavi is currently funded through a fellowship training grant supported by the National Heart, Lung, and Blood Institute of the National Institutes of Health under grant number F30HL147486. Disclosures {#jah34809-sec-0027} =========== Dr Kinchen, who contributed to metabolomics data collection, is employed by Metabolon Inc and did not participate in the study design, statistical analysis, or scientific interpretation of the current research. The remaining authors have no disclosures to report. Supporting information ====================== ###### **Table S1.** Comparison of Bogalusa Heart Study Samples **Table S2.** Novel Metabolites Significantly Associated With Echocardiographic Measures of Diastolic Function, Stratified by Race **Table S3.** Correlations of Nine Metabolite Modules From WGCNA Analysis With the E/e' Ratio **Table S4.** Correlations of Nine Metabolite Modules From WGCNA Analysis With the E/A Ratio **Table S5.** Correlations of Nine Metabolites Modules From WGCNA Analysis With Isovolumic Relaxation Time (IVRT) **Table S6.** Correlations of Nine Metabolite Modules From WGCNA Analysis With Deceleration Time (DT) **Table S7.** Correlations of Nine Metabolite Modules With Left Atrial Maximum Volume Index (LAVI) **Table S8.** Novel Metabolites Significantly Associated With Echocardiographic Measures of Diastolic Function in 1035 BHS Participants, Excluding Individuals With CHF or MI **Table S9.** Novel Metabolites Significantly Associated With Echocardiographic Measures of Diastolic Function in 1050 BHS Participants, Additionally Adjusting for Left Ventricular Mass **Table S10.** Novel Metabolites Significantly Associated With Echocardiographic Measures of Diastolic Function in 1050 BHS Participants, Additionally Adjusting for Gait Speed **Table S11.** Novel Metabolites Significantly Associated With Echocardiographic Measures of Diastolic Function in 804 BHS Participants, Adjusting for Mean Lifetime Systolic Blood Pressure **Figure S1.** Flow diagram of study participant inclusion. **Figure S2.** −Log10(P) probability plot of the multivariable‐adjusted association of serum metabolites with the natural logarithm of the E/A ratio: all participants (left), African Americans (middle), whites (right). **Figure S3.** −Log10(P) probability plot of the multivariable‐adjusted association of serum metabolites with the natural logarithm of the E/e' ratio: all particpants (left), African Americans (middle), whites (right). **Figure S4.** −Log10(P) probability plot of the multivariable‐adjusted association of serum metabolites with isovolumic relaxation time: all participants (left), African Americans (middle), whites (right). **Figure S5.** −Log10(P) probability plot of the multivariable‐adjusted association of serum metabolites with deceleration time: all participants (left), African Americans (middle), whites (right). **Figure S6.** −Log10(P) probability plot of the multivariable‐adjusted association of serum metabolites with left atrial volume index: all participants (left), African Americans (middle), whites (right). **Figure S7.** Heat map displaying pairwise correlation coefficients for the network of metabolites underlying primary and secondary bile acid metabolism pathways. ###### Click here for additional data file. The authors thank all staff members and study personnel who help conduct, sustain, and continue the BHS. The authors are especially grateful to the BHS study participants.